abdul sahli bin fakharudin - psasir.upm.edu.mypsasir.upm.edu.my/id/eprint/68747/1/fsktm 2018 8 -...

35
UNIVERSITI PUTRA MALAYSIA MODELLING OF BIOGAS PRODUCTION PROCESS WITH EVOLUTIONARY ARTIFICIAL NEURAL NETWORK AND GENETIC ALGORITHM ABDUL SAHLI BIN FAKHARUDIN FSKTM 2018 8

Upload: others

Post on 29-Jul-2020

8 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: ABDUL SAHLI BIN FAKHARUDIN - psasir.upm.edu.mypsasir.upm.edu.my/id/eprint/68747/1/FSKTM 2018 8 - IR.pdf · Prestasi model telah dinilai menggunakan punca kuasa min ralat dan pekali

UNIVERSITI PUTRA MALAYSIA

MODELLING OF BIOGAS PRODUCTION PROCESS WITH EVOLUTIONARY ARTIFICIAL NEURAL NETWORK AND

GENETIC ALGORITHM

ABDUL SAHLI BIN FAKHARUDIN

FSKTM 2018 8

Page 2: ABDUL SAHLI BIN FAKHARUDIN - psasir.upm.edu.mypsasir.upm.edu.my/id/eprint/68747/1/FSKTM 2018 8 - IR.pdf · Prestasi model telah dinilai menggunakan punca kuasa min ralat dan pekali

© COPYRIG

HT UPMMODELLING OF BIOGAS PRODUCTION PROCESS WITH

EVOLUTIONARY ARTIFICIAL NEURAL NETWORK ANDGENETIC ALGORITHM

By

ABDUL SAHLI BIN FAKHARUDIN

Thesis Submitted to the School of Graduate Studies, Universiti PutraMalaysia, in Fulfilment of the Requirements for the Degree of Doctor

of Philosophy

July 2017

Page 3: ABDUL SAHLI BIN FAKHARUDIN - psasir.upm.edu.mypsasir.upm.edu.my/id/eprint/68747/1/FSKTM 2018 8 - IR.pdf · Prestasi model telah dinilai menggunakan punca kuasa min ralat dan pekali

© COPYRIG

HT UPM

COPYRIGHT

All material contained within the thesis, including without limitation text,logos, icons, photographs and all other artwork, is copyright material ofUniversiti Putra Malaysia unless otherwise stated. Use may be made of anymaterial contained within the thesis for non-commercial purposes from thecopyright holder. Commercial use of material may only be made with theexpress, prior, written permission of Universiti Putra Malaysia.

Copyright ©Universiti Putra Malaysia

Page 4: ABDUL SAHLI BIN FAKHARUDIN - psasir.upm.edu.mypsasir.upm.edu.my/id/eprint/68747/1/FSKTM 2018 8 - IR.pdf · Prestasi model telah dinilai menggunakan punca kuasa min ralat dan pekali

© COPYRIG

HT UPM

Abstract of thesis presented to the Senate of Universiti Putra Malaysia infulfilment of the requirement for the degree of Doctor of Philosophy

MODELLING OF BIOGAS PRODUCTION PROCESS WITHEVOLUTIONARY ARTIFICIAL NEURAL NETWORK AND

GENETIC ALGORITHM

By

ABDUL SAHLI BIN FAKHARUDIN

July 2017

Chair: Associate Prof. Md Nasir bin Sulaiman, PhDFaculty: Computer Science and Information Technology

In recent years, several researchers have actively pursued the application ofmachine learning to biogas production processes. The application of artificialneural network (ANN) to generate the production model is used to improve themodelling accuracy. The model output optimisation by genetic algorithm (GA)produces higher biogas production compared to the optimisation usingstatistical methods.

This study utilised the evolutionary artificial neural network (EANN) modellingto improve the model accuracy. The EANN modelling was used to representthe biogas production process. One of the issues of ANN implementation is tocorrectly select the output activation function in achieving higher output. TheEANN used a modified activation function to meet the optimisation requirement.

To evaluate the EANN model, 19 samples of experimental data from Zainol on theregression modelling of biogas production from banana stem waste were selected.Thirteen samples were used for training (70%) and six samples were used fortesting (30%). The second dataset fromMahanty which consisted of 36 samples onthe modelling and optimisation of biogas production from industrial sludge weredivided into 25 training samples and 11 testing samples. Meanwhile, 34 samplesfrom Tedesco on the optimisation of mechanical pretreatment of Laminariaceaespp. biomass for the production of biogas were divided into 24 training samplesand 10 testing samples. The last dataset from the domain expert containing 143samples were divided into 100 training samples and 43 testing samples.

The model performance was evaluated using root mean square error (RMSE)and coefficient of determination (R2) and the maximum output from the

i

Page 5: ABDUL SAHLI BIN FAKHARUDIN - psasir.upm.edu.mypsasir.upm.edu.my/id/eprint/68747/1/FSKTM 2018 8 - IR.pdf · Prestasi model telah dinilai menggunakan punca kuasa min ralat dan pekali

© COPYRIG

HT UPM

optimisation was compared to the mathematical modelling. The experiment wasconducted with 50 trial runs on each dataset and EANN method producedbetter modelling results compared to the mathematical modelling. The modeloutput from the optimisation using GA also produced better results than themathematical model and able to limit the maximum output of theback-propagation and Levenberg-Marquardt ANN models which used linearfunction output.

ii

Page 6: ABDUL SAHLI BIN FAKHARUDIN - psasir.upm.edu.mypsasir.upm.edu.my/id/eprint/68747/1/FSKTM 2018 8 - IR.pdf · Prestasi model telah dinilai menggunakan punca kuasa min ralat dan pekali

© COPYRIG

HT UPM

Abstrak tesis yang dikemukakan kepada Senat Universiti Putra Malaysia sebagaimemenuhi keperluan untuk ijazah Doktor Falsafah

PERMODELAN PROSES PENGHASILAN BIOGASMENGGUNAKAN RANGKAIAN NEURAL BUATAN

TEREVOLUSI DAN ALGORITMA GENETIK

Oleh

ABDUL SAHLI BIN FAKHARUDIN

Julai 2017

Pengerusi: Prof. Madya Md Nasir bin Sulaiman, PhDFakulti: Sains Komputer and Teknologi Maklumat

Sejak kebelakangan ini, terdapat penyelidik yang aktif menggunakan aplikasipembelajaran mesin di dalam proses penghasilan biogas. Penggunaan rangkaianneural buatan untuk memodelkan proses penghasilan tersebut telahmeningkatkan kejituan model. Pengoptimuman model output oleh algoritmagenetik menghasilkan biogas yang lebih tinggi berbanding pengoptimumanmenggunakan kaedah statistik.

Kajian ini telah menggunakan rangkaian neural buatan terevolusi untukmeningkatkan kejituan model. Rangkaian neural buatan terevolusi digunakanuntuk mewakilkan proses penghasilan biogas. Salah satu isu di dalampenggunaan rangkaian neural buatan ialah pemilihan fungsi pengaktifan yangbetul dalam mencapai output yang lebih tinggi. Rangkaian neural buatanterevolusi telah menggunakan fungsi pengaktifan yang telah diubahsuai untukmemenuhi keperluan proses pengoptimuman.

