基于遗传神经网络的天津市公共机构能耗数据分析模型研究

发布时间：2018-06-02 14:03

本文选题：遗传算法 + BP神经网络　；参考：《天津理工大学》2014年硕士论文

【摘要】：随着我国经济社会的快速发展，社会的能耗问题日益突出，公共机构作为社会能源消耗的重要群体，其能耗的分析与预测尤其重要。公共机构能耗影响因素众多，除去便于统计的人数、建筑面积、车辆数目等因素还包括能耗管理制度、用能方式、财政拨款等其他非确定因素给能耗分析带来一定难度。本文依据天津市公共机构能耗统计平台历史数据对天津市公共机构能耗进行了分析和预测，主要工作有以下几个方面。 1）对天津市公共机构概况进行分析，从公共机构数量、组成等方面得出近年来的变化趋势。分析了公共机构能耗种类及支出类型，，得出公共机构能耗的特点：非营利性、缺少控制动因和相对稳定性。从影响公共机构能耗的因素包括：建筑面积、公用车辆数量、用能人数、公共机构类型等方面对天津市2005至2010年公共机构的能耗数据进行了详细描述。公共机构总能耗趋于稳定变化不大，人均能耗逐年降低。 2）确定影响公共机构能耗的影响因素。影响公共机构能耗的因素众多，不仅包括建筑类型、暖通结构、照明系统等硬件设施还包括非确定性因素。本文根据获得的能耗数据，运用灰色关联理论对现有能耗指标进行灰色关联分析，得出影响公共机构电耗的关键指标有：建筑面积、用能人数、编制人数和机构类型。 3)建立基于遗传神经网络的公共机构能耗分析模型。公共机构能耗组成具有高度的非线性特点，而人工神经网络具有很好的非线性、自学习与自适应能力，并且适用于处理多变量系统和很好的容错能力，选取BP神经网络进行能耗的预测。BP神经网络自身的缺陷，初始权值选择的盲目性会导致网络陷入局部最小，而基于遗传学与自然选择的遗传算法拥有全局寻优的能力，选取遗传算法对BP神经网络进行优化。通过遗传算法初始种群的生成，选择、交叉和变异操作确定神经网络的初始权值和阈值，并训练了网络结构，克服了BP神经网络的缺陷。 4）运用MATLAB语言完成能耗预测模型的仿真，选取100组天津市公共机构的能耗统计数据对遗传神经网络进行训练，验证模型的有效性，并将模型与标准BP神经网络进行比较，得出该模型优于标准BP神经网络，并运用模型对5家公共机构的能耗进行了预测。
[Abstract]:With the rapid development of our country's economy and society, the problem of energy consumption is becoming more and more prominent. As an important group of social energy consumption, it is very important to analyze and predict the energy consumption of public institutions. There are many factors affecting energy consumption in public institutions. Besides the factors such as the number of people, the building area, the number of vehicles and so on, such factors as energy consumption management system, energy use mode, financial allocation and other uncertain factors bring some difficulties to the energy consumption analysis. Based on the historical data of Tianjin public institution energy consumption statistical platform, this paper analyzes and forecasts the energy consumption of public institutions in Tianjin. The main work is as follows. 1) the general situation of Tianjin public institutions is analyzed, and the change trend in recent years is obtained from the number and composition of public institutions. The types of energy consumption and the types of expenditure of public institutions are analyzed. The characteristics of energy consumption of public institutions are as follows: non-profit, lack of control motivation and relative stability. The energy consumption data of public institutions in Tianjin from 2005 to 2010 are described in detail from the following factors: the building area, the number of public vehicles, the number of energy users and the types of public institutions. The total energy consumption of public institutions tends to change steadily, and the per capita energy consumption decreases year by year. 2) determine the factors that affect the energy consumption of public institutions. There are many factors that affect the energy consumption of public institutions, including not only the types of buildings, HVAC structures, lighting systems and other hardware facilities, but also non-deterministic factors. According to the energy consumption data obtained, the grey correlation analysis of the existing energy consumption index is carried out by using the grey correlation theory, and the key indexes affecting the power consumption of public institutions are obtained: building area, number of energy users, number of persons compiled and type of mechanism. 3) the energy consumption analysis model of public institutions based on genetic neural network is established. The energy consumption composition of common mechanism is highly nonlinear, while the artificial neural network has good nonlinear, self-learning and adaptive ability, and is suitable for multivariable systems and fault-tolerant. Selecting BP neural network to predict energy consumption. The defects of BP neural network itself, the blindness of initial weight selection will lead to the local minimum of the network, and the genetic algorithm based on genetics and natural selection has the ability of global optimization. Genetic algorithm is selected to optimize BP neural network. The initial weights and thresholds of neural networks are determined by genetic algorithm (GA) initial population generation, selection, crossover and mutation operations, and the network structure is trained to overcome the defects of BP neural networks. 4) using MATLAB language to simulate the energy consumption prediction model, 100 groups of energy consumption statistics of Tianjin public institutions are selected to train the genetic neural network to verify the validity of the model, and the model is compared with the standard BP neural network. It is concluded that the model is superior to the standard BP neural network, and the energy consumption of five public institutions is predicted by using the model.
【学位授予单位】：天津理工大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP183;F206

【参考文献】