The Application Research Of Strong Tracking Filter And Wavelet Analysis Theory In Short-term Power Load Forecasting

In automatic control, aerospace, computer science, network management, power system and many other fields of engineering practices and scientific researches, the forecasting and estimation for random loading have both theoretical and practical importance. The random loading signals are generally non-stationary, which can be further regarded as the sum of a stationary signal and an unsteady signal. However, the non-steady part is often influenced by other external unsteady factors, which adds many difficulties to the forecasting and estimation of the non-stationary random loading signal.To overcome the difficulties, the following research works are developed:(1) On account of the effects of some unstable random factor, such as weather condition on the measurements of the random loading signals, the measurement equation is re-modeled as the nonlinear equation of state parameters. Furthermore, the model of state equation is not accurate. Thus, the extended Kalman filter (EKF) and strong tracking filter (STF) are, respectively, used to estimate the system load and the influence factor. The simulation results are presented to verify that the real-time tracking performance of breaks of states and parameters of STF is better than that of EKF.(2) Most non-stationary random signals in power, networks, economic and other observation systems have an obvious periodicity. Based on (1) and using the random walk characteristics of their wavelet coefficients, a new algorithm, named as wavelet - STF hybrid estimation and prediction algorithm is developed to forecast the power load signals. The wavelet analysis is introduced to make a multi-scale transform of the random signals so that the obtained signals can be decomposed into multi-scale space. Because the new algorithm can update all state vectors on the same period based on each measurement, the prediction accuracy and length are improved. The simulation results indicate that this algorithm is applicable for short-term load prediction, and also has good robustness to model uncertainty and an excellent tracking capability of the break of parameter and state.