Models of power quality improvement devices based on neural networks

The paper examines aspects of the use of neural networks to improve the quality of electricity based on an active filter controller. An extensive analysis of power quality (PQ) indicators and existing PQ enhancement systems and devices was carried out. Modern domestic and foreign works devoted to research in the field of design and imple-mentation of active filters, development of algorithms for generating a compensating signal of an active filter are considered. Aspects of the development of control systems based on neural networks are analyzed. As a result of the analysis, a simulation model of a parallel active filter was designed for a power supply network of limited pow-er, which is a ship power system. A distinctive feature of such a system is the high internal resistance of the source, due to its relatively low power and commensurately high-power consumption of the nonlinear load. The active filter operating algorithm is implemented using a fairly well-known standard instantaneous power technique. A distinc-tive feature of the algorithm is the use of a trained two-layer neural network instead of a typical PID controller that controls the charge level of a capacitor that takes or releases reactive power from the network. The neural network is implemented using standard built-in functions of the MATLAB/Simulink modeling system. Training was carried out using compiled databases in the form of generated vector-column matrices during preliminary modeling, tak-ing into account correction coefficients. As a result of modeling the operation of an active filter with a regulator based on neural networks, the following results were obtained: the deviation of the obtained data on the total har-monic coefficient of voltage components in the process of harmonic compensation was 0.1% (4.2% without com-pensation, 0.12% and 0.13% for compensation with a PID controller and a neural network, respectively). The re-sults obtained allow us to judge the effectiveness of using neural network control as a component of control sys-tems for power active filters. Further development of control systems based on neural networks in power quality improvement systems is possible at various stages of the filtering process, starting with the detection of harmonic and interharmonic distortion, processing signals from current and voltage sensors, and ending with the generation of a control signal for the power components of the filter. One of the significant disadvantages of control systems with the addition of neural networks is the need for microprocessor devices with high computing power and per-formance. However, given the rate of growth in the productivity of modern microprocessors, the data minus is tem-porary and will be eliminated in the near future. Thus, the research carried out showed the need for further in-depth study of the issues of neural network control in power electronics.