论文部分内容阅读
Minimax probability machine regression (MPMR) was proposed for chaotic load time series global prediction. In MPMR, regression function maximizes the minimum probability that future predication will be within an ε to the true regression function. After exploring the principle of MPMR, and verifying the chaotic property of the load series from a certain power system, one-day-ahead predictions for 24 time points next day were done with MPMR. The results demonstrate that MPMP has satisfactory prediction efficiency. Kernel function shape parameter and regression tube value may influence the MPMR-based system performance. In the experiments, cross validation was used to choose the two parameters.