Untuk menilai model rangkaian neural buatan terevolusi, 19 sampel dataeksperiman Zainol yang menggunakan permodelan regresi untuk penghasilanbiogas dari sisa batang pisang telah dipilih. Tiga belas sampel digunakan untukset latihan (70%) dan enam sampel digunakan untuk set pengujian (30%). Setdaata yang kedua adalah dari Mahanty yang mengandungi 36 sampelpermodelan dan pengoptimuman proses penghasilan biogas dari enap cemarindustri, telah dipecahkan kepada 25 set latihan dan 11 sampel pengujian.Manakala, 34 sampel dari Tedesco bagi pengoptimuman proses pra-rawatanmekanikal biomas Laminariaceae spp. untuk peghasilan biogas telahdibahagikan kepada 24 sampel latihan dan 10 samapel pengujian. Data terakhirdaripada pakar domain yang mengandungi 143 sampel dan telah dibahagikankepada 100 sampel latihan dan 43 sampel pengujian.

iii

Page 7: ABDUL SAHLI BIN FAKHARUDIN - psasir.upm.edu.mypsasir.upm.edu.my/id/eprint/68747/1/FSKTM 2018 8 - IR.pdf · Prestasi model telah dinilai menggunakan punca kuasa min ralat dan pekali

© COPYRIG

HT UPM

Prestasi model telah dinilai menggunakan punca kuasa min ralat dan pekalipenentuan dan maksimum output dari proses pengoptimuman telahdibandingkan dengan permodelan matematik. Ujikaji telah dijalankan dengan50 larian bagi setiap set data dan kaeda rangkaian neural buatan terevolusimenghasilkan keputusan model lebih baik dari permodelan matematik. Modeloutput melalui pengoptimuman algoritma genetik juga menghasilkan keputusanlebih baik daripada model matematik dan dapat menghadkan output maksimumdari rangkaian neural buatan perambatan balik dan Levenberg-Marquardt yangmenggunakan fungsi linear pada output.

iv

Page 8: ABDUL SAHLI BIN FAKHARUDIN - psasir.upm.edu.mypsasir.upm.edu.my/id/eprint/68747/1/FSKTM 2018 8 - IR.pdf · Prestasi model telah dinilai menggunakan punca kuasa min ralat dan pekali

© COPYRIG

HT UPM

ACKNOWLEDGEMENTS

Alhamdulillah, with His Rahman and Raheem this journey is complete. I wouldlike to thank the following people and institutions for their support. I would liketo express my utmost appreciation to my supervisor Associate Prof. Dr. Md.Nasir Sulaiman and my co-supervisor Associate Prof. Dr. Norwati Mustapha fortheir guidance, support and encouragement in this journey. My thanks also goesto the Ministry of Higher Education for their support in scholarship award.

To my mother Saniah Jalir and father Fakharudin Zakaria, my wives NorazwinaZainol and Syarifah Fazlin Seyed Fadzir and to all my children, thank you for allthe motivation.

v

Page 9: ABDUL SAHLI BIN FAKHARUDIN - psasir.upm.edu.mypsasir.upm.edu.my/id/eprint/68747/1/FSKTM 2018 8 - IR.pdf · Prestasi model telah dinilai menggunakan punca kuasa min ralat dan pekali

© COPYRIG

HT UPM

I certify that a Thesis Examination Committee has met on 20 July 2017 toconduct the final examination of Abdul Sahli Fakharudin on his thesis entitledModelling of Biogas Production Process with Evolutionary Artificial NeuralNetwork and Genetic Algorithm in accordance with the Universities andUniversity Colleges Act 1971 and the Constitution of the Universiti PutraMalaysia [P.U.(A) 106] 15 March 1998. The Committee recommends that thestudent be awarded the Doctor of Philosophy.

Members of the Thesis Examination Committee were as follows:

Hamidah binti Ibrahim, PhDProfessorFaculty of Computer Science and Information TechnologyUniversiti Putra Malaysia(Chairperson)

Abu Bakar bin Md Sultan, PhDProfessorFaculty of Computer Science and Information TechnologyUniversiti Putra Malaysia(Internal Examiner)

Razali bin Yaakob, PhDAssociate ProfessorFaculty of Computer Science and Information TechnologyUniversiti Putra Malaysia(Internal Examiner)

Christopher Hinde, PhDProfessorDepartment Computer Science, School of ScienceLoughborough UniversityUnited Kingdom(External Examiner)

NOR AINI AB. SHUKOR, PhDProfessor and Deputy DeanSchool of Graduate StudiesUniversiti Putra Malaysia

Date: 28 December 2017

vi

Page 10: ABDUL SAHLI BIN FAKHARUDIN - psasir.upm.edu.mypsasir.upm.edu.my/id/eprint/68747/1/FSKTM 2018 8 - IR.pdf · Prestasi model telah dinilai menggunakan punca kuasa min ralat dan pekali

© COPYRIG

HT UPM

This thesis was submitted to the Senate of Universiti Putra Malaysia and hasbeen accepted as fulfilment of the requirement for the degree of Doctor ofPhilosophy.

The members of the Supervisory Committee were as follows:

Md Nasir bin Sulaiman, PhDAssociate ProfessorFaculty of Computer Science and Information TechnologyUniversiti Putra Malaysia(Chairperson)

Norwati binti Mustapha, PhDAssociate ProfessorFaculty of Computer Science and Information TechnologyUniversiti Putra Malaysia(Member)

Aida binti Mustapha, PhDAssociate ProfessorFaculty of Computer Science and Information TechnologyUniversiti Tun Hussein Onn Malaysia(Member)

Jailani bin Salihon, PhDProfessor, IrFaculty of Chemical EngineeringUniversiti Teknologi Mara(Member)

ROBIAH BINTI YUNUS, PhDProfessor and DeanSchool of Graduate StudiesUniversiti Putra Malaysia

Date:

vii

Page 11: ABDUL SAHLI BIN FAKHARUDIN - psasir.upm.edu.mypsasir.upm.edu.my/id/eprint/68747/1/FSKTM 2018 8 - IR.pdf · Prestasi model telah dinilai menggunakan punca kuasa min ralat dan pekali

© COPYRIG

HT UPM

Declaration by graduate student

I hereby confirm that:� this thesis is my original work;� quotations, illustrations and citations have been duly referenced;� this thesis has not been submitted previously or concurrently for any otherdegree at any other institutions;

� intellectual property from the thesis and copyright of thesis are fully-ownedby Universiti Putra Malaysia, as according to the Universiti Putra Malaysia(Research) Rules 2012;

� written permission must be obtained from supervisor and the office of DeputyVice-Chancellor (Research and Innovation) before thesis is published (in theform of written, printed or in electronic form) including books, journals,modules,proceedings,popular writings, seminar papers, manuscripts, posters,reports,lecture notes, learning modules or any other materials as stated in theUniversiti Putra Malaysia (Research) Rules 2012;

� there is no plagiarism or data falsification/fabrication in the thesis, andscholarly integrity is upheld as according to the Universiti Putra Malaysia(Graduate Studies) Rules 2003 (Revision 2012-2013) and the Universiti PutraMalaysia (Research) Rules 2012. The thesis has undergone plagiarismdetection software.

Signature: Date:

Name and Matric No.:

viii

Page 12: ABDUL SAHLI BIN FAKHARUDIN - psasir.upm.edu.mypsasir.upm.edu.my/id/eprint/68747/1/FSKTM 2018 8 - IR.pdf · Prestasi model telah dinilai menggunakan punca kuasa min ralat dan pekali

© COPYRIG

HT UPM

Declaration by Member of Supervisory Committee

This is to confirm that:� the research conducted and the writing of the thesis was under our supervision;� supervision responsibilities as stated in the Universiti Putra Malaysia (GraduateStudies) Rules 2003 (Revision 2012-2013) are adhered to

Signature:Name ofChairman ofSupervisoryCommittee:

Signature:Name ofMember ofSupervisoryCommittee:

Signature:Name ofMember ofSupervisoryCommittee:

Signature:Name ofMember ofSupervisoryCommittee:

ix

Page 13: ABDUL SAHLI BIN FAKHARUDIN - psasir.upm.edu.mypsasir.upm.edu.my/id/eprint/68747/1/FSKTM 2018 8 - IR.pdf · Prestasi model telah dinilai menggunakan punca kuasa min ralat dan pekali

© COPYRIG

HT UPM

TABLE OF CONTENTS

Page

ABSTRACT i

ABSTRAK iii

ACKNOWLEDGEMENTS v

APPROVAL vi

DECLARATION viii

LIST OF TABLES xiii

LIST OF FIGURES xvi

LIST OF ABBREVIATIONS xviii

CHAPTER

1 INTRODUCTION 11.1 Background 11.2 Problem Statements 21.3 Objectives 21.4 Scope of Study 21.5 Research Benefits 31.6 Organisation of the Thesis 3

2 LITERATURE REVIEW 52.1 Bioenergy and Biogas Production 52.2 Mathematical and Statistical Modelling in Biogas

Production from Waste 62.3 Artificial Neural Network 10

2.3.1 History 102.3.2 Biological Base 112.3.3 Back-propagation Training Algorithm 112.3.4 Levenberg - Marquardt Training Algorithm 132.3.5 Artificial Neural Network Modelling 14

