中文摘要：

在深入研究预警指标的充要条件、必要条件和充分条件的基础上，提出基于指标时差分析和指标组合筛选的预警指标2步筛选方法，建立指标时差分析的数学模型，给出有效的求解方法。针对电力需求预警指标组合的时差特性，提出了级联预警与预警概率的新理念，给出了预警概率的估计方法。鉴于行业电量之间隐含着一定的电能消耗链式关系，将预警指标筛选范围锁定为行业电量。基于2001-2004年的全社会用电量和行业电量月度数据，从27个细分行业中筛选出全社会用电量增长率的预警指标组合，计算了相应的预警概率，并用2005年的数据验证了预警效果，充分说明所提出的预警指标筛选方法的有效性。

英文摘要：

Based on the necessary and/or sufficient conditions of warning indicators,their screening method is proposed,which includes two steps ：index time difference analysis and indicator combination screening. The mathematical model of index time difference analysis is established with effective solution. According to the time difference characteristics of indicator combination,new concept of cascaded warning and warning probability is put forward. The estimation method of warning probability is also given. As there is a connotative power consumption chain among different sectors,the scope of indicator screening is restricted to power consumption of sectors. Based on the monthly power consumptions of entire society and sectors from 2001 to 2004,the warning indicator combination for the growth rate of entire society power consumption is screened from 27 sectors and its warning probability is calculated. The data of 2005 is used to validate the effect of this indicator combination ,which shows the effectiveness of proposed screening method.