基于多参数代价敏感系数学习及数据驱动模型的电力能耗预测

Power consumption prediction based on multi-parameter cost-sensitive coefficient learning and data-driven model

企业节能减排是我国社会发展的前沿问题和研究热点,对企业用户的电力能耗现状进行综合分析与评估是进行节能改造或节能设计的前提和基础。阐述了现阶段常用的电力能耗预测方法,分析了分类与回归树(CART)算法作为弱学习器构建预测模型的缺点,针对原始AdaBoost算法只关注预测误差率最小的不足,在算法实质基础上研究并提出一种基于多参数代价敏感系数学习的改进AdaBoost算法。建立基于改进AdaBoost算法的回归预测模型,通过真实数据进行短期电力能耗预测,验证了改进算法对模型性能的提升。

Energy conservation and emission reduction in enterprises are the frontier issues and research hot-spots in the way of China's development.Comprehensive analysis and evaluation on the current situation of power consumption in enterprises is the premise and foundation for energy-saving transformations or energy conservation designs.Having described the common forecasting methods for electricity consumption,the shortcomings of constructing a prediction model taking Classification and Regression Tree(CART)algorithm as the weak learner are analyzed.To deal with the deficiency of the traditional AdaBoost algorithm focusing on the minimum prediction error rate only,an improved AdaBoost algorithm based on multi-parameter cost-sensitive coefficient learning is studied and proposed based on the essence of the algorithm.A regression prediction model constructed based on the improved AdaBoost algorithm can make short-term power consumption prediction according to real data,which verifies the improvement of the model's performance.