基于长短期记忆网络与注意力机制的短期光伏发电预测

Short-term Photovoltaic Power Generation Prediction Based on Long-short-term Memory Network and Attention Mechanism

准确的光伏发电预测对于电网的安全稳定运行与资源优化调配具有重要作用。提出了一种基于长短期记忆网络与注意力机制的光伏出力预测模型(ATT-LSTM)。首先将光伏功率、光伏组件温度以及环境湿度三个时间序列用作模型输入,接着通过长短期记忆网络进行特征提取,然后通过注意力机制进一步提取关键特征,最后通过Adagrad算法进行对神经网络的权值进行寻优。通过实际的光伏系统历史数据集,验证了所提模型在各种天气条件下的鲁棒性,相比传统预测算法具有更好的预测准确度。

Accurate photovoltaic power generation prediction plays an important role in safe,stable grid operation and optimal allocation of resources.A photovoltaic output prediction model based on the long-short-term memory network and attention mechanism(ATT-LSTM)was proposed in this paper.Firstly,three time series,namely photovoltaic power,photovoltaic module temperature and ambient humidity,were used as model input,and feature extraction was carried out through the long-short-term memory networks.Then,key features were further extracted through the attention mechanism.Finally,the weight of the neural network was optimized by means of Adagrad algorithm.Historical data set collected from an existing photovoltaic system verified the robustness of the proposed model under various weather conditions and indicated that the presented approach had a higher prediction accuracy than the traditional prediction algorithm.