Performance Comparison of 1D-CNN and LSTM Deep Learning Models for Time Series-Based Electric Power Prediction

Abstract

ABSTRAK

Prediksi daya listrik yang akurat sangat penting untuk efisiensi energi, terutama bagi institusi dalam memenuhi kebutuhan energi yang dinamis. Penelitian ini membandingkan kinerja deep learning 1D-CNN dan LSTM untuk memprediksi daya listrik berbasis time series, menggunakan dataset dari Building Automation System (BAS) gedung STMKG. Metrik evaluasi Mean Squared Error (MSE) dan Mean Absolute Error (MAE) digunakan untuk mengukur akurasi. Hasil penelitian menunjukan untuk LSTM nilai MSE rata-rata dari 10 percobaan pada proses pengujian 3,35E-04±0,00013 dan MAE 0,01312±0,0079, sedikit lebih baik dari 1DCNN yang memiliki nilai MSE rata-rata 4,68E-04±0,0003 dan MAE 0,01855±0,00586. Walaupun perbedaannya tidak signifikan, 1D-CNN menawarkan keunggulan dalam efisiensi waktu komputasi 63,08s, 1D-CNN lebih cepat sekitar 84.19%.

Kata kunci: prediksi, daya listrik, time series, CNN, LSTM

ABSTRACT

Accurate electrical power prediction is essential for efficient energy management, especially in institutions with dynamic energy needs. This study compares the performance of 1D-CNN and LSTM for time series based electrical power prediction, using a dataset from the Building Automation System (BAS) of STMKG building. The evaluation metrics Mean Squared Error (MSE) and Mean Absolute Error (MAE) are used to measure accuracy. The results show that the LSTM had an average MSE value of 3.35E-04±0.00013 and an MAE of 0.01312±0.0079 across 10 trials. This is slightly better than the 1D-CNN, which had an average MSE value of 4.68E-04±0.0003 and an MAE of 0.01855±0.00586. Despite the marginal difference, 1D-CNN provides a computational time efficiency advantage of 63.08s, 1D-CNN is about 84.19% faster.

Keywords: prediction, electrical power, time series, CNN, LSTM