摘要： 如何对企业生产过程中能源排放进行预测一直是企业管理和政府监督时重点关注的问题，而且随着信息采集能力增强，能源排放预测过程中会涉及越来越多跨界数据，这一情况导致了预测模型面临着数据量庞大和数据关联性较低的挑战，从而增加了模型训练的难度，降低了预测的准确性。针对上述问题，本文提出了基于纵向联邦学习的能源排放智能预测模型，模型分为两部分：(1)针对跨领域联合建模过程中数据源分散、信息密度小的问题提出基于纵向联邦学习的异步网络更新方法，通过该方法保证本地数据的安全和多方建模的质量，异步更新方式还能降低多方建模的时间和空间开销；(2)针对模型间通讯数据的安全高效传递问题提出基于同态加密的数据跨平台通讯算法，利用数据加密保障通讯网络安全的同时使用数据压缩技术减小加密数据的体积，进一步提高了模型间通讯的效率。最后设计实验证明该模型具有良好的性能，相比于基准方法，本文提出的能源排放预测模型的R2值最多提升了16%，并能够降低约40%的联合建模时间，充分证明本能源排放跨界智能分析模型解决了跨界数据难以共享共用的问题，并且提高了跨界联合建模的速度和准确率。

Abstract: Energy emissions prediction in enterprise production processes has always been a focal point for enterprise management and government supervision. However, with the increasing use of cross-border data in the prediction process due to enhanced information collection capabilities, prediction models face challenges like large data volume and low data relevance, leading to difficulties in training and lower prediction accuracy. To address these issues, this paper proposes an intelligent prediction model for energy emissions based on vertical federated learning. The model is divided into two parts: (1) To handle scattered data sources and low information density during cross-domain joint modeling, an asynchronous network update method based on Vertical Federated Learning is proposed to ensure the safety of local data and the quality of multi-party modeling while increasing time and space efficiency. (2) To achieve safe and efficient communication between models, a data cross-platform communication algorithm based on Homomorphic encryption is introduced. This algorithm uses data encryption for secure communication and data compression techniques to enhance communication efficiency. Experimental results demonstrate that the proposed model outperforms traditional methods, with up to a 16% R2 improvement and about 40% reduction in joint modeling time. This cross-border intelligent analysis model effectively solves the sharing and utilization issues of cross-border data and enhances the speed and accuracy of cross-border joint modeling.