基于CSO优化深度信念网络的园区能源需求预测方法  

Park Energy Demand Forecasting Based on CSO Optimized Deep Belief Network

作　　者：吴伟杰 吴杰康[2] 雷振 郑敏嘉 张伊宁 李猛 黄欣 李逸欣 WU Weijie;WU Jiekang;LEI Zhen;ZHENG Minjia;ZHANG Yining;LI Meng;HUANG Xin;LI Yixin(Grid Planning&Research Center,Guangdong Power Grid Corporation,Guangzhou 510000,Guangdong Province,China;School of Automation,Guangdong University of Technology,Guangzhou 510006,Guangdong Province,China)

机构地区：[1]广东电网有限责任公司电网规划研究中心,广东省广州市510000 [2]广东工业大学自动化学院,广东省广州市510006

出　　处：《电网技术》2021年第10期3859-3868,共10页Power System Technology

基　　金：广东省基础与应用基础研究基金项目(2020B1515130001);广东省科技计划项目项目(202002030463);广东电网有限责任公司科技计划项目(037700KK 52190004)。

摘　　要：针对目前能源需求预测影响因素繁多、构建模型复杂、预测精度不足的问题,提出了一种改进关联分析和纵横交叉优化深度信念网络的多能互补系统能源需求预测方法。首先,分析了园区多能互补系统冷热电能源需求的影响因素,并采用互信息和误差最小的方法对其进行确定。其次,基于传统灰色关联分析的不足,建立了距离相似度和趋势相似度的综合相似度的相似日选取方法。囿于深度信念网络初始权重的随机化,采用纵横交叉优化深度信念网络对园区冷热电负荷进行预测。以园区为仿真计算实例,分析冷热电负荷变化对能源需求预测的影响,验证了所提预测方法有效地提高了预测精度,具有较高的准确性和实用性。Aiming at the problems of various influencing factors,complex construction model and insufficient prediction accuracy in energy demand forecasting,a energy demand forecasting method for the multi-energy complementary system is proposed,in which the improved correlation analysis and the crisscross optimization algorithm is used to optimize deep belief network.Firstly,the influencing factors of the energy demand of the park's multi-energy complementary system are analyzed,which are then determined with the mutual information and the minimum error methods.Secondly,as for the deficiency of the traditional grey correlation analysis,the the similar day selection method based on the comprehensive similarity of the distance similarity and trend similarity is established.Finally,limited by the randomness of the deep belief initial weight of the network,the crisscross optimization algorithm is used to optimize the deep belief network to forecast the cooling and heating loads of the park.?Taking an actual park as an example,this paper analyzes the influence of cooling,heating and power load changes on the energy demand forecasting,and verifies that the proposed forecasting method can effectively improve the prediction accuracy and has high accuracy and practicability.

关 键 词：园区能源需求预测 综合能源系统 相似日分析 改进灰色关联分析 纵横交叉算法 

分 类 号：TM721[电气工程—电力系统及自动化]