2.4 Biogas Modelling using Artificial Neural Network 152.5 Issues in Artificial Neural Networks Modelling 172.6 Evolutionary Artificial Neural Network 19

2.6.1 History 192.6.2 NeuroEvolution of Augmenting Topologies (NEAT) 20

2.7 Genetic Algorithm 242.7.1 Genetic Algorithm Optimisation in Biogas

Production 242.8 Summary 25

x

Sahli
Rectangle
Sahli
Placed Image
Sahli
Typewriter
1
Sahli
Typewriter
2
Sahli
Rectangle
Page 14: ABDUL SAHLI BIN FAKHARUDIN - psasir.upm.edu.mypsasir.upm.edu.my/id/eprint/68747/1/FSKTM 2018 8 - IR.pdf · Prestasi model telah dinilai menggunakan punca kuasa min ralat dan pekali

© COPYRIG

HT UPM

3 RESEARCH METHODOLOGY 263.1 Research Framework 263.2 Datasets and Data Pre-processing 263.3 Modelling of Biogas Process using Evolutionary Artificial

Neural Network 303.4 Model Output Optimisation using Genetic Algorithm 313.5 Experimental Setup 323.6 Model Evaluation 323.7 Optimisation Evaluation 333.8 Summary 34

4 BIOGAS PRODUCTION PROCESS MODELLINGAND OUTPUT OPTIMISATION 354.1 Evolutionary Artificial Neural Network Biogas Process

Modelling Specifications 354.2 Modelling Benchmark Specifications 394.3 Model Output Optimisation using Genetic Algorithms 484.4 Summary 50

5 RESULTS AND DISCUSSIONS 515.1 Results of Evolutionary Artificial Neural Network

Modelling 515.1.1 Discussion 57

5.2 Results of Benchmark Modelling 615.2.1 Results of Benchmark Modelling Using BP Training 615.2.2 Results of Benchmark Modelling Using LM Training 68

5.3 Modelling Results 745.4 Results of Output Optimisation 81

5.4.1 Output Optimisation for Zainol Dataset 825.4.2 Output Optimisation for Mahanty Dataset 835.4.3 Output Optimisation for Tedesco Dataset 835.4.4 Output Optimisation for Domain Dataset 84

5.5 Summary 85

6 CONCLUSIONS AND FUTURE WORKS 866.1 Conclusion of the Study 866.2 Future Works 87

REFERENCES 89

APPENDICES 99

BIODATA OF STUDENT 126

LIST OF PUBLICATIONS 127

xi

Sahli
Typewriter
3
Sahli
Typewriter
4
Sahli
Typewriter
5
Sahli
Typewriter
6
Sahli
Rectangle
Sahli
Rectangle
Sahli
Rectangle
Sahli
Rectangle
Page 15: ABDUL SAHLI BIN FAKHARUDIN - psasir.upm.edu.mypsasir.upm.edu.my/id/eprint/68747/1/FSKTM 2018 8 - IR.pdf · Prestasi model telah dinilai menggunakan punca kuasa min ralat dan pekali

© COPYRIG

HT UPM

LIST OF TABLES

Table Page

2.1 List of research on mathematical and statistical modelling of biogasproduction 9

2.2 The summaries of application of ANN and GA to model andoptimise biogas production 16

3.1 Zainol dataset specification 283.2 Mahanty dataset specification 283.3 Tedesco dataset specification 283.4 Domain dataset specification 293.5 Zainol Training Set 293.6 Zainol Testing Set 303.7 Mahanty Training Set 303.8 Mahanty Testing Set 313.9 Tedesco Training Set 313.10 Tedesco Testing Set 32

4.1 Normal Hyperbolic Tangent results 364.2 Modified Hyperbolic Tangent maximum range selection 374.3 Modified Hyperbolic Tangent results 384.4 ANN using Modified Hyperbolic Tangent 384.5 Initial NEAT parameters 394.6 Zainol dataset NEAT parameters search 404.7 Zainol dataset final NEAT parameters 404.8 Mahanty dataset final NEAT parameters 404.9 Tedesco dataset final NEAT parameters 404.10 Domain dataset final NEAT parameters 414.11 Zainol dataset BP preliminary network arhitecture 424.12 Zainol dataset BP learning parameters selection 434.13 Zainol dataset BP learning parameters 434.14 Zainol dataset LM preliminary network arhitecture 434.15 Mahanty dataset BP preliminary network arhitecture 444.16 Mahanty dataset BP learning parameters selection 444.17 Mahanty dataset BP learning parameters 444.18 Mahanty dataset LM preliminary network arhitecture 454.19 Tedesco dataset BP preliminary network arhitecture 454.20 Tedesco dataset BP learning parameters selection 464.21 Tedesco dataset BP learning parameters 464.22 Domain dataset BP preliminary network arhitecture 464.23 Domain dataset BP learning parameters selection 474.24 Domain dataset BP learning parameters 474.25 Domain dataset LM preliminary network arhitecture 484.26 GA and RSM Optimisation Comparison 484.27 Parameters for GA optimisation process 50

xii

Page 16: ABDUL SAHLI BIN FAKHARUDIN - psasir.upm.edu.mypsasir.upm.edu.my/id/eprint/68747/1/FSKTM 2018 8 - IR.pdf · Prestasi model telah dinilai menggunakan punca kuasa min ralat dan pekali

© COPYRIG

HT UPM

5.1 Modelling comparison for Zainol dataset 525.2 Modelling comparison for Mahanty dataset 545.3 Modelling comparison for Tedesco dataset 555.4 Modelling comparison for domain dataset 575.5 Difference between training and testing RMSE for all datastet using

EANN algorithm 595.6 Comparison of the best model for all dataset using EANN algorithm 605.7 Difference between training and testing RMSE for all datastet using

BP algorithm 675.8 Comparison of the best model for all dataset using BP algorithm 675.9 Difference between training and testing RMSE for all datastet using

LM algorithm 735.10 Comparison of the best model for all dataset using LM algorithm 735.11 Zainol models output optimisation 825.12 Zainol models optimisation predicted biogas output 825.13 Mahanty models output optimisation 835.14 Mahanty models optimisation predicted biogas output 835.15 Tedesco models output optimisation 845.16 Tedesco models optimisation predicted biogas output 845.17 Domain models output optimisation 855.18 Domain models optimisation predicted biogas output 85

A.1 Training Set 99A.2 Testing Set 102B.1 Crossover (CO) selection with mutation set to 0.1 103B.2 Crossover (CO) selection with mutation set to 0.2 103B.3 Population (Pop) selection with geneneration set to 100 103B.4 Population (Pop) selection with geneneration set to 200 104B.5 Population (Pop) selection with geneneration set to 500 104C.1 EANN Run35 105C.2 EANN Run36 106C.3 EANN Run37 107C.4 EANN Run38 108C.5 EANN Run39 109C.6 EANN Run40 110C.7 EANN Run41 111C.8 EANN Run42 112C.9 EANN Run43 113C.10 EANN Run44 114C.11 EANN Run45 115C.12 EANN Run46 116C.13 EANN Run47 117C.14 EANN Run48 118C.15 EANN Run49 119C.16 EANN Run50 120D.1 Zainol BP Model 121D.2 Zainol LM Model 121D.3 Zainol EANN Model 121

xiii

Page 17: ABDUL SAHLI BIN FAKHARUDIN - psasir.upm.edu.mypsasir.upm.edu.my/id/eprint/68747/1/FSKTM 2018 8 - IR.pdf · Prestasi model telah dinilai menggunakan punca kuasa min ralat dan pekali

© COPYRIG

HT UPM

D.4 Mahanty BP Model 122D.5 Mahanty LM Model 122D.6 Mahanty EANN Model 123D.7 Tedesco BP Model 123D.8 Tedesco EANN Model 123D.9 Domain BP Model 124D.10 Domain LM Model 124D.11 Domain EANN Model 125

xiv

Page 18: ABDUL SAHLI BIN FAKHARUDIN - psasir.upm.edu.mypsasir.upm.edu.my/id/eprint/68747/1/FSKTM 2018 8 - IR.pdf · Prestasi model telah dinilai menggunakan punca kuasa min ralat dan pekali

