Perancangan Model Prediksi Daya Output Generator Listrik Biogas Berbasis Backpropagation Neural Network Dan Genetic Algorithm

Syahputra, Heydhy Maulana (2026) Perancangan Model Prediksi Daya Output Generator Listrik Biogas Berbasis Backpropagation Neural Network Dan Genetic Algorithm. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 2042221095-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (5MB) | Request a copy

Abstract

Prediksi daya keluaran pada generator listrik berbahan bakar biogas merupakan tantangan yang kompleks karena fluktuasi parameter operasional seperti kandungan metana, laju aliran biogas, dan kecepatan putaran mesin (RPM). Penelitian ini merancang model prediksi daya listrik berbasis Backpropagation Neural Network (BPNN) yang dioptimalkan menggunakan Genetic Algorithm (GA) guna menghasilkan model yang efisien dan akurat dengan kompleksitas arsitektur yang minimal. Tiga parameter utama yang digunakan sebagai input model adalah kandungan CH₄, flowrate biogas, dan RPM, sedangkan output model berupa daya listrik (Watt). Akuisisi data dilakukan menggunakan sensor MQ-4, sensor mass flowrate F1031V, dan sensor IR RPM, serta sensor PZEM-004T sebagai sumber data referensi. Dataset sebanyak 385 sampel dilatih menggunakan MATLAB dengan rasio 70% data training, 15% validation, dan 15% testing. Pengujian terhadap 24 konfigurasi BPNN tanpa optimasi menghasilkan arsitektur terbaik berupa 2 hidden layer [10, 5] dengan fungsi aktivasi Sigmoid (R² testing 0,8783, RMSE 70,48 W, MAE 51,50 W, 101 parameter). Penerapan optimasi GA dengan populasi 100 kromosom selama 30 generasi berhasil menemukan arsitektur yang jauh lebih ringkas, yaitu 2 hidden layer [4, 2] dengan fungsi aktivasi Sigmoid, yang menghasilkan R² testing 0,8878, RMSE 71,86 W, dan MAE 47,92 W hanya dengan 29 parameter trainable (sekitar 29% dari model konvensional) serta performa yang setara atau sedikit lebih baik. Analisis korelasi menunjukkan flowrate sebagai parameter paling dominan (r = 0,7893), sedangkan RPM tidak berkontribusi signifikan (r = −0,0605). Pada pengujian prediksi terhadap data aktual PZEM-004T di empat level beban (40 titik pengamatan), model memperoleh MAE 88,69 W, RMSE 109,26 W, dan R² 0,8616. Model final diimplementasikan pada Raspberry Pi 5 untuk prediksi daya secara real-time, sehingga lebih sesuai untuk embedded system pada pembangkit listrik berbasis biogas.
=====================================================================================================================================
Predicting the output power of a biogas-fueled electric generator is a complex challenge due to fluctuations in operational parameters such as methane content, biogas flow rate, and engine rotation speed (RPM). This study designs an electric power prediction model based on Backpropagation Neural Network (BPNN) optimized using Genetic Algorithm (GA) to produce an efficient and accurate model with minimal architectural complexity. Three main parameters used as model input are CH₄ content, biogas flow rate, and RPM, while the model output is electric power (Watt). Data acquisition was carried out using an MQ-4 sensor, an F1031V mass flow rate sensor, and an IR RPM sensor, as well as a PZEM-004T sensor as a reference data source. A dataset of 385 samples was trained using MATLAB with a ratio of 70% training data, 15% validation, and 15% testing. Testing 24 BPNN configurations without optimization resulted in the best architecture of 2 hidden layers [10, 5] with Sigmoid activation function (R² testing 0.8783, RMSE 70.48 W, MAE 51.50 W, 101 parameters). The application of GA optimization with a population of 100 chromosomes for 30 generations succeeded in finding a much more compact architecture, namely 2 hidden layers [4, 2] with Sigmoid activation function, which produced R² testing 0.8878, RMSE 71.86 W, and MAE 47.92 W with only 29 trainable parameters (about 29% of the conventional model) and equivalent or slightly better performance. Correlation analysis showed flowrate as the most dominant parameter (r = 0.7893), while RPM did not contribute significantly (r = −0.0605). In the prediction test against actual PZEM-004T data at four load levels (40 observation points), the model obtained MAE 88.69 W, RMSE 109.26 W, and R² 0.8616. The final model was implemented on Raspberry Pi 5 for real-time power prediction, making it more suitable for embedded systems in biogas-based power plants.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Biogas, Prediksi Daya, Backpropagation Neural Network, Genetic Algorithm, Sistem Tertanam, Biogas, Power Prediction, Backpropagation Neural Network, Genetic Algorithm, Embedded System
Subjects:	T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK7878 Electronic instruments T Technology > TP Chemical technology > TP359 Biogas
Divisions:	Faculty of Vocational > Instrumentation Engineering
Depositing User:	Heydhy Maulana Syahputra
Date Deposited:	12 Aug 2026 00:38
Last Modified:	12 Aug 2026 00:38
URI:	http://repository.its.ac.id/id/eprint/144320

Actions (login required)

View Item