基于物理信息的时间卷积神经网络风电功率预测
2024,32(11):101-108
摘要：由于风力的不确定性和随机性，风电功率预测对电力系统的稳定运行至关重要；为提高风电功率模型的预测精度；对风力发电机的数学模型进行研究后，将物理建模和数据驱动建模相结合，提出一种基于物理信息的时间卷积神经网络模型用于风力发电机的功率预测；采用将风力发电机的转子运动方程嵌入时间卷积神经网络的损失函数，从而提高模型的预测能力，泛化性和物理可解释性；并在Simulink仿真软件中搭建风力发电机物理模型以获取实验数据样本，经同工况实验和外推实验表明，基于物理信息的时间卷积神经网络模型相较于原时间卷积神经网络模型的同工况实验均方根误差下降50.8%,外推实验的均方根误差下降55.2%，显著提高了风力功率预测的准确性。
关键词：风力发电机;功率预测;物理信息;时间卷积神经网络;数据驱动建模
Temporal Convolutional Neural Network for Wind Power Prediction based on Physical Information
Abstract：Due to the uncertainty and randomness of wind power, wind power prediction is very important for the stable operation of power system. To improve the prediction accuracy of wind power model; After studying the mathematical model of wind turbine, combining physical modeling and data-driven modeling, a time convolutional neural network model based on physical information was proposed for the power prediction of wind turbine. The rotor motion equation of the wind turbine is embedded into the loss function of the temporal convolutional neural network, so as to improve the prediction ability, generalization and physical interpretability of the model. The physical model of wind turbine is built in Simulink simulation software to obtain experimental data samples. The same working condition experiment and extrapolation experiment show that compared with the original time convolutional neural network model, the root mean square error of the time convolutional neural network model based on physical information is reduced by 50.8%, and the root mean square error of the extrapolation experiment is reduced by 55.2%. The accuracy of wind power prediction is significantly improved.
Key words：Wind turbine; Power prediction; Physical information; Temporal convolutional neural network; Data-driven modeling
收稿日期：2024-04-24
基金项目：温州市科研项目(ZF2022003);工业控制技术国家重点实验室开放课题(ICT2022B65)；温州市高水平创新团队项目(温委人(2020〕3 号):电气数字化设计技术国家地方联合工程
