Short-Term Load Forecasting Based on Pattern-Guided Convolutional Neural Networks Under Different Electricity Consumption Profiles

Power load forecasting is a critical aspect of ensuring the safe operation of modern power systems. The accuracy of power load forecasting directly affects the rationality of power system development planning. Therefore, this paper proposed a short-term load forecasting method based on pattern-guided convolutional neural networks. This method extracted typical load profiles by segmental downscaling reconstruction method, and adopted the improved k-means algorithm to divide the unlabeled load data into different centered load patterns, so as to realize the high-precision prediction of short-term loads under different power consumption patterns. Finally, we validated the effectiveness of the method on two datasets, and the coefficient of determination of the proposed model improved by 4.595% on average over the best comparison model.