基于数字孪生和深度学习的风力与光伏发电预测方法研究  

Research on Wind and Photovoltaic Power Generation Forecasting Method Based on Digital Twin and Deep Learning

作　　者：齐勇[1] 门泽木 解思源 成润泽 QI Yong;MEN Zemu;XIE Siyuan;CHENG Runze(School of Electronics and Information Engineering,Artificial Intelligence,Shaanxi University of Science and Technology,Xi′an 710021,China)

机构地区：[1]陕西科技大学电子信息与人工智能学院,陕西西安710021

出　　处：《软件工程》2025年第3期57-63,共7页Software Engineering

基　　金：陕西省教育服务厅地方专项计划项目“面向智慧电网的数字孪生物理信息融合平台”(22JC019)。

摘　　要：近年来,随着可再生能源发电逐步替代化石能源成为主流的发电方式,其发电过程中的不稳定性为电力运营带来了诸多挑战。为了应对挑战,文章结合使用了数字孪生技术和深度学习模型,提出了一种新的可再生能源功率预测方法。通过构建基于数字孪生的监控平台,实现了对可再生能源发电系统的实时监控,并构建了深度学习模型ELNet(ElectricNet)预测未来特定时间段的发电量。同时,采用网格搜索法自动优化超参数,有效地减少了人工调参的时间成本。通过4组数据集的测试验证,本研究所提出的模型在均方误差(MSE)评估标准下,相较于其他模型的性能平均提升了25.246百分点,能够更精准地预测发电量,有效降低损失。In recent years,as renewable energy generation gradually replaces fossil fuels as the mainstream power generation method,the instability during the generation process poses numerous challenges for power operation.To address these challenges,this paper combines digital twin technology and deep learning models to propose a new renewable energy power forecasting method.By constructing a digital twin-based monitoring platform,real-time monitoring of renewable energy generation systems is achieved,and a deep learning model,ELNet(ElectricNet),is developed to predict power generation for specific future time periods.Additionally,a grid search method is employed to automatically optimize hyperparameters,effectively reducing the time cost of manual parameter tuning.Testing on four datasets demonstrates that,compared to other models,the proposed model improves performance by an average of 25.246 percentage points under the Mean Squared Error(MSE)evaluation standard,enabling more accurate power generation predictions and effectively reducing losses.

关 键 词：可再生能源发电预测 深度学习 数字孪生技术 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]