A Neural Network Data Set Regression Prediction Algorithm Based on K-means

Among the high-dimensional data regression analysis and prediction model in the research of smart grid, traditional statistical models cannot balance information values between different dimensions, which has an impact on prediction validity of the data set. In order to solve this problem, a neural network data set regression prediction algorithm based on K-means is proposed. First, at the feature level, multi-layer recurrent neural network (RNN) extract data features of different dimensions and train responses. Then, at the algorithm level, classifier model of K-means is used to classify and fuse feature responses of the RNN according to the dimensional characteristics of the data. Subsequently, the combination forecasting model is constructed for the data set of output response, thereby improving the reliability of the prediction algorithm. The test was carried out on the public regression data set. The experimental results showed that the mean absolute error was reduced by 14.45% compared with Gated Recurrent Unit (GRU).