Smart Grid Peak Shaving with Energy Storage: Integrated Load Forecasting and Cost-Benefit Optimization

This paper presents a solution for energy storage system capacity configuration and renewable energy integration in smart grids using a multi-disciplinary optimization method. The solution involves a hybrid prediction framework based on an improved grey regression neural network (IGRNN), which combines grey prediction, an improved BP neural network, and multiple linear regression with a dynamic weight allocation mechanism to enhance prediction accuracy. Additionally, an improved cuckoo search (ICS) algorithm is designed to empower the neural network model, incorporating a gamma distribution disturbance factor and adaptive inertia weight to balance global exploration and local exploitation, achieving a 40% faster convergence rate. A multi-objective snake optimization algorithm is also developed to optimize economic cost, grid stability, and energy utilization efficiency using energy storage capacity as the decision variable. The experimental results, based on a 937-day load dataset from a chemical park in Jiangsu Province, show that the IGRNN model has better prediction accuracy than traditional models, with an RMSE of 11.1361, an MAE of 8.264, and an R2 of 96.90%. The optimized energy storage system stabilizes the daily load curve at 800 kW, reduces the peak-valley difference by 62%, and decreases grid regulation pressure by 58.3%. This research provides theoretical and practical support for energy storage planning in high renewable energy proportion grids. Future work will focus on integrating weather data and dynamic optimization strategies under policy constraints to improve system applicability in real-world scenarios.