Optimizing LSTM Hyperparameters for household electrical consumption Forecasting

The effectiveness of power management is directly linked to the accuracy of the forecasting method employed. Consequently, the utilization of optimization methods can significantly enhance forecasting accuracy. This paper presents a novel hybrid method for short-term time series forecasting of household electrical consumption. By combining the Particle Swarm Optimization (PSO) algorithm with Long Short-Term Memory (LSTM) layers, we aim to enhance the accuracy of electrical consumption predictions. The PSO algorithm is employed to optimize the hyperparameters of the forecasting model, which consists of stacked LSTM layers. The effectiveness of the proposed model is evaluated on electrical consumption data from a household in France with a one-minute resolution. Simulation results demonstrate the superior performance of the proposed method compared to other benchmark methods, including LSTM, RNN, CNN-LSTM, RNN-LSTM, and ConvLSTM models.