为提高光伏发电功率预测的精度和时效性，降低电网调度的安全隐患，提出了一种基于数字孪生模型、联合神经网络以及融合预测模型的光伏发电功率预测技术。该技术以针对常态预测的CNNLSTM网络和针对超短期预测的集成学习融合预测模型为核心，以光伏发电系统的数字孪生模型为基础框架，以某光伏电站的实测数据为基础进行了分析，实现了实时的多模式光伏发电功率精确预测。结果表明：改进的CNNLSTM联合网络模型能够实现较高预测精度，相比于现有的主流预测算法精度提高了约36%~58%；针对超短期的发电功率预测这一难点，集成学习融合框架可以进一步将预测精度提高25%左右。

In order to improve the accuracy and timeliness of photovoltaic power prediction and reduce the security risks of power grid dispatching，this paper proposed a photovoltaic power prediction technology based on digital twin model，joint neural network and fusion prediction model. This technology took CNNLSTM network for normal prediction and integrated learning fusion prediction model for ultrashort term prediction as the core，and took the digital twin model of photovoltaic power generation system as the basic framework to realize the realtime accurate prediction of multimode photovoltaic power generation based on the measured data of a photovoltaic power station. The experimental results show that the improved CNNLSTM joint network model can achieve high prediction accuracy，which is about 36% to 58% higher than the existing mainstream prediction algorithm; aiming at the difficulty of ultrashort term electricity generation power prediction，the integrated learning fusion framework can further improve the prediction accuracy by about 25%.

TP391