© COPYRIG

HT UPM

LIST OF FIGURES

Figure Page

2.1 Global energy sources 2011 62.2 Biomass and waste conversion technologies 72.3 Neuron cell 112.4 Neuron model by McCulloch dan Pitts 122.5 Back-propagation process 132.6 Comparison of two different ANN architectures 182.7 A typical cycle of the evolution of connection weights 202.8 A typical cycle of the evolution of architectures 212.9 A typical cycle of of the learning rules 212.10 NEAT genetic encoding 222.11 The two types of structural mutation in NEAT 232.12 The crossover operation in NEAT 23

3.1 Research Framework 273.2 Research Activities Flowchart 27

4.1 Hyperbolic tangent function 364.2 Modified hyperbolic tangent function pseudo-code 384.3 ANJI parameters for EANN modelling 394.4 Smallest architectures 42

5.1 RMSE for Zainol dataset EANN algorithm 515.2 R2 for Zainol dataset EANN algorithm 525.3 RMSE for Mahanty dataset EANN algorithm 535.4 R2 for Mahanty dataset EANN algorithm 535.5 RMSE for Tedesco dataset EANN algorithm 545.6 R2 for Tedesco dataset EANN algorithm 555.7 RMSE for domain dataset EANN algorithm 565.8 R2 for domain dataset EANN algorithm 565.9 EANN training RMSE comparison 585.10 EANN testing RMSE comparison 585.11 EANN training R2 comparison 595.12 EANN testing R2 comparison 605.13 RMSE for Zainol BP training 615.14 R2 for Zainol BP training 615.15 RMSE for Mahanty dataset BP training 625.16 R2 for Mahanty dataset BP training 625.17 RMSE for Tedesco dataset BP training 635.18 R2 for Tedesco dataset BP training 635.19 RMSE for domain dataset BP training 645.20 R2 for domain dataset BP training 645.21 BP training RMSE comparison 655.22 BP testing RMSE comparison 65

xv

Page 19: ABDUL SAHLI BIN FAKHARUDIN - psasir.upm.edu.mypsasir.upm.edu.my/id/eprint/68747/1/FSKTM 2018 8 - IR.pdf · Prestasi model telah dinilai menggunakan punca kuasa min ralat dan pekali

© COPYRIG

HT UPM

5.23 BP training R2 comparison 665.24 BP testing R2 comparison 665.25 RMSE for Zainol LM training 685.26 R2 for Zainol LM training 695.27 RMSE for Mahanty dataset LM training 695.28 R2 for Mahanty dataset LM training 705.29 RMSE for domain dataset LM training 705.30 R2 for domain dataset LM training 715.31 LM training RMSE comparison 715.32 LM testing RMSE comparison 715.33 LM training R2 comparison 725.34 LM testing R2 comparison 725.35 Training RMSE for Zainol dataset on all modelling algorithms 745.36 Testing RMSE for Zainol dataset on all modelling algorithms 745.37 Training R2 for Zainol dataset on all modelling algorithms 755.38 Testing R2 for Zainol dataset on all modelling algorithms 755.39 Training RMSE for Mahanty dataset on all modelling algorithms 765.40 Testing RMSE for Mahanty dataset on all modelling algorithms 765.41 Training R2 for Mahanty dataset on all modelling algorithms 775.42 Testing R2 for Mahanty dataset on all modelling algorithms 775.43 Training RMSE for Tedesco dataset on all modelling algorithms 785.44 Testing RMSE for Tedesco dataset on all modelling algorithms 785.45 Training R2 for Tedesco dataset on all modelling algorithms 795.46 Testing R2 for Tedesco dataset on all modelling algorithms 795.47 Training RMSE for domain dataset on all modelling algorithms 805.48 Testing RMSE for domain datasets on all modelling algorithms 805.49 Training R2 for domain dataset on all modelling algorithms 815.50 Testing R2 for domain dataset on all modelling algorithms 81

xvi

Page 20: ABDUL SAHLI BIN FAKHARUDIN - psasir.upm.edu.mypsasir.upm.edu.my/id/eprint/68747/1/FSKTM 2018 8 - IR.pdf · Prestasi model telah dinilai menggunakan punca kuasa min ralat dan pekali

© COPYRIG

HT UPM

LIST OF ABBREVIATIONS

ANN Artificial Neural NetworkBP Back-propagationCOD Chemical Oxygen DemandEANN Evolutionary Artificial Neural NetworkGA Genetic AlgorithmHRT Hydraulic Retention TimeLM Levenberg−MarquardtMSE Mean Square ErrorNEAT NeuroEvolution of Augmenting TopologiesOLR Organic Loading RateR2 Coefficient of DeterminationRMSE Root Mean Square ErrorRSM Response Surface Methodology

xvii

Page 21: ABDUL SAHLI BIN FAKHARUDIN - psasir.upm.edu.mypsasir.upm.edu.my/id/eprint/68747/1/FSKTM 2018 8 - IR.pdf · Prestasi model telah dinilai menggunakan punca kuasa min ralat dan pekali

© COPYRIG

HT UPM

Page 22: ABDUL SAHLI BIN FAKHARUDIN - psasir.upm.edu.mypsasir.upm.edu.my/id/eprint/68747/1/FSKTM 2018 8 - IR.pdf · Prestasi model telah dinilai menggunakan punca kuasa min ralat dan pekali

© COPYRIG

HT UPM

CHAPTER 1

INTRODUCTION

This chapter will provide the background of previous studies, problemstatements, objectives and scope of the study including the predicted output. Italso contained research benefits and overall organisations of the thesis to give ageneral perspective of the research flow.

1.1 Background

The consumption of renewable energy from 2000 to 2011 has increased by 30% andthe highest absolute increase among the renewable energy consumption was thebioenergy source (WBA, 2014). The bioenergy categories into the solid biomass,liquid biofuels, wastes and biogas. Interest in converting biomass resources to analternative fuel such as biogas, have received more attention in recent times (Yanget al., 2013). A feasibility study on biogas production and utilisation as a sourceof renewable energy in Malaysia by Hosseini and Wahid (2013) discussed the usedof palm oil industry by-product as the sources to extract biogas. A review byAhmed et al. (2015) presented the biogas production and performance evaluationof the treatment process from palm oil mill effluent.

This biogas yield can be improved with better process design, which include themodelling, simulation and optimisation process as an integrated part of moderndesign practice (Betiku and Ajala, 2014). Work by Sendjaja et al. (2015)mentioned about two main types of modelling approach in anaerobic digestion,including biogas production, which are the mathematical based model derivedfrom mathematical equations and data or statistical approach using multivariateregression and machine learning approach.

Modelling of biogas generation using mathematical and statistical approach was aproven knowledge used by many researchers (Zainol et al., 2009; Mahanty et al.,2014; Tedesco et al., 2014). They used a regression model to represent theirprocess and the model output was being optimised using statistical method toobtain the maximum biogas output. The predicted maximum biogas output fromthe generated model was presented and its successfully improved the biogas outputfrom the actual production.

The improvement of the biogas production using modern computer science fieldalso has advantages than the mathematical modelling. Such area that is beingexplored (Behera et al., 2015; Dhussa et al., 2014; Yetilmezsoy et al., 2012) was aspecific intelligent computing area, which used artificial neural networks (ANN) tomodel their process. These researchers had reported the application of ANN hadsucceeded to model their biogas production process. Most of the ANN training(Yetilmezsoy et al., 2012; Abu Qdais et al., 2010) was set from a small number ofhidden neurons to the maximum number according to each specification.

Page 23: ABDUL SAHLI BIN FAKHARUDIN - psasir.upm.edu.mypsasir.upm.edu.my/id/eprint/68747/1/FSKTM 2018 8 - IR.pdf · Prestasi model telah dinilai menggunakan punca kuasa min ralat dan pekali

© COPYRIG

HT UPM

1.2 Problem Statements

This study addressed the problem of the model output optimisation done byAkbas et al. (2015) which unable to generate higher output (Abu Qdais et al.,2010; Gueguim Kana et al., 2012) than the actual process. He used ANNmodelling with hyperbolic tangent as the activation function and themaximisation of biogas output was limited to highest function range. Theoptimal biogas output from the model could not achieve the actual biogasoutput let alone more than it.

Gueguim Kana et al. (2012) used ANN modelling with linear activation functionand the optimisation process produced higher maximum output than the actualprocess because the linear function was not limited to a certain maximum range.The implementation of linear activation on output layer made it as a thresholdlayer even the hidden layer used sigmoid activation function. If the linear functionoutput being optimised by an efficient optimiser, then the maximum output valuecould produce a very high, unrealistic output, because the unlimited range of thelinear activation function.

