Application of machine learning-based BIM in green public building design

Public activities are mostly carried out in large public buildings, which are closely related to social management. At present, people’s demand for public building facilities is increasing, its shape evolution is becoming more complex, and the scientific and technological content of construction-related technology is also increasing. The development trend of green public buildings is more and more strong. The traditional building design cannot effectively deal with the energy consumption of public buildings and people’s demand for their performance. This paper introduces BIM and machine learning technology to study their practical application in the design of green public buildings and tests the perfect machine learning algorithm. According to the experimental test results, the building energy consumption decreased by 14.3%, the carbon emission decreased by 11.39%, and the absolute value of PMV thermal comfort decreased by 34.7%, which obviously achieved the optimization effect. BIM technology parametric design can enable the design model formed by conceptual design research to automatically draw construction drawings, detailed drawings and other drawings according to the drawing requirements and standards, thus saving the designer's time and enabling him to transfer the drawing time to the program design. Finally, through experiments, the economy, rationality and operability of using BIM technology to design green public buildings are confirmed. In this paper, machine learning and BIM technology are introduced, so as to carry out design research for green public building design.


Introduction
Nowadays, the construction industry has made obvious achievements both in terms of construction technology and the total value of construction. As a place for people's daily activities and work, public building facilities are increasing (Bennett and Iossa 2006). Public buildings can bring great convenience to people's daily life, thus promoting economic development, but there will also be problems in the practical application process. For example, the energy consumption of public buildings is high and the functionality of some buildings is poor (Baek 2021). Although the total area of public buildings accounts for only 5% compared with that of civil buildings, the energy consumption of this part is as high as 10% of the total energy consumption, and the annual power consumption per square meter is even as high as 100-300 kwh (Petri et al. 2014). Although there are many problems in the development of public buildings, there are also many development opportunities in the current era. The biggest opportunity is the national level to introduce a large number of policies to support the development of green public buildings, in order to solve the problem of energy consumption (Andalas et al. 2018). Compared with traditional public buildings, green public buildings can greatly save energy, water, land, environmental protection and reduce pollution in the whole life cycle, thus providing more superior space for people's daily use and symbiosis with nature (Nguyen and Kostarakis 2018). From the architectural requirements and design concepts, it can be seen that the green building design process has increased the requirements for building information processing. The traditional design methods and processes, such as information processing methods, are difficult to meet the relevant requirements of green building design (Sparrevik et al. 2018). In this context, information technology can perfectly meet the relevant needs of green building design. Based on this, this paper introduces BIM technology and machine learning technology to apply it to the green building design process. Through BIM parametric modeling, the building information contained in it can be calculated in a coordinated manner. These information about buildings can make it easier for machine learning energy-saving analysis and environmental analysis, not only bring about the optimization of technology, but also realize the optimization of design methods for the entire construction industry. BIM and machine learning technology provides a good platform for different types of building collaboration and design research concepts, which is also essential for integrating green building design. Therefore, it is of great significance to study the application of machine learning and BIM in the design of green public buildings to overcome various information barriers of existing public buildings, so as to achieve the improvement in building design and overall building performance. This is a very realistic, important, and urgent issue before us.

