CASIA OpenIR  > 毕业生  > 硕士学位论文
基于自适应动态规划的智能空调能源系统自学习最优控制
廖泽华
Subtype硕士
Thesis Advisor魏庆来
2020-05-28
Degree Grantor中国科学院大学
Place of Conferral中国科学院自动化研究所
Degree Name工程硕士
Degree Discipline控制工程
Keyword冰蓄冷空调 空调冷负荷预测 自适应动态规划 执行依赖启发式动态规划 平行自适应动态规划
Abstract

近年来,我国的建筑能耗逐年大幅上升,解决建筑能源问题十分迫切。建筑能耗是指建筑内各种用能系统和设备的运行能耗,在一些大中城市,空调的耗电量甚至已经占到了整个城市用电量的20%以上。使用带有蓄冷设备的空调通过在夜间用电低谷时制冰蓄冷,白天用电高峰时融冰释放冷量,能够实现用电的移峰填谷,起到开发低谷用电、优化资源配置、降低建筑能耗的作用。因此研究智能空调能源系统是解决建筑能耗问题的关键之一,而如何最优化地控制空调能源系统,使得用户在经济上达到最大的效益,已经成为了发展冰蓄冷空调的一个重要问题。同时,目前冰蓄冷空调系统常用的控制策略本身的局限导致用户无法获得最高的经济效益。因此,研究智能空调能源系统的最优控制,从而实现建筑节能与节省用电费用的目标,具有重要的理论意义与实用价值。
本论文以冰蓄冷空调系统为研究对象,引入自适应动态规划方法,在对空调冷负荷进行预测的前提下,研究空调能源系统自学习最优运行策略。论文的主要工作和创新点归纳如下:
1. 建立了空调冷负荷预测模型
准确的空调冷负荷预测是冰蓄冷空调系统优化控制的前提,我们根据实际的气温、日照以及空调冷负荷等数据,采用人工神经网络,通过仿真建立了空调冷负荷预测模型,并得到了较好的冷负荷预测结果,为冰蓄冷空调系统自学习最优控制方法研究提供了负荷数据。
2. 提出了一种基于执行依赖启发式动态规划的冰蓄冷空调系统自学习最优控制方法
冰蓄冷空调系统优化控制问题是复杂非线性系统控制问题,自适应动态规划是解决该类问题的一种有效方法,而执行依赖启发式动态规划是其设计类型之一,该类型不需要构造动态系统模型。因此,我们提出了一种基于执行依赖启发式动态规划的冰蓄冷空调系统自学习最优控制方法。在不需要冰蓄冷空调系统数学模型的情况下,所提出的方法能够根据空调冷负荷需求数据以及实时电价数据实现自学习优化,从而得到最优控制结果。我们通过MATLAB进行仿真验证,结果表明,所提出的执行依赖启发式动态规划方法可以显著地降低冰蓄冷空调系统的运行费用。同时,与目前常用的几种优化控制方法相比,所提出方法具有更高的经济效益。
3. 提出了一种基于平行自适应动态规划的冰蓄冷空调系统自学习最优控制方法
为了进一步研究冰蓄冷空调系统的最优控制问题,我们考虑了控制动作连续的情况,基于双迭代自适应动态规划提出了一种新的平行自适应动态规划方法来解决的冰蓄冷空调的最优控制问题。所提出的平行自适应动态规划方法采用了传统自适应动态规划方法中的动作网络和评价网络。通过粒子群优化算法对两个神经网络的权值进行预训练,以加快算法的收敛速度,并得到平行自适应动态规划方法的初值函数,而不是传统自适应动态规划方法中随机初始化权重矩阵。在平行自适应动态规划方法的实现中,使用粒子群优化算法获得每次迭代的目标控制,避免了传统自适应动态规划方法求解非线性方程的缺点。本文还证明了所提出的自学习最优控制算法的收敛性,保证了控制策略的最优性。实验结果表明,所提出的平行自适应动态规划方法能显著地降低冰蓄冷空调系统的运行费用,与传统自适应动态规划方法相比具有更快的收敛速度。
 