It is important to develop a predictive model for engineering process that canmaximise the production. A proper implementation of specialised ANN with amodified activation function should be able to regulate the network output fromproducing an unlimited output and should be able to produce maximum outputmore than one to ensure the model output is better than the actual output.

1.3 Objectives

The main objective of this research is to model the biogas process withevolutionary artificial neural network (EANN) for optimum production. Thespecific objectives of this research are as follows:

1. To propose an implementation of evolutionary artificial neural networkmodelling to improve the mathematical modelling accuracy of biogasprocess.

2. To propose a modified activation function in evolutionary artificial neuralnetwork modelling to find the best biogas representation and to fulfil theoutput optimisation requirements.

3. To propose an appropriate parameter for genetic algorithm to optimise theselected biogas process representation.

1.4 Scope of Study

The study used a dataset from Zainol et al. (2009) in order to model the biogasprocess where she used mathematical modelling and optimisation of biogasproduction from banana stem waste. Three additional datasets also being usedto determine the model accuracy improvement and the output optimisation

2

Page 24: ABDUL SAHLI BIN FAKHARUDIN - psasir.upm.edu.mypsasir.upm.edu.my/id/eprint/68747/1/FSKTM 2018 8 - IR.pdf · Prestasi model telah dinilai menggunakan punca kuasa min ralat dan pekali

© COPYRIG

HT UPM

comparison. These four datasets will be normalised and divided into trainingand testing set.

The results from the modelling of biogas process representation will be evaluatedusing root mean square errors (RMSE) and correlation determination (R2) forperformance evaluation. The accuracy measurements were based on traditionalANN evaluation with an additional engineering process modelling validation.

The best model will be selected and it will be used to find the optimal biogasoutput using the genetic algorithm (GA). The predicted optimised output willbe collected from the model output and will be compared with the mathematicaloptimisation.

1.5 Research Benefits

There are two main benefits of this research:1. The EANN modelling for biogas production representation; it will be

proposed as an alternative to ANN which reduce the guesswork andcomplexity.

2. The modified activation function; it will present the important of specificANN design and architecture in solving a problem even though the ANNgeneralisation mostly works.

1.6 Organisation of the Thesis

The thesis consists of six chapters. Chapter 1 discussed the introduction tobiogas production modelling and optimisation. The purpose of this chapter wasto explain the problem statements, objectives of the study, the scopes of thestudy and the benefits. It concluded by the organisation of the thesis.

In Chapter 2 it discussed the literature review of the research. This chaptercontains the information on the biogas production and research related to thetopic. Followed by the previous study of modelling of biogas production fromwaste using a mathematical approach. The next part discussed about the ANNand the used of ANN for modelling and followed by previous study that usedANN for modelling biogas production. The next topic discussed about issues inANN modelling and suggested solutions. The final part of the chapter discussedabout GA and the previous study which utilising GA optimisation.

The next Chapter 3 discussed about the research methodology to perform themodelling and optimisation process. It started with the research frameworkfollowed by the biogas production data processing. The brief information ofbiogas modelling and optimisation was presented next. The experimental setup

3

Page 25: ABDUL SAHLI BIN FAKHARUDIN - psasir.upm.edu.mypsasir.upm.edu.my/id/eprint/68747/1/FSKTM 2018 8 - IR.pdf · Prestasi model telah dinilai menggunakan punca kuasa min ralat dan pekali

© COPYRIG

HT UPM

was discussed later followed by the evaluation on both modelling andoptimisation results.

In chapter 4, the details of the modelling and optimisation of biogas productionwere presented. The EANN modelling was discussed first and followed by thebenchmark modelling. The optimisation of the model using GA algorithms wasdiscussed last on this chapter.

Chapter 5 presented the results of the modelling using EANN and the benchmarkmodelling. It followed by the details on the result of GA optimisation from themodels generated by EANN and ANN.

Finally, chapter 6 concluded the findings. The purpose of this chapter is to makethe conclusion of the research and the future research that can be continued fromthis research.

4

Page 26: ABDUL SAHLI BIN FAKHARUDIN - psasir.upm.edu.mypsasir.upm.edu.my/id/eprint/68747/1/FSKTM 2018 8 - IR.pdf · Prestasi model telah dinilai menggunakan punca kuasa min ralat dan pekali

© COPYRIG

HT UPM

REFERENCES

Abdurahman, A. and Jiang, H. (2016). New results on exponentialsynchronization of memristor-based neural networks with discontinuousneuron activations. Neural Networks, 84:161–171.

Abu Qdais, H., Bani Hani, K., and Shatnawi, N. (2010). Modeling andoptimization of biogas production from a waste digester using artificial neuralnetwork and genetic algorithm. Resources, Conservation and Recycling,54(6):359–363.

Ahmed, Y., Yaakob, Z., Akhtar, P., and Sopian, K. (2015). Production of biogasand performance evaluation of existing treatment processes in palm oil milleffluent (pome). Renewable and Sustainable Energy Reviews, 42:1260–1278.

Akbas, H., Bilgen, B., and Turhan, A. M. (2015). An integrated prediction andoptimization model of biogas production system at a wastewater treatmentfacility. Bioresource Technology, 196:566–576.

Appels, L., Lauwers, J., Degrve, J., Helsen, L., Lievens, B., Willems, K.,Van Impe, J., and Dewil, R. (2011). Anaerobic digestion in global bio-energyproduction: Potential and research challenges. Renewable and SustainableEnergy Reviews, 15(9):4295–4301.

Atallah, N. M., El-Fadel, M., Ghanimeh, S., Saikaly, P., and Abou-Najm, M.(2014). Performance optimization and validation of adm1 simulations underanaerobic thermophilic conditions. Bioresource Technology, 174:243–255.

Bagheri, M., Mirbagheri, S., Ehteshami, M., and Bagheri, Z. (2015). Modeling ofa sequencing batch reactor treating municipal wastewater using multi-layerperceptron and radial basis function artificial neural networks. Process Safetyand Environmental Protection, 93:111 – 123.

Bao, G. and Zeng, Z. (2016). Global asymptotical stability analysis for akind of discrete-time recurrent neural network with discontinuous activationfunctions. Neurocomputing, 193:242–249.

Batstone, D. J., Keller, J., Angelidaki, I., Kalyuzhnyi, S. V., Pavlostathis, S. G.,Rozzi, A., Sanders, W. T., Siegrist, H., and Vavilin, V. A. (2002). The iwaanaerobic digestion model no 1 (adm1). Water Sci Technol, 45(10):65–73.Batstone, D J Keller, J Angelidaki, I Kalyuzhnyi, S V Pavlostathis, S GRozzi, A Sanders, W T M Siegrist, H Vavilin, V A Journal Article ResearchSupport, Non-U.S. Gov’t England Water Sci Technol. 2002;45(10):65-73.

Battiti, R. (1992). First- and second-order methods for learning: Between steepestdescent and newton’s method. Neural Computation, 4(2):141–166.

Baxter, J. (1992). The evolution of learning algorithms for artificial neuralnetworks. In in Complex Systems, pages 313–326. IOS Press.

89

Page 27: ABDUL SAHLI BIN FAKHARUDIN - psasir.upm.edu.mypsasir.upm.edu.my/id/eprint/68747/1/FSKTM 2018 8 - IR.pdf · Prestasi model telah dinilai menggunakan punca kuasa min ralat dan pekali

© COPYRIG

HT UPM

Behera, S., Meher, S., and Park, H.-S. (2015). Artificial neural network modelfor predicting methane percentage in biogas recovered from a landfill uponinjection of liquid organic waste. Clean Technologies and EnvironmentalPolicy, 17(2):443–453.

Betiku, E. and Ajala, S. O. (2014). Modeling and optimization of thevetiaperuviana (yellow oleander) oil biodiesel synthesis via musa paradisiacal(plantain) peels as heterogeneous base catalyst: A case of artificial neuralnetwork vs. response surface methodology. Industrial Crops and Products,53(0):314–322.

Bobbili, R., Madhu, V., and Gogia, A. K. (2014). Neural network modeling toevaluate the dynamic flow stress of high strength armor steels under highstrain rate compression. Defence Technology, 10(4):334–342.

