Comparative assessment of electricity losses in elec-trical networks using ANN and GMDH methods

Abstract

Relevance: in the modern world, where technology is developing rapidly, artificial intelligence (AI) is becoming not just a popular trend, but a powerful tool capable of transforming almost any industry. The electric power industry, like other sectors of the economy, does not remain aloof from this revolution. AI-based solutions help automate everyday tasks, including equipment health monitoring, energy consumption forecasting, power distribution management, and power grid optimization. Intelligent neural networks (ANN) are one of the key technologies in the field of artificial intelligence. AI processes large volumes of data, which helps identify hidden patterns and improve the use of resources. This can include both optimizing energy consumption and reducing losses in networks, which in turn increases the efficiency of power plants and reduces the cost of electricity production. It is also capable of analyzing data on weather conditions, seasonal changes, power demand and other factors to predict demand and optimize power production. This avoids supply interruptions and guarantees the stability of energy supply.

Aim:  forecasting electrical energy losses is one of the most important tasks in the work of energy sales organizations. Neural networks are used to solve problems that require analyzing large amounts of data. The main problem when predicting electricity losses using ANNs is the lack of statistical data for their training. To solve this problem, it is proposed to train an ANN based on data obtained as a result of mathematical modeling of the operation of the electrical network for certain calculation periods. The model is created taking into account the known generalized parameters of the functioning of the network and its loads over a long period of time.

Methods: The use of artificial neural networks, especially the multilayer perceptron (MLP), opens up new horizons in solving the problem of predicting electricity losses in distribution networks.

Results: The study showed that multilayer perceptron’s are a powerful tool for solving problems of predicting electricity losses, provided that a competent approach to choosing a model and setting parameters is used.