基于Copula联合概率密度模型和BP神经网络的风电场选址研究

A Study on Wind Farm Siting Based on Copula Joint Probability Density Models and Backpropagation Neural Networks

摘要: 为提升风能资源评估准确性并为风电场选址提供科学依据，本研究综合联合概率模型与神经网络方法，选取内蒙古自治区通辽、阿尔山、阿巴嘎旗、东胜及鄂托克旗五个气象站点2020~2024年的气象观测数据，依据不同下垫面特征选取对应风切变指数将地面风速与空气密度统一外推至50米轮毂高度后分析各站点风向、风速及空气密度的统计特征，发现风速与空气密度分别服从广义极值分布与混合正态分布且通过Frank Copula函数构建二者联合概率模型以捕捉相依结构，在此基础上构建以风矢量分量、气压、空气密度和海拔为输入的BP神经网络模型并应用于新站点锡林浩特的预测验证，结果显示模型具备良好泛化能力与预测精度，所构建的“概率模型–神经网络预测”融合框架可系统评估区域风能资源，为风电场科学选址提供可靠理论基础与技术支撑。

Abstract: To enhance the accuracy of wind energy resource assessment and provide scientific basis for wind farm siting, this study integrates probabilistic models with neural network methods. Meteorological observation data are from 2020 to 2024 at five stations: Tongliao, Aarshan, Abaga Banner, Dongsheng, and Etuoke Banner in Inner Mongolia Autonomous Region. Based on different land surface characteristics, corresponding wind shear indices were selected to uniformly extrapolate surface wind speed and air density to a hub height of 50 metres. The statistical characteristics of wind direction, wind speed, and air density at each site were then analysed. It was found that wind speed and air density respectively followed a generalized extreme value distribution and a mixed normal distribution. A joint probability model was constructed using the Frank Copula function to capture their interdependent structure. Building upon this, a BP neural network model was developed with wind vector components, barometric pressure, air density, and altitude as inputs. This model was applied to predict and validate data for the new site in Xilinhot. Results demonstrate the model’s robust generalisation capability and predictive accuracy. The developed “probabilistic model-neural network prediction” integrated framework enables systematic assessment of regional wind energy resources, providing a reliable theoretical foundation and technical support for the scientific siting of wind farms.