Research On Electric Vehicle Load Model Based On Improved Kernel Density Estimation And Latin Hyper Cube Sampling

As a new generation of transportation, electric vehicles, which have huge advantages of low energy consumption and low pollution, will develop very rapidly in the future. Different from the traditional load, electric vehicles have the character of spatial-temporal random dynamics. Establishing a more accurate electric vehicle charging load model has a very significant influence for providing more accurate data for studing the impacts of electric vehicles on the grid, infrastructure planning and operation of power system. Currently, the traditional parameter estimation method is applied to build probability distributions of state of charge and charging start time. However, low precision and poor adaptability of the traditional parameter estimation method makes its application more limited in establishing the load model of electric vehicles. Therefore, focus on this point, this paper deal with the problem in the following three aspects:(1) An adaptive kernel density estimation with boundary kernel algorithm is proposed to build probability distribution models. In order to solve the problems of boundary bias and lacking of local adaptability which exist in traditional parameter estimation method, the proposed algorithm combines boundary Kernel with adaptive bandwidth. In this way, the proposed algorithm can greatly improves the precision and adaptability of probability distribution. Besides, the proposed algorithm does not require any assumptions about the probability distribution and can also effectively dig the statistical information in the sample data. Therefore, this algorithm overcomes the shortcomings like low precision and poor adaptability of the traditional parameter estimation method. Finally, traditional parameter estimation method, kernel density estimation method and adaptive kernel density estimation with boundary kernel algorithm are applied to build probability distribution models of state of charge and charging start time. The simulation result shows the accuracy of the proposed method.(2) Latin hypercube sampling with cubic spline interpolation algorithm is proposed. Considering that conventional Latin hypercube sampling method cannot be directly used in sampling of nonparametric kernel density estimation, the proposed algorithm combines cubic spline interpolation method with conventional Latin hypercube sampling method, which makes up the limitation of conventional Latin hypercube sampling method. In addition, this proposed method retains the advantages of the conventional Latin hypercube sampling method and has significant advantages in sampling precision and efficiency compared with the acceptance-rejection sampling algorithm which is widely used in sampling of nonparametric kernel density estimation. Finally, acceptance-rejection sampling algorithm and Latin hypercube sampling with cubic spline interpolation algorithm are applied to get the sample data of state of charge and charging start time. The simulation result demonstrates the accuracy and effectiveness of the proposed method.(3) Electric public bus charging load model is set up. Based on the measured data of a certain charging station, combined with the improved kernel density estimation algorithm and Latin hypercube sampling algorithm, the charging load model of the electric bus is established. Compared the result with traditional parameter estimation method and measured data, the charging load curve derived by the proposed algorithm is much more close to the measured data, which certificates the accuracy and effectiveness of the proposed method.