A rapid prediction method for urban energy consumption based on building type identification

Abstract

The present invention discloses a method for rapid prediction of urban energy consumption based on building type identification, which relates to the technical field of urban energy consumption management. The building type identification model of the present invention is constructed using a deep learning algorithm, with multi-source information, namely building feature data and surrounding environment information data, as input, and the deep learning algorithm is used to automatically identify the building type of each building, ensuring that the classification accuracy is still high even when the data is incomplete or inaccurate. Using typical building energy consumption data of different types of buildings, a unit area energy consumption calculation standard is provided for each building type, and regional building energy consumption is predicted based on the building type obtained by the building type identification model, and an urban energy consumption distribution map is generated for more accurate energy consumption management. This method applies the building type identification model to regional energy consumption prediction, and has data universality and scalability, making it applicable in a variety of urban environments.

Description

Technical Field

[0001] The present invention relates to the technical field of urban energy consumption management, and in particular to a method for quickly predicting urban energy consumption based on building type identification. Background Art

[0002] With the rapid development of modern society, building energy consumption has become a major issue in urban planning and energy management. However, current energy consumption prediction methods often rely on aggregated macro data or static analysis of single building characteristics. Top-down energy consumption models are usually based on overall economic and demographic data for prediction, which makes it difficult to capture the specific energy consumption differences of individual buildings, resulting in inaccurate prediction results. While bottom-up energy consumption models can provide more detailed energy consumption estimates based on the physical characteristics of specific buildings, they require a large amount of building data as support. In practice, they are limited by incomplete data, especially the lack of building types, which directly affects the model's prediction accuracy.

[0003] Currently, building type classification lacks detailed and dynamic classification, making it difficult to adapt to the complex urban built environment. While some methods offer simple classification based on static features such as building shape, height, and area, these features have limited accuracy in identifying building types, particularly in multi-functional buildings and complex building complexes, leading to significant errors in energy consumption estimates. Furthermore, existing classification methods are often rigid and unable to adapt to the diverse needs of different building types in real-world applications within urban areas.

[0004] Therefore, how to improve the accuracy of energy consumption prediction based on building type segmentation and realize dynamic building energy consumption prediction under limited data has become a technical problem that needs to be solved urgently. Summary of the Invention

[0005] In response to the shortcomings of the existing technology, the present invention provides a method for rapid prediction of urban energy consumption based on building type identification. By integrating and processing multi-source data, it can achieve accurate prediction of urban building energy consumption, provide scientific decision-making support for urban planning and energy management, and solve the problem of low accuracy in urban building energy consumption prediction in the existing technology, especially when the building feature data is incomplete, which makes it difficult to achieve efficient energy consumption assessment.

[0006] A method for rapid prediction of urban energy consumption based on building type identification includes the following steps:

[0007] S1. Establish a multi-source information dataset; the multi-source information dataset includes a plurality of samples, each of which includes architectural feature data, surrounding environment information, and building type labels of a certain building;

[0008] S1.1. Acquire architectural feature data of a plurality of buildings and preprocess the architectural feature data to obtain preprocessed architectural feature data;

[0009] The building feature data includes a building outline image X geo , building height H, building area A geo , number of floors p, shape coefficient SCB, building density BD of the plot where the building is located and floor area ratio FAR of the plot where the building is located;

[0010] The method for preprocessing the building feature data is as follows: binarizing the building outline image and converting it into a single-channel image format;

[0011] S1.2. Obtain the surrounding environment information of several buildings in S1.1;

[0012] The surrounding environment information includes the type of m points of interest (POIs) closest to the building. type , the distance between the building and the m nearest points of interest (POIs), the types of the q roads nearest to the building, and the land use type of the plot where the building is located; the distance between the building and the m nearest points of interest (POIs) is the Euclidean distance;

[0013] S1.3. Obtain the building type label of each building, and use the pre-processed building feature data, surrounding environment information, and building type label corresponding to each building as a sample to construct a multi-source information dataset including several samples;

[0014] S2: Construct a building type recognition model and train it using a multi-source information dataset to obtain a trained building type recognition model;

[0015] The building type recognition model includes an image feature and vector feature extraction module, a feature fusion module and a high-level feature extraction and classification module;

