城镇供热系统层级热量结算点中短期热负荷预测方法研究

发布时间：2018-01-22 15:06

本文关键词： 热负荷预测层级热量结算点相关性分析组合预测信息熵 Adaboost　出处：《太原理工大学》2017年博士论文　论文类型：学位论文

【摘要】：供热是关系国计民生的重要基础行业和公用事业。随着供热的商品化,供热系统热负荷预测作为供热规划、生产、调度和交易等工作的基础,在供热系统安全和经济运行中起着至关重要的作用。供热负荷预测精度的高低直接影响到供热系统的供热质量、安全性和经济性。随着供热系统节能减排进程的不断推进和智慧供热的需求,使供热负荷预测越来越成为该领域研究的前沿和热点问题,其研究对节能减排、治理雾霾具有重要的意义。由于滞后性、管网的热损失、用户种类复杂程度的差异性等导致了城镇供热系统不同热量结算点热负荷具有不同的规律特点,本文将供热负荷按层级热量结算点来划分,分析各自的影响因素,引入智能算法以及相关组合理论预测技术,以城镇供热系统实测数据为基础,对各层级供热负荷预测的理论与方法进行深入研究,为供热系统运行管理提供较为科学的决策依据。主要研究工作和创新成果如下:(1)对不同层级热量结算点热负荷的特点、影响热负荷的因素及导致热负荷预测误差的相关因素进行了分析;针对历史数据样本的离群数据进行纵向和横向预处理,使其能够更加与热负荷实际运行趋势一致,进而为后期利用这些历史数据样本进行各级热量结算点短期热负荷预测奠定了基础;将相关性分析应用在各层级热量结算点热负荷预测模型输入维数的选择上,使得输入变量与各层级预测热负荷相关性更强,为提高预测结果的准确性和改善预测性能做好进一步的准备;此外,对进入模型的各参数进行归一化处理,避免进入模型的各参数因数值差异大而导致预测性能下降。(2)基于结构风险最小化原则,提出粒子群(pso)优化支持向量机(svm)模型热源热负荷预测方法。该方法对解决系统大热惯性、大时滞性导致热源热负荷随室外温度变化的非线性问题有较好的效果。建立了遗传(ga)优化支持向量机(svm)、标准支持向量机(svm)及粒子群(pso)优化支持向量机(svm)热源热负荷三种预测模型,通过相关性分析并确定预测模型输入变量的维数,证明了粒子群(pso)优化支持向量机(svm)模型在预测精度和泛化能力方面均优于其他两种预测模型。(3)针对一般热交换站用户类型较单一、样本容量大的问题,提出基于adaboost组合多个弱预测器构建出一个强预测器的热交换站热负荷预测方法。弱预测器采用处理大样本、容错能力强的bp神经网络模型,其网络阈值和权值的优化选用经过筛选出的粒子群算法(pso)。利用adaboost理论对9个粒子群(pso)优化bp神经网络预测进行组合构建出一个强预测模型。针对热交换站热负荷及其相关参数历史数据样本进行相关性分析,筛选出与热负荷最相关的影响因素作为预测模型的输入变量维数,最后通过与粒子群(pso)优化bp神经网络方法和未经优化的传统bp神经网络方法进行实验比较,证明本文提出的预测模型有效提高了热交换站热负荷的预测精度和泛化能力。(4)针对建筑热负荷样本数量少及热计量引起的用户调节规律不确定问题,提出两种组合预测方法。将解决小样本非线性问题的粒子群(PSO)优化支持向量机(SVM)模型和容错能力强的粒子群(PSO)优化BP神经网络模型作为组合方法中单一预测模型,基于信息熵理论提取单一预测模型中的有用信息,将提取的有用信息进行融合产生出更强预测能力的组合方法;在基于Adaboost组合粒子群(PSO)优化BP神经网络模型思想上,激发了将处理小样本和非线性问题的支持向量机(SVM)模型作为弱预测器,结合Adaboost理论构建出由8个弱预测器组成的建筑热负荷强预测模型。对建筑热负荷及其相关参数进行相关性分析,找出适合各自预测模型的输入变量,通过实例验证,以上两种组合预测方法均较单一预测模型有较高的预测精度,其中基于信息熵权组合方法更胜一筹,能更好地对住宅建筑热负荷进行预测。
[Abstract]:Heating is an important basic industry and public utilities. The relationship beneficial to the people's livelihood with heating commercialization, heating load forecasting as heating planning, production scheduling, and trading based work, plays an important role in the safe and economical operation of heating system. For heat load forecasting accuracy directly affects the quality of heat supply system, safety and economy. With the heating system energy saving and emission reduction in the progress of the wisdom and the demand for heating, the heating load forecasting has increasingly become the frontier and hot issues in the research field, the research on energy saving and emission reduction, plays an important role in the governance haze. Due to the lag, the heat loss of the pipe network, the user types the complexity of the differences lead to different urban heating system heat clearing heat load has different characteristic points, the heating load according to the level of heat According to settlement, analysis of the impact of each factor, the intelligent algorithms and the related theory of combination prediction technology to urban heating system is based on the measured data, in-depth study of the theory and method of the level of the heating load forecasting, provide scientific decision-making basis for the operation and management of heating system. The main research work and innovation are as follows: (1) the characteristics of heat load settlement of different levels of heat, analyzed the related factors of heat load forecast error cause and influence factors of heat load; according to the outlier data historical data samples for vertical and horizontal pretreatment, to make it more consistent with the heat load of the actual operation, and for later use these history the sample data has laid the foundation for short-term load forecasting at all levels of thermal heat settlement; the application of correlation analysis in all levels of heat clearing point pre heat load Measuring the input