Machine learning-based digital rapid prediction method for urban low-carbon update potential

Abstract

The present application relates to a kind of city low-carbon update potential digitalization fast prediction method based on machine learning, comprising the following steps: constructing ideal residential model;Performance simulation is carried out based on ideal residential model;Energy-saving reconstruction is carried out to ideal residential model;Performance simulation is carried out to the residential model after reconstruction;Based on the residential model before and after reconstruction and its performance simulation data set is constructed, the building parameters before and after reconstruction are used as the input of the energy-saving amount prediction proxy model based on XGBoost, and the residential energy-saving amount after reconstruction is used as the model output, after the model is trained using data set, it is used for building group low-carbon reconstruction potential prediction.Compared with prior art, the present application has the advantages of high prediction accuracy, can quickly predict and the like.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of building energy consumption prediction, and particularly relates to a method for digital rapid prediction of urban low-carbon renovation potential based on machine learning. BACKGROUND

[0002] Existing building renovation research mainly focuses on single buildings, on the one hand ignoring the potential influence of interaction between building groups, and on the other hand building simulation needs to input a large number of basic parameters, and there is a problem of low efficiency. Based on the low-energy template driven building renovation strategy research, there have been many explorations in the past, including the renovation of residential single buildings, office buildings, etc. However, the simulation in these studies has the problem of high time consumption, and the building scheme evaluation takes a long time, which is not conducive to the optimization of the scheme. SUMMARY

[0003] The purpose of the present application is to provide a method for digital rapid prediction of urban low-carbon renovation potential based on machine learning, which can quickly and accurately predict the energy saving amount after renovation.

[0004] The purpose of the present application can be achieved by the following technical solutions:

[0005] A method for digital rapid prediction of urban low-carbon renovation potential based on machine learning, comprising the following steps:

[0006] S1, constructing an ideal residential model;

[0007] S2, performance simulation based on the ideal residential model;

[0008] S3, energy saving renovation of the ideal residential model;

[0009] S4, performance simulation of the renovated residential model;

[0010] S5, constructing a data set based on the residential model before and after renovation and its performance simulation data, taking the building parameters before and after renovation as the input of the energy saving amount prediction proxy model based on XGBoost, taking the energy saving amount of the renovated residential model as the output of the model, and using the data set to train the model for building group low-carbon renovation potential prediction.

[0011] The S1 comprises the following steps:

[0012] S11, analyzing the actual residential building type and determining the residential building basic form;

[0013] S12, dividing the residential area into multiple block units in combination with the actual residential scale, and offsetting the block unit boundary to the interior of the residential area by a preset distance to obtain the road boundary and the ideal residential developable land range;

[0014] S13, the residential building basic form is combined in multiple ways, and a residential area element model is generated by combining the ideal residential area construction land range, and an ideal residential area model is constructed.

[0015] The residential building basic form includes a point type residential group, a plate type residential group, and a closed type residential group.

[0016] The point type residential group is divided into five types, wherein the P-1 type is a one-floor two-house small house type residential building in the early stage of urban construction, the simplified P-1 type building is 14m wide in the south-north direction and 11m wide in the east-west direction, and the distribution in the residential area scale is 6 rows of row type arrangement, and the building layer number fluctuation range is 3-7 layers; the P-2 type residential building is 27m wide in the south-north direction and 12m wide in the east-west direction, which is a typical one-floor two-house large house type residential building, and is distributed in a row type in the residential area, and the building layer number fluctuation range is 3-16 layers; the P-3 type building layout is staggered, the building is Y-shaped for south light, and the building layer number fluctuation range is 8-25 layers; the P-4 type is a square volume staggered layout, the building is 20m wide in the south-north direction and 13m wide in the east-west direction, and the layer number fluctuation range is 8-25 layers; the P-5 type residential building is a single point type layout hotel type apartment building, which is 30m wide in the south-north direction and 26m wide in the east-west direction.

