WO2025090040A1

WO2025090040A1 - Pre-application sustainable building estimation system

Info

Publication number: WO2025090040A1
Application number: PCT/TR2024/050919
Authority: WO
Inventors: Muhammet Fatih AK; Setenay UÇAR
Original assignee: Antalya Bilim University
Current assignee: Antalya Bilim University
Priority date: 2023-10-23
Filing date: 2024-08-06
Publication date: 2025-05-01
Anticipated expiration: 2026-04-23

Abstract

In the construction of sustainable buildings, it is aimed to reduce the energy consumed by maximising the comfort of the users in the interior space and to use natural materials instead of materials that emit harmful emissions. All sustainable buildings are inspected after the construction is completed to determine whether they provide energy consumption, carbon emission or user comfort. In this case, when a building is planned, it is not possible to check whether the desired result will be obtained from the building before the interior space is used. Our invention relates to a prediction system that enables the planning of how the sustainable building should be before the intended building is designed and the method that will provide the least consumption of energy is decided during the architectural design phase of the building.

Description

PRE-APPLICATION SUSTAINABLE BUILDING ESTIMATION SYSTEM

TECHNICAL FIELD

In the construction of sustainable buildings, it is aimed to reduce the energy consumed by maximising the comfort of the users in the interior space and to use natural materials instead of materials that emit harmful emissions. All sustainable buildings are inspected after the construction is completed to determine whether they provide energy consumption, carbon emission or user comfort. In this case, when a building is planned, it is not possible to check whether the desired result will be obtained from the building before the interior space is used. Our invention is related to a prediction system that enables the planning of how the sustainable building should be before the intended building is designed and the method that will provide the least consumption of energy is decided while the building is in the architectural design phase.

BACKGROUND

Construction in the concept of "Passive House", there are officially two in Turkey. The Passive House concept is a preferred method within sustainable buildings, not because it is in certain moulds, but because it is generally preferred in terms of its focus on energy efficiency. It helps to reduce the amount of energy consumed without the demand for an additional mechanical need. In regions with hot climates, it is especially beneficial to reduce the electricity costs consumed due to the need for air conditioning. In Turkey and in the regions with hot climates in the world, there is no need to use insulation materials due to the high cost, and there is no need for application on the grounds that there is not much heating load in buildings.

When we look at the countries with hot climates, the high cooling load of the hot climate causes the mechanical systems such as air conditioners, coolers, etc. to operate more and thus increase the energy required. For this reason, it is questioned in the literature whether the construction concept that saves energy for cold climates is also valid for hot climates. In this case, calculations and findings are put forward to reduce the energy requirement and operating costs by applying the Passive House concept in hot climates. In this context, our invention aims to direct the designer within the framework of the targeted parameters of artificial intelligence in architectural design by applying neural fuzzy logic method and to reach the optimum result. Thus, it is possible to include the decision stages of the energy-saving construction concept, which is extremely important in terms of predetermining and evaluating the factors that will be effective in the utilisation phase of the building, into the planning while the design phase is in progress. The efficiency results of our invention show that both the applicability of the Passive House concept for hot climates is possible and the use of artificial intelligence in making multi-criteria selection in architectural designs can be used as a method to reach the right result.

With this concept, which will benefit both the urban scale and the individual users of the buildings, it is concluded that large buildings such as indoor pools provide at least 50% efficiency in winter and 13.5% efficiency in summer. In small scales, this results in higher energy efficiency. For this reason, with this system created with artificial intelligence, 268,063 TL (according to 2022 unit price) can be profited even with the lowest efficiency in summer, especially in large-scale structures such as indoor pools that consume a lot of electricity. This efficiency can also appear as natural gas or electricity consumption according to energy consumption needs. For this reason, in order to reduce the consumption of natural resources, the Passive House concept is easy to implement and has the feature of adapting to nature.

In today's buildings, the design of the building with the attention of the above- mentioned six parameters together occurs by taking into account the landscape factor and profit rates in hot climates, while the profit of the user and the construction company is ignored. In this sense, reaching the preliminary estimates of the most appropriate building design supported by artificial intelligence helps to save time and money before and after construction. It is concluded that it is possible to reduce the needs of electricity, natural gas, etc. spent depending on the heating and cooling loads in summer and winter periods in building operating costs by applying the Passive House concept.

