Energy consumption time series prediction method and device based on gmdh selective combination

Abstract

The invention discloses a method and device for forecasting energy consumption time series based on GMDH selective combination, wherein the method includes: S1. Obtaining the original energy consumption time series y t , according to the original energy consumption time series y t Get the linear part prediction result and the nonlinear subsequence u t ; S2. Using a variety of nonlinear single models as the weak learner of the AdaBoost algorithm, in the nonlinear subsequence u t Use the AdaBoost algorithm for integrated prediction, and get multiple integrated prediction results; S3. For nonlinear subsequence u t Use the GMDH neural network to perform selective combined forecasting with multiple integrated forecasting results, and find the combined forecasting model u with the optimal complexity * ; S4. Utilize the combined forecasting model u of optimal complexity * Predict the non-linear subsequence u t Corresponding non-linear prediction result S5. Adding the linear part prediction result and the nonlinear part prediction result to obtain the time series prediction value of energy consumption. Compared with the existing mixed model model, the present invention has better prediction performance.

Description

technical field [0001] The invention relates to the technical field of energy consumption forecasting, in particular to a method and device for forecasting time series of energy consumption based on GMDH selective combination. Background technique [0002] The 2016BP World Energy Statistical Yearbook pointed out that although China's economic growth has slowed down in recent years and is undergoing structural transformation, China is still the world's largest energy consumer, producer and net importer. In 2015, my country's energy consumption accounted for 23% of the global total and 34% of the net increase in global energy consumption. Among the fossil energy sources, the fastest growth in consumption is petroleum, reaching 6.7%; among the non-fossil energy sources, solar energy has the fastest growth rate, as high as 69.7%, surpassing Germany and the United States to become the largest solar power generation country in the world. Therefore, researching and establishing a ...

AI Technical Summary

Problems solved by technology

[0006] The above studies have made great contributions to energy demand forecasting, but the existing hybrid models still have shortcomings: 1) The existing divide-and-conquer methods usually use a single forecasting model for the forecasting of the decomposed quantum sequence of energy consumption

For the forecast of China's energy consumption demand, we can only obtain the annual data of energy consumption since 1978, the sample size is small, and the existing models are difficult to guarantee the performance

Images