Time-shared electricity is a non-stationary series with trend characteristics，seasonal characteristics and stochastic characteristics，and the prediction using traditional forecasting methods will cause large errors. In order to improve the prediction accuracy of time-shared power under complex features，a power prediction model based on quadratic decomposition and tuna swarm optimization-temporal convolutional network （TSO-TCN） was proposed. Firstly，the original power data were decomposed using variational nonlinear chirp mode decomposition （VNCMD），denoised based on correlation analysis and improved wavelet thresholding denoising，and then the denoised data were subjected to symplectic geometry mode decomposition （SGMD） to obtain the component sequences. Secondly，K-Mean clustering was adopted to merge the component sequences，and TSO-TCN prediction models were established to predict the clustered component sequences. Finally，the final prediction result was determined by cumulative summation. The experimental results show that the proposed method has good prediction accuracy and provides a new idea for the traditional decomposition-integration prediction method.