基于特征迁移的光伏功率短期预测

Short term prediction of photovoltaic power based on feature transfer

随着国家对可再生能源占比要求的不断提高,新光伏电站的建设需求随之增加。为解决新建光伏电站历史数据不足问题,建立基于特征迁移学习的光伏功率短期预测模型。模型采用日辐照度特征、光伏电池温度和t-SNE算法对气象数据进行特征提取,构建具有泛化能力的高识别度预测模型特征。根据迁移学习理论,将长期运行的光伏电站历史数据用于GRU神经网络预训练,少量本地运行数据对预测模型输出权重进行微调,提升预测精度实现预测模型本地化。

As the country's requirements for the proportion of renewable energy continue to increase,the demand for the construction of new photovoltaic(PV)power plants will increase accordingly.In order to solve the problem of insufficient historical data for newly-built PV power plants,a ST-PVPF model based on feature transfer learning was established.Solar irradiance characteristics,PV cell temperature and t-SNE were used by the model to extract the features of the meteorological datal,and features of high-recognition prediction models with generalization capabilities were constructed.According to transfer learning,the historical data of long-running PV power plants are used for GRUs pre-training,and a small amount of local operating data is used to fine-tune the output weight of the prediction model to improve the prediction accuracy and realize the localization of the prediction model.