Building performance prediction method and evaluation system based on hybrid neural network

Abstract

The invention relates to the technical field of intelligent buildings, and provides a hybrid neural network-based building performance prediction method, which comprises the steps of capturing material price data in real time; the hybrid neural network model comprises a price feature conversion unit which converts material price data into price features; the first long short-term memory network is used for extracting energy consumption time characteristics from the historical energy consumption data; the second long short-term memory network is used for extracting climate time characteristics from the site climate data; the three-dimensional convolutional neural network is used for extracting three-dimensional space features of the building from the three-dimensional volume data; the splicing module is used for obtaining comprehensive feature representation; the full connection module is used for extracting nonlinear features from the comprehensive feature representation; and the output module is used for carrying out regression processing on the nonlinear characteristics to obtain the energy consumption, the cost and the full-life-cycle carbon emission of the building. The invention further discloses a building performance evaluation system based on the hybrid neural network. According to the invention, the precision and the practical adaptability of building multi-performance prediction are improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent building technology, and in particular to a building performance prediction method and evaluation system based on a hybrid neural network. Background Art

[0002] Buildings consume 40% of the world's energy. Effectively controlling building energy consumption and carbon emissions is the main direction of the development of near-zero energy buildings. At the same time, accurate estimation of building costs is very important for building operations. Therefore, it is necessary to accurately predict the energy consumption, carbon emissions and costs of buildings.

[0003] In the prior art, there have been studies on multi-objective optimization methods for building energy consumption, carbon emissions, costs, etc. For example, the Chinese patent with publication number CN118551669A discloses a multi-objective energy-saving optimization method for building envelope structures based on climate prediction. The patent uses simulation parameters and simulation results to determine the sample set. According to the sample set training, a building performance prediction model based on BP neural network is obtained, and the building energy consumption, carbon emissions, thermal discomfort time and cost are predicted by the building performance prediction model. The simulation parameters include site climate data, maintenance structure design parameters, indoor thermal environment, etc. The patent combines site climate data to predict multiple building performances. However, its sample data is a simulation result rather than actual measured data, which is not close to reality, and does not introduce real material prices, resulting in errors in the predicted performance. At the same time, the performance parameters of the building, such as carbon emissions, costs, and energy consumption, are closely related to the material properties of the components in the building and the building space characteristics, but the patent does not introduce the spatial characteristics of the building and the material properties of each component in the building performance prediction, so the building performance prediction results still have room for optimization and improvement. Summary of the invention

[0004] The present invention aims to at least solve the technical problem that the sample data in the existing building performance prediction method is the simulation result rather than the actual measured data, and the spatial characteristics of the building and the material properties of each component are not introduced, so that the building performance prediction result also has the optimization and improvement space, and provide a building performance prediction method and evaluation system based on a hybrid neural network.

[0005] In order to achieve the above-mentioned purpose of the present invention, according to the first aspect of the present invention, the present invention provides a building performance prediction method based on a hybrid neural network, comprising: real-time capture of material price data from multiple building material supplier websites; obtaining historical energy consumption data, site climate data and BIM model data of the building; extracting three-dimensional volume data of the building from the BIM model data; using a pre-trained hybrid neural network model to predict the energy consumption, cost and life cycle carbon emissions of the building based on the material price data, historical energy consumption data, site climate data and three-dimensional volume data, wherein the hybrid neural network model comprises: a price feature conversion unit, used to convert the material price data into price features; a first long short-term memory network, used to extract energy consumption time features from historical energy consumption data; a second long short-term memory network, used to extract climate time features from site climate data; a three-dimensional convolutional neural network, used to extract three-dimensional spatial features of the building from the three-dimensional volume data; a splicing module, splicing energy consumption time features, climate time features, price features and three-dimensional spatial features to obtain a comprehensive feature representation; a fully connected module, including more than one cascaded fully connected layer for extracting nonlinear features from the comprehensive feature representation; an output module, performing regression processing on the nonlinear features to obtain the energy consumption, cost and life cycle carbon emissions of the building.

[0006] In order to achieve the above-mentioned purpose of the present invention, according to the second aspect of the present invention, the present invention provides a building performance evaluation system based on a hybrid neural network, which is used to implement a building performance prediction method based on a hybrid neural network described in the first aspect of the present invention, including: a first data acquisition module, which captures material price data from multiple building material supplier websites in real time; a second data acquisition module, which acquires the historical energy consumption data, site climate data and BIM model data of the building; a third data acquisition module, which extracts the three-dimensional volume data of the building from the BIM model data; a prediction module, which uses a pre-trained hybrid neural network model to predict the energy consumption, cost and life cycle of the building based on the material price data, historical energy consumption data, site climate data and three-dimensional volume data The hybrid neural network model includes: a price feature conversion unit, which is used to convert material price data into price features; a first long short-term memory network, which is used to extract energy consumption time features from historical energy consumption data; a second long short-term memory network, which is used to extract climate time features from site climate data; a three-dimensional convolutional neural network, which is used to extract three-dimensional spatial features of the building from three-dimensional volume data; a splicing module, which splices energy consumption time features, climate time features, price features and three-dimensional spatial features to obtain a comprehensive feature representation; a fully connected module, which includes more than one cascaded fully connected layer for extracting nonlinear features from the comprehensive feature representation; and an output module, which performs regression processing on the nonlinear features to obtain the energy consumption, cost and life cycle carbon emissions of the building.

[0007] The beneficial technical effects of the present invention are as follows: the present invention introduces material price data captured in real time from the websites of multiple building material suppliers to predict construction costs, fully considering the impact of material price fluctuations on construction costs and being in line with reality; the measured historical energy consumption data is used to extract energy consumption time characteristics through the first long short-term memory network, and the climate time characteristics are obtained by using the second long short-term memory network, which can capture the changing laws of building performance (especially energy consumption) under different seasons and meteorological conditions, and improve the prediction accuracy and practical adaptability; in the three-dimensional volume data, the cubic voxel units are mapped and associated with the material type and the material physical properties, so that the three-dimensional convolutional neural network can better extract the geometric spatial characteristics and material distribution of the building, so that the cost and carbon emissions predicted by the hybrid neural network model are more accurate and more practically adaptable; the present invention realizes high-precision and high-practical adaptability of building multi-performance prediction through the integration of multi-source data and intelligent optimization technology, and improves the prediction and optimization capabilities of building design costs, energy consumption and carbon emissions throughout the life cycle. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Figure 1 It is a flow chart of a method for predicting building performance based on a hybrid neural network in a preferred embodiment of the present invention;

[0009] Figure 2 It is a structural schematic diagram of a hybrid neural network model in a preferred embodiment of the present invention;

[0010] Figure 3 is a structural block diagram of an evaluation system in a preferred embodiment of the present invention;

[0011] Figure 4 It is a specific flow chart of a building performance prediction method based on a hybrid neural network in an application scenario of the present invention. DETAILED DESCRIPTION

[0012] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0013] In the description of the present invention, it is necessary to understand that the terms "longitudinal", "lateral", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", etc., indicating the orientation or position relationship, are based on the orientation or position relationship shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention.

