Abstract :

Renewable energy schemes have proven to be a viable alternative source of energy generation to off-grid communities. Self-excited induction generators (SEIG) are commonly used as a low-cost energy source; however, their output frequency and voltage must be regulated. This paper therefore proposes a Neural Network (NN)-based power electronics frequency regulation of a SEIG. The NN architecture was designed to manage the solid-state load controller (sLC) to ensure a near-constant load on the SEIG. A reference frequency stability set-point was introduced with the rotor frequency to produce a differential frequency signal which was subsequently trained in the NN using Levenberg-Marquardt model. The surrogate signal from the NN is then compared with a constant saw tooth waveform from the signal generator via a relational operator to generate a pulse width modulation (PWM) signal. The PWM signal controls the firing angle of the insulated gate bipolar transistor (IGBT). Similarly, the duty cycle of the PWM signal from the NN regulates the gate of the IGBT which estimates the magnitude of the dumped power. The set-up was carried out in MATLAB R2018a with a 75 kW, 415 V hydro-driven SEIG under different loading conditions. The feedforward neural network (FFNN) was employed to ensure swift load variation management and frequency control via the power electronic devices. The results from the simulation show that the composite approach of FFNN-sLC was able to manage the load dynamics of the SEIG and subsequently control the load frequency with a fast response time.