LSTM network power generation multi-step prediction method based on correlation analysis

Abstract

The invention discloses an LSTM network power generation multi-step prediction method based on correlation analysis. The method comprises the following steps: firstly, calculating a Hirst index of each variable through an R / S analysis method, and removing variables without correlation; calculating the correlation degree between each variable and the power generation capacity by using a grey correlation method, further removing the variables with small correlation degree with the photovoltaic power generation capacity, and selecting the variables with large correlation degree for training; andfinally, preprocessing the training variables by using a W.Pedrycz granulation method, and constructing an improved long-short-term memory (LSTM) network to perform multi-step prediction on the photovoltaic power generation capacity. By analyzing and utilizing the variables before and after screening to perform photovoltaic power generation multi-step prediction simulation diagram and mean squareerror, the correlation analysis method and the effectiveness of constructing the improved LSTM network provided by the invention are proved, and the method is of great significance for solving the multi-step prediction problem with correlation variables.

Description

technical field [0001] The invention belongs to the field of prediction analysis, and in particular relates to a multi-step prediction method for LSTM network photovoltaic power generation based on variable correlation analysis. Background technique [0002] Solar photovoltaic power generation contains huge energy and is a renewable and clean energy. However, photovoltaic power generation is affected by multiple variables, showing strong volatility and intermittency. If it is possible to accurately predict the amount of photovoltaic power generation and realize the multi-step prediction of the effective length in the future, it will be of great significance for the control, optimization and future scheduling of electricity. [0003] There are many common time series forecasting methods and models, based on intelligent forecasting algorithms, typically wavelet analysis, support vector machine, artificial neural network, etc. Although this type of network has played an import...

AI Technical Summary

Problems solved by technology

Because the correlation of multi-feature and multi-dimensional variables cannot be clearly expressed, the prediction accuracy is not ideal

Images