北京市气温对电力负荷影响的计量经济分析

以温度派生变量度日指数为解释变量构建了气温与电力负荷的计量经济模型。模型证明了天气对电力负荷的季节性影响, 且影响显著。通过引入序列相关AR结构和解释变量的动态结构, 模型得到逐步优化, 调整的拟合优度达95%。为了检验模型的预测能力, 利用历史数据对其进行了评估, 评估结果表明模型有较好的中期电力负荷预测能力。该模型对电力企业电力调度、电力建设有较大的参考价值。

Econometric Analysis on Beijing Temperature Influence on Electricity Load

The variation of electricity load is influenced by the weather, especially the temperature. Relationship model between them is established by using of econometrics method, and its predictive power is assessed by forecasting a monthly and a quarterly load.The mean daily temperature from Beijing Weather Observatory and daily maximum electricity load from Beijing Electricity Company during 2002—2004 are collected to form the model. Because of the temperature's nonlinear affect on load, the heating degree days (HDD) and the cooling degree days (CDD), which are the derivation index of temperature, are used as explanatory variables to establish a concise linear model. Degree days index is defined analogically as the accumulated Celsius degrees between a threshold temperature and the daily mean temperature. The HDD is a good estimation of an accumulated cold during the cold season and the CDD estimates an accumulated warmth during the warm season. The 18 ℃ threshold temperature is chosen in Beijing.The development trend, the different influence on load of different month, different day (holiday and workday), as well as the lag-effect on load of HDD and CDD are fully considered in the model. The errors autoregressive structure is introduced. The test results and actual data have a good fitting degree, R2 is up to 95%, and DW is 2.The CDD has a stronger influence on electricity load than the HDD. If the CDD increases 1 ℃, the electricity load will increase 3%; the HDD increases 1 ℃, the load will only increase 0. 4%. CDD's lag-effect is also stronger than HDD's. The electricity load on holidays, such as Saturdays, Sundays, the May Day holiday and the National Day, is 3%—4% lower than workdays, in the Spring Festival, it is even lower. Assessment of its predictive power shows that it works good for the medium prediction of electricity load, systematic errors of seasonal forecasting is 5. 4% at 99. 7% confidence level.