| Energy efficiency level of the cold source system has a great impact on the operation of the central air conditioning system and energy saving of public buildings.Based on the above background,Chinese government proposed the goal of building an efficient cold source device room in 2019,the optimization of control strategy for existing large-scale public building’ s cold source systems is an economic and effective way to achieve this goal.The rapid development of the Internet of Things technology has accumulated huge operational data for public buildings,establish the data-driven optimization control strategy of cold source system can fully use the data and promote building intelligence.Reinforcement learning can be used in intelligent control systems.It gets the best strategy through agent’ s continuous attempts in the environment,which is a data-driven control method.The research on the energy-saving optimization control strategy of building cold source system based on reinforcement learning can make full use of the large amount of operation data accumulated in the existing cold source management system and follow the trend of intelligent control.In this paper,the central air-conditioning cold source system of office building in the hot summer and warm winter area was chosen as the research object and a simulation method for optimal control strategy of cold source system based on reinforcement learning was proposed.The main research work is as follows:(1)The main theory and algorithm of reinforcement learning was introduced.Based on the study of reinforcement learning theory and method,the operation process of the cold source system was abstracted into an MDP model and the cold source system MDP model was proposed;Based on the research of reinforcement learning system,determined the form of each element in the system,an cold source reinforcement learning system was proposed and the operation process of the system was determined.(2)For the cold source system of office building is usually difficult to use for practical training of controller,the problem of system environment modeling is studied,in order to improve the prediction accuracy of system environment model,the black box prediction model of indoor temperature,indoor relative humidity and energy consumption of cold source system based on model stacking theory was proposed.Xgboost,RF and SVR was selected as the base model and Ridge Regression was selected as the meta model,by compared the prediction results of different based model combinations,the structure of three prediction models was determined and the system environment simulation platform was established.(3)For the problems such as lack of self-learning ability and dependence on modeling accuracy of conventional control strategies,the cold source energy-saving optimization control strategy based on the Deep Deterministic Policy Gradient was proposed.Firstly,the main super parameters that affect the algorithm were analyzed,the heuristic search method was used to optimize the hyper parameters and the value range of the main hyper parameters was determined;Secondly,based on the policy guidance of controller,the simulation research of the energy-saving optimal control strategy was completed and the simulation results were analyzed;Thirdly,the comparation of simulation results of PSO control strategy and rule based control strategy showed that the total energy consumption of the cold source system decreased by 6.47% and 14.42%,the average indoor thermal comfort increased by 5.59% and 18.71%,and the proportion of non-comfort time decreased by 5.22% and 76.70% under the reinforcement learning control strategy.(4)For the engineering application of control strategy optimization method,the "Intelligent Control Platform for Office Building Cold Source System" was developed,which can realize the functions of system monitoring,system simulation,policy optimization and policy operation. |
PDF Full Text Request