Research On Prediction Model Of Extreme Learning Machine And Application Of Power Load Prediction

The continuous development of artificial intelligence fields provides more strategies for scientific prediction,but traditional prediction methods still have some deficiencies.As a new type of single hidden layer feedforward neural network,the extreme learning machine algorithm is applied to the prediction,which overcomes some deficiencies of the traditional prediction method.The training speed of the algorithm is improved,the tuning time of parameters is reduced,and the disadvantage that the traditional neural network is easy to fall into the local optimum is avoided.On the basis of in-depth study of the extreme learning machine algorithm theory,in order to improve the performance of the extreme learning machine prediction model,two different improved methods are proposed in this paper.The finished research contents in this paper are as follows:First,aiming at the limitation of the single kernel function,a multi-scale wavelet kernel function with strong local learning ability and a polynomial kernel function with strong global generalization ability are combined,so a new combined kernel extreme learning machine prediction model is proposed.At the same time,from the perspective of kernel parameter optimization,the quantum particle swarm optimization algorithm is used to optimize the parameters of combined kernel.The simulation results on the UCI datasets show that the model proposed in this paper has much more better fitting ability.Second,in order to improve the fitting ability of KELM and make it stable under noisy conditions,a weighted combined kernel extreme learning machine prediction model is proposed.The whale optimization algorithm is applied to solve the problem that the model has too many kernel parameters,which realizes the automatic optimization of the parameters.It can be concluded from the test results of sinc function and UCI datasets that the model has better prediction accuracy.Finally,the two models proposed in this paper are applied to the actual power load prediction,which predict the coming 24 hours and 48 hours power load of a region in the northeast China.The experimental results show that the two prediction models proposed in this paper have good application prospects.