Research And Application Of Energy Consumption Prediction And Optimization Methods For Cloud-Based Smart Buildings

"The fourteenth five-year plan" of building energy efficiency and green buildings by national ministry of housing and urban-rural development put forward that China will complete energy saving renovation of existing buildings with an area of more than 350 million square meters,and new buildings in cities and towns will be fully built for green buildings by 2025.The green and smart building industry has developed rapidly,but there are still many prominent problems in the current building energy consumption control system,such as complex site configuration,confusing management,network delays,low operation efficiency and poor energy-saving effects.The application of cloud computing technology to the field of green smart building will be an effective way to solve the above problems.Therefore,this paper conducts in-depth research on building energy consumption prediction and energy-saving optimization methods of cloud-based smart buildings.The specific innovative work is as follows.Aiming at the problems such as network delay,low operation efficiency and poor energy saving effect of the traditional centralized cloud service system of smart buildings,a novel cloud platform architecture of "cloud+edge" based dual service pool is designed,which not only ensures the global,non-real-time and long-period big data processing and analysis,but also better supports the real-time intelligent decision-making and execution of local businesses.Based on the analytic hierarchy process(AHP)based cloud service quality evaluation method and normal distribution selection method,a dynamic service deployment strategy is proposed,which realizes dynamic coverage of cloud services in the edge service pool,and greatly improves the resource utilization rate of edge side computing power and service response speed.On this basis,a dynamic inertial particle swarm optimization(DI-PSO)based task scheduling method is proposed,which dynamically updates the inertial weight factors according to the particle fitness,solves the imbalance problem of local and global search in task scheduling process,and significantly improves the task allocation efficiency.Experimental results show that the cloud platform with dual service pool architecture designed in this paper and the proposed service deployment and scheduling method can significantly improve the average response time of different cloud services,which verifies the feasibility and effectiveness of the design.Aiming at the problem of feature redundancy caused by building energy consumption prediction by artificial intelligence methods using multiple features,on the edge side,a random network based integrated energy consumption prediction method(EEMD-RVFL+-SVR)is proposed.The method combines the integrated empirical mode decomposition,a privileged information paradigm-based random vector functional link network and support vector regression,and selects features with high relevance to building energy consumption to guide modeling.More abundant information is used to build prediction models,which improves the generalization ability and robustness of the model.On the central side,a parallel time convolutional neural network(PTCN)based building energy prediction method is proposed.This method extracts the prediction features of indoor and outdoor environmental factors respectively for energy consumption prediction,combines them by linear fitting,and optimizes the parameters by least square method,which ensures the rationality and good approximation performance of the PTCN model.Aiming at the contradiction between indoor thermal comfort and high energy consumption during the operation of building air-conditioning systems,and the difficulty to accurately model the internal thermal environment affected by complex factors,a novel optimization strategy of building energy consumption combining the PTCN and the improved chimp optimization algorithm(ICHOA)is proposed.Firstly,the PTCN model is used to predict the air-conditioning energy consumption and the indoor temperature at the corresponding time,and the thermal comfort control is transformed into a cost minimization problem.The objective function of thermal comfort and energy consumption is constructed,and multiple air-conditioning temperature set points in the future day are optimized.Then the ICHOA is proposed to solve the problems that existing intelligent algorithms have many adjustment parameters,tend to fall into local optimal and lack of diversity in the solution process.In this method,individuals with larger Euclidean distance from the optimal chimp are selected for opposition learning to increase the population diversity,and adaptive factors are introduced to assist in optimization during the chimp position updating process to improve the convergence of the algorithm.The optimization experiment results of a building air conditioning system show that the proposed energy consumption optimization strategy can quickly search the optimal set temperature,and the thermal comfort index is kept within the range of[-0.5,0.5]during the test period,and the total energy consumption of air conditioning can be reduced by about 7%,thus saving building energy consumption and improving the thermal comfort.Based on the dual service pool based cloud platform architecture,energy consumption prediction mothods and energy consumption optimization strategies proposed in this paper,a visualization platform for energy consumption optimization of smart buildings is developed,and comparative experiments and application validation of the cloud platform architecture are carried out.Experimental results show that compared with the traditional centralized cloud platform architecture,the average service response time of the cloud platform with dual service pool architecture proposed in this paper can be improved by about 17.3%～37.4%,the average occupancy of computing resources can be reduced by about 5%,the earliest finish time of a single task list can be reduced by 11.3%to 20.9%,and the maximum completion time of 100 task lists can be improved by 30.4%.Since September 2021,the results of this paper have been applied in three typical cases of Longao Jinzuo at Jinan,Yinfeng Wealth Square and Linyi Municipal Party School,improving the energy-saving operation efficiency of smart buildings.Through comprehensive calculation and comparison,the average annual electricity consumption of the three cases is about 358,900 kWh,the comprehensive energy saving rate is about 4.61%,and a total of 16 property management staff are reduced.