中文摘要：

降温负荷持续增长已成为中国南方夏季最大负荷屡创新高的重要原因。提出了一种基于信息熵和变精度粗糙集优化的不确定支持向量机方法,用于中长期降温负荷预测。方法通过挖掘数据中的相互关系去除冗余信息,从输入属性变量集中寻找核心变量。该方法利用基于信息熵改进的变精度粗糙集对支持向量机的条件属性进行约简,得到最小决策表,并将该最小决策表中对应的变量作为支持向量机预测模型的输入属性变量,进行年最大降温负荷预测。且随着预测年份的推移,该支持向量机预测模型的输入属性变量亦将随之滚动更新,能够为电网规划与运行人员提供不同预测时期降温负荷预测需重点关注的影响因子。最后,利用广东省实际数据对广东电网＂十二五＂和＂十三五＂年最大降温负荷进行预测,结果表明,所提的预测方法预测效果良好,预测精度稳定,对于中长期预测过程中的各种不确定因素的影响具有较好的鲁棒性,真正实现了中长期降温负荷的动态预测。

英文摘要：

Continuous cooling load growth becomes the most important factor breaking annual maximum load records in South China. More accurate cooling load forecasting will help to improve maximum load forecasting accuracy. An uncertain support vector machine（SVM） method optimized with entropy and variable accuracy roughness set（VARS） is proposed for mid-long term cooling load forecasting. It excavates relationship between data for eliminating redundant information to search key SVM input variables. Based on variable accuracy roughness set and entropy, the method reduces SVM condition attribute set. According to reduction result, SVM makes use of corresponding variable for cooling load forecasting. The input variables are updated as forecasting year changes, making power grid operating and planning staff pay more attention to key factors in different periods. Finally, 2012~2020 annual maximum cooling load of Guangdong province is forecasted. The result shows that the forecasting method is effective and forecasting accuracy is stable. It is also of strong robustness on impacts of various uncertainties. The method can really realize dynamic mid-long term cooling load forecasting.