Relevant work
The literature reviews the applicability of BIM technology to green building design quality management and the specific ways and methods of applying green building design, improves and optimizes the organizational structure and design process of green building design quality control, summarizes and emphasizes the application of BIM technology, and expounds the importance of BIM standard in technical design quality management (Huang et al. 2021;Guo et al. 2021). The contents of integrated design in different periods, the implementation of BIM collaborative design method, BIM-integrated design method and the basic stages of integrated design are discussed and explained in the literature (Maskil-Leitan et al. 2020). Subsequently, the construction process of the optimization scheme is analyzed and the research method is described. Through the orthogonal theory and Ecotect software, the thermal environment and light environment energy consumption of public buildings are simulated and tested (Wei and Chen 2020). The literature uses genetic algorithm to study the lighting comfort, indoor ventilation and energy saving of public buildings, and considers other contents and projects that affect energy consumption (Javadpour et al. 2023;Sangaiah et al. 2022;Yan and Zhu 2020). Through the simplified model of temperature improvement frequency, natural pressure time and average natural light, the performance prediction and evaluation indexes for public buildings are constructed (Himmetoglu et al. 2022).
In the literature, BIM technology is used to optimize the design and planning of buildings, and the optimization process is analyzed. Finally, genetic algorithm is proposed to optimize the energy consumption, lighting, ventilation and comfort of buildings (Elshafei et al. 2021;Sangaiah et al. 2021). Based on the multi-objective algorithm optimization theory, the paper controls the environmental factors of green buildings and analyzes the main factors affecting the establishment of environmental parameters by using the gray correlation degree. The above factors are input into the model, and SVM and K nearest neighbor (KNN) theory are applied in this process (Sanchez et al. 2020). The algorithm optimizes the parameters by introducing adaptive genetic algorithm (AGA) and optimized adaptive particle swarm optimization (APSO), and then builds a prediction model and designs experiments to compare the prediction effect of the model (Ascione et al. 2019).

Basic theory
Machine learning (ML) is an important part of artificial intelligence, and it is an interdisciplinary subject in many fields. Since the 1950s, machine learning has received extensive attention and development. Machine learning uses machines to imitate the human learning process, deduce rules, and allow the system to learn automatically to improve from experience. Most of the machine learning methods are to preprocess the data set, create the corresponding machine learning algorithm model, determine the rules of training samples by using the built model, extract valuable information from the data set, and predict the results as accurately as possible.
Supervised learning refers to assigning labels to the input data in advance, training the model with data, using likelihood function, algebraic function or artificial neural network as the bottom function of the model, and then obtaining the output data through iteration; unsupervised learning refers to the use of clustering and other technologies to extract valuable potential relationships in data without preassigned input tags and classify the data according to the potential relationships.
With the development of the construction industry, the existing data analysis, modeling or simulation methods cannot completely solve the needs and challenges faced by the construction industry. As a powerful technical tool, machine learning can be used to improve building performance, such as predicting building energy consumption, improving building control, designing building energy systems and so on. Among them, SVM is a common algorithm, which is often used to construct energy consumption prediction, pattern recognition and other issues. Support vector machines can be divided into linear problems and nonlinear problems. The linear problem can directly find lines or planes to classify the sample data. However, in practical problems, many problems are nonlinear problems. The final optimization problem becomes: Sigmoid kernel function: Based on the optimal parameters, the multi-objective prediction is further carried out, and the prediction curves corresponding to the training set and the test set are drawn, respectively. When R 2 is close to 0, the model matching effect is not good.
where n-number of samples; y-target analog value.

Algorithm optimization
The goal of genetic algorithm is to find the optimal solution through simulation. The core idea of genetic algorithm is to code the studied problem, create an initial population, and use the principle of species evolution in nature to evolve generation by generation to finally obtain the approximate value of the optimal target solution. The biggest characteristic of genetic algorithm is that it has unique global optimization ability. It can get the approximate range of optimal solution in the whole range of study parameters, and there is no local optimization. And because it adopts the natural evolutionary computation method of random iteration, it is stable, practical and efficient. When facing complex optimization problems, genetic algorithms can usually achieve the optimal results quickly and accurately.
After decades of development, genetic algorithms have been widely used in various fields.
In the actual optimization problem, it often involves several objectives that need to be optimized at the same time. These goals often have different dimensions and conflict with each other. Therefore, the multi-objective optimization problem (MOP) arises at the historic moment. Its purpose is to generate multiple objective optimal solution sets within a given parameter range, which is called ''Pareto optimal solution sets.'' The multi-objective optimization problem is composed of two or more objective functions and many constraints. The multi-objective optimization problem can be summarized as the following formula. The multi-objective optimization problem can balance the optimal solutions of various objectives in different properties. The Pareto optimal solution set obtained contains more effective information, which is worth further exploration. It can solve many practical problems, and is widely concerned and applied.
The linear membership function lfi is defined for n groups of solutions from the Pareto optimal solution set, and its expression is as follows: Max (f)-the maximum value of the ith objective function in the solution set.
Calculation distance: The point (x pareto . y areto , z pareto ) is the coordinate corresponding to each point in the Pareto optimal solution set, and the point with the shortest distance is the optimal solution in the Pareto optimal solution set.

