Curtain wall lighting performance simulation method and system based on deep learning

Abstract

The application provides a curtain wall lighting performance simulation method and system based on deep learning, and relates to the field of building design and simulation technology. First, the curtain wall design elements (including curtain wall structure arrangement and curtain wall material light transmission performance) and target area light action elements (covering light direction at different times and light intensity change at different times) are integrated, a pre-trained lighting correlation network construction model is input, the light transmission performance distribution of each part of the curtain wall and the indoor space light diffusion performance distribution are generated, the dynamic correlation network of the curtain wall lighting is constructed to present the dynamic action relationship between each element, interactive simulation operation is performed through the deep learning simulation module, the simulation result containing the indoor area lighting balance and continuous performance is generated, and the curtain wall design optimization guide is generated to adjust the curtain wall design elements and optimize the indoor lighting performance.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of architectural design and simulation, in particular, to a curtain wall lighting performance simulation method and system based on deep learning. BACKGROUND

[0002] In the field of architectural design and construction, curtain wall, as an important facade element of modern architecture, not only bears the function of beautifying the appearance of the building, but also has a key influence on indoor lighting. Reasonable curtain wall design can effectively utilize natural light to provide a comfortable, uniform and continuous lighting environment for the indoor, thereby reducing the use of artificial lighting and energy consumption, and meeting the development concept of green building.

[0003] Currently, in the field of curtain wall lighting performance analysis, traditional methods mainly rely on empirical formulas and simple numerical simulation. Empirical formulas are usually based on some simplified assumptions and statistical data, and it is difficult to accurately consider the complex dynamic relationship between curtain wall design elements and lighting action elements. For example, for different curtain wall structural arrangements and material light transmission performance, empirical formulas may not accurately predict the actual lighting effect under different time period lighting directions and intensity changes. Although simple numerical simulation can simulate the propagation and distribution of light to some extent, it often ignores the interaction mechanism between curtain wall design elements and lighting action elements, resulting in a large deviation between simulation results and actual situation. In addition, traditional methods are difficult to comprehensively and dynamically present the correlation between curtain wall design elements, lighting action elements and indoor lighting performance, and cannot provide targeted and forward-looking optimization guidance for curtain wall design. SUMMARY

[0004] In view of the above-mentioned problems, in combination with the first aspect of the present application, the present application embodiment provides a curtain wall lighting performance simulation method based on deep learning, which comprises:

[0005] Integrating curtain wall design elements and target area lighting action elements, the curtain wall design elements including curtain wall structural arrangement, curtain wall material light transmission performance, and the lighting action elements including different time period lighting directions and different time period lighting intensity changes;

[0006] Inputting the curtain wall design elements and the lighting action elements into a pre-trained lighting correlation network construction model to generate curtain wall part lighting acceptance performance distribution and indoor space lighting diffusion performance distribution;

[0007] Constructing a curtain wall lighting dynamic correlation network according to the lighting acceptance performance distribution and the lighting diffusion performance distribution, the curtain wall lighting dynamic correlation network being used to present the dynamic action relationship between curtain wall design elements, lighting action elements and indoor lighting performance;

[0008] The deep learning simulation module is started to interactively simulate and calculate the curtain wall lighting dynamic correlation network, to generate a curtain wall lighting simulation result containing indoor area lighting balance performance and indoor area lighting persistence performance.

[0009] A curtain wall design optimization guide is generated according to the curtain wall lighting simulation result, and the curtain wall design optimization guide is used to adjust curtain wall design elements to optimize indoor lighting performance.

[0010] In another aspect, the embodiment of the present application also provides a curtain wall lighting performance simulation system based on deep learning, which is characterized by comprising:

[0011] A processor; a machine readable storage medium for storing machine executable instructions of the processor; wherein the processor is configured to execute the machine executable instructions to perform the above-mentioned curtain wall lighting performance simulation method based on deep learning.

[0012] In another aspect, the embodiment of the present application also provides a computer program product, which comprises machine executable instructions stored in a computer readable storage medium, and a processor of a computer device reads the machine executable instructions from the computer readable storage medium, and the processor executes the machine executable instructions to enable the computer device to perform the above-mentioned curtain wall lighting performance simulation method based on deep learning.

[0013] Based on the above aspects, by integrating curtain wall design elements and target area lighting action elements, curtain wall structure arrangement, curtain wall material light transmission performance, and light illumination direction and intensity change in different time periods are covered, the integrated elements are input into a pre-trained lighting correlation network construction model, curtain wall part light receiving performance distribution and indoor space light diffusion performance distribution are generated, the distribution of light in the curtain wall and indoor space is accurately described from two aspects of micro and macro, the curtain wall lighting dynamic correlation network constructed according to the curtain wall part light receiving performance distribution and indoor space light diffusion performance distribution presents the dynamic action relationship between the curtain wall design elements, the lighting action elements and the indoor lighting performance in an intuitive and scientific way, the deep learning simulation module is started to interactively simulate and calculate the dynamic correlation network, the generated curtain wall lighting simulation result not only contains indoor area lighting balance performance, but also covers indoor area lighting persistence performance, fully and carefully reflects the actual situation of indoor lighting, the curtain wall design optimization guide generated based on the simulation result can accurately adjust the curtain wall design elements, thereby effectively optimizing the indoor lighting performance, and significantly improving the accuracy of the curtain wall lighting performance simulation. BRIEF DESCRIPTION OF DRAWINGS

[0014] Figure 1is an execution flow diagram of a curtain wall lighting performance simulation method based on deep learning provided by an embodiment of the present application.

[0015] Figure 2 is a schematic diagram of exemplary hardware and software components of a curtain wall lighting performance simulation system based on deep learning provided by an embodiment of the present application. DETAILED DESCRIPTION

[0016] The present application will be described in detail below with reference to the accompanying drawings of the specification, Figure 1 is a flow diagram of a curtain wall lighting performance simulation method based on deep learning provided by an embodiment of the present application, which will be described in detail below.

[0017] Step S110: integrate curtain wall design elements and target area lighting elements, the curtain wall design elements include curtain wall structure arrangement, curtain wall material light transmission performance, and the lighting elements include different time period light incidence direction and different time period light intensity change.

