Research On Measurement-based Load Modeling And Time-variation

As an important part of the power system, the influence of load models should not be underestimated. It has been nearly a hundred years since load modeling was carried out in foreign countries while in our country the deserved attention was obtained in the last three decades, and the process is very slow as its complexity. Two sections are included in this paper. Firstly models are built to subscribe the property of past loads, which is fundamental, Secondly models are defined considering the time-variation, which focuses on practical research.Data sources, model structure and parameter identification method are the three necessary conditions to establish a load model. Three kinds of currently available data sources are analyzed and a method for the selection of fault data file is presented. Commonly used load model structures are introduced and the time-variation adaptable one is chosen. Adaptive genetic algorithm is used for the identification of parameters with the improvement of crossover and mutation operators. Then a load modeling software platform is developed.For the load model varies as load characteristics changes, the law behind the time-variation must be found for practical. Some measured data were analyzed and it is found that some rules do exist. The data in same seasons can be divided into two kinds according to whether belong to rest time. When comes to the data of three different seasons, it is reasonable to be split into three kinds. What is more, data of similar time in two different years can also be considered identical. However, the load characteristics are affected by too many factors, so the load time-variation studies need to be the basis of statistical analysis of large amounts measured data.Cluster analysis is an effective method for the time-variation research, with which the measured data can be clustered into several groups described by only one model. The characteristic vector based on the response space is necessary and ideal for the cluster analysis, which is constituted by the response active and reactive power excited by the same voltage. As the instability of power output caused by the rectangular voltage previously used, the trapezoidal voltage excitation is presented which having a higher numerical stability. In order to reduce the dimension of feature vectors, piecewise aggregate approximation is proposed which could improve the efficiency of clustering with no harm to the accuracy.