A Predictive Energy Management Framework for Malaysian Residential Microgrid Using LSTM-Driven Adaptive Rule-Based Control

Abstract

The rapid increase in global energy demand and the growing emphasis on environmental sustainability have accelerated the integration of renewable energy sources (RES) into modern microgrids. However, the stochastic nature of RES, the unpredictable behavior of load demand, and the variability of utility electricity tariffs under the Time of Use (ToU) pricing mechanism increase the operational complexity of microgrid energy management systems (EMS). To address these challenges, this research focuses on developing and implementing a hybrid EMS that integrates machine learning (ML)-based Long Short-Term Memory (LSTM) predictive models with conventional Rule-based Control (RBC) strategies for tertiary-level control of a centralized, grid-connected residential microgrid. The proposed EMS approach combines one-hour-ahead forecasts of solar Photovoltaic (PV) energy generation and residential load demand with predefined Load Following (LF) and Cycle Charging (CC) control strategies, enabling the EMS to anticipate future operating conditions while preserving the simplicity of the conventional RBC framework. The main objective of this research is to evaluate whether integrating LSTM predictive models into conventional RBC strategies can reduce the total operating cost of a Malaysian residential microgrid under the applicable ToU electricity tariff and Solar Accelerated Transition Action Program (Solar ATAP) scheme. The microgrid is modeled in MATLAB Simulink as a grid-connected system comprising a solar PV system, battery energy storage system (BESS), and three residential load profiles representing terrace, apartment, and condominium households in Kuala Lumpur, Malaysia. The data used to develop the proposed hybrid EMS are sourced from various trusted platforms, including Tenaga Nasional Berhad (TNB) energy smart meters, Huawei Solar Fusion, and Solcast. The simulation results showed that the developed LSTM predictive models achieved higher forecasting accuracy for both solar PV energy generation and residential load demand than the baseline Feedforward Neural Network (FFNN) models. Leveraging these forecasts, the proposed LSTM-RBC-based EMS effectively coordinated the distributed energy resources (DERs) under the ToU electricity tariff and Solar ATAP scheme by strategically scheduling BESS charging and discharging operations and managing grid energy import and export based on economic operating conditions. The proposed hybrid EMS achieved a significant reduction in total operating cost by approximately 30% to 40% compared with conventional RBC strategies under different initial BESS State of Charge (SoC) scenarios. These results demonstrate that integrating ML-based LSTM predictive models into conventional RBC strategies enables more economical microgrid operation, highlighting the potential of predictive control to improve the operational performance of residential microgrids.

ABSTRAK: Peningkatan pesat dalam permintaan tenaga global serta penekanan yang semakin meningkat terhadap kelestarian alam sekitar telah mempercepat integrasi sumber tenaga boleh baharu (RES) ke dalam mikrogrid moden. Walau bagaimanapun, sifat stokastik RES, ketidakpastian tingkah laku permintaan beban, serta perubahan tarif elektrik utiliti di bawah mekanisme harga Masa Penggunaan (ToU) telah meningkatkan kerumitan operasi sistem pengurusan tenaga (EMS) mikrogrid. Bagi menangani cabaran ini, kajian ini memberi tumpuan kepada pembangunan dan pelaksanaan EMS hibrid yang mengintegrasi model ramalan Ingatan Jangka Pendek Berpanjangan (LSTM) berasaskan pembelajaran mesin (ML) dengan strategi konvensional Kawalan Berasaskan Peraturan (RBC) bagi kawalan peringkat tertier untuk mikrogrid kediaman berpusat yang disambungkan kepada grid. Pendekatan EMS ini menggabungkan ramalan satu jam ke hadapan bagi penjanaan tenaga solar foto volta (PV) dan permintaan beban kediaman dengan strategi kawalan Pengikutan Beban (LF) dan Kitar Pengecasan (CC) yang telah ditetapkan, bagi membolehkan EMS menjangka keadaan operasi pada masa hadapan sambil mengekalkan kesederhanaan kerangka RBC konvensional. Objektif utama kajian ini adalah bagi menilai sama ada integrasi model ramalan LSTM ke dalam strategi RBC konvensional dapat mengurangkan jumlah kos operasi mikrogrid kediaman di Malaysia di bawah tarif elektrik ToU yang berkuat kuasa dan skim Program Solar Accelerated Transition Action Programme (Solar ATAP). Mikrogrid tersebut dimodelkan dalam MATLAB Simulink sebagai sistem yang disambung kepada grid dan terdiri daripada sistem solar PV, sistem penyimpanan tenaga bateri (BESS), serta tiga profil beban kediaman yang mewakili rumah teres, apartmen, dan kondominium di Kuala Lumpur, Malaysia. Data yang digunakan dalam pembangunan EMS hybrid ini diperoleh dari beberapa platform yang dipercayai, termasuk meter pintar tenaga Tenaga Nasional Berhad (TNB), Huawei FusionSolar, dan Solcast. Hasil simulasi menunjukkan bahawa model ramalan LSTM yang dibangunkan mencapai ketepatan ramalan lebih tinggi bagi kedua-dua penjanaan tenaga solar PV dan permintaan beban kediaman berbanding model asas Rangkaian Neural Suap ke Depan (FFNN). Melalui ramalan ini, EMS berasaskan LSTM-RBC berjaya menyelaras sumber tenaga teragih (DERs) dengan berkesan di bawah tarif elektrik ToU dan skim Program Solar ATAP melalui penjadualan strategik operasi pengecasan dan nyahcas BESS, serta pengurusan import dan eksport tenaga grid berdasarkan keadaan operasi yang lebih ekonomik. EMS hibrid ini mencapai pengurangan ketara dalam jumlah kos operasi, iaitu kira-kira 30% hingga 40% berbanding strategi RBC konvensional di bawah senario keadaan cas awal (SoC) BESS berbeza. Dapatan ini menunjukkan bahawa pengintegrasian model ramalan LSTM berasaskan ML ke dalam strategi RBC konvensional membolehkan operasi mikrogrid yang lebih ekonomik, sekaligus menonjolkan potensi kawalan ramalan dalam meningkatkan prestasi operasi mikrogrid kediaman.