Next day load curve forecasting using recurrent neural network structure

A recurrent neural network approach for next day load curve forecasting using the concept of similar days is presented. Neural networks applied in traditional prediction methods all use similar days´ data to learn the trend of similarity. However, learning all the similar days´ data is a complex task, and does not suit the training of a neural network. The authors use the Euclidean norm with weighted factors to evaluate the similarity between a forecast day and a searched previous day. The load correction method for the generation of new similar days is proposed in which the new load data that are suitable to train the neural network, are generated by correcting the historical days´ data with the help of load correction rates. In the proposed prediction method, the forecast load is obtained by adding a correction to the selected similar days´ data, the correction being obtained from the neural network. For training the neural network the back-propagation algorithm along with online learning are adopted. The performance of the developed method for load forecasting has been tested using the actual load data and temperature data of the Okinawa Electric Power Company, Japan. The results indicate that accurate load forecasting can be achieved by adopting the proposed method. This paper is an extension of the one-hour-ahead load forecasting method.