Exploration Of Intelligent Optimization Method Of Building Energy System Based On Information Utilization

In the operation phase of the building,how to use the limited energy storage equipment to achieve the enhancement of renewable energy consumption is one of the key issues to focus on.However,in the building energy system integrating renewable energy,the uncertainty caused by the volatility of renewable energy supply and the randomness of the load restricts the further enhancement of renewable energy consumption.While the uncertainty can be eliminated through the use of information which making the transfer,transformation and utilization of energy more orderly.To this end,based on thermodynamic theory and information theory,this paper explores the evolution path from the uncertainty of the micro-state to the macro-state,and analyzes the relationship between information and energy of the macro system.The intelligent optimization method of the building energy system based on information utilization is studied around the influence of information utilization on the optimization of the operation phase of the building energy system.First,by analyzing the energy-information relationship of microprocesses in the Carnot cycle,the heat engine work process that includes information utilization is clarified.And with the help of information entropy theory,this paper further elucidate the intrinsic mechanism by which the information utilization means represented by data-driven model can realize the improvement of energy system’s ability to do work.The information utilized in this process is the key that drives the process,and the system does not increase the output work out of thin air.Information erasure still requires external energy input to the memory,which does not violate the second law of thermodynamics.Aiming to enhance the use of information,a control strategy screening method that considers the robustness of the information interaction between supply and demand sides is investigated.The uncertainty of bilateral information interaction due to the stochastic and fluctuating nature of both supply and demand causes the renewable energy consumption and energy cost of the building energy system in the operation phase to fail to meet the expectations in the planning phase.In this paper,the deviation of mutual information(DMI)is defined as a parameter to characterize the deviation of information interaction,so that the robustness of different control strategies to the deviation of information interaction can be quantified.Moreover,out of 100 simulation sets with added errors,the strategy with direct control of the room temperature exhibited10% greater acceptable simulation sets than the other strategies.Aiming to improve the accuracy of observation information,a data-driven model construction method considering information utilization enhancement is studied,and a parameter screening method with information relevance as a quantitative index is proposed.The data-driven models of room temperature,room humidity,room power consumption,demand-side return water temperature,ground source heat pump water supply temperature and ground source heat pump power consumption are also constructed based on the actual data acquired from the established building energy system data collection platform using the JANET algorithm that can model time-series data.The mean absolute percentage errors of the six data-driven models constructed were 0.08%,0.53%,0.16%,2.85%,3.94%,and 9.04%,respectively.The multi-agent collaboration-based deep reinforcement learning control method is constructed based on the above research,using D3 QN as a single-agent control algorithm and VDN as a multi-agent collaboration algorithm,to collaboratively optimize the energy cost,renewable energy consumption and indoor comfort of the system.With the control strategy and data-driven model obtained from the above study,the system’s energy cost,renewable energy consumption,and indoor comfort are collaboratively optimized using a multi-agent collaboration-based deep reinforcement learning algorithm.In this paper,D3QN(Dueling Double-Deep Q Network)was used to control a single agent and VDN(Value-Decomposition Networks)was used to control multiple agents to collaborate.And incorporating prioritized experience replay and feasible action screening mechanisms with the controlled characteristics of the energy system.Compared with the baseline control model,the multi-agent collaboration algorithm reduces the uncomfortable hours by 84%,the unconsumed amount of renewable energy by 43%,and the energy cost by 8%.The demand response process under electric heating mode is further analyzed for a district building energy system containing building clusters,with the objective of renewable energy consumption and economic optimization.And the impact of disorderly electricity demand on the safe operation of the grid is effectively reduced by actively regulating the electricity price.A quantitative demand response model of users is constructed using Weber-Fechner law and k-means clustering algorithm.The deep Q network was used to build a dynamic subsidy price generation framework for load aggregators.Through simulation analysis based on the evolutionary game model of a project in a rural area in Tianjin,China,the following conclusions were drawn:compared with the benchmark model,regenerative electric heating users can save up to8.7% of costs,power grid companies can save 56.6% of their investment,and wind power plants can increase wind power consumption by 17.6%.