Enhancing renewable energy load forecasting through deep data analysis and feature extraction techniques

Renewable energy power and user load exhibit significant volatility and uncertainty. By forecasting both supply and load, the energy system’s capacity for renewable energy integration and operational stability can be enhanced. This paper investigates forecasting models for system electric load and photovoltaic power within the context of a hybrid energy system in commercial building scenarios. While ensuring the accuracy of the data prediction models, it is also essential to enhance prediction speed. An improved backpropagation prediction model is employed, where the LMBP@anlz. IMFs training function optimizes the training speed and convergence of the learning model, making it well-suited for nonlinear time-domain data prediction. Additionally, the data augmentation technique is enhanced using the EMD algorithm. To reduce the volatility of energy data, the EMD algorithm preprocesses the data, and the analysis of the IMF components reveals the mechanisms influencing energy fluctuations. This data-driven approach strengthens the energy data structure. Subsequently, amplitude-frequency analysis methods extract effective features from the analyzed data, enhancing the prediction model’s focus on target features from various dimensions. This significantly improves the prediction accuracy for extreme data in short-term load forecasting. The correlation between predicted and observed electric load power values ranges from 0.9984 to 0.9997, while the correlation between predicted and observed PV power values ranges from 0.9978 to 0.9994. The research results demonstrate, through in-depth analysis of energy data, the impact mechanism of extreme load data on short-term load forecasting accuracy. This provides theoretical and technical support for operational forecasting research of integrated systems coupled with renewable energy.