[0018] In this embodiment, taking the curtain wall lighting performance simulation of a certain commercial building as an example, the acquisition of curtain wall design elements needs to consider the architectural design drawings, technical parameters of curtain wall materials, etc. In terms of curtain wall structure arrangement, the overall layout of the curtain wall needs to be determined, for example, the curtain wall of the commercial building is composed of multiple rectangular panels, and the position and size of each panel have detailed design parameters, which can be extracted from the CAD drawings of architectural design to form curtain wall structure arrangement data containing coordinate attributes and size parameters. In terms of curtain wall material light transmission performance, the light transmittance, reflectivity, and absorptivity of the curtain wall material need to be obtained, which can be obtained through technical documents provided by the material supplier or laboratory test reports, for example, the curtain wall material of the commercial building is a certain type of Low-E glass, and its light transmittance, reflectivity, and absorptivity have clear test data.

[0019] The acquisition of lighting elements needs to be combined with the geographical location and meteorological data of the target area. In terms of different time period light incidence direction, according to the geographical location of the commercial building, astronomical algorithm or meteorological software is used to calculate the solar altitude angle and azimuth angle in different seasons, different dates, and different times, so as to determine the incidence direction of light and form light incidence direction data in different time periods. In terms of different time period light intensity change, meteorological data of the region is collected, including solar radiation intensity data in different time periods, which can be obtained from local weather stations or weather databases, so as to obtain data of light intensity change in different time periods. The above obtained curtain wall structure arrangement, curtain wall material light transmission performance, different time period light incidence direction, and different time period light intensity change are integrated to form a complete data set.

[0020] Step S120: input the curtain wall design elements and the light action elements into the pre-trained daylighting association network construction model to generate the light acceptance performance distribution of each part of the curtain wall and the light diffusion performance distribution of the indoor space.

[0021] Step S121: perform spatial digital conversion on the curtain wall structure arrangement in the curtain wall design elements to form a curtain wall structure digital form with coordinate attributes, each structure part in the curtain wall structure digital form having corresponding position parameters and size parameters.

[0022] In this embodiment, for the curtain wall structure arrangement of the above commercial building, a spatial coordinate system is first determined, for example, taking a certain point on the ground floor of the building as the origin, the horizontal direction as the x and y axes, and the vertical direction as the z axis. Then, according to the design drawings of the curtain wall structure, the position and size parameters of each curtain wall panel and support frame and other structure parts are converted into digital representations in this coordinate system. For example, the lower left corner coordinates of a certain curtain wall panel are (x1, y1, z1), and the upper right corner coordinates are (x2, y2, z2). The position parameters of the panel can be represented by these two coordinate points, and the size parameters can be obtained by calculating the distance between the two coordinate points, such as length x2-x1, width y2-y1, and height z2-z1. By performing the above conversion on all curtain wall structure parts, a curtain wall structure digital form with coordinate attributes is finally formed, which is stored in the form of a data structure, for example, a three-dimensional array can be used to represent the position and size parameters of each structure part, and each element of the array corresponds to the coordinate and size information of a structure part.

[0023] Step S122: perform feature conversion on the light transmission performance of the curtain wall material in the curtain wall design elements to form a material light transmission feature form recognizable by the daylighting association network construction model, the dimension of the material light transmission feature form matching the input layer dimension of the daylighting association network construction model.

[0024] In this embodiment, for the light transmission performance of the curtain wall material of the commercial building, the original data includes transmittance, reflectance, and absorption rate. In order to convert the above parameters into a feature form that can be recognized by the daylighting correlation network construction model, the above parameters need to be standardized. For example, assuming that the input layer dimension of the daylighting correlation network construction model is 3, corresponding to transmittance, reflectance, and absorption rate. First, the values of transmittance T, reflectance R, and absorption rate A of the curtain wall material are obtained, and then the above values are normalized so that their value range is between [0, 1]. The normalization method can use the minimum-maximum normalization, that is, for each parameter, calculate its proportion relative to the possible value range of the parameter. For example, the possible value range of transmittance is [0, 1], so the normalized transmittance T' = T / 1; the same for reflectance and absorption rate. The normalized T', R', and A' are combined into a three-dimensional vector in order, and the dimension of the three-dimensional vector matches the input layer dimension of the daylighting correlation network construction model, thereby forming a material light transmission feature form.

[0025] Step S123: direction conversion is performed on the light illumination directions of different time periods in the light illumination element to form a light illumination direction set, each direction information in the light illumination direction set corresponding to the light illumination direction of a specific time period.

[0026] In this embodiment, for the light illumination direction data of different time periods in the area where the commercial building is located, it needs to be converted into a light illumination direction set. First, determine the time division granularity, for example, divide a day into 24 time periods in units of hours. For each time period, according to the calculated solar elevation angle and azimuth angle, the light illumination direction is converted into a vector form. For example, in a certain time period, the solar elevation angle is h and the azimuth angle is a, then the light illumination direction vector can be represented as (sin(h)cos(a), sin(h)sin(a), cos(h)), where the x component represents the projection in the horizontal direction, the y component represents the projection in the horizontal direction perpendicular to the x axis, and the z component represents the projection in the vertical direction. The light illumination direction vectors of each time period are arranged in time sequence to form a light illumination direction set, each element in the light illumination direction set corresponding to a light illumination direction of a specific time period, and stored in the form of a vector for subsequent model processing.

[0027] Step S124: time sequence conversion is performed on the light illumination intensity changes of different time periods in the light illumination element to form a light illumination intensity time sequence feature, the time interval of the light illumination intensity time sequence feature being consistent with the time period division of the light illumination direction set.

[0028] In this embodiment, for the light intensity variation data of different time periods of the area where the commercial building is located, it needs to be converted into light intensity time sequence features. First, according to the same time division granularity as the light action direction set, that is, in units of hours, the solar radiation intensity data of each period is extracted. Then, the above data is standardized, for example, Z-standardization is used, the difference between the solar radiation intensity of each period and the mean and standard deviation of the entire period sequence is calculated, and then divided by the standard deviation to obtain the standardized solar radiation intensity data. The above standardized data is arranged in chronological order to form a time sequence, and the time interval of the time sequence is consistent with the time period division of the light action direction set, thereby forming the light intensity time sequence features.

[0029] Step S125: input the curtain wall structure digital form, the material light transmission feature form, the light action direction set, and the light intensity time sequence features into the input layer of the pre-trained daylight correlation network construction model.

