基于簇负荷特性曲线的“聚类-回归”电力大用户短期负荷预测

Short-term Power Load Forecasting for Large Consumers Based on Cluster Load Characteristic Curve and Clustering-regression Model

针对电力大用户的精准负荷预测对于配电网发展规划、调控运行、安全可靠供电具有重要意义,电力负荷预测是泛在电力物联网中实现用户精准用电感知的基础工作。为了在负荷预测模型中引入用户用电特征,提出了簇负荷特性曲线的概念,进而提出了一种基于簇负荷特性曲线的“聚类-回归”电力大用户短期负荷预测方法。首先,对区域内电力大用户用电特征进行聚类分析,用电特征类似的用户聚为一簇,用簇负荷特性曲线表征簇内用户用电特征;其次,回归阶段将簇负荷特性曲线作为区域总负荷的属性因子,训练深度神经网络,进行负荷预测。在TensorFlow深度学习框架下实现了“聚类-回归”负荷预测模型,以我国西南某区域电网实际数据设计实验,验证了模型的准确性;为了达到最佳预测效果,对模型超参数进行了优化;进一步考虑到电力大数据的应用环境,设计了压力测试,验证了模型的有效性。方法可以良好应用于电力大数据环境下的大用户负荷预测。

Accurate power load forecasting for large consumers,which is important to distribution network development planning,dispatching and operation,safety and reliable power supply,is basic in terms of perceiving consumer power usage in the Ubiquitous Power Internet of Things.Therefore,this paper introduced users’electricity consumption characteristics and proposed a large-consumer-oriented short-term power load forecasting method based on cluster load characteristic curve and clustering-regression model.We firstly clustered and analyzed regional large users according to electricity consumption,then obtaining the“cluster load characteristic curves”,which represent the features of various types of users and reveal the properties of a regional total load.Furthermore,we built a“clustering-regression”power load forecasting model under the framework of TensorFlow deep learning and verified its accuracy by experiments designed with the actual data of a regional power grid in southwest China.In order to achieve the best forecasting effect,we optimized the hyperparameters of the model and designed a stress testing for considering the application environment of power big data,thus verifying the effectiveness of the model.The proposed scheme can be applied to large consumer power load forecasting in power big data environment.