[0014] In the description of the present invention, unless otherwise specified and limited, it should be noted that the terms "installed", "connected" and "connected" should be understood in a broad sense. For example, it can be a mechanical connection or an electrical connection, or it can be the internal connection between two components. It can be a direct connection or an indirect connection through an intermediate medium. For ordinary technicians in this field, the specific meanings of the above terms can be understood according to the specific circumstances.

[0015] The execution subject of the building performance prediction method based on a hybrid neural network provided by the present invention includes but is not limited to at least one of the electronic devices such as a server, a terminal, etc. that can be configured to execute the method provided in the embodiment of the present application. In other words, the building performance prediction method based on a hybrid neural network can be executed by software or hardware installed on a terminal device or a server device, and the software can be a blockchain platform. The server includes but is not limited to: a single server, a server cluster, a cloud server or a cloud server cluster, etc. The server can be an independent server, or it can be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (Content Delivery Network, CDN), and big data and artificial intelligence platforms.

[0016] The present invention provides a building performance prediction method based on a hybrid neural network. In a preferred embodiment, Figure 1 As shown, the method includes:

[0017] Step S1, real-time acquisition of material price data from multiple building material supplier websites; acquisition of historical energy consumption data, site climate data and BIM model data of the building.

[0018] In this embodiment, target building material supplier websites are selected, preferably including but not limited to China Building Materials Network and Wanlong Building Materials Network. Web Scraping technology is used regularly to crawl web page data from the target building material supplier website, and by analyzing the web page structure, key data fields stored in HTML tags are identified to obtain price data of various types of building materials. In an example of material price data crawling, Python's Scrapy framework is used to automatically crawl price information in web pages. This information is usually contained in specific HTML elements (such as HTML tags) on the product details page. In the tag<spanclass="price"> ). The crawling process includes handling the website's paging logic to ensure that a complete data set is collected from each material product category. Material price data is numerical data.

[0019] In this embodiment, the historical energy consumption data of the building is not limited to the total power consumption of the air conditioning system, lighting, other equipment, etc. of the building in multiple time periods in the past, and can be collected and obtained through the energy detection system of the building (such as the total electric meter). The historical energy consumption data is stored in the form of time series, indicating the energy consumption changes of the building in different time periods, for example, the energy consumption trends of the day, week, and month.

[0020] In this implementation, climate factors will have a significant impact on the energy consumption and carbon emissions of buildings. Site climate data includes the temperature, humidity, solar radiation, etc. of the geographical location of the building. Site climate data can be obtained through a meteorological API or a meteorological monitoring system, and site climate data of different time periods are recorded to form time series data as input to capture the impact of climate conditions on building performance (such as energy consumption).

[0021] In this embodiment, a BIM model of a building is pre-built, and the BIM model is usually stored in a standard format (such as IFC or Revit format) to obtain BIM model data. The BIM model data includes geometric shape, material type, location, physical properties, etc.

[0022] Step S2, extracting the three-dimensional volume data of the building from the BIM model data.

[0023] In this embodiment, in one example, the processing of step S2 is preferably, but not limited to: voxelizing the BIM model of the building, converting the three-dimensional building design into a voxel grid, the grid size can be preset, the BIM model is discretized with a grid of the same size, and the obtained voxel grid data is used as three-dimensional volume data, so as to be able to express the three-dimensional spatial structure of the building in detail. In another example, the method of obtaining the three-dimensional volume data of the building is: projecting the building BIM model into multiple two-dimensional views (such as plan views and elevation views), and then inputting the multiple views as three-dimensional volume data into a two-dimensional CNN, or directly inputting them into a 3DCNN to utilize three-dimensional information.

[0024] Step S3, using a pre-trained hybrid neural network model to predict the energy consumption, cost and life cycle carbon emissions of the building based on material price data, historical energy consumption data, site climate data and three-dimensional volume data, where Figure 2 As shown, the hybrid neural network model includes:

[0025] The price feature conversion unit is used to convert material price data into price features. The price feature conversion unit is mainly used to standardize the price data of various types of materials, such as linearly mapping all prices to a certain numerical range for subsequent feature extraction and processing. The price feature is a one-dimensional vector, including the material code of each type of material and the standardized price value of each type of material.

[0026] The first long short-term memory network is used to extract energy consumption time features from historical energy consumption data. The first long short-term memory network is an LSTM network, which can learn and predict the time series changes of historical energy consumption data and obtain the energy consumption change trend.

[0027] The second long short-term memory network is used to extract climate time characteristics from the site climate data. The second long short-term memory network is an LSTM network, which can learn and predict the climate and enable the hybrid neural network model to learn how the changes in site climate over time affect the performance of the building, such as cost, energy consumption, and carbon emissions over the entire life cycle.

[0028] The 3D convolutional neural network is used to extract the 3D spatial features of the building from the 3D volume data. The 3D convolutional neural network is a 3D CNN network. The 3D spatial features obtained by the 3D convolutional neural network represent the geometric spatial features of the 3D voxel grid and material distribution.

[0029] The price feature conversion unit, the first long short-term memory network, the second long short-term memory network and the three-dimensional convolutional neural network can be executed in parallel to improve processing efficiency.

[0030] The splicing module splices the energy consumption time feature, climate time feature, price feature and three-dimensional space feature to obtain a comprehensive feature representation. The three-dimensional space feature output is a multi-dimensional feature tensor. The splicing module performs:

[0031] 1) Flatten the three-dimensional spatial features and convert them into a one-dimensional vector, which is expressed as follows:

[0032] flatten(x) = x.reshape(-1)

[0033] Here, x represents the three-dimensional spatial feature, and the reshape(-1) function means automatically calculating the row or column values ​​according to the total number of elements in x.

