Short-term forecasting method for net load considering the impact of high proportion of distributed renewable energy

This is the authors' abstract. We don't add key points for this paper.

• Sequence-to-point NILM approach decomposes net load into photovoltaic, wind, and actual load components. • Entropy correlation analysis quantifies the relationships among net load, actual load, wind power, and photovoltaic outputs. • Hybrid CNN-Transformer-DenseNet, stacked GRU variants, and BIGRU models forecast load, wind, and PV outputs. With the increasing penetration of distributed renewable energy sources, the volatility and multi-energy coupling of power system loads has become an increasingly important factor, making it particularly difficult to accurately forecast net loads with a high share of distributed renewable energy sources. The limitations of the metering equipment in the distribution network make it difficult to measure data from distributed generation units individually. This hinders the realization of indirect net load forecasting. In this paper, a combined short-term load forecasting model based on non-intrusive decomposition theory is proposed. The model first employs a neural network-based sequence-to-point learning non-intrusive load decomposition method to decompose the net load data into constituent components with different characteristics. Subsequently, the most appropriate forecasting method is selected based on the load characteristics of each component. Finally, the accurate prediction of net load is realized by aggregating the prediction results of each component. Example validation shows that the proposed combination algorithm not only enhances the prediction accuracy, but also exhibits excellent generalization capabilities.