The energy consumption of an enterprise energy management system (EMS) is a complex process with nonlinearity, time-variance, larger delay, greater inertia and other dynamic characteristics, resulting in the failure of a single-item prediction model to achieve satisfactory prediction results. In this paper, a combination prediction method, based on BP neural network, was proposed to predict the energy consumption of an enterprise EMS for improving the prediction accuracy. The energy consumption of enterprise energy management system (EMS) was predicted and analyzed using gray combination models, i.e., GM (1.1) and pGM (1.1), gray Markov chain, and BP neural network prediction model. These single-item models and their prediction processes were constructed and successfully applied to predict the energy consumption of iron and steel enterprises. The data pertaining to energy consumption of these enterprises from January to December 2018 and January to March 2019 were used for predicting the simulation and testing, respectively. The results showed that the prediction results of our approach has an average relative error of 3.327% and 1.298% respectively, which are extremely lower than the existing approaches for improving the prediction accuracy.

企业能源管理系统（EMS）的能耗是一个复杂的过程，具有非线性，时变，较大的延迟，较大的惯性和其他动态特性，导致单项预测模型无法获得令人满意的预测结果。提出了一种基于BP神经网络的组合预测方法来预测企业EMS的能耗，以提高预测精度。利用灰色组合模型GM（1.1）和pGM（1.1），灰色马尔可夫链和BP神经网络预测模型对企业能源管理系统（EMS）的能耗进行了预测和分析。构建了这些单项模型及其预测过程，并将其成功应用于钢铁企业的能耗预测。这些企业2018年1月至12月和2019年1月至2019年3月的能耗数据分别用于模拟和测试预测。结果表明，我们的方法的预测结果的平均相对误差分别为3.327％和1.298％，远低于现有的提高预测精度的方法。