[0016] The image feature and vector feature extraction module is used to extract features from the input vector features and building outline image to obtain a high-dimensional space representation and a high-dimensional feature map respectively;

[0017] The feature fusion module is used to fuse the high-dimensional feature map and high-dimensional space representation output by the image feature and vector feature extraction module by superposition to obtain fused features;

[0018] The high-level feature extraction and classification module first downsamples the fused features through a convolutional layer and a ReLU activation function layer to obtain a comprehensive feature vector; then it is input into a fully connected layer and a Softmax activation function to calculate the score of each building type and convert it into the classification probability of each building type, thereby obtaining the identified building type;

[0019] Furthermore, the image feature and vector feature extraction module includes two processing units, the first processing unit performs one upsampling on the input vector feature to obtain the upsampled vector feature, and then performs several downsampling on the upsampled vector feature, and generates vector features of different spatial dimensions after each downsampling, and the vector feature obtained after the last downsampling is called a high-dimensional space representation; the second processing unit performs one preliminary downsampling on the building outline image to obtain a feature map, and the spatial dimension of the feature map is the same as the spatial dimension of the vector feature obtained by one upsampling in the first processing unit, and then the feature map obtained by the downsampling is added and fused with the vector feature obtained by one upsampling in the first processing unit to obtain a fused feature; the fused feature continues to be downsampled several times, and the fused feature after each downsampling is fused with the vector feature obtained by the corresponding downsampling in the first processing unit, and finally a high-dimensional feature map is obtained;

[0020] Furthermore, the upsampling is implemented by a deconvolution layer and a ReLU activation function layer; the downsampling is implemented by a convolution layer and a ReLU activation function layer;

[0021] Furthermore, the spatial dimension of the fused features fused in the second processing unit is the same as the spatial dimension of the vector features obtained after downsampling; the spatial dimension of the high-dimensional space representation is the same as the spatial dimension of the high-dimensional feature map X D The spatial dimensions are the same;

[0022] Furthermore, the input vector features include building height H, building area A geo , number of floors p, building density BD of the plot where the building is located, volume ratio FAR of the plot where the building is located, shape coefficient SCB, type of the q roads closest to the building, land use type of the plot where the building is located, type of the m points of interest POI closest to the building type and the distance between the building and its nearest m points of interest (POIs);

[0023] X v ={H,A geo ,p,BD,FAR,SCB,R type ,L type ,(POI typei ,D POIi |i=1,2,……,m)}

[0024] Among them, X v is the input vector feature, R type is the type of road, L type is the land use type, POI typei Indicates the type of the i-th interest point among the m interest points closest to the building, DPOIi Represents the distance between a building and the i-th interest point among the m nearest interest points, where i is the number of the interest point;

[0025] Furthermore, the classification probability of each building type is:

[0026]

[0027] in, is the classification probability of the Tth building type, T is the building type number, W k and W T is the weight vector, V final is the comprehensive feature vector, M is the number of building types;

[0028] S3: Obtain architectural feature data and surrounding environment information of the building to be identified and input them into the trained building type recognition model to obtain the identified building type;

[0029] S3.1: Collect architectural feature data and surrounding environment information of the building to be identified, construct vector features based on the architectural feature data and surrounding environment information, and standardize the building outline images in the vector features and architectural feature data;

[0030] The standardization specifically includes: performing binarization processing on the building outline, converting it into a single-channel image format, and scaling the vector features to the input range of the building type recognition model;

[0031] S3.2: Input the normalized vector features and the building outline image into the trained building type recognition model to obtain the recognized building type;

[0032] S4: For each building type, collect the actual unit area energy consumption data of several buildings of that building type, calculate the average unit area energy consumption value of each building type, and use it as the energy consumption benchmark for typical buildings;

[0033]

[0034] in, is the average energy consumption per unit area of ​​each building type, Z is the number of buildings of each building type, e Tz is the actual energy consumption per unit area of ​​the zth building in the Tth building type;

[0035] S5: Based on the identified building type and typical building energy consumption benchmark, calculate the energy consumption prediction result of the area where the building to be identified is located;

[0036] S5.1: Calculate the total energy consumption E for each building based on the building type and the typical building energy consumption benchmark corresponding to that building type. n ;

[0037]

[0038] Among them, A geon is the floor area of ​​the nth building; p n is the number of floors of the nth building; Building type T identified for the nth building n Typical building energy consumption benchmark, n is the building number;

[0039] S5.2: Take the total energy consumption of the building E n Sum up to calculate the regional overall building energy consumption prediction result E total ;

[0040]

[0041] Where N is the number of buildings.