[0034] 2) The splicing module splices the one-dimensional vector obtained after flattening the three-dimensional spatial features with the energy consumption time features, climate time features, and price features to form a longer one-dimensional feature vector, which is recorded as a comprehensive feature representation. The comprehensive feature representation can contain both spatial and temporal information. The splicing module processing process can be expressed as:

[0035] Z = concatenate(Flatten(x CNN ),x LSTM1 ,x LSTM2 ,x cost )

[0036] Among them, x LSTM1 represents the energy consumption time characteristic of the first long short-term memory network output, x LSTM2 represents the climate time characteristics output by the second long short-term memory network, x cost represents the price feature, Flatten(x CNN ) represents the three-dimensional spatial features X of the building extracted by the three-dimensional convolutional neural network CNN Flattening is performed.

[0037] A fully connected module includes one or more cascaded fully connected layers for extracting nonlinear features from comprehensive feature representation. A fully connected module may include one fully connected layer or multiple cascaded fully connected layers. Each fully connected layer can be regarded as a linear transformation followed by a nonlinear activation function, which helps the model capture complex nonlinear relationships. The processing of the fully connected module can be expressed as:

[0038] Y=f(W·Z+b)

[0039] Wherein, Y represents a nonlinear feature, W and b represent a weight matrix and a bias matrix of a fully connected module, respectively, and f(·) represents an activation function of the fully connected module, preferably but not limited to a ReLU function.

[0040] The output module performs regression processing on nonlinear features to obtain the energy consumption, cost and carbon emissions of the building throughout its life cycle.

[0041] The output module can use the regression model to predict the energy consumption, cost and carbon emissions of the building at the current time. The output module may include a fully connected layer and an activation function processing unit.

[0042] In this embodiment, preferably, in order to improve the prediction accuracy, a material price data cleaning step is also included: using Python's Pandas library to clean the captured data, including removing non-numeric characters, processing missing values, removing duplicates, and unifying the data format for subsequent analysis and model training.

[0043] In a preferred embodiment, in order to facilitate the three-dimensional convolutional neural network to simultaneously extract the geometric space characteristics and material distribution characteristics of the building, the material and physical properties are simultaneously mapped to the voxel unit. In step S2, the three-dimensional volume data of the building is extracted from the BIM model data, including:

[0044] Step S21, discretize the BIM model of the building into a plurality of cubic voxel units by voxelization. In one example, the BIM model of the building is discretized into a plurality of cubic voxel units with the same cubic voxel size.

[0045] Step S22: extracting the material type of the material contained in each cubic pixel unit from the BIM model data, and mapping the material code corresponding to the material type and the material physical property vector to the cubic unit.

[0046] Material type and material physical properties are key parameters for calculating building costs, energy consumption and carbon emissions. Mapping them with cubic units helps three-dimensional convolutional neural networks accurately extract building geometric spatial characteristics and material distribution characteristics.

[0047] Material types are usually represented by discrete categorical variables, which can be integer-coded. Each material type is assigned a unique identifier (i.e., material code).

[0048] Material_type(x,y,z)=m

[0049] Where m is an integer corresponding to the material type at the center point of the cubic voxel (x,y,z). In an example, m=0 might represent air, m=1 might represent concrete, m=2 might represent steel, and so on.

[0050] Each type of material corresponds to a material physical property vector, and the material physical property vector is not limited to including material density, material thermal conductivity, material specific heat capacity, and material carbon emission factor.

[0051] Material density is a physical property that describes the relationship between a material's mass and volume and can be expressed as follows:

[0052] Density(x,y,z)=ρ

[0053] Where ρ represents the material density at the center point (x, y, z) of the cubic unit, usually in kg / m 3 .

[0054] The thermal conductivity of a material is the ratio of the heat flux per unit area of ​​the material to the temperature difference on both sides of the material, which is expressed by the following formula:

[0055] Thermal_Conductivity(x,y,z)=κ

[0056] Where κ represents the thermal conductivity of the cubic element with the center point at (x, y, z) and its unit is W / (m\cdotK).

[0057] The specific heat capacity of a material, expressed in J / (kg·K), is used in thermal performance analysis, such as evaluating the thermal buffering capacity of a material in energy optimization.

[0058] Carbon emission factor of the material (e c ) as numerical data, can be input into the model after being standardized.

[0059] In this implementation, each physical property in the material physical property vector may be stored in a corresponding cubic pixel unit in the form of a floating point number.

[0060] Step S23, each cubic voxel unit is represented as a three-dimensional array, wherein each three-dimensional array includes the coordinate position of the center point of the cubic voxel unit, the material code mapped to the cubic voxel unit, and the material physical property vector. Preferably, in order to reduce the redundancy of data storage and improve the processing efficiency of the model, the three-dimensional array in the three-dimensional volume data is represented as a sparse matrix:

[0061] SparseMatrix={(i,j,k,M,P|V i,j,k )}

[0062] Where (i, j, k) represents the cubic pixel unit V i,j,k The center point coordinate position, M represents the cubic pixel unit V i,j,k The mapped material type code, P represents the cubic voxel unit V i,j,k The mapped material physical property vector.

[0063] This implementation not only reduces redundant data but also speeds up data reading and processing by using a sparse matrix.

[0064] Step S24, combining the three-dimensional arrays of all cubic pixel units to obtain the three-dimensional volume data of the building.

[0065] In this embodiment, in order to further reduce the amount of data storage and improve data processing efficiency, when storing the three-dimensional volume data of the building, the region formed by multiple continuous cubic pixel units with the same material physical property vector is compressed and represented as:

[0066] CompressedSparseMatrix={(i1,j1,k1,i2,j2,k2,M,P)}

[0067] Among them, (i1, j1, k1) and (i2, j2, k2) respectively represent the coordinate positions of the center points of the two boundary cubic voxel units of the region formed by multiple continuous cubic voxel units with the same material physical property vector. In an example, (i1, j1, k1) and (i2, j2, k2) can respectively represent the coordinate positions of the center points of the cubic elements of the lower left vertex and the upper right vertex of the above region (which is a rectangular region or a cubic region). (i1, j1, k1) and (i2, j2, k2) can also respectively represent the coordinate positions of the center points of the cubic elements of the upper left vertex and the lower right vertex of the above region (which is a rectangular region or a cubic region). Through compressed representation, voxel data of homogeneous materials or blank areas can be effectively stored, storage overhead can be optimized, and computing performance can be improved.

