Implementation of the spatial voltage vector modulation algorithm using an artificial neural network

Abstract

An algorithm for space vector pulse width modulation (SVPWM) based on an artificial neural network is proposed.In calculating the action time of the voltage vectors according to the traditional SVPWM algorithm, the values of the relative phase angle on the sectors are used as variables, which causes difficulties in their calculation.To overcome these shortcomings, mathematical models have been compiled that determine the opening time of the power element by the absolute phase angle on the inverter converter. Since mathematical models have different types of calculation formulas for sectors and it is impossible to express them in an unambiguous formula, to eliminate these difficulties, a function was built using an artificial neural network, with which you can generally determine the opening time of a power element in phases A, B and C.Regardless of the sectors, the SVPWM algorithm sets the opening time of the power element according to the phase angle θ on the inverter converter.The generalized ability of the artificial neural network and its accuracy, as well as the simulation operation time in the MATLAB environment, were tested.The proposed method makes it possible to achieve high speed control accuracy and significantly reduce the operation time.When controlling the traction drive of an electric rolling stock with an asynchronous traction motor, it is possible to reduce the regulation period, as a result, the traction capacity of electric locomotives and metro electric trains with vector control or with direct torque control increases by about 1.1–1.2 times.