[0042] Compared with the prior art, the present invention has the following beneficial effects:

[0043] Existing methods often rely on large-scale, high-quality datasets for training. However, in practice, data loss and incomplete classification are common problems, affecting the reliability of energy consumption prediction. This paper proposes a rapid urban energy consumption prediction method based on building type identification, which comprehensively utilizes multi-source information to improve prediction accuracy. Specifically:

[0044] 1. The building type recognition model of the present invention is constructed using a deep learning algorithm. It takes multi-source information, namely building feature data and surrounding environment information data, as input, and uses the deep learning algorithm to automatically identify the building type of each building, ensuring high classification accuracy even when the data is incomplete or inaccurate.

[0045] 2. Using typical building energy consumption data of different types of buildings, a unit area energy consumption calculation standard is provided for each building type. Regional building energy consumption is predicted based on the building type obtained by the building type identification model, and an urban energy consumption distribution map is generated for more accurate energy consumption management. This method applies the building type identification model to regional energy consumption prediction, and has data universality and scalability, making it applicable in a variety of urban environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 This is a flow chart of a method for rapid prediction of urban energy consumption based on building type identification in an embodiment of the present invention;

[0047] Figure 2 Schematic diagram of the structure of the building type recognition model in an embodiment of the present invention. DETAILED DESCRIPTION

[0048] The present invention is described in detail below with reference to the accompanying drawings and embodiments.

[0049] A rapid prediction method for urban energy consumption based on building type identification, such as Figure 1 As shown, the following steps are included:

[0050] S1. Establish a multi-source information dataset; the multi-source information dataset includes a plurality of samples, each of which includes architectural feature data, surrounding environment information, and building type labels of a certain building;

[0051] S1.1. Acquire architectural feature data of a plurality of buildings and preprocess the architectural feature data to obtain preprocessed architectural feature data;

[0052] The building feature data includes a building outline image X geo , building height H, building area A geo , number of floors p, shape coefficient SCB, building density BD of the plot where the building is located and floor area ratio FAR of the plot where the building is located;

[0053] The method for preprocessing the building feature data is as follows: binarizing the building outline image and converting it into a (128×128×1) single-channel image format;

[0054] The building density BD is calculated as follows:

[0055]

[0056] Among them, q is the total number of buildings on the plot where the building is located, f is the building number, S f Represents the floor area of ​​the building; A represents the land area of ​​the plot where the building is located;

[0057] The volume ratio FAR is calculated as follows:

[0058]

[0059] Among them, N f Represents the number of floors of the building;

[0060] The shape coefficient SCB is used to characterize the geometric characteristics of different building types, and the calculation formula is:

[0061]

[0062] Among them, AREA geo is the building surface area, V is the building volume;

[0063] S1.2. Obtain the surrounding environment information of several buildings in S1.1;

[0064] The surrounding environment information includes the type of m points of interest (POIs) closest to the building. type , the distance between the building and the m nearest points of interest (POIs), the types of the q roads nearest to the building, and the land use type of the plot where the building is located; the distance between the building and the m nearest points of interest (POIs) is the Euclidean distance;

[0065] Type of point of interest (POI) in this embodiment type Including residential (D res ), Office (D off ), Business (D com )、Medical (D hos ), Shops(D shop ) and school (D edu ) etc.; Type of road (R type ) including main roads (R major ) and secondary roads (R minor ) etc.; the land use types of the plots include residential (L res ), Office(L off ), Business (L com )、Medical(L hos ), Shops(L shop ) and school (L edu )wait.

