NEURAL NETwork ALGORITHMS FOR ELECTRICAL ENGINEERING SYSTEMS: THE TASKS SOLVING BRIEF CLASSIFICATION

The use of artificial neural networks and machine learning methods in electrical engineering and electric power industry is actively developing. There are reviews of individual applications, but a systematic classification of the tasks being solved is difficult. The traditional division into regression and classification algorithms is insufficient. The paper proposes a hierarchical classification of tasks solved using neural network algorithms in electrical engineering, taking into account the interconnections between them. The classification includes the following sections. Signal processing provides for analysis and signal processing of electrical engineering systems using artificial neural networks. Multidimensional data processing involves analysis and interpretation of multidimensional data typical of complex electrical engineering systems. Regression tasks are aimed at predicting continuous quantities, for example, signal approximation, forecasting electricity consumption. Classification tasks involve division of data into classes, for example, diagnostics and fault identification. The features of the application in the «real-time» mode and in the «delayed» processing mode are considered. Combined use of algorithms provides for application of hybrid approaches combining various neural network architectures and methods. Approaches based on artificial neural networks are particularly effective for tasks that allow «delayed» processing. Tasks requiring real-time solutions (for example, relay protection and automation) often need additional monitoring using other methods that ensure the necessary reliability.