The development of new energy represents a significant area of interest and activity at the national and international levels. The advent of new energy prediction technology has the potential to revolutionize real-time scheduling and maintenance, among other processes. In this research, we put forth a hybrid prediction model based on variational mode decomposition (VMD) of the Rime optimization algorithm (RIME) in conjunction with the attention mechanism of the short-term and long-term neural network (LSTM). The historical wind power data of the offshore wind farms in Fujian Province in February 2022 are used as the basis for training and prediction. Subsequently, the Intrinsic Mode Functions (IMFs) components obtained from the decomposition are inputted into the prediction model one by one. Ultimately, the component prediction results are superimposed to yield the final results. On the basis of single-step prediction, multi-step prediction validation is carried out, and the accuracy of the obtained prediction results (MAPE, RMSE) is more accurate compared with that of the benchmark prediction model.