Spatial Parameters Optimization Research Based On BP Neural Network And Multi-objective Optimization Algorithm

With the continuous development of building optimization field,an increasing number of optimization methods have sprung up like mushrooms after rain,including various intelligent optimization algorithms.Such intelligent optimization algorithms usually rely on traditional building performance simulation methods to obtain building performance indicators during the optimization process.However,Intelligent optimization algorithms usually require large-scale calculations,and the time of building performance simulation is often limited by the complexity of the building model and the configuration of the computer,which leads to a great difficulty for the designers to abtain the feedback with much efficiency and accuracy in engineer because of the long-time optimization,the building performance optimization method based on the intelligent optimization algorithm mainly exists in scientific research instead of the practical project.Therefore,based on BP neural network and SPEA-II multi-objective optimization algorithm,this paper builds an accurate and efficient building performance prediction and optimization platform to assist designers in decision-making.Firstly,this paper establishes a mathematical model for building performance prediction and optimization with spatial parameters as optimization variables.Taking the small and medium-sized high-speed railway stations in cold areas as an example,this paper studies the modeling method of the small and medium-sized high-speed railway stations,and summarizes the decision variables and objective functions of the prediction and optimization model,as well as the building performance simulation method and data processing method appropriately.In the end,taking building space parameters as decision variables and building energy consumption and indoor thermal comfort as objective functions,a building performance prediction and optimization platform is built on the parametric design platform.Secondly,the training process and prediction results of building performance prediction model are analyzed and verified.In this paper,the process of BP neural network model selection is analyzed in detail,and the appropriate BP neural network topology is finally determined.then,the performance of BP neural network is quantitatively analyzed and verified,including the accuracy and stability of model prediction.The final conclusion shows that the performance prediction model based on BP neural network has better performance which meets the engineering requirements.Finally,this paper combines BP neural network prediction model and SPEA-Ⅱ algorithm model to make quantitative analysis and qualitative analysis of the optimization results of the case and presents them in the form of visualization.Quantitative analysis includes evolutionary process,convergence process and comprehensive quality evaluation of solution set;qualitative analysis includes Pareto frontier and optimal scheme analysis.The final conclusion shows that the prediction and optimization results of the platform are accurate and reliable,the optimization scheme is reasonable,and the engineering application value is high.The innovations of this paper are mainly in the following two aspects:1、 With the direction of building performance design and the premise of efficient and accurate feedback,this paper builds the parametric platform which provides designers with high credibility design guidance in the initial stage of the design.2 、 Based on the advantages of parametric design,this paper models the spatialparameters which is one of the most important parameters in the initial stage of the design with the support of the algorithm.this paper also verifies the optimization feasibility of building air conditioning load and buildingindoor thermal comfort which provides a certain theoretical guidance for passive building optimization.