Load Index Forecasting Method Based on Nonlinear Dimensionality Reduction of Correlated Factors and Generalized Regression Neural Network Fitting

Abstract

The power load characteristic index is an industry index that describes the characteristics of the load and the law of load change, and is an important reference basis for the decision-making of the power grid dispatching and planning department. However, the medium and long-term load characteristic index data points are sparse, and the forecasting method that only analyzes the data trend has great limitations. Therefore, it is necessary to consider the external influence factors of the load in the forecasting model to improve the effectiveness and accuracy of the forecast. First, the weight-improved gray correlation analysis method is used in the article to evaluate the degree of influence of external factors such as weather, economy, and society on the load characteristic indicators. The factors with low correlation are removed, and then t-SNE is used. Reduce the dimensions of multiple influencing factors to reduce data redundancy. Then build multiple linear and nonlinear regression models of mid and long term load indicators through generalized regression neural network, and determination of optimal super parameters by one dimensional optimization to achieve mid and long term load characteristic indicators prediction. Finally, the feasibility of the method is verified by using relevant data such as load in a certain area of southwest.