[0030] In this embodiment, the pre-trained daylight correlation network construction model is a deep learning model, and the structure of the input layer needs to match the dimension of the input data. The curtain wall structure digital form, material light transmission feature form, light action direction set, and light intensity time sequence features obtained above are combined according to the input requirements of the model to form an input data set. For example, the curtain wall structure digital form can be input as a three-dimensional tensor, the material light transmission feature form as a three-dimensional vector, the light action direction set as a two-dimensional tensor (time x direction vector dimension) input, and the light intensity time sequence features as a one-dimensional vector (time dimension) input.

[0031] Step S126: perform spatial analysis on the curtain wall structure digital form through the structure feature processing layer of the daylight correlation network construction model to generate spatial position features of each part of the curtain wall.

[0032] In this embodiment, the structure feature processing layer of the daylighting correlation network construction model includes multiple convolution layers and pooling layers, which are used for spatial analysis of the digital form of the curtain wall structure. First, the digital form of the curtain wall structure is input to the first convolution layer of the structure feature processing layer in the form of a three-dimensional tensor. The convolution layer uses multiple convolution kernels to perform convolution operations on the input tensor to extract local spatial features of the curtain wall structure. For example, the size of the convolution kernel can be set according to the detail level of the curtain wall structure. Through the convolution operation, multiple feature maps are obtained, each of which corresponds to a local spatial feature. Then, the above feature maps are subjected to a pooling operation, such as maximum pooling or average pooling, to reduce the dimension of the feature maps while retaining the main spatial features. After processing by multiple convolution layers and pooling layers, the spatial position features of each part of the curtain wall are finally obtained, which are represented in the form of high-dimensional vectors. Each vector corresponds to a part of the curtain wall, and the dimension of the vector reflects the spatial position information of the part and the spatial relationship information with the surrounding structure.

[0033] Step S127: The material transmittance feature form is subjected to attribute analysis by the material feature processing layer of the daylighting correlation network construction model to generate material transmittance attribute features of each part of the curtain wall.

[0034] In this embodiment, the material feature processing layer of the daylighting correlation network construction model includes multiple fully connected layers, which are used for attribute analysis of the material transmittance feature form. First, the material transmittance feature form is input to the first fully connected layer of the material feature processing layer in the form of a three-dimensional vector. The fully connected layer performs linear transformation and nonlinear activation on the input vector to extract basic attribute features of the material transmittance. For example, the three-dimensional vector is mapped to a higher-dimensional space through linear transformation, and then nonlinear features are introduced through a nonlinear activation function (such as a ReLU function) to enhance the expression ability of the model. After processing by multiple fully connected layers, the material transmittance attribute features of each part of the curtain wall are finally obtained, which are represented in the form of high-dimensional vectors. Each vector corresponds to a part of the curtain wall, and the dimension of the vector reflects the material transmittance, reflectivity, and absorption rate of the part, as well as the correlation information between these attributes.

[0035] Step S128: The light illumination feature processing layer of the daylighting correlation network construction model performs spatio-temporal correlation on the set of light action directions and the light intensity time sequence features to generate light action features of each time period.

[0036] In this embodiment, the light feature processing layer of the daylighting correlation network construction model comprises multiple recurrent neural network layers (such as LSTM layers) and convolutional layers, which are used to correlate the light direction set and the light intensity time sequence feature in space and time. First, the light direction set is input to the convolutional layer of the light feature processing layer in the form of a two-dimensional tensor, and the convolutional layer performs convolution operation on the light direction vector of each time period to extract the spatial features of the light direction. At the same time, the light intensity time sequence feature is input to the recurrent neural network layer in the form of a one-dimensional vector, and the recurrent neural network layer analyzes the time sequence change of the light intensity to extract the time features of the light intensity. Then, the spatial features obtained by the convolutional layer and the time features obtained by the recurrent neural network layer are fused, for example, the two feature vectors are combined into a new feature vector by splicing, which contains the spatial information of the light direction and the time information of the light intensity. After multiple processing steps, the light action features of each time period are finally obtained, which are represented in the form of high-dimensional vectors, each vector corresponds to a time period, and the dimension of the vector reflects the space-time correlation information between the light direction and the light intensity of the time period.

[0037] Step S129: The spatial position feature, the material light transmission property feature, and the light action feature are cross-element fused by the element interaction layer of the daylighting correlation network construction model to generate the light receiving comprehensive feature of each part of the curtain wall.

[0038] In this embodiment, the element interaction layer of the daylighting correlation network construction model comprises multiple attention mechanism layers and fully connected layers, which are used to cross-element fuse the spatial position feature, the material light transmission property feature, and the light action feature. First, the spatial position feature, the material light transmission property feature, and the light action feature are input to the attention mechanism layer, which calculates the attention weight between each feature, i.e., determines the importance of each feature in the fusion process. For example, the similarity between the spatial position feature and the material light transmission property feature is calculated to obtain the attention weight between them; similarly, the attention weights between the spatial position feature and the light action feature, and the material light transmission property feature and the light action feature are calculated. Then, according to the above attention weights, each feature is weighted, for example, the spatial position feature is multiplied by its corresponding attention weight to obtain the weighted spatial position feature; similarly, the weighted material light transmission property feature and the light action feature are obtained. Finally, the three weighted features are spliced to form the light receiving comprehensive feature of each part of the curtain wall, which is represented in the form of a high-dimensional vector, each vector corresponds to a part of the curtain wall, and the dimension of the vector reflects the comprehensive influence information of the spatial position, the material light transmission property, and the light action of the part.

[0039] Step S1210: generating a light acceptance performance distribution of each part of the curtain wall according to the spatial distribution of the light acceptance comprehensive features of each part of the curtain wall.

[0040] In this embodiment, the light acceptance comprehensive features of each part of the curtain wall are represented in the form of high-dimensional vectors, and each vector corresponds to a part of the curtain wall. First, the above high-dimensional vectors are decoded and converted into a form that can intuitively represent the light acceptance performance, for example, some dimensions in the vector can be mapped to the indicators of light acceptance intensity, light acceptance time, light incidence angle, etc. Then, according to the spatial position information of each part of the curtain wall, the above decoded indicators are distributed according to the spatial position to form a two-dimensional or three-dimensional distribution map. For example, in a two-dimensional plane, based on the plane layout of the curtain wall, each position point corresponds to a part of the curtain wall, and different colors or numerical values are used to represent the indicators of light acceptance intensity, light acceptance time, light incidence angle, etc. of the part, thereby generating a light acceptance performance distribution of each part of the curtain wall.

