A Day-Ahead Wind Power Forecasting Method Based on Physical Limitation Characteristics and Wind Speed Trend Clustering

Wind power is characterized by significant nonlinearity and uncertainty, which is jointly affected by wind speed fluctuation and wind turbine operation mechanism, bringing challenges to high-precision prediction. In this paper, we propose a short-term prediction method that integrates the constrained characteristics of wind power and wind speed trend clustering. Firstly, a dual-objective function based on location consistency and shape consistency is constructed, and the Nondominated Sorting Genetic Algorithm II (NSGA-II) multiobjective optimization algorithm is introduced to automatically classify the low, medium and high speed intervals of the wind speed to reflect the operational characteristics of wind farms. Subsequently, fuzzy C-mean clustering (FCM) is further employed within the medium-speed zone to identify typical wind speed trend patterns based on cosine similarity. The Long Short-Term Memory (LSTM) network model is trained separately in each wind speed sub-interval to realize multi-model partition modeling. Finally, the most matching model is automatically selected for prediction in combination with the test wind speed sequence. The experimental results show that the method in this paper has higher fitting ability in the wind power change drastic or restricted intervals, which is significantly better than the traditional global modeling and the prediction method based on trend division only.