随机配置网络在短时电力负荷曲线预测中的应用

Application of Stochastic Configuration Network in Short-term Power Load Curve Forecasting

传统梯度类神经网络负荷预测模型在面对高维度、大规模负荷数据集时,存在模型构建复杂、训练时间长等问题.为提高负荷曲线预测模型训练的时效性和预测准确性,提出了一种基于随机配置网络的短时电力负荷曲线预测方法.首先针对弱局部负荷波动对预测模型的影响,利用Savitzky-Golay滤波器对负荷时序平滑进行处理,将时序滤波处理后的负荷序列、节假日、气象等数据作为预测模型的输入组成部分.在此基础上,发挥随机配置网络模型的随机增量学习优势,完成负荷曲线预测模型的训练.利用某电厂采集的短时负荷数据及其影响因素数据对模型的预测效果进行验证,仿真结果表明,随机配置网络预测模型相较于深度神经网络模型在模型训练的时间效率方面更具优势,预测的效果基本与深度神经网络模型接近.

The traditional neural network prediction models based on gradient descent algorithm has some problems,such as complex model construction and long training time,when facing high dimension and large scale load data set.In order to improve the timeliness of model training and accuracy of prediction for load curve forecasting model,a short-term power load curve forecasting method based on stochastic configuration network is proposed.Firstly,the Savitzky-Golay filter is used to smooth the load sequence to weaken the adverse effects of local load fluctuations on the forecast model.Then,the load series processed by time series filtering,holidays,weather and other data were used as the input components of the forecasting model.On this basis,the advantage of stochastic incremental learning by the stochastic configuration network model is utilized to complete the training of the load curve prediction model.The short-term load data collected by a power plant and its influencing factors are used to verify the prediction effect of the model.The simulation results show that the prediction model based on stochastic configuration network has more advantages in the time efficiency of model training.At the same time,the overall prediction effect of the model is close to that of the deep neural network model.