基于自适应深度长短时记忆网络的电力负荷预测

Power Load Forecasting Based on Adaptive Deep Long and Short Time Memory Network

电力负荷数据有着明显的时序依赖关系。针对电力负荷的时序依赖特性,提出了一种基于自适应深度长短时记忆网络模型来进行电力负荷的预测。该模型通过深度长短时记忆网络有效的提取负荷序列的时序依赖关系。另外,该模型输入的自适应度量可以解决幅度变化和趋势确定的问题,避免了网络的过拟合。新的混合输出机制可以通过相对误差调整预测结果,使预测结果更加准确。实验结果表明,该模型优于BP神经网络、自回归模型、灰色系统、极限学习机模型和K-近邻模型。自适应深度LSTM网络为电力负荷预测提供了一种新的有效方法。

Power load data have obvious timing dependence. Aiming at the time-dependent characteristics of power load, an adaptive depth long-term and short-term memory network model is proposed to predict power load. The model can extract sequential dependencies of load sequenceseffectively through deep memory networks. In addition, the input adaptive measurement of the model can solve the problem of amplitude change andtrend determination, and avoid over-fitting of the network. The experimental results show that the model is superior to BP neural network, autoregressivemodel, grey system, limit learning machine model and K-nearest neighbor model. Adaptive depth LSTM network provides a new effective method forpower load forecasting.