[0066] S1.3. Obtain the building type label of each building, and use the pre-processed building feature data, surrounding environment information, and building type label corresponding to each building as a sample to construct a multi-source information dataset including several samples;

[0067] S2: Construct a building type recognition model and train it using a multi-source information dataset to obtain a trained building type recognition model;

[0068] like Figure 2 As shown, the building type recognition model includes an image feature and vector feature extraction module, a feature fusion module and a high-level feature extraction and classification module;

[0069] The image feature and vector feature extraction module is used to extract the input vector feature X V Perform feature extraction on the building outline image to obtain high-dimensional space representation and high-dimensional feature map respectively;

[0070] Specifically, the image feature and vector feature extraction module includes two processing units. The first processing unit processes the input vector feature X. VPerform one upsampling to obtain the upsampled vector features, and then perform several downsampling on the upsampled vector features, and generate vector features of different spatial dimensions after each downsampling, and the vector features obtained after the last downsampling are called high-dimensional space representation; the second processing unit performs one preliminary downsampling on the building outline image to obtain a feature map, and the spatial dimension of the feature map is the same as the spatial dimension of the vector features obtained by one upsampling in the first processing unit, completing the preliminary alignment, and then adding and fusing the feature map obtained after downsampling with the vector features obtained by one upsampling in the first processing unit to obtain a fused feature; the fused feature continues to be downsampled several times, and the fused features after each downsampling are fused with the vector features obtained by the corresponding downsampling in the first processing unit, and finally a high-dimensional feature map X is obtained. D ;

[0071] The upsampling is achieved by a deconvolution layer and a ReLU activation function layer; the downsampling is achieved by a convolution layer and a ReLU activation function layer;

[0072] The spatial dimension of the fused feature to be fused in the second processing unit is the same as the spatial dimension of the vector feature obtained after downsampling; the spatial dimension of the high-dimensional space representation is the same as the spatial dimension of the high-dimensional feature map X D The spatial dimensions are the same;

[0073] In this embodiment, the convolution kernels of the convolution layer of the second processing unit are 1×1, 3×3, and 5×5. The sizes of the fused features after three feature extractions are (32×32×192), (16×16×480), and (8×8×832), respectively.

[0074] The input feature vector X V Including building height H, building area A geo , number of floors p, building density BD of the plot where the building is located, volume ratio FAR of the plot where the building is located, shape coefficient SCB, type of the q roads closest to the building, land use type of the plot where the building is located, type of the m points of interest POI closest to the building type and the distance between the building and its nearest m points of interest (POIs);

[0075] X v ={H,A geo ,p,BD,FAR,SCB,R type ,L type ,(POI typei ,D POIi |i=1,2,3,4)}

[0076] Among them, R type is the type of road, Ltype is the land use type, POI typei Indicates the type of the i-th interest point among the m interest points closest to the building, D POIi Represents the distance between a building and the i-th interest point among the m nearest interest points, where i is the number of the interest point;

[0077] In this embodiment, m=4, and the 1×16×1 vector feature X is first extracted in the image feature and vector feature extraction module. V Expanded to 32×32×192 vector features To ensure consistency with the image features in spatial dimensions, the vector features are then downsampled to 16×16×480, and further downsampled to 8×8×832.

[0078] The feature fusion module is used to fuse the high-dimensional feature map and high-dimensional space representation output by the image feature and vector feature extraction module by superposition to obtain the fused feature F fusion , ensuring that the two feature types are fused in the same spatial dimension, providing more comprehensive information support for the final building type identification;

[0079] The high-level feature extraction and classification module first uses the convolution layer and the ReLU activation function layer to extract the fused feature F fusion Downsampling is performed to obtain a 4×4×1024 comprehensive feature vector V final ;Then it is input into the fully connected layer and the Softmax activation function to calculate the score of each building type and convert it into the classification probability of each building type, thereby obtaining the identified building type;

[0080] For example, if the classification probability of a residence is If the value is the largest, then the building type of the input building is predicted to be residential;

[0081] The comprehensive feature vector is:

[0082] V final =Conv(F fusion )

[0083] Among them, Conv means convolution;

[0084] The classification probability of each building type is:

[0085]

[0086] in, is the classification probability of the Tth building type, T is the building type number, W k and W Tis a weight vector, which is a parameter learned by the deep learning model during the training process. It is used to calculate the classification probability of building types. M is the number of building types. This patent claims to support M building types, but in this case, M = 6 is the total number of classifications, including residential (T res ), Office(T off ), Business (Tc om )、Medical(T hos) 、Shops(T shop ) and school (T edu ).