[0068] In this embodiment, in order to improve the mapping rate, the above-mentioned material coding and material physical property vectors can be associated and mapped with various parts of the building (such as walls, floors, roofs, etc.), which can be specifically achieved through the following steps and mapping relationships: Extract the information of various parts of the building from the BIM model, such as walls, floors, roofs, columns, etc. Each part corresponds to a specific geometric area, and its material type information is usually included in the BIM model. Based on the part material information stored in the BIM model, the material type (such as concrete, steel) can be mapped to the corresponding material coding by querying the material database. The material coding is stored in each voxel unit so that 3D CNN can extract these features. According to the material type coding, the physical property vectors (such as density, thermal conductivity, carbon emission factor, etc.) of the associated material are obtained from the material property database of the BIM model and mapped to the corresponding cubic voxel unit. For example, assuming that the wall is composed of concrete, all cubic voxel units in the wall area are assigned the physical property data of concrete.

[0069] In a preferred embodiment, since the size of the cubic voxel unit directly affects the accuracy of the grid and the computational complexity of the subsequent model, the size of the cubic voxel unit is set in an adaptive manner for each region. Therefore, step S21, discretizing the BIM model of the building into multiple cubic voxel units through voxelization processing, includes:

[0070] Step S211, setting a three-dimensional geometric solid bounding box circumscribed to the BIM model. The three-dimensional geometric solid bounding box may be composed of an outer frame of at least one stacked cube or cuboid, and the number and size of the stacked cubes or cuboids are determined according to the geometric shape of the BIM model. The occupied space area of ​​the BIM model is estimated by the geometric solid bounding box to determine the overall range of the cubic voxel grid.

[0071] Step S212, dividing the three-dimensional geometric solid bounding box into multiple spatial regions.

[0072] In one example, the size of a spatial region is more than 10 times the size of a cubic pixel unit. The three-dimensional geometric solid bounding box can be evenly or non-evenly divided into a plurality of spatial regions, and the geometric shape of the spatial region can be a cube or a cuboid or an irregular geometric shape.

[0073] Step S213, determining the geometric complexity and material property change rate of each spatial region, and calculating the cubic pixel unit size of each spatial region according to the geometric complexity and material property change rate of each spatial region. The higher the geometric complexity in the spatial region and the greater the material property change rate, the smaller the cubic pixel unit size is. The cubic pixel unit size is preferably, but not limited to, the diagonal length of the cube and the side length of the cube.

[0074] Step S214, discretize each spatial region into more than one cubic voxel unit according to the size of the cubic voxel unit of each spatial region. The discretization process can refer to the existing Marching Cubes algorithm or uniform gridding algorithm. The Marching Cubes algorithm is mainly used to decompose complex surface geometry (such as curved building exterior walls) into multiple cubes.

[0075] This embodiment can use smaller cubic voxels in complex geometric areas and spatial areas with large material property changes to improve feature extraction precision and building performance prediction accuracy. Generally, the cubic voxel volume can be set to 1cm according to the required precision. 3 Up to 10cm 3 Between units.

[0076] In this embodiment, in order to delete the redundant cubic pixel units in step S21 and improve the accuracy of the obtained cubic pixel units, preferably, the method further includes:

[0077] Determine whether each cubic pixel unit is covered by the geometric shape of the building one by one, and remove the cubic pixel units that are not covered by the geometric shape of the building.

[0078] In this embodiment, if the center point of the cubic pixel unit is located within the geometric shape of the building, the cubic pixel unit is considered to be covered by the building and the coverage mark is set to 1; if the center point of the cubic pixel unit is not located within the geometric shape of the building, the cubic pixel unit is considered to be not covered by the building and the coverage mark is set to 0.

[0079] In a preferred embodiment, the cubic pixel unit size L in each spatial region is calculated according to the following formula:

[0080]

[0081] Among them, C g represents the geometric complexity of each spatial region; ΔM represents the rate of change of material properties of each spatial region; L represents the diagonal length or side length of the cubic pixel unit; l represents an auxiliary constant, which is greater than or equal to 1 cm. S 表 Represents the sum of the surface areas of the components of the BIM model contained in each spatial area, V 容 The sum of the volumes of the components of the BIM model contained in each space area, C g The unit is cm. If a space area is located inside a component of the BIM model and does not contain any surface of the component, then C g is equal to 0. If a spatial region accommodates the local surfaces of more than two components, then C g It is equal to the ratio of the sum of the local surface areas of all the components it accommodates to the sum of the volumes of the parts of all the components it accommodates within this spatial area. represents the number of physical properties in the material physical property vector, s represents the index of the physical property in the material physical property vector, and q s(max) represents the maximum value of the sth physical property in the spatial region, q s(min) Represents the minimum value of the sth physical property in the spatial region.

[0082] The above formula allows the use of smaller cubic voxel units in complex geometric areas and spatial areas with large changes in material properties to improve the expression accuracy of three-dimensional volume data.

[0083] In a preferred embodiment, to facilitate subsequent simplification of discrete operations, step S213 includes:

[0084] Step a, determine the geometric complexity and material property change rate of each spatial region.

[0085] The BIM model of a building can be considered as consisting of multiple components, and the geometric complexity of each spatial area is C g It can be the number of components in the spatial region. The material property change rate calculation formula for each spatial region is:

[0086]

[0087] Among them, Q represents the number of material types in the spatial region; S represents the number of physical properties in the material physical property vector, s represents the index of the physical property in the material physical property vector, and q s(max) represents the maximum value of the sth physical property in the spatial region, q s(min) Represents the minimum value of the sth physical property in the spatial region.

