Application of neural networks in generation scheduling with fuzzy data

In this paper a new method which is noted as FANN+ES is proposed to solve unit commitment problems. We adopt an artificial neural network (ANN) enhanced by the fuzzy set concept to obtain an initial solution near the optimal operating point, and use an expert system (ES) to refine the initial solution. The ANN in this study is a three-layer network with back-propagation and supervised learning schemes. The training information in ANN is the typical daily unit commitment schedule. As the output of ANN consists of various possibilities for unit states and the load curve contains forecasting errors, the fuzzy set concept is employed successfully in this algorithm. The fuzzy variables are load demands, strategy costs and ANN output unit combinations. The application of expert system which has 26 rules in the knowledge base is an efficient way to improve the initial solution and obtain an economical operation schedule. Although the inclusion of fuzzy set will increase the execution time of the analytical methods for unit commitment it can be merged with the training process of ANN which is off-line and yet provide the scheduling output in seconds. The case studies indicate that the FANN ES can present a moremore » economical solution than ANN ES, without increasing the computation time.« less