GP-based temperature forecasting for electric load forecasting

This paper proposes a new probabilistic method for maximum temperature forecasting in short-term electrical load forecasting. The proposed method makes use of Gaussian process (GP)of the kernel machine to evaluate the predicted temperature. In recent years, electric power markets become more deregulated and competitive. The power system players are concerned with maximizing a profit while minimizing a risk in the power markets. To improve the accuracy of load forecasting model, it is a key to predict the weather conditions of input variables precisely. In other words, it is meaningful to consider the uncertainty of the predicted temperature in short-term load forecasting. The proposed method aims at extending temperature forecasting for the average point into that for the posterior distribution to handle the uncertainty of temperature forecasting. The proposed method is successfully applied to real data of temperature in Tokyo. A comparison is made between the proposed and the conventional methods such as MLP (multi-layered perceptron), RBFN (radial basis function network) and SVR (support vector regression).