[0087] S3: Obtain architectural feature data and surrounding environment information of the building to be identified and input them into the trained building type recognition model to obtain the identified building type;

[0088] S3.1: Collect architectural feature data and surrounding environment information of the building to be identified, construct vector features based on the architectural feature data and surrounding environment information, and standardize the building outline images in the vector features and architectural feature data;

[0089] The standardization specifically includes: binarizing the building outline, converting it into a single-channel image format of (128×128×1), and scaling the vector features to the input range of the building type recognition model to ensure compatibility with the building type recognition model;

[0090] S3.2: Input the normalized vector features and the building outline image into the trained building type recognition model to obtain the recognized building type;

[0091] S4: For each building type, collect the actual unit area energy consumption data of several buildings of this building type. T , calculate the average energy consumption per unit area for each building type As a benchmark for typical building energy consumption;

[0092] In this implementation, 100 buildings of each type, including residential, office, commercial, medical, shop and school, are collected, and the actual energy consumption data per unit area of ​​each building is recorded. T , where T∈{residential, office, commercial, medical, shop, school}

[0093]

[0094] Among them, e Tz is the actual energy consumption per unit area of ​​the zth building in the Tth building type;

[0095] S5: Based on the identified building type and typical building energy consumption benchmark, calculate the energy consumption prediction result of the area where the building to be identified is located;

[0096] S5.1: Calculate the total energy consumption E for each building based on the building type and the typical building energy consumption benchmark corresponding to that building type. n ;

[0097]

[0098] Among them, A geon is the floor area of ​​the nth building; p n is the number of floors of the nth building; Building type T identified for the nth building n Typical building energy consumption benchmark, n is the building number;

[0099] S5.2: Take the total energy consumption of the building E n Sum up to calculate the regional overall building energy consumption prediction result E total ;

[0100]

[0101] Where N is the number of buildings. Through this step, a regional energy consumption map can be drawn and high-density energy consumption areas can be effectively identified to provide support for urban planning and energy management.

Claims

1. A method for rapid prediction of urban energy consumption based on building type identification, characterized in that: The steps include: S1. Establish a multi-source information dataset; the multi-source information dataset includes a plurality of samples, each of which includes architectural feature data, surrounding environment information, and building type labels of a certain building; S2: Construct a building type recognition model and train it using a multi-source information dataset to obtain a trained building type recognition model; The building type recognition model includes an image feature and vector feature extraction module, a feature fusion module and a high-level feature extraction and classification module; The image feature and vector feature extraction module is used to extract features from the input vector features and building outline image to obtain a high-dimensional space representation and a high-dimensional feature map respectively; The feature fusion module is used to fuse the high-dimensional feature map and high-dimensional space representation output by the image feature and vector feature extraction module by superposition to obtain fused features; The high-level feature extraction and classification module first downsamples the fused features through a convolutional layer and a ReLU activation function layer to obtain a comprehensive feature vector; then it is input into a fully connected layer and a Softmax activation function to calculate the score of each building type and convert it into the classification probability of each building type, thereby obtaining the identified building type; The input vector features include building height H, building area A geo , number of floors p, building density BD of the plot where the building is located, volume ratio FAR of the plot where the building is located, shape coefficient SCB, type of the q roads closest to the building, land use type of the plot where the building is located, type of the m points of interest POI closest to the building type and the distance between the building and its nearest m points of interest (POIs); X v ={H,A geo ,p,BD,FAR,SCB,R type ,L type ,(POI typei ,D POIi |i=1,2,……,m)} Among them, X v is the input vector feature, R type is the type of road, L type is the land use type, POI typei Indicates the type of the i-th interest point among the m interest points closest to the building, D POIi Represents the distance between a building and the i-th interest point among the m nearest interest points, where i is the number of the interest point; The image feature and vector feature extraction module includes two processing units. The first processing unit performs one upsampling on the input vector feature to obtain the upsampled vector feature, and then performs several downsampling on the upsampled vector feature, and generates vector features of different spatial dimensions after each downsampling. The vector feature obtained after the last downsampling is called a high-dimensional space representation; the second processing unit performs one preliminary downsampling on the building outline image to obtain a feature map, and the spatial dimension of the feature map is the same as the spatial dimension of the vector feature obtained by one upsampling in the first processing unit, and then the feature map obtained by the downsampling is added and fused with the vector feature obtained by one upsampling in the first processing unit to obtain a fused feature; the fused feature continues to be downsampled several times, and the fused feature after each downsampling is fused with the vector feature obtained by the corresponding downsampling in the first processing unit, and finally a high-dimensional feature map is obtained; The upsampling is achieved by a deconvolution layer and a ReLU activation function layer; the downsampling is achieved by a convolution layer and a ReLU activation function layer; The spatial dimension of the fused feature to be fused in the second processing unit is the same as the spatial dimension of the vector feature obtained after downsampling; the spatial dimension of the high-dimensional space representation is the same as the spatial dimension of the high-dimensional feature map X D The spatial dimensions are the same; S3: Obtain architectural feature data and surrounding environment information of the building to be identified and input them into the trained building type recognition model to obtain the identified building type; S4: For each building type, collect the actual unit area energy consumption data of several buildings of that building type, calculate the average unit area energy consumption value of each building type, and use it as the energy consumption benchmark for typical buildings; S5: Based on the identified building type and typical building energy consumption benchmark, calculate the energy consumption prediction result of the area where the building to be identified is located.

