基于并行深度信念网络的电力负荷预测

Application of deep learning in power load analysis

针对传统电力负荷预测算法的训练速度慢、预测准确度不高等问题,提出了一种并行的基于深度信念网络的电力负荷预测方法。该方法基于并行计算框架和深度信念网络,对历史电力负荷和天气信息数据进行并行训练并预测负荷值。实验结果表明,该方法预测的电力负荷值与实际值之间的平均误差较低,预测精度高于传统方法,有效减少了算法训练和预测的耗时,可适应大规模电力数据场景下的预测需求。

In view of the traditional power load forecasting algorithm model of slow training speed and the prediction problem of poor effect, a parallel load forecasting method is proposed based on the deep belief network. Based on parallel computing framework and deep belief network, the method is parallel train the history power load and weather information data, and load values is forcasted through the training model. The experimental results show that the average error between the predicted power load value and the actual value is low and the prediction accuracy is higher than the traditional method. It effectively reduces the elapsed time of consuming training and prediction, and can adapt to the prediction demand in largescale power data scenarios.