Short-term electricity load online forecasting based on extreme learning machine
In order to meet the demand for real-time power load forecasting in smart grids,a sparse recur-sive least squares-based extreme learning machine(SRLS-ELM)online learning algorithm was proposed and used for short-term power load online forecasting.Compared with the online sequential ELM,the SRLS-ELM algorithm did not need to select off-line samples to initialize the network output weights,combined the squared error of the network learning with the sparse regularization term of the output weights,used the l1-norm as a sparse method to simplify the hidden layer structure of the network,em-ployed the sub-gradient strategy to solve the problem that the cost function in the solving process cannot be minimized everywhere,completed the online learning by the recursive least squares training method,and could be adaptively adjusted to find the optimal regularization parameters according to the estimation error.Simulation experiments showed that the proposed algorithm could effectively simplify the network structure,and the short-term power load online prediction model based on SRLS-ELM had higher predic-tion accuracy and learning efficiency with high robustness compared with ELM,kernel based ELM batch learning,online sequential ELM semi-online learning and accurate online support vector regression models.
short-term power load forecastingextreme learning machineonline learningregularization
万方数据
维普