Simulation analysis
The orthogonal test table is designed with allpairs software, and L54 (66) orthogonal table is selected with reference to the above orthogonal design. Some data of energy consumption, carbon emission and thermal comfort value of 50 sample groups simulated by design builder are shown in Application of machine learning-based BIM in green public building design 9033 Table 1; 50 groups of sample data in the table are used as the training set of SVM multi-objective prediction, and 15 groups of data are randomly selected as the SVM multiobjective prediction test.
In the Pareto optimal solution set composed of these 60 points, the ideal point method is used to find the optimal solution. According to the actual engineering situation, the distance between P (2418.183953.96, 0.17) and the ideal point is the shortest. Therefore, the design parameters corresponding to this point are the best optimization method. The comparison of design parameters and target values between the original design and the optimized design is shown in Table 2. At this time, 3 mm LOE general double-layer transparent glass is selected as the external window material, 0.1 lm brick external wall ? 0.031 mxps extruded polystyrene ? 0.1 M concrete block ? 0.013 m gypsum plaster is selected as the external wall material, and 0.01 M asphalt ? 0.185 mmw glass wool ? 0.2 m gap ? 0.013 m gypsum board is selected as the roof material. Through multi-objective optimization, the building energy consumption is reduced by 14.3%, the construction carbon emission is reduced by 11.39%, and the PMV thermal absolute value and comfort value are reduced by 34.7%, achieving the optimization effect.

BIM technology
BIM is the abbreviation of building information model, which is based on three-dimensional information technology and expresses the building entity and functional characteristics of construction projects through digital models. BIM is not a single type of software or a combination of multiple types of software, but a form of conceptual combination. Parametric BIM and visualization software create 3D models designed directly in the software, instead of generating from multiple 2D drawings, so BIM technology promotes the collaborative work among multiple design disciplines. In this regard, although different disciplines can cooperate through 2D drawing, 2D drawing is inherently more difficult and time-consuming than 3D models. BIM model has good coordination among different disciplines, which can reduce errors and shorten design time. Through BIM technology, design problems can be found as early as possible and designers can modify them, which is more cost-effective than modifying the design when the project is almost completed.

BIM evaluation and application
In the design stage, public buildings must meet the requirements of all control elements in the standard and are divided into three green levels according to the satisfaction of general projects and expected projects.
Whether it is an ordinary project or a preferred project, there are assessment points in the assessment process, and relevant supporting materials must be submitted. For example, the terms of saving land and external environment ''do not bring light pollution to surrounding buildings and do not affect the sunshine required by surrounding residential buildings.'' In the evaluation, the first thing to consider is whether to use high reflection mirror aluminum alloy decorative exterior wall or glass curtain wall; whether the outdoor landscape light is directly into the air, etc. The supporting materials to be submitted include drawings, design descriptions and inspection reports of the day provided by the design unit. In addition, there are also relevant requirements for certification materials, such as the spatial relationship between surrounding residential buildings and new projects through the general plan; when it is necessary to adopt glass curtain wall design, the light pollution caused by the curtain wall to surrounding buildings shall be provided in relevant design documents and analysis reports. It can be said that the evaluation points and requirements of the test materials explain the provisions of the green building evaluation standards.
The method of using BIM technology in the assessment terms: The clause mentioned that if there is a residential building around the project, the daily analysis report needs to be submitted to determine whether the project will affect the sunshine needs of the surrounding residential buildings. The method of this evaluation uses BIM to export the architectural design model through the format supported by Rizhao analysis software through BIM design software. After simple adjustment of Rizhao analysis software model, you can easily analyze and simulate it, and use the simulation results and other related data to write the sunlight analysis report. This avoids designers' waste of time and energy to re -model the current green architectural design process.
The method of applying BIM technology in the evaluation clause: The evaluation point refers to the wind speed requirements of the building pedestrian area and must be judged through the simulation results of the external wind environment of the building. Therefore, the BIM technical method used in this evaluation clause is similar to Rizhao analysis, that is, the format of the building design model supported by the wind environment analysis software through the BIM design software, and the outdoor air environment analysis simulation is performed after simply adjusting the wind environment analysis software model. Determine whether the wind speed of the pedestrian area is less than 5 m/s by analyzing the simulation results. It also avoids time and energy of designers in the current green architectural design process of re-modeling.

