Ultra short-term wind power prediction based on lightweight learning machine with error compensation

The wind power prediction model has been improved in order to obtain higher prediction accuracy, but this model structure then becoming complicated and the training time is prolonged. Therefore, this paper proposes a Lightweight Learning Machine with Error Compensation (LLM-EC), which consists of two parts: prediction and error compensation. The Lightweight Learning Machine (LLM) accomplishes the prediction part by learning the historical patterns of wind energy and related factors. To improve prediction accuracy, this paper incorporates an Improved Temporal Attention Mechanism (ITAM) into LLM. In the error compensation part, the prediction results of the LLM are re-compensated using the Error Compensation Machine (ECM) to reduce the error accumulation during the rolling prediction process. Finally, a comparison of the benchmark model with LLM-EC in terms of prediction accuracy, training time, and memory usage reveals that LLM-EC has significantly less prediction error; less training time; and less memory occupied by the model.