Hybrid physical and data driven modeling for dynamic operation characteristic simulation of wind turbine

Abstract

Accurate simulation of wind turbines is the key to achieve efficient control of wind turbines and optimal scheduling of wind power systems. Most of the existing methods are physical simulation models, which have heavy computational burden, especially for large-scale wind turbines, and the simulation accuracy is poor under complex dynamic operating conditions. In response, this paper proposes a hybrid physical and data driven model for dynamic operation characteristic simulation of wind turbines. Firstly, a database of dynamic characteristics of wind turbine operation under various wind conditions is constructed, and then the sequence learning (Seq2Seq) model is used to mine the time series characteristics of wind turbine operation parameters. Combining with long short-term memory (LSTM) network, the mapping relationship between various operating parameter time series to establish. On this basis, an attention mechanism is added to the model to learn the dynamic response characteristics of rotor inertia to natural wind process. Through numerical example analysis, the method proposed in this paper can accurately excavate and learn the time sequence dynamic characteristics of wind turbine operating conditions and parameters under various wind conditions, thereby improving the simulation accuracy and calculation efficiency of wind turbine output power and load dynamic characteristics. Taking the actual wind turbine operation data as an example, the simulation accuracy of wind turbine output power reaches 99.8%. Compared with traditional methods, and the simulation accuracy of tower root load reaches 88.5%.