Design method
During the BIM-based green architectural design process, professionals from various tasks such as designers, engineers, and other types of work participated in each design process and coordinated people to provide a basis for decision -making. In different design stages, the role and importance of design participants are constantly adjusted. Figure 1 lists the development of the main participants in the design of the BIM-based green building design.
Building a collaborative design platform is an important guarantee and method for the designer's participation in the design process. The collaborative platform is an important foundation for various professional designers to complete model design and real-time information interaction. Design platforms here include two meanings: effective communication mechanisms and design and communication technology platforms between design teams.
The system construction broke the original work method of architects and other designers in the working method, strengthened communication in the design process, and improved the overall thinking of the design members. If architectural design is designed as a simple collection of various technical measures of green buildings, it will bring high construction costs while losing the high economic benefits of architectural design under the overall project. During the BIM-based green building design process, the system was formed at the beginning of the project, which outlined the duties of each member and the schedule of regular discussion conferences. For example, the same professional designer discusses interior design every day at a certain time; designers of different majors should communicate with each other within a reasonable time range; during the planning stage, members of the relevant teams should be invited to participate in the discussion of the plan.
Technology platform is another important guarantee for successful green design. The technology platform includes design tools and information interactive platforms, namely software and hardware technical support.
Concept design is the stage of the entire design scheme, so there are usually many different design solutions at this stage. The concept design determines the basic design structure of the next stage. The conceptual design stage is the last link of the entire design process. It mainly determines the spatial function distribution of the entire building, the design method and construction method of the structural equipment. The concept design includes the quality, appearance and refinement of architectural design, which determines the location of the building and the direction of the venue, as well as the natural structure and inherent quality.
The concept design almost depends on the knowledge and experience of architects in conventional design. After the plan is formed, it is passed on to other members of the design team for refinement and analysis. If the analysis results do not meet the construction goals, the plan should be modified according to the feedback to form an alternative plan before analyzing. After the project is finalized, it takes a lot of time to deepen the design and drawing construction drawings. This is clearly seen from Fig. 2.
In traditional design, the payment plan form proposed by architects is usually 15%of the concept design stage, 30%in the detailed design phase, and 55%in the construction diagram and detailed chart stage. This also shows that in traditional design, designers put most of their time and energy on drawing drawings. In BIM-based green architectural design, thanks to the parameterized design of BIM technology, the model design formed in conceptual design can automatically draw drawings, detailed drawings and other drawings according to the requirements of drawings and standards. Designers spend more time and effort on graphic design.