Bornholdt, S. and Graudenz, D. (1992). General asymmetric neural networks andstructure design by genetic algorithms. Neural Networks, 5(2):327–334.

Bos, A., Bos, M., and van der Linden, W. E. (1992). Artificial neural networks asa tool for soft-modelling in quantitative analytical chemistry: the predictionof the water content of cheese. Analytica Chimica Acta, 256(1):133–144.

Bourquin, J., Schmidli, H., van Hoogevest, P., and Leuenberger, H. (1998).Pitfalls of artificial neural networks (ann) modelling technique for datasets containing outlier measurements using a study on mixture propertiesof a direct compressed dosage form. European Journal of PharmaceuticalSciences, 7(1):17–28.

Broomhead, D. and Lowe, D. (1988). Radial basis functions, multi-variablefunctional interpolation and adaptive networks. RSRE-MEMO-4148.

Cairns, D. S., Huff, R. D., and Clayton, C. D. (1990). Multivariate processmodelling of advanced composite materials. Composites Manufacturing,1(4):229–234.

Caudell, T. P. and Dolan, C. P. (1989). Parametric connectivity: Trainingof constrained networks using genetic algorithms. In Proceedings of the3rd International Conference on Genetic Algorithms, pages 370–374, SanFrancisco, CA, USA. Morgan Kaufmann Publishers Inc.

Caudill, M. (1988). The polynomial adaline algorithm. Comput. Lang. (SanFrancisco, CA), 5(12):53–59.

Chalmers, D. J. (1990). The evolution of learning: an experiment in geneticconnectionism. In Proceedings of the 1990 Connectionist Models SummerSchool, pages 81–90. Morgan Kaufmann.

Chinese, D., Patrizio, P., and Nardin, G. (2014). Effects of changes in italianbioenergy promotion schemes for agricultural biogas projects: Insights froma regional optimization model. Energy Policy, 75:189–205.

Dhussa, A. K., Sambi, S. S., Kumar, S., Kumar, S., and Kumar, S. (2014).Nonlinear autoregressive exogenous modeling of a large anaerobic digesterproducing biogas from cattle waste. Bioresource Technology, 170:342–349.

90

Page 28: ABDUL SAHLI BIN FAKHARUDIN - psasir.upm.edu.mypsasir.upm.edu.my/id/eprint/68747/1/FSKTM 2018 8 - IR.pdf · Prestasi model telah dinilai menggunakan punca kuasa min ralat dan pekali

© COPYRIG

HT UPM

Djatkov, D., Effenberger, M., and Martinov, M. (2014). Method for assessing andimproving the efficiency of agricultural biogas plants based on fuzzy logic andexpert systems. Applied Energy, 134:163–175.

Donoso-Bravo, A., Mailier, J., Martin, C., Rodrguez, J., Aceves-Lara, C. A.,and Wouwer, A. V. (2011). Model selection, identification and validation inanaerobic digestion: A review. Water Research, 45(17):5347–5364.

Dreyfus, S. E. (1990). Artificial neural networks, back propagation, andthe kelley-bryson gradient procedure. Journal of Guidance, Control, andDynamics, 13(5):926–928.

Durruty, I., Zaritzky, N. E., and Gonzlez, J. F. (2013). Organic fractions influenceon biogas generation from potato residues. kinetic model generalization.Biomass and Bioenergy, 59:458–467.

Emkes, H., Coulon, F., andWagland, S. (2015). A decision support tool for landfillmethane generation and gas collection. Waste Management, 43:307–318.

Fogel, D. B., Fogel, L. J., and Porto, V. W. (1990). Evolving neural networks.Biological Cybernetics, 63(6):487–493.

Fontanari, J. F. and Meir, R. (1991). Evolving a learning algorithm for the binaryperceptron. Network: Computation in Neural Systems, 2(4):353–359.

Giam, X. and Olden, J. D. (2015). A new r2-based metric to shed greater insighton variable importance in artificial neural networks. Ecological Modelling,313:307–313.

Gibson, C. A., Meybodi, M. A., and Behnia, M. (2014). Investigation of a gasturbine chp system under the carbon price in australia considering naturalgas and biogas fuels. Applied Thermal Engineering, 68(12):26–35.

Gueguim Kana, E. B., Oloke, J. K., Lateef, A., and Adesiyan, M. O.(2012). Modeling and optimization of biogas production on saw dust andother co-substrates using artificial neural network and genetic algorithm.Renewable Energy, 46(0):276–281.

Guo, M., Song, W., and Buhain, J. (2015). Bioenergy and biofuels: History,status, and perspective. Renewable and Sustainable Energy Reviews,42:712–725.

Hagan, M. T. and Menhaj, M. B. (1994). Training feedforward networks with themarquardt algorithm. IEEE Transactions on Neural Networks, 5(6):989–993.

Hancock, P. J. B. and Smith, L. (1991). Gannet: Genetic design of a neural netfor face recognition, volume 496 of Lecture Notes in Computer Science, booksection 43, pages 292–296. Springer Berlin Heidelberg.

Harp, S. A., Samad, T., and Guha, A. (1989). Towards the genetic synthesisof neural network. In Proceedings of the Third International Conferenceon Genetic Algorithms, pages 360–369, San Francisco, CA, USA. MorganKaufmann Publishers Inc.

91

Page 29: ABDUL SAHLI BIN FAKHARUDIN - psasir.upm.edu.mypsasir.upm.edu.my/id/eprint/68747/1/FSKTM 2018 8 - IR.pdf · Prestasi model telah dinilai menggunakan punca kuasa min ralat dan pekali

© COPYRIG

HT UPM

He, G., Bluemling, B., Mol, A. P. J., Zhang, L., and Lu, Y. (2013). Comparingcentralized and decentralized bio-energy systems in rural china. EnergyPolicy, 63:34–43.

He, W., Williard, N., Chen, C., and Pecht, M. (2014). State of charge estimationfor li-ion batteries using neural network modeling and unscented kalmanfilter-based error cancellation. International Journal of Electrical Power &Energy Systems, 62:783–791.

Heaton, J. (2015). Encog: Library of interchangeable machine learning modelsfor java and c#. Journal of Machine Learning Research, 16:1243–1247.

Hejase, H. A. N., Al-Shamisi, M. H., and Assi, A. H. (2014). Modeling ofglobal horizontal irradiance in the united arab emirates with artificial neuralnetworks. Energy, 77:542–552.

Hertz, J., Krogh, A., and Palmer, R. G. (1991). Introduction to the Theory ofNeural Computation. Addison-Wesley Longman Publishing Co., Inc., Boston,MA, USA.

Hinken, L., Huber, M., Weichgrebe, D., and Rosenwinkel, K. H. (2014).Modified adm1 for modelling an uasb reactor laboratory plant treating starchwastewater and synthetic substrate load tests. Water Research, 64:82–93.

Hinton, G. E. and Nowlan, S. J. (1996). How learning can guide evolution. InBelew, R. K. and Mitchell, M., editors, Adaptive Individuals in EvolvingPopulations, pages 447–454. Addison-Wesley Longman Publishing Co., Inc.,Boston, MA, USA.

Hosseini, S. E. and Wahid, M. A. (2013). Feasibility study of biogas productionand utilization as a source of renewable energy in malaysia. Renewable andSustainable Energy Reviews, 19:454–462.

James, D. and Tucker, P. (2004). Another neat java implementation.http://anji.sourceforge.net/. Accessed: 2014-06-2.

Kaelbling, L. P., Littman, M. L., and Moore, A. W. (1996). Reinforcementlearning: A survey. J. Artif. Int. Res., 4(1):237–285.

Kandel, T. P., Gislum, R., Jrgensen, U., and Lrke, P. E. (2013). Predictionof biogas yield and its kinetics in reed canary grass using near infraredreflectance spectroscopy and chemometrics. Bioresource Technology,146:282–287.

Kim, S. E. and Seo, I. W. (2015). Artificial neural network ensemble modelingwith conjunctive data clustering for water quality prediction in rivers. Journalof Hydro-environment Research, 9(3):325 – 339.

Koch, K., Lbken, M., Gehring, T., Wichern, M., and Horn, H. (2010). Biogas fromgrass silage measurements and modeling with adm1. Bioresource Technology,101(21):8158–8165.

