Improving Energy Consumption Prediction Using LSTM and DNN Optimized by Particle Swarm Optimization

Abstract

This chapter presents a novel approach to energy consumption forecasting by integrating Particle Swarm Optimization (PSO) with Long Short-Term Memory (LSTM) networks and Deep Neural Networks (DNN). A daily climate dataset from Delhi, India, was used for training and validation. Initial results using LSTM alone were improved significantly by optimizing the model's parameters with PSO. The same optimization applied to the DNN also demonstrated notable enhancements. The effectiveness of the approach was evaluated using accuracy metrics such as Mean Squared Error (MSE) and Mean Absolute Error (MAE), confirming that PSO can substantially enhance the forecasting accuracy of energy consumption models.