Short-term Prediction of Multidimensional Data in Power System Based on CNN

With the continuous promotion of smart grid construction in China, the amount of data accumulated in the power system has increased significantly. It is important for grid state estimation and security situation assessment to mine relevant features from a large amount of time series data and predict future development trends. In this paper, a short-term prediction model of multidimensional data is established based on CNN. And one-dimensional convolution kernel is used to extract different dimensional features of data. Based on this model, single-step prediction and multi-step prediction methods are realized, respectively. In the multi-step prediction method, the direct output method is used to improve the prediction accuracy. The results show that the prediction model has high accuracy in different time scales. And the prediction error decreases with the decrease of time scales. Therefore, the prediction model can provide a reference for the short-term prediction of power systems.