Artificial Neural Network Training Algorithm for Factor-Based Prediction of Electricity Consumption in the Household Sector

An analysis of the known methods for forecasting electricity consumption in urban distribution electric networks has shown that they are all based on the availability of sources of both electric and thermal energy. Reduction or complete absence of heat sources leads to significant forecast errors, which entails a drop in the energy efficiency of urban electrical networks and degraded reliability of power supply system individual components. Theoretical, computer, and neural network models that help achieve more accurate forecasting of electricity consumption in the household sector are proposed. Based on the developed mathematical model and taking into account the factor coefficients obtained for 2020 for nine cities of the Republic of Tajikistan, monthly values of the coefficient characterizing the terrain conditions were calculated. The results obtained using the proposed mathematical model were compared with the data of computer and neural network models. A method that helps obtain more accurate forecasting of electricity consumption in the household sector is proposed. To automate the monitoring and control of electricity generated by renewable energy sources, an algorithm for training an artificial neural network for factor-based forecasting of electricity consumption is proposed, the use of which will help improve the forecasting accuracy owing to the possibility of constantly training the neural network. The algorithm efficiency is confirmed by good agreement between the results obtained both using the proposed models and based on the readings of electricity meters.