基于差分隐私和集成学习的可再生能源出力预测方法  

Renewable Generation Prediction Method Based on Differential Privacy and Ensemble Learning

作　　者：李洋 王臻懿 谈竹奎 LI Yang;WANG Zhenyi;TAN Zhukui(State Key Laboratory of Internet of Things for Smart City,University of Macao,Macao 999078,China;Electric Power Research Institute of Guizhou Power Grid Co.,Ltd.,Guiyang 550002,Guizhou,China)

机构地区：[1]澳门大学智慧城市物联网国家重点实验室,中国澳门999078 [2]贵州电网有限责任公司电力科学研究院,贵州贵阳550002

出　　处：《电力大数据》2024年第10期1-8,共8页Power Systems and Big Data

基　　金：国家重点研发计划资助项目(2022YFE0205300);澳门科学技术发展基金资助项目(0053/2022/AMJ)。

摘　　要：可再生能源在保障能源供应、改善气候环境及优化国家能源结构方面扮演着至关重要的角色。准确的可再生能源出力预测对于维持电力系统平衡和降低运行成本具有重要意义。然而,由于其高度依赖复杂的气候条件和时空相关性,传统方法难以实现精确预测。尽管机器学习模型被广泛应用于此领域,但数据不足的问题仍然限制了预测的准确性。为解决这一难题,本文提出了一种结合差分隐私与集成学习的新型可再生能源发电量预测方法。首先设计了一个基于差分隐私的数据保护机制来实现数据的聚合,从而解决数据不足的问题;然后构建了一种基于集成学习的可再生能源出力预测框架,利用多个基本模型来共同完成预测,从而提高预测的准确性和鲁棒性。案例研究全面证明了所提出方法的有效性和优越性。Renewable energy plays a crucial role in ensuring energy supply,improving the climate environment,and optimizing the national energy structure.Accurate renewable energy output forecasting is of great significance for maintaining power system balance and reducing operating costs.However,due to its high dependence on complex climate conditions and spatiotemporal correlations,traditional methods are difficult to achieve accurate predictions.Although machine learning models are widely used in this field,the problem of insufficient data still limits the accuracy of predictions.To address this challenge,this paper proposes a novel renewable energy generation prediction method that combines differential privacy and ensemble learning.Firstly,a data protection mechanism based on differential privacy was designed to achieve data aggregation and solve the problem of insufficient data;then,a renewable energy output prediction framework based on ensemble learning was constructed,utilizing multiple basic models to jointly complete the prediction,thereby improving the accuracy and robustness of the prediction.The case study comprehensively demonstrates the effectiveness and superiority of the proposed method.

关 键 词：可再生能源 出力预测 差分隐私 数据保护 集成学习 

分 类 号：TM73[电气工程—电力系统及自动化]