基于动态递归长短期记忆神经网络的光伏功率预测模型研究

Research on Photovoltaic Power Prediction Model Based on Dynamic Recurrent Long Short-term Memory Neural Network

天气过程相关性对光伏发电功率预测精度产生了较大影响,为此提出一种基于动态递归长短期记忆神经网络的光伏功率预测模型。首先通过动态提取气象因素特征进行模型训练,以捕捉光伏电站在周期性和波动性特征下的过程性气象变化为目标,进行建模;然后由双向递归神经网络和改进长短期递归神经网络构成有效的信息流动和状态更新机制,修正过程性气象因素的影响,输出光伏预测功率;最后采用历史运行数据进行仿真验证。试验结果表明,所提方法相较于传统方法,预测结果的平均绝对误差和均方根误差显著减小,证实了所提方法的精度优势,能较好地满足光伏功率预测的精度需求。

The correlation of weather processes has a significant impact on the accuracy of photovoltaic power generation prediction.Therefore,a photovoltaic power prediction model based on dynamic recursive long short-term memory neural network was proposed.Firstly,model training was conducted by dynamically extracting meteorological factor features to capture the process meteorological changes of photovoltaic power plants under periodic and fluctuating characteristics,and modeling was carried out at same time;then,an effective information flow and state update mechanism composed of a bidirectional recurrent neural network and an improved long short-term recurrent neural network were used to correct the influence of procedural meteorological factors and output photovoltaic prediction power;finally,historical operating data was used for simulation verification.The experimental results show that the proposed method significantly reduces the average absolute error and root mean square error of the prediction results compared to traditional methods,confirming that the accuracy advantage of the proposed method can better meet the accuracy requirements of photovoltaic power prediction.