Daily Load Forecasting Using Quick Propagation Neural Network with a Special Holiday Encoding

In the last decade, neural networks have been applied in daily load forecasting. Nevertheless, two main problems are still present for using neural networks in this domain: first, poor load forecasting in holidays because complex load behavior, and second, the lack of a global model for both holidays and non-holidays. To solve these two problems, we propose a new special holiday encoding that considers holidays and its preceding and following days which are also affected by the holiday. This proposed encoding is used in conjunction with quick propagation neural network. In the experiments the proposed holiday encoding is compared with other encoding based on the forecasting error of quick propagation. To evaluate their performances, we used a Peruvian load data set. The results show that the proposed holiday encoding produce better forecasting results than the results produced by other holiday encoding. Finally, these same results are also better than those results obtained by using ARIMA model which is a statistical technique also used in practice.