Research on User Adjustable Power Load Forecasting Method Based on Deep Learning and Cluster Analysis

With the rapid development of intelligent power system and demand-side management technology, accurate prediction and reasonable regulation of user adjustable load play a key role in improving the operation efficiency of power grid and promoting the consumption of new energy. This article proposes a load schedulability evaluation method based on the combination of deep learning and clustering analysis. This method utilizes the TSMixer network to model user historical load data, capturing the correlation between time series characteristics and features through time mixing and feature mixing methods. Combined with K-means clustering algorithm, the historical daily load data of users were clustered, and the adjustable load potential of users under each clustering mode was investigated. Then, this study used the TSMixer network to predict the user's load data in the future day, and classified the predicted results into the most similar clustering mode, so as to evaluate the adjustable load in the future day. The experimental results show that the TSMixer model has good predictive performance. At the same time, comparative analysis demonstrates that the load adjustability evaluation method proposed in this study, based on load prediction, outperforms existing methods.