Egyptian Unified Grid hourly load forecasting using artificial neural network

This paper presents an artificial neural Nntwork (ANN) based hourly load forecasting application to the Egyptian Unified Grid (EUG). The ANN involved is designed using the multi-layer backpropagation learning technique. The ANN input layer receives all relevant information that can significantly contribute to the prediction process, excluding the weather input information. The input layer receives information on: the class of day type; the hour in day time; the load in hour-before; the load in day-before at same hour; the average load in day-before; the peak load in day-before; the minimum load in day-before; and similar of last four measurements but in the week before. On the other hand, the ANN output layer provides the predicted hourly load. The ANN load forecasting model is trained based on an historical domain of knowledge. The required knowledge patterns are obtained for the EUG during the winter of 1993. When testing the trained ANN, it proves that it can be applied to the prediction of hourly load very efficiently and accurately. The training process scores an average error of 0.18% (absolute) with a standard deviation of 2.32%. On the other hand, the evaluation process reaches a 0.49% average error with a 2.92% standard deviation.