AIM OF INVENTION

There is a construction concept defined as "Passive House" within sustainable buildings. This construction concept ensures the use of natural methods and natural materials to ensure the highest level of harmony with nature. Passive Houses, which include parameters such as natural ventilation, orientation of the building for natural heating, window/wall ratios for the most efficient use of lighting, the use of insulation materials according to climatic conditions, and the use of shading elements to naturally cut excess light, were first implemented in Germany with a cold climate. Over time, it continued to be applied in other countries with cold climates and continued to be applied in buildings with all kinds of functions due to energy saving. In the last 10 years, it has been applied in buildings with warm climates, but it has been revealed in the measurements of the applied buildings that they are not as efficient as the buildings in cold climates. With our invention, it is aimed to determine the parameters by considering the climate in regions with hot climates and to predict the appropriate structure before the application and to facilitate the decision in the architectural design phase. The parameters to be considered in buildings to be applied in hot climates are determined as follows;

- Temperature difference between indoor and outdoor

- Insulation (window frame, glass, wall, ceiling, floor) material thermal transmittance values

- Availability/non-availability of natural ventilation

- Having/not having shading elements

- Orientation of the building (North, south, east, west)

- Window to wall ratio

With these parameters, with the help of artificial intelligence, it is ensured that the annual energy requirement of a building can be found in the lowest possible way. The parameters and output are the desired and obtained data required for a Passive House. Obtaining this data is also called training data to be taught to artificial intelligence by collecting real data by comparing the houses measured in the literature. In order to obtain the correct results with the help of artificial intelligence, the results were compared by performing a trial test. The data obtained for the test test are data that are not in real measurements but the results are known. With these data defined as test data, it was started to test whether there was learning or not. First, the error value was measured and the real data and the learnt data were compared. The training root mean square error (RMSE) value of the data learnt within our invention is 0.0033174 and the test root mean square error (RMSE) value is 0.17755. Since the error value is low, the learning of the data that will verify the result is completed. Thus, in order to reach the result in neural fuzzy logic, the rules are created and through the rules, the most ideal ratios are extracted in order to meet the lowest energy requirement. Regression analysis was performed for the accuracy of the results obtained and information about the existence and strength of the relationship between the variables was obtained and as a result, it was determined that the R² value was 0.9997 (99%).

With the invention, "Passive House" can be applied in hot climates by looking at the requirements of our own climate zone. In this way, if we correctly apply what nature offers us for sustainable structures, it will provide a change in the construction sector in terms of economic sense and unnecessary consumption of energy resources, as the least energy consumption can be achieved.

FIGURE LIST

Figure 1 . ANFIS Error Graph

Figure 2. Comparison of the output obtained in training and estimated outputs

DETAILED DESCRIPTION OF THE INVENTION

Neural fuzzy logic is a method presented by synthesising the learning ability of artificial neural networks with the decision-making ability of fuzzy logic method. In fuzzy logic method, since there is no specific method for situations such as determining the input parameters, this situation is determined by trial and error; In neural fuzzy logic, they differ from each other due to the application of learning methods to determine, regulate or determine the structure of the parameters. To explain through the subject of our invention; while adjusting the structure, situations such as the number of variables to be calculated, the number of rules in the system, the range of membership functions of the inputs are determined. In the variable adjustment process, conditions such as membership function centres, start and end points, weights of the connections are determined. Therefore, different learning types are applied in different layers within the neural fuzzy logic method.

Learning methods in neural fuzzy logic systems enable the development of a model that learns by training using the data set together with fuzzy logic modelling principles. Combining the strength of fuzzy logic in working with verbal variables and decision making with the ability of artificial neural networks to learn, flexibility, speed and adaptability constitutes neural fuzzy logic. While fuzzy logic directly utilises expert knowledge and experience, neural fuzzy logic has the ability to classify and adapt to the problem thanks to its learning ability, and to make decisions like a human being thanks to its decision-making ability. While fuzzy logic has verbal variables, neural fuzzy logic has verbal and numerical variables. While the rules are created by the machine as a result of learning the data in neural fuzzy logic, in fuzzy logic the rules are created with the help of expert knowledge.

