基于特征排序与深度学习的母线负荷预测方法

Bus load forecasting based on feature ranking and deep learning

负荷预测是指导电力系统规划和安全经济运行的重要依据。传统的负荷预测一般指区域负荷总量的预测,不能够体现底层母线负荷水平,无法满足电网精益化管理的要求,母线负荷预测是解决这一问题的关键途径。文章提出了基于特征排序与深度学习的母线负荷预测模型。首先,针对各区域母线负荷差异性较大的现状,使用随机森林算法对预测目标影响较大因素进行排序,选择特征贡献度较高的特征属性;其次,在模型训练阶段选择了深度置信网络,学习并跟踪母线负荷变化趋势;最后,采用北京电网某条110 kV母线负荷进行实例验证。结果表明,文章所建立的预测模型具有良好的预测精度和稳定度。

Load forecasting is an important issue for guiding power system planning,scheduling,and economic operations.The traditional load forecasting generally refers to the prediction of the total regional load,which cannot sufficiently reflect the underlying bus load level and cannot meet the requirements of modern grid management.Therefore,bus load forecasting becomes the key way to solve this problem.In this paper,the bus load forecasting model based on feature ranking and deep learning is proposed.At first,considering that different types of bus load possess the distinctive characteristic,the random forest model is used to rank the input features according to the contribution to forecasting target.Then,deep belief network is deployed to track the tendency of bus load curve at the training stage.At last,the case verification is carried out on actual bus load data of Beijing power grid company,the results show that the prediction model established in this paper has good prediction accuracy and stability.