[0017] The plate type residential group is divided into two types, and the residential building layout density and layer number limit are different, wherein the S-1 type plate type residential building is 64m wide in the south-north direction and 11m wide in the east-west direction, which is a typical two-floor four-house low multi-storey residential building, and the layer number fluctuation range is 3-7 layers; the S-2 type is a typical row type residential building, which is 56m wide in the south-north direction and 11m wide in the east-west direction, and the layer number fluctuation range is 8-25 layers.

[0018] The closed type residential group is divided into two types, wherein the C-1 type is an open type enclosure, and the building layer number fluctuation range is 3-16 layers; the C-2 type is a back-to-back type residential building, and the building layer number fluctuation range is 3-16 layers.

[0019] The performance simulation includes outdoor microclimate, street block energy consumption and solar radiation simulation.

[0020] The energy saving reconstruction includes using energy saving lamps, using low energy consumption electrical appliances, using low solar heat gain coefficient value glass, and adding solar photovoltaic panels on the roof.

[0021] The data set includes building form parameters, residential lighting before and after reconstruction, equipment power density difference, glass SHGC difference, photovoltaic panel power, and residential energy use intensity before and after reconstruction, and the building form parameters include building volume rate, building density, open space index, average building layer number, and staggered degree.

[0022] The input of the energy-saving amount prediction proxy model includes building volume ratio, building density, open space index, average building story, staggered degree, residential lighting before and after reconstruction, equipment power density difference, glass SHGC difference, and photovoltaic panel power.

[0023] Compared with the prior art, the present application has the following beneficial effects:

[0024] (1) The present application accelerates simulation through a proxy model, and the XGBoost model trained in the present application is hundreds of times faster than an energy consumption simulation plug-in in predicting population activity-related indicators, has high prediction accuracy performance, saves a large amount of time for energy consumption simulation, and improves the efficiency of evaluation.

[0025] (2) The present application proposes a data-driven existing residential area reconstruction and optimization workflow, and the trained model can be combined with a genetic algorithm and the like to quickly optimize the best reconstruction indicators, so as to help designers obtain the best reconstruction strategy and reconstruction degree under different morphological indicators, quickly predict the low-carbon reconstruction potential of urban blocks, determine the adaptability of low-carbon reconstruction strategies, and further guide urban renewal and improve design work efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0026] Figure 1 is a method flowchart of the present application;

[0027] Figure 2 is a schematic diagram of a residential building basic form and an ideal residential model generated by the present application. DETAILED DESCRIPTION

[0028] The present application will be described in detail below in combination with the drawings and specific embodiments. The present embodiment is implemented on the premise of the technical solution of the present application, and gives a detailed implementation manner and specific operation process, but the protection scope of the present application is not limited to the following embodiments.

[0029] The present embodiment provides a machine learning-based digital rapid prediction method for urban low-carbon renewal potential, as shown in Figure 1 The method comprises the following steps:

[0030] S1, constructing an ideal residential model.

[0031] Specifically, S1 comprises the following steps:

[0032] S11, analyzing actual residential building types and determining a residential building basic form.

[0033] In the present embodiment, some most common residential building basic forms in Shanghai are collected, and then the plan is generalized through parameterization, and as a result, 9 typical Shanghai residential building forms are obtained, as shown in Figure 2As shown, the residential building base form includes point-type residential groups (P-1-P-5, a total of 5 types), plate-type residential groups (S-1, S-2, a total of 2 types), and enclosed-type residential groups (C-1, C-2, a total of 2 types).

[0034] The point-type residential group is divided into five types. The P-1 type is a one-floor two-house small house residential building in the early stage of urban construction. The simplified P-1 type building is 14m wide in the north-south direction and 11m wide in the east-west direction. In the residential scale, it is distributed in a 6-row array. The building layer fluctuation range is 3-7 layers. The P-2 type residential building is 27m wide in the north-south direction and 12m wide in the east-west direction. It is a typical one-floor two-house large house. It is distributed in a row array in the residential area. The building layer fluctuation range is 3-16 layers. The P-3 type building layout is staggered. The building is Y-shaped to strive for south light. The building layer fluctuation range is 8-25 layers. The P-4 type is a square volume staggered layout. The building is 20m wide in the north-south direction and 13m wide in the east-west direction. The layer fluctuation range is 8-25 layers. The P-5 type residential building is a single-point layout hotel-type apartment building. The north-south face is 30m wide, and the east-west face is 26m wide.