[0088] Step b, calculate the cubic voxel unit size of each spatial region according to the geometric complexity and material property change rate of each spatial region. The cubic voxel unit size of each spatial region can be represented by the number of cubic voxels Num included in the discretized spatial region. Under the condition that the spatial regions have the same size, the larger the Num, the smaller the cubic voxel unit size of the spatial region, and the smaller the Num, the larger the cubic voxel unit size of the spatial region. The number of cubic voxels Num is:

[0089] Num=f b (1+C g +ΔM)

[0090] In the above steps, f b (·) means to find the sum of 1+C g +ΔM is the power of 2 that is closest to its value. For example, if 1+C g +ΔM is 7.6, so the nearest power of 2 is 8, so Num = 8; if 1+C g The value of +ΔM is 2.6, so Num=2.

[0091] In a preferred embodiment, in order to improve the prediction accuracy of the hybrid neural network model, the training process of the hybrid neural network model includes:

[0092] Step B1, obtaining material price data, site climate data, historical energy consumption data of multiple buildings in different time periods, and obtaining BIM model data of multiple buildings.

[0093] Specifically, material price data, site climate data, and historical energy consumption data of multiple buildings in each time period are obtained, and pre-processing such as cleaning and normalization is performed on the obtained material price data, site climate data, and historical energy consumption data of multiple buildings.

[0094] Step B2, extracting three-dimensional volume data based on the BIM model data of different buildings to obtain different three-dimensional volume data; each building obtains corresponding three-dimensional volume data.

[0095] Step B3, combining the three-dimensional volume data of each building with the material price data, historical energy consumption data and site climate data of the building in different time periods to obtain samples of each building in different time periods, and setting corresponding labels for the samples to obtain a sample data set.

[0096] In this implementation, each building can correspond to multiple samples, but the time period of each sample is different. Different buildings in the same time period correspond to different samples. A sample includes material price data, site climate data, historical energy consumption data of the building, and three-dimensional volume data of the building within a time period. The label corresponding to each sample includes three true values, namely, the true value of the building cost, the true value of the building energy consumption, and the true value of the carbon emissions of the building throughout its life cycle.

[0097] The material price data in the sample is input into professional cost estimation software (such as Glodon). The cost estimation software can provide detailed cost estimation in the construction phase, including material, labor and machinery costs, etc. The construction cost obtained after these detailed cost data are calculated by the cost estimation software is taken as the true value of the construction cost. The material price corresponding to time point t is:

[0098]

[0099] Among them, p i and q i are the unit price and quantity of the i-th material respectively.

[0100] The actual energy consumption at the last moment of the sample corresponding to the time period is taken as the true value of building energy consumption. According to the building BIM model data and actual energy consumption data, in accordance with relevant carbon emission calculation standards, such as the "Guidelines for Building Carbon Emission Calculation", these standards comprehensively calculate the carbon emissions of the entire process of building from construction to operation to demolition based on the carbon emission factors used by different materials and energy, and use this carbon emission as the true value of carbon emissions. This can improve the accuracy and availability of the hybrid neural network model.

[0101] Step B4, constructing the network structure of the hybrid neural network model.

[0102] Step B5, using the sample data set to iteratively train the constructed hybrid neural network model until the training stop condition is reached, and in the iterative training, the network parameters in the hybrid neural network model are optimized according to the gradient back propagation of the loss function, where the loss function L(θ) is:

[0103] L(θ)=λ1·L cost (θ)+λ2·L energy (θ)+λ3·L carbon (θ)

[0104] Wherein, θ represents the network parameters of the hybrid neural network model; λ1, λ2, and λ3 represent the first weight coefficient, the second weight coefficient, and the third weight coefficient, respectively; L cost (θ) represents the mean square error between the cost output by the hybrid neural network model during training and the true value of the building cost in the sample label; L energy (θ) represents the mean square error between the energy consumption output by the hybrid neural network model during training and the true value of the building energy consumption in the sample label; L carbon (θ) represents the mean square error between the full life cycle carbon emissions output by the hybrid neural network model during training and the true value of the full life cycle carbon emissions in the sample label.

[0105] In a preferred embodiment, in order to optimize the design of new building projects and optimize the renovation of existing buildings to achieve multi-objective optimization, the method also includes optimizing the design parameters of new buildings based on the NSGA-III algorithm, including:

[0106] Step A1, initialize a population including N solutions, each solution represents a set of design parameters for a new building; set a target performance set, the target performance set includes target energy consumption, target cost and target life cycle carbon emissions, N is a positive integer; each set of design parameters is limited to but not limited to the geometric characteristics, number of floors, orientation, etc. of the building. The size of N is preferably but not limited to 500. Each solution represents a potential building design scheme.

[0107] Step A2, based on the population, crossover and mutation operations are performed to generate a progeny population, the progeny population includes N solutions, and the progeny population and the population are merged to obtain a candidate population. The candidate population includes 2N solutions. Binary crossover (crossover rate is 0.9) and multi-point mutation (mutation rate is 0.1) are used as genetic operations to maintain genetic diversity.

[0108] Step A3, based on Pareto-dominated non-dominated sorting, the candidate population is divided into several different non-dominated layers. In each generation, individuals are selected to enter the next generation through non-dominated sorting and crowding comparison to ensure the diversity and quality of solutions.

[0109] Step A4: calculate the distance between each solution and the target performance set in each non-dominated layer, and select N solutions with the closest distance to the target performance set from multiple non-dominated layers as a new population.

[0110] Step A5, repeating steps A2 to A5 until an iteration stop condition is reached, and then proceeding to step A6. The iteration stop condition is preferably, but not limited to, when the number of iterations reaches a maximum number of iterations.

[0111] Step A6: Select the solution closest to the target performance set from the new population as the Pareto optimal solution, and output the design parameters of the Pareto optimal solution.

[0112] In each generation of the genetic algorithm, the solution set will continue to evolve and eventually generate a set of Pareto optimal solutions. Pareto optimal solutions represent different trade-offs, in which the improvement of any one objective (cost, energy consumption or carbon emissions) may lead to the deterioration of other objectives. NSGA-III improves the breadth and accuracy of the solution space search by optimizing algorithm parameters such as population size, crossover probability and mutation probability.

[0113] In this embodiment, preferably, in step A4, calculating the distance between each solution and the target performance set in each non-dominated layer includes:

[0114] Step A41, obtaining current material price data and site climate data.