dimension of model selection, the input variables and the level of prediction of heat load correlation is stronger, make further preparations for accuracy and improve the prediction performance to improve the prediction results; in addition, the parameters into the model are normalized to avoid the parameters into the model for numerical difference caused the prediction performance decline. (2) based on structural risk minimization principle of particle swarm optimization (PSO) support vector machine (SVM) prediction model of heat source heat load. The method to solve the system thermal inertia, large time delay in heat load nonlinear problems with the outdoor temperature change has a better effect. To establish a genetic optimization support (GA) support vector machine (SVM), standard support vector machine (SVM) and particle swarm optimization (PSO) support vector machine (SVM) and heat load of three kinds of prediction model, through the correlation analysis and determine the model input Variable dimension (PSO) proved that particle swarm optimization support vector machine (SVM) model is superior in the prediction accuracy and generalization ability of the other two kinds of forecasting model. (3) for the heat exchange station user type is relatively single, large sample size problems, put forward a combination of AdaBoost to construct a weak predictor a strong predictor of heat exchange station heat load forecasting method based on weak predictor. Using large sample processing, BP neural network model for fault tolerant ability, optimize the network weights and threshold selection after selected particle swarm algorithm (PSO). By using the AdaBoost theory of 9 Particle Swarm (PSO) to build a prediction a strong prediction model of BP neural network optimized by correlation analysis. According to the heat exchange station heat load and parameters related to historical data sample, selected factors related to thermal load as input variables of predictive model Finally, dimension, and particle swarm optimization (PSO) method of BP neural network and the traditional BP neural network method to optimize the experimental results show that the model proposed in this paper effectively improved the station heat load forecasting accuracy and generalization ability of heat exchange. (4) for the number of building heat load and heat metering caused by small sample the user control law uncertain problems, put forward two kinds of forecast methods. Combined particle swarm will solve small sample nonlinear problems (PSO) optimization of support vector machine (SVM) particle swarm models and fault tolerant ability (PSO) to optimize the BP neural network model as a single method in combination forecasting model, information entropy theory to extract useful the single information in the prediction model based on the useful information extracted from fusion to produce more combination prediction method based on particle swarm combination; Adaboost (PSO) on the idea of optimization BP neural network model, Stimulate the support vector machine will deal with small samples and nonlinear problem (SVM) model as a weak predictor is constructed, consisting of 8 weak predictor building heat load forecast model combined with the Adaboost theory. The correlation analysis of building heat load and its related parameters, to find suitable for their respective input variables in the model, through the examples, more than two kinds of combination prediction methods were compared with the single prediction model has higher information entropy, the combination method based on edge, can predict residential building heat load better.

【学位授予单位】：太原理工大学
【学位级别】：博士
【学位授予年份】：2017
【分类号】：TU995

【相似文献】