[0035] The plate-type residential group is divided into two types, and the residential building layout density and layer number limit are different. The S-1 type plate-type residential building is 64m wide in the north-south direction and 11m wide in the east-west direction. It is a typical two-floor four-house low-rise residential building. The layer fluctuation range is 3-7 layers. The S-2 type is a typical row house. The north-south face is 56m wide, and the east-west face is 11m wide. The layer fluctuation range is 8-25 layers.

[0036] The enclosed-type residential group is divided into two types. The C-1 type is an open-type enclosure. The ventilation and lighting effect is good. There are few east-west directions. The building layer fluctuation range is 3-16 layers. The C-2 type is a back-to-back house. The building layer fluctuation range is 3-16 layers.

[0037] S12, in combination with the actual residential scale, divides the residential area into multiple block units, and offsets the block unit boundary to the inside of the residential area by a preset distance to obtain the road boundary and the ideal residential construction land range.

[0038] Specifically, in combination with the actual residential scale, an ideal residential area with a size of 240m x 240m is planned on the Grasshopper platform. Then, the land is divided into a 3x 3 grid to generate 9 80m x 80m block units. Then, the block unit boundary is offset inward by 5m to obtain the road boundary. Therefore, the ideal residential construction land range is 70m x 70m, and the road width is 10m, as shown in Figure 2 .

[0039] S13, multiple combinations of residential building basic form are made, and ideal residential construction land range is combined to generate residential area element model, and ideal residential model is constructed.

[0040] The residential building basic form module is embedded in the Grasshopper platform. The construction range of the block grid is randomly selected by type to make multiple combinations of the 9 types of residential buildings in the basic form module, and a variety of residential area element models are automatically generated, as shown in the ideal residential model of Figure 2 , which is further used for energy consumption simulation and modification. Among them, the residential element model building type composition, building height, and building density, volume rate, etc. are different, which ensures the universality of the residential sample. Among them, the way to automatically generate a variety of residential area element models is to randomly jump out of a small square residential type, and continuously jump out of nine to form a nine-square residential area, that is, a residential area element model. The algorithm in the Grasshopper platform of the basic form module is coded with tags, but it will be displayed as a basic block in the model visualization and simulation. Figure 1

[0041] S2, performance simulation based on ideal residential model, including outdoor microclimate, block energy consumption and solar radiation simulation.

[0042] In this embodiment, the residential building height is controlled at 3 to 25 layers, and the layer height is uniformly set to 3m. The number of layers and the building type appearing in each grid are generated randomly by parameterization. The building function ratio is as follows: bedroom ratio 40%, living room ratio 20%, kitchen ratio 10%, bathroom ratio 10%, and corridor space ratio 20%.

[0043] In addition, according to the General Code for Building Energy Efficiency and Renewable Energy Utilization GB 55015-2021, the building lighting and equipment power density, building internal personnel density, air conditioning, envelope, window-wall ratio and other building basic indicators are set as shown in Table 1. Among them, the lighting power density of the unmodified building is 5.0W / ㎡; the equipment power density is 3.8W / ㎡; the personnel density is 0.04 people / ㎡; the air conditioning summer refrigeration temperature is 26℃, and the winter heating temperature is 18℃; the building envelope heat transfer coefficient of the outer wall is 0.8W / (㎡·k), the roof heat transfer coefficient is 0.5W / (㎡·k), the heat transfer coefficient of the partition wall and the interior partition wall is 1.5W / (㎡·k); the glass solar heat coefficient is 0.84, and the heat transfer coefficient is 2.7W / (㎡·k); the north-south window-wall ratio is 0.4, and the east-west window-wall ratio is 0.1.

