| The digital simulation of power system based on the device modeling is the main tool for power system operation, optimization, design and control. However, the load modeling, as one of the four foundation devices, is backward development because of its particularity and difficulty. So, its development is the deciding factor of the precision of the digital simulation. The traditional load modeling makes achievement in many scenarios. However, as the injection of wind power, it has great impact on the power system. From the perspective of the load modeling, wind power injection changes the load composition and the power flow. What’s more, the uncertainty of the generalized load rises on account of the random of the wind power and the time-variation of the load, which brings the new challenge to the load modeling. This paper discusses some problems in the process of the generalized load modeling considering the uncertainty of the wind power injection as follows:Firstly, a novel method based on probability statistics is proposed to learn and abstract steady state generalized load characteristics. To analyze the change of the power direction, the node characteristic is divided into source characteristics and load characteristics. To solve the uncertain variation problem, the active power is used to represent characteristics and segment according to the measured data. The range of the segments is determined adaptively and the probability distribution is got by probability statistics. Levenberg-Marquardt neural network is used to abstract the node characteristics, then the united model is built and it is applied to the actual case about the risk analysis. It is verified that the new method can not only build accurate model, but also analyze the uncertain problem making use of probability statistics according to the different scenarios. So the new method is the extension and supplement of the traditional method in the application.Secondly, using the experience of the clustering analysis and comprehensive of the traditional load modeling, the time information is imported to analyze the generalized load characteristics on the basis of the new method of the load modeling, which can improve the accuracy and the practicability of the load modeling analysis. The clustering analysis needs reasonable clustering method and strategy to solve the complex problem considering the wind power and the load. So, the AP clustering is used, which doesn’t need input the parameters and is more reasonable. The simulation results show the effectiveness and feasibility of our algorithm.Thirdly, the longitudinal clustering is proposed, which is used to achieve the longer time scale clustering. The clustering result reflects the seasonal characteristics, making use of the AP clustering and the longitudinal clustering. The longitudinal clustering need determine the minimal time interval T firstly and realize the intraday pre-clustering according to the statistics of the active power in every T. Then, the similar historical days are clustered together based on the daily property getting from the step forward. At last, the longitudinal time unit clustering is realized on the basis of the ratio of every category in the time units TZH· The similarity of the clustering result between different years shows the effectiveness of the clustering strategy.At last, on the basis of the longitudinal clustering, the horizontal clustering strategy is proposed, which can achieve the shorter time scale clustering. The horizontal clustering lines up all the data and clusters the similar periods, which reflects the daily time property. The strategy can achieve the shorter time scale clustering and the shorter time scale clustering in the unified time frame. The generalized load modeling with the probabilistic information is used to testify the effectiveness of the clustering strategy. The simulation results show the rationality of the new strategy, which is useful for dispatch and control considering wind power integration. |
PDF Full Text Request