2. The method for rapid prediction of urban energy consumption based on building type identification according to claim 1 is characterized in that: S1 specifically includes: S1.

1. Acquire architectural feature data of a plurality of buildings and preprocess the architectural feature data to obtain preprocessed architectural feature data; The building feature data includes a building outline image X geo , building height H, building area A geo , number of floors p, shape coefficient SCB, building density BD of the plot where the building is located and floor area ratio FAR of the plot where the building is located; S1.

2. Obtain the surrounding environment information of several buildings in S1.1; The surrounding environment information includes the type of m points of interest (POIs) closest to the building. type , the distance between the building and the m nearest points of interest (POIs), the types of the q nearest roads to the building, and the land use type of the plot where the building is located; the distance is the Euclidean distance; S1.

3. Obtain the building type label of each building, and use the pre-processed building feature data, surrounding environment information, and building type label corresponding to each building as a sample to construct a multi-source information dataset including several samples.

3. The method for rapid prediction of urban energy consumption based on building type identification according to claim 2 is characterized in that: The method for preprocessing building feature data described in S1.1 is: binarizing the building outline image and converting it into a single-channel image format.

4. The method for rapid prediction of urban energy consumption based on building type identification according to claim 1 is characterized in that: The classification probability of each building type is: in, is the classification probability of the Tth building type, T is the building type number, W k and W T is the weight vector, V final is the comprehensive feature vector, and M is the number of building types.

5. The method for rapid prediction of urban energy consumption based on building type identification according to claim 1 is characterized in that: S3 specifically includes: S3.1: Collect architectural feature data and surrounding environment information of the building to be identified, construct vector features based on the architectural feature data and surrounding environment information, and standardize the building outline images in the vector features and architectural feature data; The standardization specifically includes: performing binarization processing on the building outline, converting it into a single-channel image format, and scaling the vector features to the input range of the building type recognition model; S3.2: Input the standardized vector features and building outline image into the trained building type recognition model to obtain the recognized building type.

6. The method for rapid prediction of urban energy consumption based on building type identification according to claim 1 is characterized in that: The average energy consumption per unit area for each building type described in S4 is: in, is the average energy consumption per unit area of ​​each building type, Z is the number of buildings of each building type, e Tz is the actual energy consumption per unit area of ​​the zth building in the Tth building type.

7. The method for rapid prediction of urban energy consumption based on building type identification according to claim 1 is characterized in that: S5 specifically includes: S5.1: Calculate the total energy consumption E for each building based on the building type and the typical building energy consumption benchmark corresponding to that building type. n ; Among them, A geon is the floor area of ​​the nth building; p n is the number of floors of the nth building; Building type T identified for the nth building n Typical building energy consumption benchmark, n is the building number; S5.2: Take the total energy consumption of the building E n Sum up to calculate the regional overall building energy consumption prediction result E total ; Where N is the number of buildings.

Images