[0044] Table 1

[0045]

[0046] ​After generating a large number of residential meta-models, the first batch of outdoor microclimate, block energy consumption and solar radiation simulation is performed on the sample block model using Urban Weather Generator (UWG), Honeybee and OpenStudio.

[0047] S3, energy-saving reconstruction is performed on the ideal residential model.

[0048] According to the existing research on low-carbon reconstruction of residential buildings and actual energy-saving reconstruction application, four modifications are performed on the residential model in this embodiment, which are: using energy-saving lamps, using low-energy appliances, using low-solar heat gain coefficient (SHGC) value glass and adding solar photovoltaic panels on the roof. The lighting power density of the reconstructed residential building is reduced in the range of 0-3.55w / ㎡, the equipment power density is reduced in the range of 0-1.5w / ㎡, the glass SHGC is reduced in the range of 0-0.7, and the solar photovoltaic panel power changes in the range of 0.275-0.6kW. After randomly selecting values within the adjustment range of the reconstruction strategy through parameterization, the building performance simulation is performed.

[0049] S4, performance simulation is performed on the reconstructed residential model, including outdoor microclimate, block energy consumption and solar radiation simulation.

[0050] The simulation of this step is synchronous with step S2, which will not be described in detail in this embodiment.

[0051] S5, based on the residential model before and after the reconstruction and the performance simulation data, a data set is constructed, the building parameters before and after the reconstruction are taken as the input of the energy-saving amount prediction proxy model based on XGBoost, the residential energy-saving amount after the reconstruction is taken as the model output, and the model is trained using the data set and used for building group low-carbon reconstruction potential prediction.

[0052] This embodiment records the corresponding building form and the related parameters after the reconstruction, such as: building form parameters such as building volume rate (FAR), building density (BD), open space index (OSR), average building floor number (AF) and staggered degree (SD), the difference between the lighting and equipment power density of the residential building before and after the reconstruction, the difference between the glass SHGC before and after the reconstruction, the power of the photovoltaic panel and the energy use intensity (EUI) of the residential area before and after the reconstruction, and is packaged into an Excel file and imported into jupyter as a data set for training the energy-saving amount prediction proxy model.

[0053] Among them, the calculation formulas (1)-(5) of each form parameter are as follows:

[0054]

[0055]

[0056]

[0057]

[0058] Skewness (SD) = Maximum building height - Average building height (5)

[0059] After the dataset is prepared, it is divided into a training set and a test set in a ratio of 0.6:0.4.

[0060] The inputs of the energy saving amount prediction proxy model include morphological parameters such as building floor area ratio (FAR), building density (BD), open space ratio (OSR), average building floor number (AF), and skewness (SD), residential lighting before and after the reconstruction, equipment power density difference, glass SHGC difference, and photovoltaic panel power, and the output is the energy saving amount of the residential area after the reconstruction. The XGBoost energy saving amount prediction proxy model before and after the reconstruction is obtained through training, and the prediction result R 2 reaches 0.82. The rapid prediction of the energy saving amount after the reconstruction is realized while the accuracy of the prediction result is ensured, so as to further guide the selection of the reconstruction strategy and promote the development of low-carbon residential areas.

[0061] The above describes the preferred embodiments of the present application in detail. It should be understood that those skilled in the art can make many modifications and changes without creative labor according to the concept of the present application. Therefore, any technical solution obtained by logical analysis, reasoning, or limited experiment on the basis of the prior art according to the concept of the present application should be within the protection scope determined by the claims.