[0115] Step A42, generating BIM model data of the new building based on the design parameters of each solution, obtaining current energy consumption simulation data of the new building using energy consumption simulation software based on the BIM model data of the new building, and using the energy consumption simulation data as historical energy consumption data of the new building;

[0116] Step A43, extracting the three-dimensional volume data of the new building from the BIM model data of the new building;

[0117] Step A44, inputting the three-dimensional volume data, material price data, site climate data and historical energy consumption data of the new building into the pre-trained hybrid neural network model, obtaining the predicted energy consumption, cost and life cycle carbon emissions of the new building, and composing the predicted energy consumption, cost and life cycle carbon emissions of the new building into a performance set of each solution;

[0118] Step A45, calculate the distance between the performance set of each solution and the target performance set. Specifically, calculate the ratio of the absolute value of the difference between the predicted energy consumption of the new building and the target energy consumption of the target performance set to the target energy consumption as the energy consumption ratio; calculate the ratio of the absolute value of the difference between the predicted cost of the new building and the target cost of the target performance set to the target cost as the cost ratio; calculate the ratio of the absolute value of the difference between the predicted full life cycle carbon emissions of the new building and the target full life cycle carbon emissions of the target performance set to the target full life cycle carbon emissions as the carbon emission ratio; sum the cost ratio, energy consumption ratio and carbon emission ratio, and obtain the total as the distance between the performance set of each solution and the target performance set. This distance represents the closeness of the performance set of each solution to the target performance set.

[0119] In this implementation, the scheme can also generate a set of Pareto optimal solutions through the NSGA-III multi-objective optimization algorithm. These solutions can provide the best compromise between the cost, energy consumption and carbon emissions of the building, helping decision makers to select the optimal design that meets the needs of a specific project. The generation of the Pareto front makes the optimization process more transparent. Decision makers can intuitively understand the performance of different design solutions in multiple dimensions and make balanced choices based on project priorities such as budget, energy saving or environmental protection. This multi-objective optimization provides more flexibility for building design, making the design process more efficient and controllable.

[0120] The optimized solution set will be presented through visualization technology to help decision makers understand the performance of different building design schemes in terms of cost, energy consumption and carbon emissions. Use Python's Matplotlib library to plot the dynamic changes of the optimization process, such as the downward trend of cost and energy consumption with the number of iterations. Use Bokeh for interactive visualization to show the performance of different design schemes in three dimensions: cost, energy consumption and carbon emissions. In addition, the optimization results are integrated into Power BI to build an interactive dashboard that displays the performance indicators of each design scheme in real time and allows users to adjust and compare different schemes according to different priorities.

[0121] In this embodiment, in terms of visualization, all optimization results can be intuitively presented through visualization tools, including predictions and comparisons of key indicators such as building costs, energy consumption, and carbon emissions. Using interactive visualization platforms (such as Power BI or Tableau), management can track changes brought about by adjustments to the design plan in real time and make accurate decisions in a data-driven manner. This visualization function not only facilitates the work of architects and project managers, but also improves the efficiency of communication with investors or other stakeholders, allowing each participant to clearly understand the optimization results and make wise choices.

[0122] From the perspective of long-term benefits, the solution of this implementation also has the ability to be dynamically updated. As the market environment changes, such as material price fluctuations, technological advances or policy changes, the system can flexibly adjust the optimization process to ensure that the design of the building project always has the best economy and sustainability. At the same time, during the operation phase of the building, the real-time input of relevant data can also provide continuous optimization support for the long-term maintenance and energy management of the building, ensuring that the project shows the best benefits throughout its life cycle.

[0123] In short, this technical solution ensures that the building design has efficient cost control, energy consumption optimization and carbon emission reduction functions in the initial planning, construction and operation stages through multi-dimensional intelligent integration and optimization, which meets the needs of green buildings and low-carbon development. This systematic and dynamically optimized solution provides a forward-looking and sustainable solution for the construction project, greatly improving the economic benefits and environmental friendliness of the project.

[0124] Figure 4 A specific flow chart of the building performance prediction method based on hybrid neural network provided by the present invention in an application scenario is shown.

[0125] The present invention also discloses a building performance evaluation system based on a hybrid neural network, which is used to implement the above-mentioned building performance prediction method based on a hybrid neural network. Figure 3 As shown, including:

[0126] The first data acquisition module captures material price data from multiple building material supplier websites in real time;

[0127] The second data acquisition module acquires the historical energy consumption data, site climate data and BIM model data of the building;

[0128] The third data acquisition module extracts the three-dimensional volume data of the building from the BIM model data;

[0129] The prediction module uses a pre-trained hybrid neural network model to predict the energy consumption, cost and life cycle carbon emissions of a building based on material price data, historical energy consumption data, site climate data and 3D volume data. The hybrid neural network model includes:

[0130] A price feature conversion unit, used to convert material price data into price features;

[0131] The first long short-term memory network is used to extract energy consumption time features from historical energy consumption data;

[0132] The second long short-term memory network is used to extract climate temporal features from site climate data;

[0133] 3D convolutional neural network, used to extract 3D spatial features of buildings from 3D volume data;

[0134] A splicing module is used to splice energy consumption time characteristics, climate time characteristics, price characteristics and three-dimensional space characteristics to obtain a comprehensive feature representation;

[0135] A fully connected module, comprising one or more cascaded fully connected layers for extracting nonlinear features from the comprehensive feature representation;

[0136] The output module performs regression processing on nonlinear features to obtain the energy consumption, cost and carbon emissions of the building throughout its life cycle.

[0137] In this embodiment, the first data acquisition module and the second data acquisition module correspond to step S1 in the above-mentioned building performance prediction method based on hybrid neural network, the third data acquisition module corresponds to step S2 in the above-mentioned building performance prediction method based on hybrid neural network, and the prediction module corresponds to step S3 in the above-mentioned building performance prediction method based on hybrid neural network. The specific execution process of each module will not be repeated here.

[0138] The present invention provides a building performance prediction method and evaluation system based on a hybrid neural network, which can significantly improve the comprehensive performance of building design in terms of cost, energy consumption and carbon emissions throughout the life cycle through multi-dimensional integrated optimization.

[0139] First, in terms of cost optimization, the technical solution combines the material market data obtained by Web Scraping with the cost data of the cost software, and uses 3D CNN neural network analysis to extract the parameters of the building design, which can generate accurate construction cost forecasts. Through the continuous update of market data, the construction cost forecast results after design adjustment can also be output dynamically in real time. In this way, project managers can make reasonable decisions in the design stage to ensure that the construction budget is strictly controlled, while maximizing the savings in material and labor costs and improving economic benefits.

