The Upper Limit Estimation Of Forecast Accuracy For Electric Load In "Mechanism Model+Identification Model" Strategy

Scientific forecasting is foundation and assurance of correct decision. Short-term load forecasting of electric power systems is a traditional important subject, which is important to economy and stability of system operation.What is the upper limit of forecast accuracy for the special time series? The estimation method of the upper limit is given, basing on the mathematical statistics and time series theory. According to the different time series, using intelligent method, which are SVM, wavelet method and so on, demonstrate the validity and efficiency of the method by numerical experiments.The main contents and results are:(1)Experience of short-term load forecasting has proved that prediction accuracy is different for different power network and which is related to historical data length for the given power network. According to the experience, build mathematical statistics models.(2)According to Cramer's decomposition theorem, a given non-stationary time series can be decomposed into two uncorrelated different parts: signals and white noise. The confidence interval for estimation of white noise variance is inversely proportional to sample size. It indicates that the upper limit of prediction accuracy is existent.(3)When the rate of signals is lower than exponential growth, the signals will be gradually weakened with the increase of differential times. On the contrary, the white noise will be quickly strengthened. After multi-difference, only contains white noise. So the noise variance can be approximately estimated.(4)Demonstrate the validity and efficiency of the upper limit of prediction accuracy by numerical experiments, using the SVM, wavelet and power spectrum analysis, etc.(5)Demonstrate the validity and efficiency of the upper limit of prediction accuracy by numerical experiments, using the theory of optimization of load forecasting models linear combination based on robust statistics.