DFSO-LSTM-Based Market Demand Forecasting and Resource Scheduling for Independent Energy Storage in Power Grid

Abstract

Forecasting market demand and scheduling energy storage scheduling are critical challenges in power grids, particularly due to data irregularities and renewable energy uncertainty. This research proposes a hybrid DFSO-LSTM model combining Demand-based Fish Swarm Optimization (DFSO) and Long Short-Term Memory (LSTM) to enhance demand prediction and operational efficiency. The model was evaluated using a large-scale dataset comprising hourly power consumption, climate variables, and system parameters collected from public sources between January 2018 and June 2023. DFSO dynamically tunes key LSTM hyperparameters including time steps and hidden units by minimizing RMSE across validation sets. Experiments were conducted using python and obtained comparative analysis results against GA–BP and GAN-NetBoost shows that the proposed model achieves superior accuracy of 96.15%, with MAPE of 0.0569, RMSE of 1.085, MAE of 1.1025, MSE of 1.895 and R² of 0.957. A one-minute reduction in execution time demonstrates practical deployment viability. Statistical tests confirm that improvements are significant. These results validate the model’s effectiveness in enabling scalable, real-time energy storage scheduling and peak load management in smart grid environments.