Predictive Modeling of Demand Response Impact on Solar-Integrated Power Systems Using Bayesian Optimisation Long Short-Term Memory Neural Networks

Abstract

Maintaining grid stability is made more difficult by the growing integration of solar energy into contemporary power systems, particularly when supply and demand are fluctuating. Demand Response (DR) programs provide opportunities for dynamic load management, but in order to measure their impact in real time, they need sophisticated forecasting tools. This study models and assesses the influence of DR signals on a solar-integrated power system by developing a predictive framework with a Bayesian-optimised Long Short-Term Memory (LSTM) neural network using the MATLAB Platform. The model was trained with carefully designed features, such as environmental, grid, and consumption parameters, using multivariate time-series data. It was then assessed under various DR intensities. Key hyper-parameters were adjusted using Bayesian optimisation, which greatly enhanced forecasting performance. The model demonstrated strong generalisation across low and high DR scenarios, achieving general performance metrics like RMSE of 0.101 kW, MAE of 0.062 kW, and R2 of 0.9426. The results obtained in this study indicate that the suggested model is a useful instrument for strategic demand-side management and intelligent energy forecasting.