[0140] In terms of energy consumption prediction and optimization, the solution can more accurately capture the actual energy consumption of buildings under different climate conditions by introducing measured energy consumption data and combining it with the LSTM neural network model. The solution can simulate energy use in the future operation phase, help optimize the building's energy management system, and provide solutions to reduce energy consumption, such as improving ventilation design, selecting insulation materials, or optimizing the operation and scheduling of energy equipment. This is particularly important for the design of green buildings, which can enable buildings to perform more energy-efficiently during operation and reduce long-term energy consumption.

[0141] At the same time, the solution follows China's carbon emission calculation standards and uses a neural network model to accurately estimate the carbon emissions of the building's entire life cycle. This includes carbon emissions from the production of materials, the construction of the building, energy consumption in the operation phase, and waste treatment in the demolition phase. Through this estimate, building designers can make adjustments to material selection or design optimization based on the carbon emission prediction results. For example, low-carbon building materials can be selected or the proportion of clean energy used can be increased, thereby effectively reducing the impact of construction projects on the environment. This solution also meets the country's requirements for low-carbon buildings and green buildings, ensuring that construction projects can meet environmental protection standards during the design phase and contribute to global carbon reduction goals.

[0142] The key technical points of the present invention are:

[0143] Through the integration of multi-source data and intelligent optimization technology, the prediction and optimization capabilities of building design costs, energy consumption and carbon emissions throughout the life cycle are improved. The present invention integrates multiple data sources, including building material market data obtained through Web Scraping, measured energy consumption data, China's carbon emission calculation standards, and building cost data generated by cost software. By cleaning, standardizing and integrating these data, the model can handle complex input data, providing strong data support for subsequent analysis and prediction.

[0144] The present invention uses a hybrid neural network model that combines 3D CNN (three-dimensional convolutional neural network) and LSTM (long short-term memory network) groups. 3D CNN is used to process spatial features in architectural design, such as geometric structure and material usage; LSTM is used to analyze the time series characteristics of building energy consumption. This combination of models can more accurately predict the cost, energy consumption and carbon emissions of buildings, especially when processing large-scale data and complex features.

[0145] The introduction of measured energy consumption data is a notable feature of this invention. Unlike traditional simulation data, measured data reflects the actual energy consumption of the building during operation. By using this data, the model can more accurately capture the fluctuations and trends of building energy consumption, significantly improving the realism and reliability of energy consumption forecasts.

[0146] At the same time, the present invention calculates the carbon emissions of the entire life cycle of a building based on China's current carbon emission calculation standards, combined with the carbon emission factors and energy consumption data of building materials. Such standardized processing ensures that the prediction complies with the requirements of the country's environmental protection policies and can help design buildings that are more in line with low-carbon and environmentally friendly standards.

[0147] In terms of cost calculation, the present invention integrates data from cost estimation software. These software are widely used in the construction industry and can provide accurate construction cost estimates. After inputting this data into the model, combined with the design parameters, the overall cost of the building can be predicted more accurately. This is particularly critical when dealing with market dynamics (such as material price fluctuations).

[0148] The application of the NSGA-III multi-objective optimization algorithm is another key point. The algorithm allows multiple objectives to be optimized, such as the cost, energy consumption and carbon emissions of a building, and generates a set of Pareto optimal solutions through a genetic algorithm. These solutions can help decision makers find the best balance between different optimization objectives and ensure that the design achieves the best combination in terms of economy, energy saving and environmental protection.

[0149] To support decision-making, the invention uses visualization technology to display multi-dimensional optimization results, helping decision makers quickly understand the pros and cons of different design solutions. Visualization makes the design process more transparent and improves communication efficiency between different stakeholders. Ultimately, the entire optimization process is dynamic and can be continuously updated as market conditions, material supply, and technology change, ensuring that the building design remains optimal throughout its life cycle.

[0150] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", "an implementation", "a preferred implementation" or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0151] Although the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the claims and their equivalents.

Claims

1. A building performance prediction method based on hybrid neural network, characterized in that: include: Real-time scraping of material price data from multiple building material supplier websites; Obtain the building’s historical energy consumption data, site climate data, and BIM model data; Extract the building's three-dimensional volume data from the BIM model data; Predict the energy consumption, cost and life cycle carbon emissions of a building using a pre-trained hybrid neural network model based on material price data, historical energy consumption data, site climate data and 3D volume data. The hybrid neural network model includes: A price feature conversion unit, used to convert material price data into price features; The first long short-term memory network is used to extract energy consumption time features from historical energy consumption data; The second long short-term memory network is used to extract climate temporal features from site climate data; 3D convolutional neural network, used to extract 3D spatial features of buildings from 3D volume data; A splicing module is used to splice energy consumption time characteristics, climate time characteristics, price characteristics and three-dimensional space characteristics to obtain a comprehensive feature representation; A fully connected module, comprising one or more cascaded fully connected layers for extracting nonlinear features from the comprehensive feature representation; The output module performs regression processing on nonlinear features to obtain the energy consumption, cost and carbon emissions of the building throughout its life cycle.

2. A building performance prediction method based on a hybrid neural network as claimed in claim 1, characterized in that: The step of extracting the three-dimensional volume data of the building from the BIM model data comprises: The BIM model of the building is discretized into multiple cubic voxel units through voxelization; Extract the material type of each cubic element from the BIM model data, and map the material code and material physical property vector corresponding to the material type to the cubic element; Represent each cubic pixel unit as a three-dimensional array, wherein each three-dimensional array includes the coordinate position of the center point of the cubic pixel unit, the material code mapped to the cubic pixel unit, and the material physical property vector; The three-dimensional arrays of all cubic voxel units are combined to obtain the three-dimensional volume data of the building.

3. A building performance prediction method based on hybrid neural network as claimed in claim 2, characterized in that: The voxelization process discretizes the BIM model of the building into a plurality of cubic voxel units, including: Set up a 3D geometric solid bounding box that is external to the BIM model; Divide the three-dimensional geometric solid bounding box into a plurality of spatial regions; Determine the geometric complexity and material property change rate of each spatial region, and calculate the cubic voxel unit size of each spatial region according to the geometric complexity and material property change rate of each spatial region; Each spatial region is discretized into more than one cubic voxel according to the cubic voxel size of each spatial region.

