Research On Case Data Driven Evaluation Method Of Air-conditioning System In Buildings

The centralized air-conditioning system is one of the most important energy consumers in public and commercial buildings,and it is also an important guarantee for providing a comfortable indoor environment for the building occupants.A large number of previous cases show that a good and proper HVAC(heating,ventilation and air-conditioning)system design has a positive impact on the performance of a building and vice versa.However,in an actual building construction project,the decision-making of design option selection is not a simple step,which always needs multiple rounds of analyses and evaluation so as to meet the requirements of all parties in the project.During this process,the decision-makers should consider a series of factors,such as energy efficiency,capital cost,construction conditions,indoor environmental quality,impact on the ambient environment,etc.The core problem to be solved can be summarized as: “selecting the most appropriate air conditioning system scheme in the architectural design stage".From the perspective of mathematical principle,it is a typical multi-factor decision-making process with high complexity,due to the large number of the influencing factors.The goal of this thesis is to solve such a complex multi-factor decision-making problem by proposing a new idea and methodology: data driven evaluation method based on actual cases of air-conditioning systems.The multi-factor decision-making process is essentially a mathematical model with single factor score as input and total evaluation result as output.Therefore,we refine the core work of realizing the decision-making process into the task of “determining the most appropriate values of the model parameters”,in which the model parameters refer to the weights of all evaluation indices.The training process of these model parameters can be solved by a data-driven method.Taking a large number of real design cases of the air-conditioning system in commercial buildings as the information source,combined with the developed estimation model for cooling/heating load and energy consumption,we can quickly extract various case data,including building thermal properties,airconditioning system performance,and design scheme information.On this basis,two types of comprehensive evaluation principles are used as the theoretical framework and the global optimization method is used as the training algorithm of the model.Finally,a set of recommended weight values of all evaluation indices are obtained.This case data-driven evaluation method can extract and quantify the decision-making influencing factors contained in the information source to the greatest extent.Meanwhile,it also realizes the function that the weight values can be updated dynamicly with the accumulation of data in the information source,which provides the possibility for further optimization of the evaluation results.Specifically,this paper includes the following parts:First it introduced synoptically the research method and idea adopted in this thesis,and focused on the following three points in detail: the application scenario of the comprehensive evaluation method,the index selection and classification,and the information source.The application scenario is that: with the inputs of the objective conditions of the actual project and the objects of the optional design schemes suitable for the current inputs,the weight factors obtained by the comprehensive evaluation method were used to judge the rating score of each single factor,so as to realize the design scheme recommendation and auxiliary decision-making in the actual project.Then,we defined the evaluation indices that contained in this thesis,and classified these indices into two categories: difference-weighting indices and function-weighting indices,according to the the current academic classification of mathematical principles related to decision-making and evaluation.As the information source that must be relied on to adopt the data-driven method to solve the decision-making problem,we established a standardized database of actual building projects as the data basis for next-step study.The information related to the HVAC systems in the case buildings is stored in this database.To solve the problem of “prediction of the cooling/heating load and the energy consumption of HVAC systems in planning or design stage”,we comprehensively considered the convenience of the method and the accuracy of calculation,and proposed a new fast estimation model of the loads and the energy consumptions of HVAC systems according to the practical scenario in the thesis.In this model,the idea of “splitting the calculation of cooling/heating loads and energy consumptions,and completing it step by step" is adopted.For load calculation,we combined forward modeling and ML(Machine Learning)algorithms to realize the rapid prediction of the annual cooling/heating loads by generating a pre-simulated database of prototypical building models with different functions(e.g.,office,hotel,shopping mall etc.)in cities located in different climate zones and KNN(nearest neighbor search)algorithm;In terms of energy consumption calculation,we determined all possible forms of airconditioning systems and operation scenarios of the objective buildings according to the full arrangement(a total of 300),and then used the pre-simulating method to obtain the annual average energy efficiency ratio of the system under each scenario.Through this parameter and the predicted values of loads,the energy consumptions of HVAC systems could be estimated.Besides,we also construted several scenarios of input parameters to test and verify the algorithm of this approach,all with less than ±5% deviation from the simulation results of detailed BPS models.Finally,we integrated this approach into the evaluation method as a functional module to fast output the physical quantities to calculate the values of each single index.In the main part we expounded the principles and the establishment process of the case data-driven comprehensive evaluation method proposed in this paper.At the mechanism level of multi-factor decision-making problem,we also adopted a new way: two main kinds of comprehensive evaluation method(based on difference driven/ function driven)were combined to constitute the underlying logic,and then to realize the structured processing of two categories of indices.Among them,the difference weighting method is RST(Rough Set Theory),and we determined the weights of all computable indices through attribute reduction algorithm and importance calculation.For the function weighting indices,AHP(Analytic Hierarchy Process)was adopted as its evaluation framework,and we used the judgment matrix to represent the empirical information from experts contained in the decision results.Based on this framework,the process of comprehensive evaluation could be abstracted into a data-driven model by introducing the idea of ML algorithm,with which we took the acquired information source — the actual building project database as the basis and the large number of decision results in this database as the training data,and adopted genetic algorithm(GA)with a new specific coding approach for genes to search the evaluation criteria that are most consistent with the decision results corresponding to the training data set.The final results are: the recommended values of the weight factors of all indices.In addition,the values of all weight factors can also be dynamically updated due to the “Data-Driven”characteristics of this new method,which means that the evaluation results will further tend to be reasonable and accurate with the continuous accumulation of training data.Based on the theoretical result,we developed a GUI(graphical user interface)software tool with the current results as the computation engine.Then,we utilized parts of the data in the actual project database again as the testing set,and applied this tool in these scenarios in batches.By comparing the evaluation results with the actual decisions,it shows that these two kinds of evauation/decision results are generally consistent.We used Deviation Values(DV)to express the degree of this consistency.For training set,the average DV reached at 1.43;and for testing set,the average DV was 1.403.These two DVs were approximately equal.Besides,the average DV of 1.4means that the normalized value of the total score is larger than +1.43σ(the quantile of +1.43σ in normal distribution is 92.4%),which matches the description — “the actually selected design scheme of the air-conditioning system also gets a high total score in the comprehensive evaluation method).Threfore,we can consider that the values of all weight factors are basically reasonable.So far,we have achieved the following core target: a general solution of“intelligent selection of HVAC designs",which can establish the corresponding comprehensive evaluation system based on quantities of decision-making related information sources,and make relatively reasonable decisions.Obviously,we have not tried to mechanically and strictly prove the universality of the established evaluation system,but focused on the “Methodology”.In fact,due to the “case datadriven” characteristics of the method,the final evaluation system will also have the characteristics of “dynamic correction” and “self evolution”.It means that the research results of this paper can be applied in a variety of practical scenarios,including:(1)to improve the efficiency of communication and discussion among all parties in construction projects as a quick quantitative evaluation tool for decisionmaking support;(2)to evaluate the retrofitting design scheme of existing buildings based on the actual performance of the air-conditioning systems in operation;(3)to automatically select the most approporiate HVAC system at the beginning of the design process,as one of the key steps of "automatic design of HVAC system".