Abstract: In order to improve the accuracy of solarpower prediction,a short-term solarpower probability prediction method based on an improved temporal convolutional network is proposed. First,recursive feature elimination is used to determine the number of features,and the EGSG method is used for feature selection;then the variational mode decomposition（VMD）is used to decompose the power sequence. Finally,an improved time convolutional network prediction model combined with attention mechanism is constructed to obtain the predicted values at different quantiles in the future,the kernel density estimation is used to obtain the probability density curve. Experimental results show that the proposed method has higher prediction accuracy and can reflect the uncertainty of photovoltaic output moreeffectively.