[0041] Step S1211: the indoor diffusion simulation layer of the daylighting correlation network construction model simulates the indoor diffusion path and energy change process of light after passing through the curtain wall according to the light acceptance performance distribution and the spatial parameters of the digital form of the curtain wall structure, and generates an indoor space light diffusion performance distribution.

[0042] In this embodiment, the indoor diffusion simulation layer of the daylighting correlation network construction model includes multiple physical simulation models and neural network layers, which are used to simulate the indoor diffusion path and energy change process of light after passing through the curtain wall. First, according to the light acceptance performance distribution of each part of the curtain wall, the light emission intensity and direction of each part of the curtain wall are determined. Then, combined with the spatial parameters of the digital form of the curtain wall structure, such as the position, size of the curtain wall, the shielding condition of the support frame, etc., the propagation path of light in the indoor space after emission from the curtain wall part is simulated. In the process of simulating the propagation path, the spatial structure of the indoor space needs to be considered, such as the shape of the room, the reflectivity of the wall, the shielding of the furniture, etc., which can be obtained through the pre-established indoor space model. At the same time, the energy change of light in the propagation process needs to be simulated, such as the reflection and absorption of light on the wall, the scattering in the air, etc., which can be calculated through physical formulas or empirical models, for example, the reflected and absorbed energy is calculated according to the reflectivity and absorption rate of the wall material, and the scattered energy is calculated according to the scattering coefficient of the air. Through the above simulation of the light emission of each part of the curtain wall, the light intensity, light coverage range, etc. of each region in the indoor space are finally obtained, and the above information is distributed according to the spatial position of the indoor region to form an indoor space light diffusion performance distribution.

[0043] Step S130: constructing a curtain wall lighting dynamic correlation network according to the light receiving performance distribution and the light diffusion performance distribution, the curtain wall lighting dynamic correlation network being used to present the dynamic action relationship between curtain wall design elements, light action elements and indoor lighting performance.

[0044] Step S131: extracting light receiving intensity performance, light receiving time performance and light incidence angle performance of each part of the curtain wall from the light receiving performance distribution as first type node information of the curtain wall lighting dynamic correlation network.

[0045] In this embodiment, for the generated light receiving performance distribution of each part of the curtain wall, key information needs to be extracted as first type node information. First, the light receiving intensity performance can be obtained by analyzing the light intensity values in the distribution, for example, for each curtain wall part, the maximum, minimum and average values of light intensity in different time periods are extracted, which can reflect the light receiving intensity of the part. The light receiving time performance can be obtained by analyzing the light time information in the distribution, for example, for each curtain wall part, the effective light time, continuous light time and the like in a day, a month and a year are extracted, which can reflect the light receiving time of the part. The light incidence angle performance can be obtained by analyzing the light incidence angle values in the distribution, for example, for each curtain wall part, the maximum, minimum and average values of light incidence angle in different time periods are extracted, which can reflect the light incidence angle of the part. The above extracted light receiving intensity performance, light receiving time performance and light incidence angle performance are arranged according to the curtain wall parts to form the first type node information, and each curtain wall part corresponds to a group of the above information.

[0046] Step S132: extracting light reaching intensity performance, light coverage range performance and light balance degree performance of each region in the room from the light diffusion performance distribution as second type node information of the curtain wall lighting dynamic correlation network.

[0047] In this embodiment, for the generated indoor space light diffusion performance distribution, the key information needs to be extracted as the second type of node information. First, the light intensity performance can be obtained by analyzing the light intensity values in the distribution, for example, for each indoor area, the maximum, minimum, average, etc. of the light intensity at different time periods are extracted, which can reflect the light intensity of the area. The light coverage performance can be obtained by analyzing the light coverage area and shape information in the distribution, for example, for each indoor area, the light coverage area and the shape of the light coverage at different time periods are extracted, which can reflect the light coverage of the area. The light balance performance can be obtained by analyzing the uniformity of the light intensity distribution in the distribution, for example, the variance, standard deviation, etc. of the light intensity in the area are calculated, the smaller the variance, the higher the light balance, and the above statistical quantities are taken as the indicators of the light balance performance. The above extracted light intensity performance, light coverage performance, and light balance performance are arranged according to the indoor area to form the second type of node information, and each indoor area corresponds to a set of the above information.

[0048] Step S133: Extract the curtain panel size performance, panel splicing gap performance, and support frame blocking performance from the curtain structure arrangement of the curtain design element as the third type of node information of the curtain daylighting dynamic correlation network.

[0049] In this embodiment, for the curtain structure arrangement of the commercial building, the key information needs to be extracted as the third type of node information. First, the curtain panel size performance can be obtained by analyzing the panel size parameters in the curtain structure arrangement, for example, for each curtain panel, the length, width, height, etc. of the size information are extracted, which can reflect the size of the panel. The panel splicing gap performance can be obtained by analyzing the panel splicing method and gap size in the curtain structure arrangement, for example, for each two adjacent curtain panels, the width, length, shape, etc. of the splicing gap information are extracted, which can reflect the splicing gap. The support frame blocking performance can be obtained by analyzing the support frame position and size in the curtain structure arrangement, for example, for each support frame, the projection area on the curtain plane, the blocked curtain panel area, etc. of the information are extracted, which can reflect the blocking of the support frame. The above extracted curtain panel size performance, panel splicing gap performance, and support frame blocking performance are arranged according to the curtain structure part to form the third type of node information, and each curtain structure part corresponds to a set of the above information.

[0050] Step S134: Extract the material light transmittance performance, material reflectivity performance, and material absorption performance from the curtain material light transmittance performance of the curtain design element as the fourth type of node information of the curtain daylighting dynamic correlation network.

