Multi-Objective Intelligent Demand-Side Management Using Explainable PSO-Optimized LSTM in Renewable Energy-Based Smart Grids

Abstract

The fast penetration of renewable energy resources, electric vehicles (EVs), and distributed energy systems has made demand-side management (DSM) in modern smart grids very complex. Traditional DSM methods usually consider load forecasting and scheduling as separate tasks, which results in suboptimal energy consumption, higher operational cost, greater carbon emissions, and lower grid stability under dynamic operating conditions. To address the above limitations, an Explainable Hybrid Particle Swarm Optimization-Long Short-Term Memory (PSO-LSTM) framework for intelligent multi-objective demand-side management in renewable energy-integrated smart grids is proposed. The proposed framework integrates the temporal learning capability of LSTM to accurately forecast the short-term electricity demand and the global optimization capability of PSO to optimally schedule appliances, coordinate EV charging, allocate renewable energy, and implement dynamic demand response. Hence, the optimization problem is formulated as a constrained multi-objective mathematical model to minimize the electricity cost, peak-to-average ratio (PAR), carbon emissions, and scheduling delay and to maximize the renewable energy utilization, consumer comfort, and grid reliability. The framework is validated on multivariate smart grid datasets, including historical electricity consumption, weather, photovoltaic generation, electricity prices, battery state-of-charge, EV charging profiles, and grid load data. Experimental results show that the proposed framework PSO–LSTM obtains an accuracy of 98.24% for forecasting and decreases RMSE and MAE to 0.081 and 0.061, respectively. Moreover, the proposed model achieves 24.76% peak-load reduction, 21.35% electricity cost savings, 91.42% renewable energy utilization, and significant reductions in carbon emissions compared with ANN, SVM, standalone LSTM, and Deep Reinforcement Learning (DRL) models. The statistical analysis also validates the robustness and superiority of the proposed framework in multiple performance metrics. The proposed explainable hybrid framework provides a scalable, computationally efficient, and real-time intelligent energy management solution for next-generation sustainable smart grids and smart city applications.