Design evaluation and influencing factors
An effective evaluation must first establish an evaluation index system, which is a unified scale to compare and measure various factors. The goal itself is a description of the general effect that can be achieved in any system.  Fig. 2 Relationship between cost changes and current design stages Generally speaking, this description is abstracted and blurred, and does not include conditions and measurable values that are directly used as the basis of assessment. The construction of the evaluation index system, as the requirements for the reasonable implementation of the entire evaluation, must follow three basic principles, namely: Target oriented: First explain the evaluation requirements and target orientation, the goal as the center of the evaluation indicator system, and guide the evaluation indicators to be closer to the direction of real and comprehensive behavior. The guidance principle requires the value evaluation direction of each indicator in the system to be equal and consistent with the goals. Reasonable structure: After the overall goal is clear, the next step is to form a reasonable indicator hierarchical structure. The indicators of each layer of the system should be unique, complete and relatively independent, to avoid each other or overlap. The indicators are the same: The final evaluation is the comparison of the indicator and the system factors. The evaluation indicators should adopt unified or recognized concepts and standards as much as possible, eliminate or correct uncertain factors and inconsistent factors, so that the evaluation indicators will always penetrate the system, similar to the evaluation model.
Finding key factors that affect the quality of green construction engineering design at various stages are important preparations for building an evaluation system. According to the analysis results of the quality target, quality status, and quality influence factors of the green building design, combined with the establishment principles of the evaluation index system, the cause and effect analysis method is used to check the rating factors. Cause and effect analysis was first used by Japanese quality management experts to find quality problems, also known as fish bone maps, and the relationship between the quality characteristics and related quality factors used to analyze the statement. The fish bone map in this article is a structural fish bone map. There is no complete causal relationship between the elements and characteristics, and it also expresses the structure and composition relationship. It can organize problems in a structured manner. It is mainly composed of design quality problems and related influencing factors, as shown in Fig. 3.
This article proposes a comprehensive meteorological indicator, that is, human comfort indicators, and its expression method is shown in 8: In the formula, D is the comfort index of the human body; T is the average daily temperature, and the unit's degree Celsius; U is a daily average relative humidity, the unit is%; V is the average daily wind speed, and the unit is M/S. It is recommended to use the temperature index to predict the comfort of the human body and the relationship between the temperature index and the indoor temperature. It can be expressed as a formula 9: When using the support vector machine algorithm to predict the modeling, the basic idea is: the input space X in the input space is mapped to the high-vitamin symbol space through nonlinear mapping /(x), and then, the high-dimensional linear space is returned.
The maximum value of the ai solution function: Here is a nuclear function K to achieve nonlinear transformation. The nuclear function meets K (xi, From the literature knowledge, it can be seen that if the prior knowledge of the model is lacking, the SVM algorithm based on the radial base function (RBF) is better than other nuclear functions. The mathematical formula is: The parameter optimization model can be expressed as: Max G (r, C), and its constraints are: The formula of the selection variance of the computing training adaptation function (MSE) is: From the data set collected by the experiment, 200 data groups are selected as a sample set in the sequential order, and the other 25 data groups are used as prediction sets.
The prediction results of different models are compared to Fig. 4.
According to the experimental data in Table 3, the model prediction results will be affected by many factors, including training methods and modeling methods.
Compared with the four prediction methods, APSO-SVM has better prediction accuracy than the other three prediction methods under the same iteration times. It is a widely used environmental parameter prediction model.

Energy consumption analysis of green buildings
Use Ecotect green building performance simulation software to analyze the annual energy consumption and monthly discomfort hours of the building, and calculate the discomfort caused by energy consumption each month (see Fig. 5). See Table 4 for the calculation results. The monthly uncomfortable energy consumption diagram is shown in Fig. 5.

Conclusion
The energy consumption of public buildings is relatively high due to their functions and characteristics. On the other hand, people's pursuit of architectural forms is diverse, and  the requirements for intelligent level and use of buildings are higher and higher, and the requirements for building construction technology are more and more complex. Therefore, in the face of the increasing building forms and construction requirements, the traditional two-dimensional expression of architectural design concept is insufficient. Based on this background, this paper combines machine learning technology and BIM technology in the design of green public buildings. BIM and machine learning are introduced in the architectural design process to break the barriers between various disciplines. The design method has certain scientific, systematic progressiveness and operability and plays a certain role in expanding and improving the green building design theory in China.
Funding The authors have not disclosed any funding.
Data availability Data will be made available on request.

Declarations
Conflict of interest The authors declare that they have no conflict of interests.
Ethical approval This article does not contain any studies with human participants performed by any of the authors. Application of machine learning-based BIM in green public building design 9039