[0051] In this embodiment, the key information of the light transmission performance of the curtain wall material of the commercial building needs to be extracted as the fourth type of node information. First, the light transmission performance of the material can be obtained by analyzing the light transmission parameters of the curtain wall material, such as extracting the light transmission values of the curtain wall material under different wavelengths of light, which can reflect the light transmission performance of the material. The material reflectivity performance can be obtained by analyzing the reflectivity parameters of the curtain wall material, such as extracting the reflectivity values of the curtain wall material under different wavelengths of light, which can reflect the reflectivity performance of the material. The material absorption performance can be obtained by analyzing the absorption parameters of the curtain wall material, such as extracting the absorption values of the curtain wall material under different wavelengths of light, which can reflect the absorption performance of the material. The above extracted material light transmission performance, material reflectivity performance, and material absorption performance are arranged according to the curtain wall material part to form the fourth type of node information, and each curtain wall material part corresponds to a set of the above information.

[0052] Step S135: Extracting the light intensity peak value performance of different time periods and the light direction change performance of different time periods from the light action elements as the fifth type of node information of the curtain wall daylighting dynamic correlation network.

[0053] In this embodiment, the key information of the light transmission performance of the curtain wall material of the commercial building needs to be extracted as the fourth type of node information. First, the light transmission performance of the material can be obtained by analyzing the light transmission parameters of the curtain wall material, such as extracting the light transmission values of the curtain wall material under different wavelengths of light, which can reflect the light transmission performance of the material. The material reflectivity performance can be obtained by analyzing the reflectivity parameters of the curtain wall material, such as extracting the reflectivity values of the curtain wall material under different wavelengths of light, which can reflect the reflectivity performance of the material. The material absorption performance can be obtained by analyzing the absorption parameters of the curtain wall material, such as extracting the absorption values of the curtain wall material under different wavelengths of light, which can reflect the absorption performance of the material. The above extracted material light transmission performance, material reflectivity performance, and material absorption performance are arranged according to the curtain wall material part to form the fourth type of node information, and each curtain wall material part corresponds to a set of the above information.

[0054] Step S136: Assembling the node group of the curtain wall daylighting dynamic correlation network, which includes curtain wall part nodes, indoor area nodes, curtain wall structure nodes, curtain wall material nodes, and light action nodes, and each type of node corresponds to the above different types of node information.

[0055] In this embodiment, according to the five types of node information extracted above, the node group of the curtain wall lighting dynamic correlation network is established. First, the curtain wall part node corresponds to the first type of node information, and each curtain wall part node contains the light intensity performance, light time performance, light incidence angle performance, etc. of the part. The indoor area node corresponds to the second type of node information, and each indoor area node contains the light intensity performance, light coverage performance, light balance performance, etc. of the area. The curtain wall structure node corresponds to the third type of node information, and each curtain wall structure node contains the curtain panel size performance, panel splicing gap performance, support frame shielding performance, etc. of the structure part. The curtain wall material node corresponds to the fourth type of node information, and each curtain wall material node contains the material transmittance performance, material reflectivity performance, material absorption performance, etc. of the material part. The light action node corresponds to the fifth type of node information, and each light action node contains the light intensity peak performance at different time periods and the light direction change performance at different time periods. The above nodes are organized according to the set logical relationship, for example, according to the spatial correspondence relationship between the curtain wall part and the indoor area, the belonging relationship between the curtain wall part and the curtain wall structure, the belonging relationship between the curtain wall part and the curtain wall material, and the action relationship between the curtain wall part and the light action, etc. to form a node group.

[0056] Step S137: Analyze the action relationship between the curtain wall part node and the indoor area node, and establish the first type of connection relationship between the curtain wall part node and the indoor area node according to the diffusion path of light from the curtain wall part to the indoor area. The action degree of the first type of connection relationship is determined according to the energy transfer in the light diffusion process.

[0057] In this embodiment, first, the indoor area node corresponding to each curtain wall part node is determined, for example, which can be obtained by analyzing the diffusion path of light from the curtain wall part to the indoor area. For example, after the light emitted from a certain curtain wall part propagates in the indoor space, it finally reaches some indoor areas, so that the curtain wall part node has an action relationship with these indoor area nodes. Then, the energy transfer in the light diffusion process is calculated, for example, which can be determined by analyzing the energy loss of light in the propagation process. For example, the energy of light emitted from the curtain wall part is E1, and the energy of light reaching the indoor area is E2, then the energy transfer efficiency is E2 / E1, which can be used as the action degree of the first type of connection relationship. According to the above action relationship and action degree, the first type of connection relationship is established between the curtain wall part node and the indoor area node, for example, represented by a directed edge, and the weight of the edge is the action degree.

[0058] Step S138: analyze the action relationship between the curtain wall part node and the curtain wall structure node, and establish a second type of connection relationship between the curtain wall part node and the curtain wall structure node according to the influence of the curtain wall structure on the light acceptance of the curtain wall part. The action degree of the second type of connection relationship is determined according to the influence of the structure shielding on the light acceptance.

[0059] In this embodiment, first, the curtain wall structure node corresponding to each curtain wall part node is determined, which can be obtained by analyzing the belonging relationship between the curtain wall structure and the curtain wall part, for example. For example, a certain curtain wall part belongs to a certain curtain wall panel, and the curtain wall panel is related to certain support frames and splicing gaps. Therefore, the curtain wall part node has an action relationship with these curtain wall structure nodes. Then, the influence of the curtain wall structure on the light acceptance of the curtain wall part is analyzed, for example, the shielding of the support frame will reduce the light acceptance area of the curtain wall part, thereby affecting the light acceptance intensity. The influence degree of the structure shielding on the light acceptance is calculated, for example, the proportion of the shielding area to the area of the curtain wall part, which can be used as the action degree of the second type of connection relationship. According to the above action relationship and action degree, the second type of connection relationship is established between the curtain wall part node and the curtain wall structure node, for example, represented by a directed edge, and the weight of the edge is the action degree.

[0060] Step S139: analyze the action relationship between the curtain wall part node and the curtain wall material node, and establish a third type of connection relationship between the curtain wall part node and the curtain wall material node according to the influence of the curtain wall material on the light absorption and reflection of the curtain wall part. The action degree of the third type of connection relationship is determined according to the influence of the material light transmittance on the light transmission.

