摘要
针对某地区黑色金属加工及冶炼行业的363家中小企业,先利用探索性因子分析技术将企业的用电大数据提炼为反映其用电规模、电压波动、负荷波动、电网冲击和生产特征五个方面的公共因子。再基于这一单一视角的用电特征数据,采用梯度提升决策树集成学习模型,对这些企业的电费偿付能力进行了风险识别建模和预测。结果表明:探索性因子分析技术是三个测试方案中最佳的数据特征选择和提取技术,且基于该技术的梯度提升决策树分类器模型的查准率、查全率、AUC和Kappa系数都接近理想值。这表明,在无法获知企业多方位信息的现实约束下,仅基于用电特征单一视角数据来识别和预测中小企业的电费偿付能力风险是切实可行且高效的。
As far as 363 medium-sized and small enterprises with respect to the ferrous metal smelting and rolling processing industry are concerned,the exploratory factor analysis is firstly adopted as a data preprocessing skill.Consequently,the enterprises＇ massive electricity data could be summarized into five common factors which individually reflect the scale of power consumption,the voltage fluctuation,the load fluctuation,the impact to power grid and the production characteristics.And then based on these single perspective feature data,gradient boosting decision tree model within the framework of ensemble learning is present to classify and predict those enterprises＇ solvency risk.The results show that exploratory factor analysis technology is the best way of feature selection and extraction among the three tested schemes,and also the model performance assessment indices including the precision,recall,AUC and Kappa coefficients are nearly close to their ideal values.In a sense,it points out a path of evaluating the solvency risk of those medium-sized and small enterprises with lack of information by using a feasible and efficient way.
出处
《技术经济》
CSSCI
北大核心
2018年第2期91-96,共6页
Journal of Technology Economics
基金
国家自然科学基金项目"基于供给使用表和考虑企业异质性的中国投入产出模型及应用研究"(71673269)
关键词
电力客户
中小企业
偿付能力
信用评价
风险识别
售电市场
electricity customer
medium-sized and small enterprise
credit evaluation
risk identification
electricity market