92

Page 30: ABDUL SAHLI BIN FAKHARUDIN - psasir.upm.edu.mypsasir.upm.edu.my/id/eprint/68747/1/FSKTM 2018 8 - IR.pdf · Prestasi model telah dinilai menggunakan punca kuasa min ralat dan pekali

© COPYRIG

HT UPM

Koza, J. R. and Rice, J. P. (1991). Genetic generation of both the weightsand architecture for a neural network. In Neural Networks, 1991.,IJCNN-91-Seattle International Joint Conference on, volume ii, pages397–404 vol.2.

Kumar, A., Kumar, N., Baredar, P., and Shukla, A. (2015). A review on biomassenergy resources, potential, conversion and policy in india. Renewable andSustainable Energy Reviews, 45:530–539.

Kuvendziev, S., Lisichkov, K., Zekovi, Z., and Marinkovski, M. (2014).Artificial neural network modelling of supercritical fluid co2 extraction ofpolyunsaturated fatty acids from common carp (cyprinus carpio l.) viscera.The Journal of Supercritical Fluids, 92:242–248.

Li, C., Champagne, P., and Anderson, B. C. (2011). Evaluating and modelingbiogas production from municipal fat, oil, and grease and synthetic kitchenwaste in anaerobic co-digestions. Bioresource Technology, 102(20):9471–9480.

Linko, P. and Zhu, Y.-H. (1992). Neural network modelling for real-time variableestimation and prediction in the control of glucoamylase fermentation.Process Biochemistry, 27(5):275–283.

Lo, H. M., Kurniawan, T. A., Sillanp, M. E. T., Pai, T. Y., Chiang, C. F.,Chao, K. P., Liu, M. H., Chuang, S. H., Banks, C. J., Wang, S. C., Lin,K. C., Lin, C. Y., Liu, W. F., Cheng, P. H., Chen, C. K., Chiu, H. Y.,and Wu, H. Y. (2010). Modeling biogas production from organic fractionof msw co-digested with mswi ashes in anaerobic bioreactors. BioresourceTechnology, 101(16):6329–6335.

Lo Schiavo, M., Prinari, B., Gronski, J. A., and Serio, A. V. (2015). An artificialneural network approach for modeling the ward atmosphere in a medical unit.Mathematics and Computers in Simulation, 116:44–58.

Lpez, I., Passeggi, M., and Borzacconi, L. (2015). Variable kinetic approach tomodelling an industrial waste anaerobic digester. Biochemical EngineeringJournal, 96:7–13.

Ma, Y. Z. and Gomez, E. (2015). Uses and abuses in applying neural networksfor predictions in hydrocarbon resource evaluation. Journal of PetroleumScience and Engineering, 133:66–75.

Mahanty, B., Zafar, M., Han, M. J., and Park, H.-S. (2014). Optimization ofco-digestion of various industrial sludges for biogas production and sludgetreatment: Methane production potential experiments and modeling. WasteManagement, 34(6):1018–1024.

Masters, T. (1993). Practical Neural Network Recipes in C++. Academic PressProfessional, Inc., San Diego, CA, USA.

McCord-Nelson, M. and Illingworth, W. T. (1991). A Practical Guide to NeuralNets. Addison-Wesley Longman Publishing Co., Inc., Boston, MA, USA.

McCulloch, W. and Pitts, W. (1943). A logical calculus of the ideas immanent innervous activity. The bulletin of mathematical biophysics, 5(4):115–133.

93

Page 31: ABDUL SAHLI BIN FAKHARUDIN - psasir.upm.edu.mypsasir.upm.edu.my/id/eprint/68747/1/FSKTM 2018 8 - IR.pdf · Prestasi model telah dinilai menggunakan punca kuasa min ralat dan pekali

© COPYRIG

HT UPM

Menczer, F. and Parisi, D. (1992). Evidence of hyperplanes in the genetic learningof neural networks. Biol. Cybern., 66(3):283–289.

Miao, P., Shen, Y., and Xia, X. (2014). Finite time dual neural networks with atunable activation function for solving quadratic programming problems andits application. Neurocomputing, 143:80–89.

Miller, G. F., Todd, P. M., and Hegde, S. U. (1989). Designing neuralnetworks using genetic algorithms. In Proceedings of the Third InternationalConference on Genetic Algorithms, pages 379–384, San Francisco, CA, USA.Morgan Kaufmann Publishers Inc.

Montana, D. J. and Davis, L. (1989). Training feedforward neural networks usinggenetic algorithms. In Proceedings of the 11th International Joint Conferenceon Artificial Intelligence - Volume 1, IJCAI’89, pages 762–767, San Francisco,CA, USA. Morgan Kaufmann Publishers Inc.

Muhlenbein, H. and Kindermann, J. (1989). The dynamics of evolutionand learning - toward genetic neural networks. In Pfeifer, R., Schreter,Z., Fogelman-Soulie, F., and Steels, L., editors, Connectionism inPerspective, pages 173–198. Elsevier Science Publishers B.V., Amsterdam,The Netherlands.

Nastos, P., Paliatsos, A., Koukouletsos, K., Larissi, I., and Moustris, K. (2014).Artificial neural networks modeling for forecasting the maximum daily totalprecipitation at athens, greece. Atmospheric Research, 144:141 – 150.Perspectives of Precipitation Science - Part {II}.

Nath, K. and Das, D. (2011). Modeling and optimization of fermentative hydrogenproduction. Bioresource Technology, 102(18):8569–8581.

Nolfi, S., Parisi, D., and Elman, J. L. (1994). Learning and evolution in neuralnetworks. Adapt. Behav., 3(1):5–28.

Ozkaya, B., Demir, A., and Bilgili, M. S. (2007). Neural network predictionmodel for the methane fraction in biogas from field-scale landfill bioreactors.Environmental Modelling & Software, 22(6):815 – 822.

Paredis, J. (1991). The evolution of behavior: Some experiments. In Meyer,J. and Wilson, S. W., editors, Proc. 1st Int. Conf. Simulation of AdaptiveBehavior: From Animals to Animat. MIT Press, Cambridge,MA,USA.

Piotrowski, A. P. and Napiorkowski, J. J. (2013). A comparison of methods toavoid overfitting in neural networks training in the case of catchment runoffmodelling. Journal of Hydrology, 476:97–111.

Podstawczyk, D., Witek-Krowiak, A., Dawiec, A., and Bhatnagar, A. (2015).Biosorption of copper(ii) ions by flax meal: Empirical modeling and processoptimization by response surface methodology (rsm) and artificial neuralnetwork (ann) simulation. Ecological Engineering, 83:364–379.

94

Page 32: ABDUL SAHLI BIN FAKHARUDIN - psasir.upm.edu.mypsasir.upm.edu.my/id/eprint/68747/1/FSKTM 2018 8 - IR.pdf · Prestasi model telah dinilai menggunakan punca kuasa min ralat dan pekali

© COPYRIG

HT UPM

Prakasham, R., Sathish, T., and Brahmaiah, P. (2011). Imperative role of neuralnetworks coupled genetic algorithm on optimization of biohydrogen yield.International Journal of Hydrogen Energy, 36(7):4332 – 4339. EmergingMaterials Technology: Materials in Clean Power System.

Rangasamy, P., Pvr, I., and Ganesan, S. (2007). Anaerobic tapered fluidizedbed reactor for starch wastewater treatment and modeling using multilayerperceptron neural network. Journal of Environmental Sciences, 19(12):1416– 1423.

Rumelhart, D. E., Hinton, G. E., and McClelland, J. L. (1986). Paralleldistributed processing: Explorations in the microstructure of cognition, vol.1. chapter A General Framework for Parallel Distributed Processing, pages45–76. MIT Press, Cambridge, MA, USA.

Russell, S. J. and Norvig, P. (2003). Artificial Intelligence: A Modern Approach.Pearson Education, 2 edition.

Scaglia, B., Confalonieri, R., DImporzano, G., and Adani, F. (2010). Estimatingbiogas production of biologically treated municipal solid waste. BioresourceTechnology, 101(3):945–952.

Schaffer, J. D., Caruana, R. A., and Eshelman, L. J. (1990). Using geneticsearch to exploit the emergent behavior of neural networks. Phys. D,42(1-3):244–248.