[0061] In this embodiment, first, the curtain wall material node corresponding to each curtain wall part node is determined, which can be obtained by analyzing the belonging relationship between the curtain wall material and the curtain wall part, for example. For example, a certain curtain wall part uses a certain type of Low-E glass, so the curtain wall part node has an action relationship with the curtain wall material node of the glass. Then, the influence of the curtain wall material on the light absorption and reflection of the curtain wall part is analyzed, for example, the absorption rate of the material will affect the amount of light absorption, and the reflectivity will affect the amount of light reflection, thereby affecting the amount of light transmission. The influence degree of the material light transmittance on the light transmission is calculated, for example, the higher the transmittance, the more the amount of light transmission, which can be used as the action degree of the third type of connection relationship. According to the above action relationship and action degree, the third type of connection relationship is established between the curtain wall part node and the curtain wall material node, for example, represented by a directed edge, and the weight of the edge is the action degree.

[0062] Step S1310: Analyzing the action relationship between the curtain wall part node and the light action node, establishing a fourth type of connection relationship between the curtain wall part node and the light action node according to the influence of the light action on the light receiving of the curtain wall part, and the action degree of the fourth type of connection relationship is determined according to the influence of the light intensity and the light direction on the light receiving of the curtain wall part.

[0063] In this embodiment, first, the light action node corresponding to each curtain wall part node is determined, which can be obtained by analyzing the action relationship between the light action and the curtain wall part. For example, the light direction and the light intensity of a certain period of time will affect the light receiving of a certain curtain wall part, so the curtain wall part node has an action relationship with the light action node of the period. Then, the influence of the light action on the light receiving of the curtain wall part is analyzed. For example, the stronger the light intensity, the higher the light receiving intensity of the curtain wall part; the more suitable the angle between the light direction and the curtain wall part, the higher the light receiving intensity of the curtain wall part. The influence degree of the light intensity and the light direction on the light receiving of the curtain wall part is calculated, for example, a function can be established, taking the light intensity and the light direction as input and the light receiving intensity as output, and the output value of the function can be used as the action degree of the fourth type of connection relationship. According to the above action relationship and action degree, the fourth type of connection relationship is established between the curtain wall part node and the light action node, for example, represented by a directed edge, and the weight of the edge is the action degree.

[0064] Step S1311: Analyzing the action relationship between the indoor area node and the curtain wall structure node, establishing a fifth type of connection relationship between the indoor area node and the curtain wall structure node according to the influence of the curtain wall structure on the light diffusion of the indoor area, and the action degree of the fifth type of connection relationship is determined according to the influence of the structure arrangement on the indoor light distribution.

[0065] In this embodiment, first, the curtain wall structure node corresponding to each indoor area node is determined, which can be obtained by analyzing the spatial relationship between the curtain wall structure and the indoor area. For example, the support frame of a certain curtain wall structure will affect the light diffusion path of the indoor area, thereby affecting the light distribution of the indoor area, so the indoor area node has an action relationship with the curtain wall structure node. Then, the influence of the curtain wall structure on the light diffusion of the indoor area is analyzed. For example, the position of the support frame will change the diffusion direction of the light, thereby affecting the light coverage range and the light intensity distribution of the indoor area. The influence degree of the structure arrangement on the indoor light distribution is calculated, for example, by comparing the light distribution difference of the indoor area with and without the curtain wall structure, and the light distribution difference can be used as the action degree of the fifth type of connection relationship. According to the above action relationship and action degree, the fifth type of connection relationship is established between the indoor area node and the curtain wall structure node, for example, represented by a directed edge, and the weight of the edge is the action degree.

[0066] Step S1312: analyze the action relationship between the indoor area node and the curtain wall material node, establish a sixth type of connection relationship between the indoor area node and the curtain wall material node according to the influence of the curtain wall material on the indoor area light intensity, and the action degree of the sixth type of connection relationship is determined according to the influence of the material light transmittance on the indoor light intensity.

[0067] In this embodiment, first, the curtain wall material node corresponding to each indoor area node is determined, for example, which can be obtained by analyzing the action relationship between the curtain wall material and the indoor area. For example, the light transmittance of a certain curtain wall material will affect the amount of light transmission, thereby affecting the light intensity of the indoor area, so that the indoor area node has an action relationship with the curtain wall material node. Then, analyze the influence of the curtain wall material on the indoor area light intensity, for example, the higher the light transmittance of the material, the higher the light intensity of the indoor area. Calculate the influence degree of the material light transmittance on the indoor light intensity, for example, by establishing a function, taking the material light transmittance as the input and the indoor area light intensity as the output, the output value of the function can be used as the action degree of the sixth type of connection relationship. According to the above action relationship and action degree, the sixth type of connection relationship is established between the indoor area node and the curtain wall material node, for example, represented by a directed edge, and the weight of the edge is the action degree.

[0068] Step S1313: integrate the above various types of nodes, various types of node information and various types of connection relationships to form a complete curtain wall daylighting dynamic correlation network, which is used to present the dynamic action relationship between the curtain wall design elements, the light action elements and the indoor daylighting performance.

[0069] In this embodiment, the curtain wall part node, the indoor area node, the curtain wall structure node, the curtain wall material node, the light action node obtained above, and the corresponding various types of node information and various types of connection relationships are integrated. First, all the nodes are organized according to their categories and action relationships to form a node set of a network. Then, all the connection relationships are organized according to their types and action degrees to form an edge set of a network. Finally, the node set and the edge set are combined into a complete network structure, for example, which can be represented by a graph data structure, wherein the nodes represent the above various types of nodes, the edges represent the above various types of connection relationships, and the weights of the edges represent the action degrees. In this way, a complete curtain wall daylighting dynamic correlation network is formed, which can directly present the dynamic action relationship between the curtain wall design elements, the light action elements and the indoor daylighting performance.

[0070] Step S140: start the deep learning simulation module to interactively simulate the curtain wall daylighting dynamic correlation network, and generate a curtain wall daylighting simulation result containing the indoor area daylighting balance performance and the indoor area daylighting persistence performance.

[0071] Step S141: transforming the network structure of the curtain wall lighting dynamic correlation network, converting the node information and connection relationship information in the curtain wall lighting dynamic correlation network into a network feature form recognizable by the deep learning simulation module, the number of rows of the network feature form corresponding to the number of nodes in the curtain wall lighting dynamic correlation network, and the number of columns corresponding to the sum of the node information dimension and the connection relationship feature dimension.