Within the scope of our invention, firstly, the passive house requirements and the factors that will affect the energy efficiency are investigated and the factors that will affect the output for the inputs are determined. With the neural fuzzy logic method, it is aimed to find some inputs and outputs that will provide passive house requirements in buildings. In the first stage, the input of the outputs to be obtained with the neural fuzzy logic method was prepared. The neural fuzzy logic inputs used are as follows;

- Outdoor temperature value

- Total energy required in the building

- Insulation U-value

- Natural Ventilation (present / absent)

- Shading element (present / absent)

- Building orientation (South / North / East / West)

- Window to wall ratio (WWR)

The following outputs were obtained as a result of processing the data with neural fuzzy logic method.

There are many examples in the literature regarding the window-to-wall ratio. Marino et al. (2017) conducted research in Italy that the correct design of the building envelope is a very important stage of the construction process, which should appropriately take into account the climatic conditions of the location of the building. Xue et al. (2019) conducted research in China on optimising the window to wall ratio of sunshades for daylight and energy saving requirements. Considering the views of hotel rooms, it was concluded that the daylight requirement determines the minimum window to wall ratio value, while the energy consumption has the maximum value. (Zolfaghari and Jones 2022), which also examines the effect of lighting on the windowwall ratio (Zolfaghari and Jones 2022) and the effect of the window-wall ratio on the energy demand for space heating and cooling on an office building in Turin and Helsinki (Chiesa et al. 2019). The parameters used for these results are insulation level, orientation of the building, shading, controlled natural ventilation. These real data obtained from the literature review were used as input. As a result of preparing the data as input and digitising the verbal data, the rules must be specified as in Table 1 in order to process the input data. The more the number of rules, the more accurate and realistic the results are. After specifying the rules, the software hosted by the inventive system evaluates the inputs according to the rules and creates the model of the problem. These rules can be changed and monitored when necessary, and new rules can be added while the system is running. These additions and changes are carried out with an input device included in the system. The device can be any device containing a processor that can run software such as a computer, phone, tablet. The input can be provided by any touch or mechanical interface device.

Table 1

In order to create the numerical equivalent of the data, the data containing the inputs must first be normalised. In order to prepare the collected data in two parts as training and test, the first step is to determine the numerical values of the inputs. The value ranges to be used for each of the inputs were normalised by linear transformation between 0 and 1 according to the formula 1 below. According to the formula applied for normalisation, the results were given values between 0 and 1 .

Formula 1

A total of 200 normalised input data, 176 training and 44 test, were prepared in the model, which was prepared as six inputs and one output. The next step for the use of the prepared inputs and numerical values is to determine the membership functions. Membership function is one of the eight membership functions with the names of triangle, trapezoid, bell-shaped, gauss, gauss2, pi-shaped, dsigmodial and psigmodial. In the graph of these membership functions, there may be differences in membership functions for the purpose of use due to reasons such as obtaining a complete result of the data on the line or having a fuzzy value between 0 and 1 on the line.

Table 2. Membership functions error values

In order to find the one with the least error value among the membership functions and to be able to proceed in that way, the training of the inputs to be used according to the function types and the comparison of the results are provided by the RMSE values given in Table 2. The root mean square error (RMSE) value shown in Formula 2 below was used as the performance evaluation criterion of the model to be created with the prepared inputs and their values;

Formula 2

Based on the comparison, gauss2mf, which has the lowest error rate among them, was determined as the membership function to be used in the modelling process. According to Table 2, the RMSE value was determined as 0.0033174. Thus, in order to reach the outputs, the training process was continued by continuing with the Gauss2 membership function. As shown in the graph in Figure 1 , the error size decreases to 0.0033174 at the end of the 16th cycle with Gaussian2 membership function.

In order to get results with the neural fuzzy logic method, which is an artificial intelligence method, it was tested by comparing whether the results were consistent in the machine learning process of two different data separated as training and test. By looking at the comparison between the outputs of the data separated as training and test, it is shown in Figure 2 that the relationship between their values and the real values is close, as shown by the round lines and the star shape overlapping on top of each other. In this case, it is determined that the correspondence relationship between the real data and the outputs of the data predicted by the neural fuzzy network is strong.

Table 3. Input error value table

The error value in the test data is shown as 0.17755 in Table 3 with the determination of the inputs. After finding the error value of the trained data, some rules were generated. These rules are verbal expressions showing the effects of inputs on output. These expressions are defined as "if / "if. For a model with a total number of inputs (six inputs for our invention) and outputs (one output for our invention), different rules were added for the membership functions determined by the number of intervals (there are three different intervals for our invention) to the power of the number of inputs (sixth power of three for our invention). In other words, a total of 3⁶=729 rules were created in the model during the trial phase of the invention. This number of rules varies according to the number of inputs, outputs and intervals.

