Probabilistic short-term load forecasting with Gaussian processes

This paper proposes a new probabilistic method for short-term load forecasting with the Gaussian processes (GP). In recent years, the degree of uncertainty increases as the power system becomes more deregulated and competitive. The power system players are concerned with maximizing the profit while minimizing the risk in the power market. As a result, it is important to consider the uncertainty of the predicted load in short-term load forecasting appropriately. The proposed method aims at extending load forecasting for the average point into that for the posterior distribution of the predicted load to handle the uncertainty of load forecasting. In this paper, the hyperparameters of the covariance function is evaluated in GP by the hierarchical Bayesian model after extending GP into the kernel-based method. The proposed method is tested for real data of one-step ahead daily maximum load forecasting in comparison with the conventional methods such as MLP, RBFN and SVR.