基于不确定性分析的电力工程项目评估建模与仿真  

Modeling and Simulation of Power Engineering Project Evaluation Based on Uncertainty Analysis

作　　者：曹伟杰 戴健 顾博渊 陈超超 CAO Weijie;DAI Jian;GU Boyuan;CHEN Chaochao(Wuxi Power Supply Branch of State Grid Jiangsu Electric Power Co.,Ltd.,Wuxi 214062,China)

机构地区：[1]国网江苏省电力有限公司无锡供电分公司,江苏无锡214062

出　　处：《微型电脑应用》2025年第2期70-73,共4页Microcomputer Applications

基　　金：2022年江苏无锡公司配网项目精准投资评价体系研究(B71030225F53)。

摘　　要：为了提升电力工程数据预测值的可用性,在考虑工程设计阶段各种不确定性的基础上进行针对项目评估的建模与仿真。将电力工程实施前的不确定因素划归为偶然不确定性及认知不确定性,再基于不同不确定性的特点分别建模。引入一种基于贝叶斯理论的深度学习网络,并使用二次采样的方法进行高维度积分求解,进而提升模型的计算效率。在建立最终预测模型时,考虑到电力工程建设的长时间特性,采用长短期记忆(LSTM)单元搭建序列传播网络。仿真结果表明:模型迭代过程中的相关参数均成正态分布,所提模型可以将传统深度学习网络的点值输出转化为预测概率输出;在评估模型的有效性时,基于预测值的均值与方差搭建了预测结果的95%置信区间,而所提模型的预测精度为93.56%,所有预测结果均落在95%的置信区间内。In order to improve the availability of the predicted value of the power engineering data,the modeling and simulation for project evaluation are carried out after considering various uncertainties in the engineering design stage.The proposed method classifies the uncertain factors before the implementation of power engineering into accidental uncertainty and cognitive uncertainty,and then models are established separately based on the characteristics of uncertainties.By introducing a deep learning network which is based on Bayesian theory,and using the method of secondary sampling to solve high-dimensional integral,the calculation efficiency of the model is improved.In building the final prediction model,by considering the long-term characteristics of power engineering construction,the long short-term memory(LSTM)unit is used to build the sequence propagation network.The simulation results show that the relevant parameters of the model are normally distributed in the iterative process,and the proposed model can convert the point value output of the traditional deep learning network into the predictive probability output.When evaluating the effectiveness of the model,a 95%confidence interval of the prediction results is built based on the mean value and variance of the predicted values.The prediction accuracy of the model is 93.56%,and all prediction results fall within the 95%confidence interval.

关 键 词：不确定性 深度学习 贝叶斯理论 造价预测 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论] TN929.5[自动化与计算机技术—计算机科学与技术]