For each output rule, verbal values and expression were determined differently. These values are named as low-medium-high, low-medium-large, small-medium- large. When these verbal values and inputs come together, the output rule is formed as low-medium-most. If the first of the rules is analysed as an example; the low temperature in outdoor conditions, the insulation material is selected to minimise the heat transfer, and the evaluation is considered as the first two inputs. In order to get natural ventilation in the space, the use of open windows, the use of shading elements on the facade, the direction of the building towards the south and the ratio of the window to the wall is expressed as small, the rule that the annual energy requirement, which is the output, will be less, is formed.

Table 4. Example of created rules for 23 out of 729

With the formation of rules, the output value becomes accessible. These results are estimated results generated by artificial intelligence. For the six inputs and results specified in Table 3, the output data are normalised and converted into values between 0 and 1 . In order to calculate the output of these numerical data with neural fuzzy logic, the results are given values between 0 and 1 according to the formula applied for normalisation of all data. For normalisation, the highest and lowest values of the input must be known. Therefore, with the help of real data, all inputs and outputs are converted into numbers between 0 and 1 . The formula used is below;

An example for obtaining the result after normalisation can be shown as follows; the first input temperature is 0.83 (hot), the second input is 0.982 (insulation U value is high), the third input is 0 (ventilation is open), the fourth input is 0 (there is a shading element), the fifth input is 0.1 (building direction is south) and the sixth input is 0.5 (window wall ratio is 50%). In this case, the output value (annual energy requirement) was found to be 0.508. By keeping some of the input variables constant and changing some of them, the data to be obtained from the result will indicate the annual energy requirement.

Regression analysis was performed to compare the actual values given during the training with the results. Regression analysis is an analysis to give information about the existence and strength of the relationship between variables. According to the result obtained, the R² value was found to be 0,9997. The high R² value indicates that the regression model fit is good. This shows that the fit between the real inputs and the inputs in the inventive algorithm is very good.

In order to obtain numerical results from verbal data, it is necessary to examine the effects on variable inputs by keeping some values constant. Eight comparison data are created and their differences are analysed. Eight different methods of changes to be made on a building are listed below; 1. Measurement of the effect on the annual energy requirement with the availability of ventilation and the use of shading in hot weather conditions

2. Measuring the effect of the lack of ventilation in hot weather conditions and the use of shading on the annual energy requirement

3. Measurement of the effect on the annual energy requirement with the presence of ventilation in hot weather conditions and without the use of shading

4. Measuring the effect of not using ventilation and shading in hot weather conditions on the annual energy requirement

5. Measuring the effect of the availability of ventilation in cold weather conditions and the use of shading on the annual energy requirement

6. Measuring the effect of the lack of ventilation in cold weather conditions and the use of shading on the annual energy requirement

7. Measurement of the effect on the annual energy requirement with the presence of ventilation in cold weather conditions and without the use of shading

8. Measuring the effect of not using ventilation and shading in hot weather conditions on the annual energy requirement

In this way, the correct building design can be reached by measuring the differences between them and the changes that affect energy consumption. If there are compulsory situations, preliminary information about how the result may be can be obtained thanks to this method. Detailed sample results are given in Table 5-12. Insulation materials U values are matched with real materials and the results are evaluated.

The verbal results obtained from the numerical data are as follows;

- When the temperature drops too much and increases too much, the effect on output is too high, but when the temperature is moderate, the effect on output decreases.

- According to the thermal transmittance value of the insulation material, if the insulation is bad or uninsulated, the effect on the output is very high, but when the insulation material is used well, the effect on the output is halved. There is not much difference between the U value of the insulation being medium and good.

- Although there is not much difference between natural ventilation being open or closed, being open creates more energy demand than being closed.

- There is not much difference between the presence and absence of shading elements on the facade, but the presence of shading increases the energy demand more than the absence of shading.

- There is not much difference between south, north, east and west in the change of the energy amount of the direction of the building. However, in numerical terms, it is determined that the building orientated to the north needs more energy and the building orientated to the south needs less energy.

- The ratio of the area where the windows are located on the facade of the building and the ratio of the wall, called PDO, causes more energy demand in the ranges where the window-wall ratio is 20-30-40-50-60%. The least energy requirement is 10% window ratio and then 90% window ratio is one of the findings extracted from the graph.

