抄録
This paper proposes a new method for the forecast of daily maximum electric load by using feedforward neural networks and recurrent neural networks. While the maximum electric load mostly depends on the weather conditions of the day, it is important for the forecast to consider the influence of the load or the weather conditions for the past few days. The proposed method consists of two steps in the learning stage. In the first step, the feedforward networks learn the standard electric load corresponding to the weather conditions of the day in which the load is to be predicted. In the second step, the recurrent networks learn the difference between the standard electric load and the measured one. The load forecast is executed as follows: the feedforward networks output the standard load after the weather conditions of the day are inputted. The final result of the load forecast is obtained as for the output of the recurrent networks which correct the standard load considering the time dependency of electric load. Computational experiments show the high abilities of the proposed method so that the annual average error of the forecast for weekdays is 1.7%.
