Analytical framework for household energy management: integrated photovoltaic generation and load forecasting mechanisms

Abstract

This research focuses on investigating predictive analytics for renewable energy systems, specifically developing advanced forecasting models for solar photovoltaic (PV) power generation and non-dispatchable load consumption. To address the challenges associated with the intermittent and variable nature of solar energy, an innovative hybrid model is proposed. Specifically, this research integrates the K-nearest neighbor (KNN) classification method and genetic algorithm (GA) to optimize a backpropagation neural network (BPNN). This novel approach significantly enhances the precision of short-term solar photovoltaic power generation forecasting, enabling more accurate predictions of power output. This study proposed a prediction algorithm for non-dispatchable loads based on an online learning long short-term memory (LSTM) network. The algorithm determines whether to update parameters in the LSTM network through an online learning strategy by evaluating the root mean square error (RMSE) between prediction results and actual power consumption. The KNN-MBP algorithm reduces the RMSE by 50.36% compared to the MBP algorithm through weather classification. The KNN-GA-MBP algorithm demonstrates the best prediction performance among the three algorithms, with an RMSE of only 0.39 kW, this represents a 43.37% improvement in RMSE over the KNN-MBP algorithm and a 71.89% improvement over the MBP algorithm.