Research On Application Of Least Square Support Vector Machines And Unscented Kalman Filter In State Estimation

Power system state estimation is the core of electric energy management system and the bases of dispatch, control, security evaluation and so on. With the structure and operation mode of electrical network becoming more complex, the modern power system is requested to master the actual condition of entire net quickly and exactly, forecast and analyze the trend of system to provide the countermeasures to problems occured in operational process. State estimation contains static state estimation and dynamic state estimation which possess characteristic advantages according to different requirements. Therefore, this paper do some research on static state estimation and dynamic state estimation, the main works are as follows:â‘ static state estimate1) Research on existing regression model of state estimation based on least square support vector machines in static state estimation. On the basis of inheriting the advantanges of primary model, this paper proposes a two-storeyed gridsearch strategy about parameters-optimization based on minimization of mean square error, the strategy can choose reasonable parameters which determines machine learning capacity of the model, to minimize the structural risk and guarantee that distribution character of samples to be trained is reasonable, for promoting estimation accuracy.2) For suppressing the passive effect of that muti-baddata acts on state estimation, according to error of fitting which samples generate in regession estimation model, this paper adopts robust method to secondly adjust kernel parameters of samples to formulate a robust regession estimation model based on least square support vector machines, thus enhance the robustness of state estimation model, so that the model is more immune to the bad data.â‘¡dynamic state estimationOn account of weak traceability and unsatisfactory robustness that the traditional extended Kalman filter type algorithms which apply to dynamic state estimation represent in handling the abnormal event of power system, this paper adopts an new method of dynamic state estimation based on adptive unscented transform Kalman filter, avoiding the introduction of linear error. Morever, in the estimation process the statistical characteristics of noise could be synchronously identified and corrected, that makes the method has a higher valuation accuracy and stability than traditional methodes.Finally, Matlab programs based on the posed method are implemented, three IEEE standard system are calculated and analyzed. The results show the feasibility, accuracy and effectiveness of the methods and programs in this paper.