- Along with the temperature value, other inputs reduce the annual energy requirement. Temperature input is an important input that affects the amount of energy to be used.

- The second most important input is insulation. Especially with shading and ventilation inputs, it has a significant effect on the output.

- Ventilation and shading inputs do not have much effect on output. They provide some change, but they are not as effective as insulation and temperature inputs..

According to the findings, energy needs may vary between buildings with cold weather conditions depending on the heating and cooling load. In addition, the absence of ventilation reduces the annual energy requirement in hot weather and cold weather conditions.

For the training data and output data process, parameters are determined by literature review. Evaluations are made by considering Passive House criteria for reducing energy consumption and efficient use of existing buildings by supporting natural conditions. Firstly, data are collected to determine the inputs and to decide what the desired result is. After the parameters obtained are determined, the process of accumulating data on what the result is by making measurements in buildings with hot and cold climates begins.

The step-by-step procedure is as follows;

1- Collection of measured values in real buildings

2- In neural fuzzy logic, a part of the real data is separated as training and a part as test in order to test whether the machine can perform learning. At this stage, if the experiment fails, the machine is continued to learn. In the invention, according to the RMSE value obtained with the prepared real and test data, the machine completed learning in the 16th trial and became ready to find the result.

3- For how many trials the accuracy rate is high, regression analysis is performed to confirm the accuracy rate (99% for the invention).

4- After completing machine learning, neural fuzzy logic has prepared 729 rules. The purpose of this is to create verbal expressions that determine the conditions and obstacles related to whatever result we want to achieve. Thus, the parameters affecting the result are also revealed.

5- Tables 5, 6, 7, 8, 9, 10, 11 and 12 were created by keeping some parameters constant and using some variable expressions. In this way, the output shows how it is affected for each constant value. By comparing the normalised data in the table with each other, it is determined which situations will affect energy consumption more and which ones will affect it less.

Table 5. The effect on the annual energy requirement with the availability of ventilation and the use of shading in hot weather conditions

Table 6. The effect on annual energy requirement with the absence of ventilation and the use of shading in hot weather conditions

Table 7. The effect on the annual energy requirement with the availability of ventilation in hot weather conditions and without the use of shading

Table 8. The effect of not using ventilation and shading in hot weather conditions on the annual energy requirement

Table 9. The effect on the annual energy requirement with the availability of ventilation and the use of shading in cold weather conditions

Table 10. The effect on the annual energy requirement with the absence of ventilation and the use of shading in cold weather conditions

Table 11. The effect on the annual energy requirement with ventilation in cold weather conditions and without the use of shading

Table 12. The effect on annual energy demand by not using ventilation and shading in cold weather conditions

Claims

1. Pre-implementation sustainable construction estimation system, characterised by comprising a device containing a processor that can run a software that enables input and changes to be made within the inputs and rules through an interface; wherein the software;

- Comprises inputs which are outdoor temperature value, total energy required in the building, insulation U value, natural ventilation (present / absent), shading element (present / absent), orientation of the building (South / North / East / West), window wall ratio,

- Prepares data as input and normalisation and quantification of the collected verbal data,

- Determines the membership functions of inputs and numerical values,

- Comprises rules for the processing of input data,

- Evaluates inputs with neural fuzzy logic according to rules,

2. The pre-application sustainable structure estimation system according to Claim 1 , characterised by comprising the software that normalises the value ranges to be used for each of the inputs according to formula 1 by linear transformation between 0 and 1 in order to prepare the collected verbal data in two parts, training and testing.

3. The pre-application sustainable structure estimation system according to Claim 1 , characterised by comprising the software comprising triangular, trapezoidal, generalized bell, gauss, gauss2, pi-shaped, dsigmodial and psigmodial membership functions.

4. The pre-application sustainable structure estimation system according to Claim 1 , characterised by comprising the software that enables the selection of the membership function with the least error rate according to formula 2 of the inputs in the step of determining the membership functions.

5. The pre-application sustainable structure estimation system according to Claim 1 , characterised by comprising the software further comprising a number of different rules equal to the number of intervals to the power of the number of inputs.

6. The pre-application sustainable structure estimation system according to Claim 1 , characterised by comprising the software which includes the rule for evaluating the output as low-medium-high, few-moderate-many, small-medium- large verbal values for each output rule, and the rule for evaluating the output as few-moderate-many when these verbal values and inputs come together.