Design of Power Load Forecasting Algorithm for Distribution Network Based on Machine Learning

Power load forecasting is a complex task. At present, the external environment of China's economic operation is complex and severe, and there are great uncertainties in internal economic growth, economic structure optimization and population growth, which further increases the difficulty of power load forecasting. In this paper, based on ML(Machine Learning), a power load forecasting algorithm for distribution network is designed. The power load forecasting algorithm sets the original data into the model in tensor form including data volume, time step and variable dimension. During supervised training, the prediction error between the predicted value and the expected value is back-propagated, and the network parameters are updated by gradient descent optimization algorithm until the model meets the convergence condition or completes the number of iterations. Finally, the research shows that when the four-layer network is adopted, the comprehensive load value is the smallest, which shows that when the number of hidden layers is 2, higher prediction accuracy can be obtained. Therefore, it can further prove the effectiveness of this algorithm in forecasting power load of distribution network.