Applications In The Smart Grid Based On Optimized Kernel Extreme Learning Machine

Wind power is growing fastest in the global power generation industry.The global installed capacity is expected to continue to grow in the future,and the uncertainty and variability of wind resources make it difficult to control the output power of wind farms.Therefore,the integration of large amount of wind power in power system poses some important challenges to the stability of power grid and the reliability of power supply.Wind power prediction plays a key role in wind power integration and operation.On the other hand,with the development of the power system and the penetration of new energy,the complexity and uncertainty of the power system increase obviously,which makes the power load forecasting become more and more important.Therefore,in the field of smart grid,wind power and power load forecasting are need to achieve more accurate results,so as to cope with the high cost of the smart grid operation and low power system reliability.Extreme Learning Machine(ELM)as a Single-hidden Layer Feedforward Neural Networks(SLFNs),a rapid machine learning method,is different from the neural network which use gradient descent iteration algorithm,it randomly determines weights between input layer and hidden layer of network and the output of the network is only obtained by matrix operations.Further combined with kernel learning method,Kernel Extreme Learning Machine(KELM)effective uses the advantages of the ELM with its fast training speed and simple process and also widely applys in the field of classification and modeling.This thesis is took KELM method as the main line,combining with strategy such as Biogeography-Based Optimization(BBO),studied its application in the smart grid,to further improve the prediction accuracy of wind power and power load for the purpose.This thesis is studied as follows:(1)Based on the ELM theory,basic KELM,KELM with RBF kernel and KELM with wavelet kernel are deeply researched,and in order to improve the stability and generalization ability of KELM method,the regularization coefficient ? is introduced as well as use the regularized least squares algorithm in the solution process.(2)Use the Genetic Algorithm(GA),Differential Evolution(DE),Simulated Annealing(SA)three algorithms to optimize the choice of the several parameters of the extreme learning machine,namely,the parameters of the RBF kernel function and the wavelet kernel function and the regularization coefficient.Then use the O-KELM method with two different kernels to apply to a region of the mid-term peak power load forecasting example,and under the same condition compare with Support Vector Machine(SVM),the Optimization Extreme Learning Machine(O-ELM)methods.The experimental results show that the O-KELM methods havebetter prediction effect than other methods,among which the O-KELM methods of wavelet kernel have the best effect,and the DE-KELM algorithm of wavelet kernel has the highest modeling accuracy.(3)Combined KELM method with Biogeography-Based Optimization(BBO)Algorithm,a kernel extreme learning machine based on BBO method optimization(BBO-KELM)is formed.Under the same condition,take the above method with the existing methods of O-ELM to apply to the optimal selection of KELM input structure and parameters of RBF kernel function and Tikhonov regularization coefficient to further improve the learning performance of KELM method.Then use BBO-KELM in different parts of the wind farms' wind power prediction examples,at the same time,compared with other methods under the same condition,thus draws the results that the BBO-KELM2 algorithm with the cosine transfer model is the best method.