Top.Mail.Ru

Perm Polytechnic University has created a “smart” voltage stabilizer

Scientists at Perm Polytechnic University have developed a “smart” voltage stabilizer based on a neural network: the system speeds up the response by 58% and significantly increases the accuracy of stabilization, the PNIPU press service reported.

The problem of unstable voltage is a common problem for modern electronic devices. Power sources based on renewable technologies (solar panels, batteries, hybrid systems) produce voltage fluctuations due to variable operating conditions and load surges. This makes it difficult to directly power low-power electronics that require strictly fixed parameters.

Traditional solutions with analog circuits on operational amplifiers do not fully cope with the task: they operate according to rigid logic, do not adapt to changes and do not guarantee safety in the face of sudden interference. The key difficulty arises from the nonlinear operation of the optocoupler in electrical isolation: the input and output are not directly connected, which leads to distortion and delays in stabilization.

Researchers at Perm Polytechnic University have proposed an intelligent voltage control system based on an adaptive neural network. It is based on an electronic DC‑DC converter controlled by a microcontroller. It measures the output voltage in real time, compares it with the reference value and corrects the operation. The neural network (built on a perceptron model) takes into account nonlinearities and errors of components, adapts to conditions, and compensates for inaccuracies due to heating or aging.

Testing in a computer model and on a physical prototype confirmed the advantages of the development. Compared to the classical scheme, a system with a neural network reduces the time it takes for the voltage to reach a given level from 125 to 79 microseconds (58% acceleration), reduces the stabilization error, bringing it closer to zero, and requires only 24 recalculation iterations instead of 57 - that is, it spends half as much computing resources.

The development opens the way to more reliable and energy-efficient power sources. It can be used in portable equipment, autonomous systems and robotics, where stability and quality of power supply are critical.