Hybrid Algorithm for Optimization of Hyperparameters of Neural Network Model of Electricity Consumption Forecasting

The paper is devoted to the problem of improving the accuracy of short-term forecasting of electricity consumption using deep machine learning tools. The novelty of the work lies in the use of genetic algorithm (GA) to select hyper-parameters of neural network (NN), which affect the quality of its work. The genetic algorithm for selection of hyperparameters of neural network was created and tested. The influence of the given hyperparameters of the NN on the error of power consumption forecasting, such as the size of the data packet - Bs; the number of NN layers - j; neuron activation functions - Fa; optimizers - O, has been investigated. The training and testing of the created network with the selected parameters on the training and test sample containing data on power consumption for 2 years (17520 hours) was carried out. The analysis of input data on power consumption showed that the optimal parameters of the predictive neural network model in manual mode are: pack-age size 250 (empirically selected), 3 hidden layers, activation function “ReLU”, optimizer “Adam”. Different ways of hy-perparameter selection (manually and by means of GAs) are considered.