Short-Term Load Forecasting Based on Kernel Conditional Density Estimation

A short-term load forecasting model based on the kernel estimation of the conditional probability density distribution is proposed. The pattern vector of the load time series sequence can be treated as the multivariate random variable whose value determines the pattern component values of the next sequence, which is forecasted. Probability density functions are obtained from historical load time series by means of nonparametric density estimation. This approach uses the product kernel estimators. The kernel function smoothing parameters are determined using cross-validation procedure. The suitability of the proposed approach is illustrated through applications to real load data.