In the scheduling process of integrated energy systems

soft sensor modeling is of great significance for new energy power prediction. However

data characteristics including multimodality

nonlinearity and time lag limit the performance of traditional soft sensor modeling. In view of this

a wind-solar power soft sensor modeling method based on the rime optimization algorithm-residual network-bidirectional long short-term memory network(RIME-ResNet-BiLSTM)was proposed. Key features were extracted by the maximum mutual information theory

and the K-means clustering algorithm was adopted to divide the dataset into four categories to improve the generalization ability of the model. The Gramian angular field(GAF)dual mapping and bidirectional long short-term memory(BiLSTM)were used to extract spatial and temporal features respectively for multimodal feature fusion. Simulation comparisons were carried out for verification. The root mean square error of photovoltaic power prediction of the proposed model was reduced by 34.61%

43.73% and 14.41% compared with the other three models

while the RMSE of wind power prediction was decreased by 19.97%

20.99% and 10.97% correspondingly. On this basis

an economic scheduling model for integrated energy systems was constructed. A thermal energy storage of concentrating solar power stations was introduced to decouple the 鈥減ower generation determined by heat load鈥?operation mode

and electric energy storage was configured to enhance the regulation capacity of the system. Experimental verifications show that the total operation cost of the proposed scheme is reduced by 26.64% and 12.27% relative to other comparison schemes.