Claims

1. A rapid digital prediction method for the potential of urban low-carbon renewal based on machine learning, characterized in that, Includes the following steps: S1, Construct an ideal residential area model; S2, performance simulation based on an ideal residential area model; S3, energy-saving renovation of an ideal residential area model; S4, to perform performance simulation on the modified residential area model; S5. Based on the residential area models before and after renovation and their performance simulation data, a dataset is constructed. The building parameters before and after renovation are used as the input of the XGBoost-based energy-saving prediction proxy model, and the energy saving of the renovated residential area is used as the model output. The model is trained using the dataset and then used for predicting the low-carbon renovation potential of building groups. S1 includes the following steps: S11, Analyze the actual building types in the residential area and determine the basic form of the residential buildings; S12, based on the actual size of the residential area, divide the residential area into multiple block units, and offset the boundary of the block unit into the interior of the residential area by a preset distance to obtain the road boundary and the ideal buildable land area of ​​the residential area; S13, the basic form of residential buildings is combined in multiple ways, and the residential meta-model is generated by combining the buildable land area of ​​the ideal residential area to construct the ideal residential area model; The basic architectural forms of the residential area include point-type residential clusters, slab-type residential clusters, and enclosed residential clusters. The inputs to the energy-saving prediction proxy model include building volume ratio, building density, open space index, average number of building floors, staggeredness, residential lighting before and after renovation, equipment power density difference, glass SHGC difference, and photovoltaic panel power. The buildable area of ​​the street grid is determined by randomly selecting the type of housing. Different types of housing in the basic form module are combined in a multi-dimensional way to automatically generate a variety of residential area meta-models. The way to automatically generate a variety of residential area meta-models is to randomly select a type of small square housing area from the parameters, and then select nine of them in a row to form a nine-square housing area, which is the residential area meta-model.

2. The method for rapid digital prediction of urban low-carbon renewal potential based on machine learning according to claim 1, characterized in that, The point-style residential complexes are divided into five types. Among them, type P-1 consists of small-sized residential buildings with one elevator serving two households in the early stages of urban construction. The simplified P-1 type buildings have a north-south width of 14m and an east-west width of 11m. They are arranged in rows of 6 buildings in the residential area, with the number of floors ranging from 3 to 7. Type P-2 residential buildings have a north-south width of 27m and an east-west width of 12m. They are typical large-sized residential buildings with one elevator serving two households. They are arranged in rows in the residential area, with the number of floors ranging from 3 to 16. Type P-3 buildings have a staggered layout. The buildings are Y-shaped to maximize south-facing sunlight, with the number of floors ranging from 8 to 25. Type P-4 buildings have a square staggered layout. The buildings have a north-south width of 20m and an east-west width of 13m, with the number of floors ranging from 8 to 25. Type P-5 residential buildings are hotel-style apartment buildings with a single-point layout. The north-south width is 30m and the east-west width is 26m.

3. The method for rapid digital prediction of urban low-carbon renewal potential based on machine learning according to claim 1, characterized in that, The slab-type residential complexes are divided into two categories, with differences in building density and floor number restrictions. The S-1 type slab-type residential complex has a north-south width of 64m and an east-west width of 11m. It is a typical low-rise and multi-story residential complex with two elevators and four households per floor, and the number of floors ranges from 3 to 7. The S-2 type is a typical row house, with a north-south width of 56m and an east-west width of 11m, and the number of floors ranges from 8 to 25.

4. The method for rapid digital prediction of urban low-carbon renewal potential based on machine learning according to claim 1, characterized in that, The enclosed residential complexes are divided into two types: C-1 is an open-style enclosed complex with a building floor range of 3-16 floors; C-2 is a U-shaped residential complex with a building floor range of 3-16 floors.

5. The method for rapid digital prediction of urban low-carbon renewal potential based on machine learning according to claim 1, characterized in that, The performance simulations include outdoor microclimate, street energy consumption, and solar radiation simulations.

6. The method for rapid digital prediction of urban low-carbon renewal potential based on machine learning according to claim 1, characterized in that, The energy-saving renovation includes using energy-saving lighting fixtures, low-energy electrical appliances, glass with low solar heat gain coefficient, and adding solar photovoltaic panels to the roof.

7. The method for rapid digital prediction of urban low-carbon renewal potential based on machine learning according to claim 1, characterized in that, The dataset includes building form parameters, residential lighting before and after renovation, equipment power density difference, glass SHGC difference, photovoltaic panel power, and residential energy use intensity before and after renovation. The building form parameters include building volume ratio, building density, open space index, average number of building floors, and staggeredness.

Images