基于多特征神经网络的日前光伏功率预测

The day-ahead photovoltaic power prediction based on multi-feature neural networks

工业互联网技术在能源领域飞速发展,准确的光伏发电功率短期预测有利于智能电网的调度管理,可以提高电力系统的运行效率以及经济性和稳定性。为了进一步提高预测方法的准确性,提高电网的运行质量,提出了一种结合深度学习和特征工程的短期光伏功率预测方法,该方法被称为衍生和基础神经网络特征融合(DBDeepFF)。该方法旨在提高多种传统预测模型的准确性,以澳大利亚公开的光伏发电功率历史数据为例进行了验证。结果表明,提出的架构能够使原始数据和衍生的特征在不同的模型上进行训练。该方法与普通预测方法相比,可以实现快速收敛,同时预测结果的平均绝对百分比误差(MAPE)降低4%左右,为区域内光伏电站的稳定运行提供了可行的数据指导。

With the rapid development of industrial Internet technology in the field of energy,accurate short-term prediction of photovoltaic power generation is conducive to the dispatching and management of the smart grid,and can improve the operation efficiency,economy,and stability of power systems.To further improve the accuracy of the prediction method and improve the operation quality of the power grid,a short-term photovoltaic power prediction method combining deep learning and feature engineering was proposed.The architecture is called Derived and Basic Neural Network Feature Fusion(DBDeepFF).This method aims to improve the accuracy of many traditional prediction models.The historical photovoltaic power data published in Australia is used as an example to verify the results.The results show that the proposed architecture enables the original data and derived features to be trained on different models.Compared with the common prediction method,this method can achieve rapid convergence,and the mean absolute percentage error(MAPE)of the prediction results is reduced by about 4 percentage points,which provides feasible data guidance for the stable operation of photovoltaic power stations in the region.