Hybrid wind power prediction method based on long-short-term memory neural network

Abstract

The invention discloses a hybrid wind power prediction method based on a long-short-term memory neural network. The invention belongs to the technical field of wind power generation power prediction.According to the prediction method, a long-short-term memory neural network is introduced into wind power prediction, firstly, a decision tree method is adopted to carry out feature selection on fan system data, importance sorting is carried out on alternative input variables, feature attributes with the low importance degree are removed, feature attributes with the high importance degree are obtained, and the accuracy of later prediction is improved; wavelet threshold noise reduction is conducted on the selected characteristic attribute values, and original signals are obtained from the signals mixed with the high noise; an LSTM method is adopted to train the hybrid wind power prediction model, and the weight matrix of the prediction model is updated by increasing the number of iterations, so that the prediction precision is improved; and finally, the prediction error of the LSTM model is corrected by using a least square method, thereby further improving the prediction precision of the LSTM hybrid wind power prediction method.

Description

technical field [0001] The invention relates to the field of active power prediction of wind turbines in the process of wind power generation in wind farms, in particular to a hybrid wind power prediction method based on long-short-term memory neural networks. Background technique [0002] The installed capacity of wind power in my country is increasing year by year, and the proportion of wind power in the power industry continues to expand. The volatility, intermittency and chaos of wind power have brought serious challenges to the safe and stable operation of the power grid. At the same time, wind farms are required to maintain a high spinning reserve capacity to stabilize the grid voltage, resulting in serious economic losses. In order to reduce economic losses, the power system requires the prediction of ultra-short-term (15min-4h) and short-term (5h-48h) wind power of wind farms, and the prediction error should generally not exceed 20%. [0003] Although traditional st...

AI Technical Summary

Problems solved by technology

[0004] 1) The traditional machine learning algorithm does not consider the time-dependent information in the sample information, and the calculation cost is high, which affects the learning efficiency of the algorithm

[0005] 2) Traditional statistical methods are difficult to mine deep feature information in data, and there is still a lot of room for improvement in predictive performance

Images