Schievano, A., Pognani, M., DImporzano, G., and Adani, F. (2008). Predictinganaerobic biogasification potential of ingestates and digestates of a full-scalebiogas plant using chemical and biological parameters. BioresourceTechnology, 99(17):8112–8117.

Schittenkopf, C., Deco, G., and Brauer, W. (1997). Two strategies to avoidoverfitting in feedforward networks. Neural Networks, 10(3):505–516.

Schmitt, L. M. (2001). Theory of genetic algorithms. Theoretical ComputerScience, 259(12):1 – 61.

Sendjaja, A. Y., Tan, Y., Pathak, S., Zhou, Y., bin Abdul Majid, M., Liu,J. L., and Ng, W. J. (2015). Regression based state space adaptive model oftwo-phase anaerobic reactor. Chemosphere, 140:159–166.

Shah, S. and Palmieri, F. (1990). Meka-a fast, local algorithm for trainingfeedforward neural networks. In 1990 IJCNN International Joint Conferenceon Neural Networks, pages 41–46 vol.3.

Shamsul, N. S., Kamarudin, S. K., Rahman, N. A., and Kofli, N. T. (2014). Anoverview on the production of bio-methanol as potential renewable energy.Renewable and Sustainable Energy Reviews, 33:578–588.

Shehu, M. S., Abdul Manan, Z., and Wan Alwi, S. R. (2012). Optimization ofthermo-alkaline disintegration of sewage sludge for enhanced biogas yield.Bioresource Technology, 114:69–74.

95

Page 33: ABDUL SAHLI BIN FAKHARUDIN - psasir.upm.edu.mypsasir.upm.edu.my/id/eprint/68747/1/FSKTM 2018 8 - IR.pdf · Prestasi model telah dinilai menggunakan punca kuasa min ralat dan pekali

© COPYRIG

HT UPM

Smith, J. M. (1987). When learning guides evolution. Nature, 329(6142):761–762.10.1038/329761a0.

Soltanali, S., Halladj, R., Tayyebi, S., and Rashidi, A. (2014). Neural networkand genetic algorithm for modeling and optimization of effective parameterson synthesized zsm-5 particle size. Materials Letters, 136:138–140.

Souza, T. S. O., Ferreira, L. C., Sapkaite, I., Prez-Elvira, S. I., and Fdz-Polanco,F. (2013). Thermal pretreatment and hydraulic retention time in continuousdigesters fed with sewage sludge: Assessment using the adm1. BioresourceTechnology, 148:317–324.

Srinivas, M. and Patnaik, L. M. (1991). Learning neural network weights usinggenetic algorithms-improving performance by search-space reduction. InNeural Networks, 1991. 1991 IEEE International Joint Conference on, pages2331–2336 vol.3.

Stanley, K. (2002). Neat c++. http://nn.cs.utexas.edu/?neat-c. Accessed:2014-06-2.

Stanley, K. O. and Miikkulainen, R. (2002). Evolving neural networks throughaugmenting topologies. Evol. Comput., 10(2):99–127.

Starr, K., Talens Peiro, L., Lombardi, L., Gabarrell, X., and Villalba, G. (2014).Optimization of environmental benefits of carbon mineralization technologiesfor biogas upgrading. Journal of Cleaner Production, 76:32–41.

Strik, D. P., Domnanovich, A. M., Zani, L., Braun, R., and Holubar, P. (2005).Prediction of trace compounds in biogas from anaerobic digestion using the{MATLAB} neural network toolbox. Environmental Modelling & Software,20(6):803 – 810.

Syaichurrozi, I., Budiyono, and Sumardiono, S. (2013). Predicting kineticmodel of biogas production and biodegradability organic materials: Biogasproduction from vinasse at variation of cod/n ratio. Bioresource Technology,149:390–397.

Tedesco, S., Marrero Barroso, T., and Olabi, A. G. (2014). Optimizationof mechanical pre-treatment of laminariaceae spp. biomass-derived biogas.Renewable Energy, 62:527–534.

Turkdogan-Aydnol, F. I. and Yetilmezsoy, K. (2010). A fuzzy-logic-based modelto predict biogas and methane production rates in a pilot-scale mesophilicuasb reactor treating molasses wastewater. Journal of Hazardous Materials,182(13):460–471.

Vierucci, U. (2002). Neat java (jneat). http://nn.cs.utexas.edu/?jneat. Accessed:2014-06-2.

WBA (2014). World Bioenergy Association (WBA) global bioenegystatistics 2014. http://www.worldbioenergy.org/content/wba-gbs. Accessed:2015-06-2.

96

Page 34: ABDUL SAHLI BIN FAKHARUDIN - psasir.upm.edu.mypsasir.upm.edu.my/id/eprint/68747/1/FSKTM 2018 8 - IR.pdf · Prestasi model telah dinilai menggunakan punca kuasa min ralat dan pekali

© COPYRIG

HT UPM

Whitley, D. (1989). The genitor algorithm and selection pressure: Why rank-basedallocation of reproductive trials is best. In Proceedings of the ThirdInternational Conference on Genetic Algorithms, pages 116–121. MorganKaufmann.

Whitley, D. (1994). A genetic algorithm tutorial. Statistics and Computing,4(2):65–85.

Whitley, D., Starkweather, T., and Bogart, C. (1990). Genetic algorithmsand neural networks: optimizing connections and connectivity. ParallelComputing, 14(3):347–361.

Wilamowski, B. M., Iplikci, S., Kaynak, O., and Efe, M. O. (2001). An algorithmfor fast convergence in training neural networks. In Neural Networks, 2001.Proceedings. IJCNN ’01. International Joint Conference on, volume 3, pages1778–1782 vol.3.

Wilson, S. W. (1990). Perceptron redux: Emergence of structure. Phys. D,42(1-3):249–256.

Xu, F., Huang, Z., Miao, H., Ren, H., Zhao, M., and Ruan, W. (2013).Identical full-scale biogas-lift reactors (blrs) with anaerobic granular sludgeand residual activated sludge for brewery wastewater treatment and kineticmodeling. Journal of Environmental Sciences, 25(10):2031–2040.

Yang, X.-S., Koziel, S., and Leifsson, L. (2013). Computational optimization,modelling and simulation: Recent trends and challenges. Procedia ComputerScience, 18(0):855–860.

Yao, X. (1999). Evolving artificial neural networks. Proceedings of the IEEE,87(9):1423–1447.

Yao, X. and Shi, Y. (1995). A preliminary study on designing artificial neuralnetworks using co-evolution. In in Proceedings of the IEEE SingaporeInternational Conference on Intelligent Control and Instrumentation, pages149–154.

Yetilmezsoy, K., Turkdogan, F. I., Temizel, I., and Gunay, A. (2012). Developmentof ann-based models to predict biogas and methane productions in anaerobictreatment of molasses wastewater. International Journal of Green Energy,10(9):885–907.

Yin, C., Rosendahl, L., and Luo, Z. (2003). Methods to improve predictionperformance of ann models. Simulation Modelling Practice and Theory,11(34):211–222.

Zainol, N., Salihon, J., and Abdul-Rahman, R. (2008). Biogas production frombanana stem waste: optimisation of 10 l sequencing batch reactor. InSustainable Energy Technologies, 2008. ICSET 2008. IEEE InternationalConference on, pages 357–359.

Zainol, N., Salihon, J., and Abdul-Rahman, R. (2009). Biogas production fromwaste using biofilm reactor: Factor analysis in two stages system. WorldAcademy of Science, Engineering and Technology, 54.

97

Page 35: ABDUL SAHLI BIN FAKHARUDIN - psasir.upm.edu.mypsasir.upm.edu.my/id/eprint/68747/1/FSKTM 2018 8 - IR.pdf · Prestasi model telah dinilai menggunakan punca kuasa min ralat dan pekali

© COPYRIG

HT UPM

Zhou, H., Lffler, D., and Kranert, M. (2011). Model-based predictions of anaerobicdigestion of agricultural substrates for biogas production. BioresourceTechnology, 102(23):10819–10828.

Zhou, Y. and Wu, Y. (2011). Analyses on influence of training data set to neuralnetwork supervised learning performance. In Jin, D. and Lin, S., editors,Advances in Computer Science, Intelligent System and Environment, pages19–25. Springer Berlin Heidelberg, Berlin, Heidelberg.

98