Enhancing the performance of Feed-Forward Neural Networks in the bus short-term load forecasting

Bus load patterns present low correlation in respect to the aggregated system load, due to their volatility and high complexity. Thus, special care should be placed in the sophisticated selection and training of the appropriate forecasting model. This paper is concerned with the Short-Term Load Forecasting on a distribution transformer that feeds a suburban area in Northern Greece. The forecaster corresponds to a modified version of the Feed-Forward Neural Network (FFNN) that has been proposed for the Greek interconnected system. Two novel FFNNs are introduced that differ with the previous one in the types of the variables of the input layer. Experimental results denote that the proposed FFNNs lead to higher prediction accuracy.