[0072] In this embodiment, the structure of the network feature form recognizable by the deep learning simulation module needs to be determined first. For example, the network feature form can be a two-dimensional matrix, where the number of rows is equal to the number of nodes in the curtain wall lighting dynamic correlation network, and the number of columns is equal to the sum of the information dimension of each node and the connection relationship feature dimension of the node. For each node, first extract its node information, such as the light acceptance intensity performance, light acceptance time performance, and light incidence angle performance of the curtain wall part node, and convert the above information into a numerical vector, with the vector dimension being the node information dimension. Then, extract the connection relationship information of the node, such as the degree of influence of the connection relationship between the node and other nodes and the type of connection relationship, and convert the above information into a numerical vector, with the vector dimension being the connection relationship feature dimension. Finally, the node information vector and the connection relationship information vector are spliced to form the feature row corresponding to the node. The feature rows of all nodes are arranged in the order of the nodes to form a two-dimensional matrix of the network feature form.

[0073] Step S142: inputting the network feature form into the network feature processing layer of the deep learning simulation module, and integrating the neighborhood features of the network feature form through network convolution operation to generate the neighborhood integrated features of each node.

[0074] Step S1421: identifying the direct neighboring node group of each node in the curtain wall lighting dynamic correlation network, which includes all nodes directly correlated with the node through the connection relationship.

[0075] In this embodiment, for each node in the curtain wall lighting dynamic correlation network, the connection relationship of the network needs to be traversed to find all nodes directly correlated with the node through the connection relationship, and these nodes form the direct neighboring node group of the node. For example, for a curtain wall part node, by checking the connection relationship between the node and other nodes, all indoor area nodes, curtain wall structure nodes, curtain wall material nodes, and light action nodes directly connected to the node are found, which are the direct neighboring node group of the curtain wall part node.

[0076] Step S1422: calculating the connection relationship influence degree between each node and each node in its direct neighboring node group, which is determined according to the influence degree parameter of the connection relationship.

[0077] In this embodiment, for each node and each node in its direct neighboring node group, the degree of action parameter of the connection relationship between them needs to be obtained, which has been determined when the curtain lighting dynamic correlation network is constructed, such as the degree of action calculated in steps S137-S1312. For each connection relationship, the degree of action parameter is directly extracted as the degree of action of the connection relationship between the node and its direct neighboring node.

[0078] Step S1423: standardizing the feature row information corresponding to each node in the network feature form and the feature row information corresponding to each node in the direct neighboring node group of the node, eliminating the dimension difference of the feature, to generate standardized feature row information.

[0079] In this embodiment, the feature row information of each node and the feature row information of each node in its direct neighboring node group need to be standardized. The standardization method can use Z-standardization, that is, for each feature dimension, the mean and standard deviation of the dimension in all related feature rows are calculated, then each feature value is subtracted from the mean and divided by the standard deviation to obtain the standardized feature value. For example, for a certain feature dimension, assuming that the feature values of the node and each node in its direct neighboring node group are x1, x2, …, xn, the mean μ=(x1+x2+…+xn) / n is calculated, and the standard deviation σ=√[((x1-μ)²+(x2-μ)²+…+(xn-μ)²) / (n-1)] is calculated, then each feature value xi is standardized as xi'=(xi-μ) / σ, which eliminates the dimension difference between different feature dimensions and generates standardized feature row information.

[0080] Step S1424: performing element-level operation on the standardized feature row information to generate a set of neighboring node feature operation results.

[0081] In this embodiment, the standardized feature row information of each node and the standardized feature row information of each node in its direct neighboring node group need to be element-level operated. The element-level operation mode can be multiplication, addition, subtraction, etc., and the specific operation mode can be determined according to the design requirements of the model. For example, multiplication operation is used to multiply the standardized feature row information of the node with the standardized feature row information of each direct neighboring node corresponding to the elements to obtain a plurality of operation result vectors, and these vectors constitute a set of neighboring node feature operation results.

[0082] Step S1425: weighting each result in the set of neighboring node feature operation results with the corresponding degree of action of the connection relationship to generate a set of weighted neighboring node features.

[0083] In this embodiment, for each result vector in the set of result vectors of the adjacent node feature operation, it is necessary to perform a weighted operation with the corresponding connection relationship action degree. The way of weighted operation is to multiply each element in the result vector by the connection relationship action degree to obtain a weighted vector. For example, a certain result vector is v=[v1, v2, …, vm], and the corresponding connection relationship action degree is w, then the weighted vector is v'=[v1xw, v2xw, …, vmxw]. All the above weighted vectors form a weighted adjacent node feature set.

[0084] Step S1426: performing element-level summation operation on all features in the weighted adjacent node feature set to generate a neighborhood feature summation result.

[0085] In this embodiment, for all vectors in the weighted adjacent node feature set, element-level summation operation is needed. The way of element-level summation operation is to add corresponding elements of each vector to obtain a new vector. For example, the vectors in the weighted adjacent node feature set are v1'=[v11, v12, …, v1m], v2'=[v21, v22, …, v2m], …, vn'=[vn1, vn2, …, vnm], then the element-level summation vector is v_sum=[v11+v21+…+vn1, v12+v22+…+vn2, …, v1m+v2m+…+vnm], which is taken as the neighborhood feature summation result.

[0086] Step S1427: performing normalization processing on the neighborhood feature summation result to eliminate the feature value difference caused by the difference in the number of adjacent nodes of different nodes, and generating a normalized neighborhood feature.

[0087] In this embodiment, since the number of direct adjacent nodes of different nodes may be different, the numerical size of the neighborhood feature summation result is different. In order to eliminate the above difference, it is necessary to perform normalization processing on the neighborhood feature summation result. The method of normalization processing can be to divide by the number of direct adjacent nodes to obtain an average weighted feature vector. For example, the number of direct adjacent nodes of a certain node is k, and the neighborhood feature summation result is v_sum, then the normalized neighborhood feature is v_reg=v_sum / k. In this way, the feature value difference caused by the difference in the number of adjacent nodes of different nodes is eliminated.

[0088] Step S1428: combining the original feature row information corresponding to each node with the normalized neighborhood feature of the node to generate the neighborhood integrated feature of each node.

[0089] In this embodiment, for each node, the original feature row information (i.e., the feature row information without standardization processing) of the node needs to be combined with the regularized neighborhood features. The combination processing can be splicing, that is, the vectors of the original feature row information and the regularized neighborhood features are spliced to form a new vector, which is used as the integrated neighborhood features of the node. For example, the vector of the original feature row information is v_original = [o1, o2, …, op], the vector of the regularized neighborhood features is v_reg = [r1, r2, …, rq], and the vector of the integrated neighborhood features is v_integrated = [o1, o2, …, op, r1, r2, …, rq].

