基于迁移学习的电能替代节能量在线估计方法

电能替代成为能源转型发展的重要趋势和关键路径,节能量的快速准确估计有利于电能替代项目的推广。为充分利用少量电能替代项目调试期数据快速估计节能量,提出了一种基于迁移学习的单位节能量在线估计方法。首先利用回归算法对大量基期样本展开训练,获得基期能耗模型;其次,利用基于迁移学习的回归算法对大量基期样本、少量调试期样本展开训练,并通过不同的权重更新策略迭代调整基期样本、调试期样本权重,获得调试期能耗模型;最后,采用归一法获得参考条件下能耗差值,即单位节能量。针对干燥领域的电能替代进行仿真分析,证明了所提方法的有效性,并研究了迭代次数、样本数目和样本组合对所提算法预测误差的影响。

Online Energy-Saving Estimation of Electric Energy Substitution Applying Transfer Learning

Electric energy substitution has become an important trend and key path for the development of energy transition. The rapid and accurate estimation of energy saving is conductive to the promotion of electric energy substitution projects. To make full use of a small amount of the operation data during the electric energy adjustment period to achieve the purpose of rapid energy-saving estimation, this paper proposes an online unit energy-saving estimation method based on transfer learning. In this method, we firstly use regression algorithm to train a large amount of base period samples to obtain a base period energy consumption model. Secondly, we use the regression algorithm based on transfer learning to train a large amount of base period samples and a small amount of adjustment period samples together, and use different weight updating strategies to iteratively adjust the weights of the based period samples and the adjustment period samples to obtain an adjustment period energy consumption model. Finally, the normalization method is used to obtain the energy consumption difference under the reference conditions, that is, the unit energy-saving amount. In this paper, the simulation analysis of electric energy substitution in the drying field proves the effectiveness of the proposed method, and shows the influence of the iteration times, sample number, the way of sample combination on the prediction errors of the proposed algorithm.