An adaptive power system load forecasting scheme using a genetic algorithm embedded neural network

The ability of load monitoring instrumentation to predict the onset of peak power demands, based on the acquired voltage and current measurements, is essential to most load management strategies. This paper describes an innovative load forecasting scheme employing the genetic algorithm (GA) embedded neural network. The new load forecasting technique is compared with the conventional artificial neural network approaches, which sometime suffer from the local minima optimization problem. Employing genetic algorithms on the design and training of artificial neural networks (ANNs) allows parameters to be easily optimized. Furthermore, the artificial neural network requires a heavy trial-and-error design and training procedure which impairs the performances of the resulting load forecasting schemes. Using the genetic algorithm approach to construct the ANN results in significantly reduced trial-and-error effort in the training phase and produces a load forecasting scheme which is more efficient, adaptive, and optimized than that provided by traditional artificial neural network based approaches. The proposed genetic algorithm embedded artificial forecasting scheme has been tested with data obtained from a sample study performed on the Taiwan power system and the experimental evaluations have demonstrated adaptability and effectiveness of the proposed load forecasting scheme