Other Abstract

In recent years, the energy consumption of buildings in our country has increased dramatically, so it is extremely urgent to solve the problem of highly building energy consumption. In some large and medium-sized cities, the power consumption of air conditioning has even accounted for more than 20% of the whole cities’ power consumption. The ice-storage air conditioning (IAC) can store cooling by making ice in the off-peak time and release cooling by melting ice in the peak time to realize peak shifting and valley filling of power consumption. It can help to exploit the use of electricity in the depressed time, optimize the resource configuration, and reduce the building energy consumption. Hence, the study of intelligent air conditioning energy systems is one of the keys to solve the problem of building energy consumption. How to optimize the control so that the user can achieve the maximum economic benefits, has become an important issue. The limitations of the current control strategies for IAC systems make the users unable to obtain the highest economic benefits. Therefore, it is of great theoretical and practical significance to study the optimal control of the IAC systems to achieve the goal of building energy saving and cost saving.
In this thesis, the IAC system is taken as the object of study, and an adaptive dynamic programming (ADP) method is utilized to research the optimal operation strategy of air conditioning energy system on the premise of predicting the cooling load. The main work and innovations of this thesis are summarized as follows:
1. A Prediction Model of Air Conditioning Cooling Load
Accurate prediction of air conditioning cooling load is the premise of optimal control for IAC systems. Based on the data of the temperature, irradiance and cooling load in an actual project, we use artificial neural networks to establish a prediction model of the cooling load, and get good prediction results of cooling load, which provide load data for the study of self-learning optimal control method of IAC systems.
2. A Self-Learning Optimal Control for Ice-Storage Air Conditioning Systems Via Data-Based Action Dependent Heuristic Dynamic Programming
The optimal control problem of IAC system is a complex nonlinear system control problem. ADP is an effective method to solve this kind of problem, and action dependent heuristic dynamic programming (ADHDP) is one of its design types, which does not need to construct a dynamic system model. Therefore, we design an ADHDP algorithm to obtain the optimal control for the IAC systems. The developed algorithm enables the IAC system to realize self-learning by the date of cooling load demand and real-time electricity rate, where the mathematical model of the IAC system is not required. We use MATLAB to carry on the simulation verification. The results show that the ADHDP method can significantly reduce the operating cost of the IAC system, and has higher economic benefits than several commonly used control methods.
3. A Novel Self-Learning Optimal Control Method for Ice-Storage Air Conditioning Systems Via Parallel Adaptive Dynamic Programming
In order to further study the optimal control of ice-storage air conditioning system, we consider the continuous control actions, and propose a parallel self-learning optimal control method based on the dual adaptive dynamic programming. The developed parallel adaptive dynamic programming (PADP) method employs two neural networks of traditional ADP method, which are the action network and the critic network, respectively. The weights of these two networks are pre-trained via particle swarm optimization (PSO) method, in order to speed up the convergence of the PADP algorithm. Then the initial value function of the PADP method will be obtained, instead of randomly initializing the weight matrices in traditional ADP methods. In the implementation of PADP, PSO algorithm is utilized to achieve the target control in each iteration, which avoids solving a nonlinear equation in traditional ADP algorithms. We have proven the convergence property of the PADP algorithm for solving the optimal control problem of the IAC system, which guarantees the optimality of the control strategy. Finally, the effectiveness of the proposed PADP method is illustrated by the numerical results and the comparisons. Compared with the traditional ADP method, the PADP algorithm has faster convergence speed.

Pages96
Language中文
Document Type学位论文
Identifierhttp://ir.ia.ac.cn/handle/173211/39593
Collection毕业生_硕士学位论文
Recommended Citation
GB/T 7714
廖泽华. 基于自适应动态规划的智能空调能源系统自学习最优控制[D]. 中国科学院自动化研究所. 中国科学院大学,2020.
Files in This Item:
File Name/Size DocType Version Access License
廖泽华硕士学位论文.pdf(2277KB)学位论文 暂不开放CC BY-NC-SAApplication Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[廖泽华]'s Articles
Baidu academic
Similar articles in Baidu academic
[廖泽华]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[廖泽华]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.