Yuni Artha Chyntia Saragih, Erma Suryani

Implementasi Metode Hybrid Backpropagation Neural Network dan Particle Swarm Optimization untuk Prediksi Konsumsi Listrik Rumah Tangga Berdasarkan Golongan Tarif

Introduction

Implementasi metode hybrid backpropagation neural network dan particle swarm optimization untuk prediksi konsumsi listrik rumah tangga berdasarkan golongan tarif . Prediksi akurat konsumsi listrik rumah tangga dengan metode hybrid Backpropagation Neural Network dan Particle Swarm Optimization (BPNN-PSO). Model ini tingkatkan akurasi peramalan untuk perencanaan listrik.

Abstract

Konsumsi listrik pada sektor rumah tangga terus meningkat seiring dengan pertumbuhan penduduk dan perkembangan sosial ekonomi. Penelitian ini mengusulkan model peramalan hybrid yang mengintegrasikan Backpropagation Neural Network (BPNN) dengan Particle Swarm Optimization (PSO) untuk meningkatkan akurasi prediksi konsumsi listrik rumah tangga berdasarkan golongan tarif. Data historis penjualan listrik bulanan periode 2020 sampai dengan 2024 digunakan sebagai dataset, yang mencakup jumlah pelanggan, daya tersambung, dan konsumsi energi. Hasil penelitian menunjukkan bahwa model BPNN–PSO memiliki kinerja yang lebih baik dibandingkan dengan BPNN murni. Proses optimasi berhasil menurunkan nilai Mean Absolute Percentage Error (MAPE) dari 57,97% menjadi 46,38% serta meningkatkan nilai koefisien determinasi (R²) dari –0,2403 menjadi 0,1831. Model yang diusulkan kemudian digunakan untuk memproyeksikan kebutuhan listrik periode 2025–2029 dan menunjukkan adanya tren pertumbuhan yang konsisten. Temuan ini membuktikan bahwa pendekatan hybrid BPNN–PSO dapat menjadi alat peramalan yang lebih andal dalam mendukung perencanaan dan pengambilan keputusan di sektor ketenagalistrikan.

Review

The paper "Implementasi Metode Hybrid Backpropagation Neural Network dan Particle Swarm Optimization untuk Prediksi Konsumsi Listrik Rumah Tangga Berdasarkan Golongan Tarif" addresses a highly relevant and practical challenge: accurately forecasting household electricity consumption, which is crucial for effective energy sector planning and resource management. The authors propose an innovative hybrid predictive model that integrates a Backpropagation Neural Network (BPNN) with Particle Swarm Optimization (PSO). This approach aims to leverage the pattern recognition capabilities of neural networks while enhancing their performance through metaheuristic optimization, offering a promising solution to improve the accuracy of electricity consumption predictions based on tariff groups. The methodology utilized historical monthly electricity sales data from 2020 to 2024, encompassing metrics such as customer count, connected power, and energy consumption. The core finding indicates that the hybrid BPNN-PSO model significantly outperforms a standalone BPNN. Quantitatively, the abstract reports a substantial reduction in the Mean Absolute Percentage Error (MAPE) from 57.97% for the pure BPNN to 46.38% for the BPNN-PSO model. Furthermore, the coefficient of determination (R²) saw a notable increase from a negative value of –0.2403 to 0.1831. These results suggest that PSO effectively optimizes the BPNN's parameters, leading to a more accurate model. The optimized model was subsequently applied to project future electricity needs for 2025–2029, demonstrating consistent growth trends and highlighting its utility for long-term strategic planning. While the demonstrated relative improvement of the BPNN-PSO hybrid model over a pure BPNN is a commendable achievement, the absolute performance metrics presented in the abstract warrant further consideration. An MAPE of 46.38% and an R² of 0.1831, though improved, still indicate a model with substantial error and limited explanatory power, which might temper its overall reliability for critical decision-making, despite the authors' claim of "more reliable." For a comprehensive journal publication, it would be beneficial to detail *which specific* BPNN parameters PSO optimizes and to include comparisons with other established forecasting methodologies (e.g., ARIMA, SARIMA, other machine learning techniques) to more robustly validate the proposed hybrid approach's superiority. Nevertheless, the study introduces a valuable hybrid modeling technique with clear potential for practical application in energy sector planning, laying a solid foundation for future research aimed at achieving even higher levels of predictive accuracy and robustness.

Full Text

Comments

You need to be logged in to post a comment.