4. A building performance prediction method based on a hybrid neural network as claimed in claim 3, characterized in that: Also includes: Determine whether each cubic pixel unit is covered by the geometric shape of the building one by one, and remove the cubic pixel units that are not covered by the geometric shape of the building.

5. The building performance prediction method based on hybrid neural network as claimed in claim 3, characterized in that: The cubic voxel size L in each spatial region is calculated as follows: Among them, C g represents the geometric complexity of each spatial region; ΔM represents the rate of change of material properties in each spatial region.

6. A method for predicting building performance based on a hybrid neural network as claimed in claim 2, 3, 4 or 5, characterized in that: Also includes: Represent a 3D array of 3D volume data as a sparse matrix: SparseMatrix={(i,j,k,M,P|V i,j,k )} Where (i, j, k) represents the cubic pixel unit V i,j,k The center point coordinate position, M represents the cubic pixel unit V i,j,k Mapped material type code, P represents the cubic voxel unit V i,j,k The mapped material physical property vector; The region compression formed by multiple consecutive cubic voxel units with the same material physical property vector is expressed as: CompressedSparseMatrix={(i1,j1,k1,i2,j2,k2,M,P)} Among them, (i1, j1, k1) and (i2, j2, k2) respectively represent the coordinate positions of the center points of two boundary cubic voxel units in the area formed by multiple continuous cubic voxel units with the same material physical property vector.

7. A building performance prediction method based on a hybrid neural network as described in claim 2 or 3 or 4 or 5, characterized in that: The training process of the hybrid neural network model includes: Obtain material price data, site climate data, historical energy consumption data of multiple buildings for different time periods, and obtain BIM model data of multiple buildings; Extract 3D volume data based on BIM model data of different buildings to obtain different 3D volume data; The three-dimensional volume data of each building is combined with the material price data, historical energy consumption data and site climate data of the building in different time periods to obtain samples of each building in different time periods, and corresponding labels are set for the samples to obtain a sample data set; Construct the network structure of the hybrid neural network model; The constructed hybrid neural network model is iteratively trained using the sample data set until the training stop condition is reached. In the iterative training, the network parameters in the hybrid neural network model are optimized according to the gradient back propagation of the loss function, where the loss function L(θ) is: L(θ)=λ1·L cost (θ)+λ2·L energy (θ)+λ3·L carbon (i Wherein, θ represents the network parameters of the hybrid neural network model; λ1, λ2, and λ3 represent the first weight coefficient, the second weight coefficient, and the third weight coefficient, respectively; L cost (θ) represents the mean square error between the cost output by the hybrid neural network model during training and the true value of the building cost in the sample label; L energy (θ) represents the mean square error between the energy consumption output by the hybrid neural network model during training and the true value of the building energy consumption in the sample label; L carbon (θ) represents the mean square error between the full life cycle carbon emissions output by the hybrid neural network model during training and the true value of the full life cycle carbon emissions in the sample label.

8. A building performance prediction method based on hybrid neural network as claimed in claim 7, characterized in that: It also includes the optimization of design parameters of new buildings based on the NSGA-III algorithm, including: Step A1, initializing a population including N solutions, each solution representing a set of design parameters for a new building; setting a target performance set, the target performance set including target energy consumption, target cost and target life cycle carbon emissions, N is a positive integer; Step A2, performing crossover and mutation operations based on the population to generate a child population, the child population includes N solutions, and merging the child population and the population to obtain a candidate population; Step A3, dividing the candidate population into several different non-dominated layers based on Pareto-dominated non-dominated sorting; Step A4, calculating the distance between each solution and the target performance set in each non-dominated layer, and selecting N solutions with the closest distance to the target performance set from multiple non-dominated layers as a new population; Step A5, repeating steps A2 to A5 until the iteration stop condition is reached, and then proceeding to step A6; Step A6: Select the solution closest to the target performance set from the new population as the Pareto optimal solution, and output the design parameters of the Pareto optimal solution.

9. A building performance prediction method based on hybrid neural network as claimed in claim 8, characterized in that: In step A4, the distance between each solution and the target performance set is calculated in each non-dominated layer, including: Obtain current material price data and site climate data; Generate BIM model data of the new building based on the design parameters of each solution, use energy consumption simulation software based on the BIM model data of the new building to obtain the current energy consumption simulation data of the new building, and use the energy consumption simulation data as the historical energy consumption data of the new building; Extracting the three-dimensional volume data of the new building from the BIM model data of the new building; Input the three-dimensional volume data, material price data, site climate data and historical energy consumption data of the new building into the pre-trained hybrid neural network model to obtain the predicted energy consumption, cost and life cycle carbon emissions of the new building, and form the performance set of each solution with the predicted energy consumption, cost and life cycle carbon emissions of the new building; Calculate the distance of each solution's performance set from the target performance set.

10. A building performance evaluation system based on a hybrid neural network, used to implement a building performance prediction method based on a hybrid neural network as claimed in any one of claims 1 to 9, characterized in that: include: The first data acquisition module captures material price data from multiple building material supplier websites in real time; The second data acquisition module acquires the historical energy consumption data, site climate data and BIM model data of the building; The third data acquisition module extracts the three-dimensional volume data of the building from the BIM model data; The prediction module uses a pre-trained hybrid neural network model to predict the energy consumption, cost and life cycle carbon emissions of a building based on material price data, historical energy consumption data, site climate data and 3D volume data. The hybrid neural network model includes: A price feature conversion unit, used to convert material price data into price features; The first long short-term memory network is used to extract energy consumption time features from historical energy consumption data; The second long short-term memory network is used to extract climate temporal features from site climate data; 3D convolutional neural network, used to extract 3D spatial features of buildings from 3D volume data; A splicing module is used to splice energy consumption time characteristics, climate time characteristics, price characteristics and three-dimensional space characteristics to obtain a comprehensive feature representation; A fully connected module, comprising one or more cascaded fully connected layers for extracting nonlinear features from the comprehensive feature representation; The output module performs regression processing on nonlinear features to obtain the energy consumption, cost and carbon emissions of the building throughout its life cycle.

Images