Short-Term Load Forecasting using Long Short-Term Memory Network on Various Sub System Load in PLN Indonesia

• Published in: 2021 International Conference on Technology and Policy in Energy and Electric Power (ICT-PEP) pp. 175 - 179
• Main Authors: Indralaksono, Rio, Firdaus, Alif Maulana, Wakhid, M. Abdul, Andreas, Novemi Uki, Wibowo, Galih Hendra, Abdillah, Muhammad
• Format: Conference Proceeding
• Language: English
• Published: IEEE 29.09.2021
• Subjects: ann-pso; Data models; forecasting; Java; Load forecasting; lstm; Predictive models; short-term; Simulation; Time series analysis; Training data
• DOI: 10.1109/ICT-PEP53949.2021.9601005

PT. Perusahaan Listrik Negara (PLN) Indonesia has a responsibility to provide and ensures the continuity of electricity on safety and reliability for all areas in Indonesia. An accurate electrical load forecasting is fundamental to deliver efficient energy scheduling management. This paper proposes a prominent technique called long-short term memory (LSTM) for short-term load forecasting. The load data for this research work is taken by each of 30 minutes in a day for each four weeks earlier in the past to predict the following week's load. To examine the efficacy of the proposed method, the Sulawesi and Java-Bali Sub System Load are utilized in this paper. The simulation result shows that the LSTM with 1-input has similar performance to ANN-PSO in MAPE value and slightly superior performance in RMSE at Java Bali Sub System Load. At Sulawesi Sub System Load, it is shown that LSTM with 1-input is superior to all performance parameters compared to 3-input LSTM, have a little lower performance to Java Bali prediction about 1.03% on MAPE value, but perform better performance up to 45.08 on RSME value. Although the performance of LSTM is different in each case, the simulation result shown the capability of LSTM as an additional reference in performing short-term load forecasting on the electricity system.