The Wind Generator’ Power Effective Forecast Method Based on Modified One-Dimensional Convolutional Neural Network and metaheuristics

Abstract

The method for predicting of wind generator’ power based on a modified one-dimensional convolutional neural network was proposed in the article. The model of a modified one-dimensional convolutional neural network was created; it allows to extract the most significant features and increase the forecast accuracy due to the automatic calculation of the convolution, pooling and dense (fully connected) layers’ number and sizes. The method for neural network model parametric identification based on local search was developed; according to the absence of recurrent connections it allows to use a batch learning mode for the learning rate’ increase. The method for the neural network model parametric identification was created; according to the swarm of cats’ adaptive optimization and using of simulated annealing it makes possible to make the search global at the first iterations, and to make the search local at the last iterations; it increases of the forecast accuracy too. The method for predicting the wind generator’ power based on a modified one-dimensional convolutional neural network and metaheuristics increases the forecast efficiency and can be used in various intelligent systems for analyzing the characteristics of technical objects with high dynamics.