[0090] Step S143: performing network structure analysis of different levels on the integrated neighborhood features by the multi-level network analysis layer of the deep learning simulation module, and extracting local correlation features and overall correlation features of the curtain wall lighting dynamic correlation network respectively to generate multi-level network correlation features.

[0091] Step S1431: setting a first-level analysis range, the first-level analysis range corresponding to a direct neighboring range of nodes in the curtain wall lighting dynamic correlation network, and being used for extracting local correlation features.

[0092] In this embodiment, the setting of the first-level analysis range is based on the direct neighboring relationship of the nodes in the curtain wall lighting dynamic correlation network. For example, for each node, the direct neighboring range of the node is the direct neighboring node group of the node and the node itself. In this range, the correlation features between the nodes are analyzed, and the features reflect the network structure relationship in the local area.

[0093] Step S1432: performing standardization processing on the integrated neighborhood features of each node in the first-level analysis range, performing neighboring node feature difference analysis on the standardized integrated neighborhood features, extracting the feature difference between the node and the direct neighboring nodes, and generating a local feature difference situation.

[0094] In this embodiment, first, the integrated neighborhood features of each node are standardized, and the processing method is similar to that in step S1423, that is, the mean and standard deviation of the integrated neighborhood features of the node and each node in the direct neighboring node group of the node are calculated in each dimension, and then Z-standardization is performed. Then, the standardized integrated neighborhood features are analyzed for neighboring node feature difference, for example, the Euclidean distance or cosine similarity between the standardized integrated neighborhood features of the node and the standardized integrated neighborhood features of each direct neighboring node is calculated, and these distances or similarities reflect the feature difference between the node and the direct neighboring nodes. The above difference situation is arranged according to the set format to generate a local feature difference situation.

[0095] Step S1433: feature screening is performed on the local feature difference situation, key feature content capable of reflecting the local correlation relationship is reserved, and local correlation features are generated.

[0096] In this embodiment, the method for performing feature screening on the local feature difference situation can be based on the importance or correlation of the features. For example, the correlation of each feature with the local correlation relationship is calculated, and features with higher correlation are reserved. Alternatively, a principal component analysis method is used to extract main feature components, which can reflect the key feature content of the local correlation relationship. The screened feature content is arranged to generate the local correlation features.

[0097] Step S1434: a second-level analysis range is set, the second-level analysis range corresponds to an indirect neighboring range of nodes in the curtain wall lighting dynamic correlation network, i.e., a direct neighboring range of direct neighboring nodes, and is used to extract intermediate-level correlation features.

[0098] In this embodiment, the setting of the second-level analysis range is based on the indirect neighboring relationship of the nodes in the curtain wall lighting dynamic correlation network. For example, for each node, its indirect neighboring range is the direct neighboring node group of the direct neighboring nodes of the node (excluding the node itself). In this range, the correlation features between the nodes are analyzed, and the above features reflect the network structure relationship in the intermediate-level region.

[0099] Step S1435: in the second-level analysis range, the neighborhood integration features of each node are standardized, the standardized neighborhood integration features are analyzed for indirect neighboring node feature transmission, the indirect neighboring node feature influence transmitted by the direct neighboring nodes of the node is extracted, and intermediate-level feature transmission situations are generated.

[0100] In this embodiment, first, the neighborhood integration features of each node are standardized, and the processing method is similar to that in step S1423. Then, the standardized neighborhood integration features are analyzed for indirect neighboring node feature transmission, for example, by analyzing the connection relationship and feature transmission path between the direct neighboring nodes and the indirect neighboring nodes, the indirect neighboring node feature influence transmitted by the direct neighboring nodes of the node is calculated. For example, for each direct neighboring node, the feature transmission coefficient between it and the indirect neighboring node is calculated, and then the coefficient is multiplied by the standardized neighborhood integration features of the indirect neighboring node to obtain the indirect neighboring node feature influence transmitted by the direct neighboring nodes of the node. The above influence situation is arranged according to the set format to generate the intermediate-level feature transmission situation.

[0101] Step S1436: feature extraction is performed on the intermediate-level feature transmission situation, key feature content capable of reflecting the intermediate-level correlation relationship is reserved, and intermediate-level correlation features are generated.

[0102] In this embodiment, the method of feature extraction for the intermediate level feature transmission is similar to step S1433, that is, based on the importance or relevance of the features, the key feature content reflecting the association relationship of the intermediate level is screened out. For example, the relevance of each feature to the association relationship of the intermediate level is calculated, and the features with higher relevance are retained; or the principal component analysis method is used to extract the main feature components. The extracted feature content is arranged to generate the intermediate level association feature.

[0103] Step S1437: setting a third level analysis range, the third level analysis range corresponding to the whole node range of the curtain wall lighting dynamic association network, used for extracting overall association features.

[0104] In this embodiment, the third level analysis range is set to the whole node range of the curtain wall lighting dynamic association network. In this range, the association features between all nodes are analyzed, and the above features reflect the structural relationship of the entire network.

[0105] Step S1438: in the third level analysis range, the neighborhood integration features of all nodes are standardized, the global feature statistics of the standardized neighborhood integration features are performed, the overall distribution of the global node features and the mutual influence are extracted, and the global feature carding result is generated.

[0106] In this embodiment, first, the neighborhood integration features of all nodes are standardized, and the processing method is similar to step S1423. Then, the global feature statistics of the standardized neighborhood integration features are performed, for example, the mean, standard deviation, maximum value, minimum value and other statistics of the neighborhood integration features of all nodes in each dimension are calculated, and these statistics reflect the overall distribution of the global node features. At the same time, the covariance or correlation coefficient between nodes is calculated, and these coefficients reflect the mutual influence of the global node features. The above statistics and coefficients are arranged to generate the global feature carding result.

[0107] Step S1439: feature integration is performed on the global feature carding result, the global feature distribution and mutual influence are converted into a feature form that can reflect the overall association relationship, and the overall association feature is generated.

[0108] In this embodiment, the method of feature integration of the global feature carding result is to comprehensively process the global feature distribution and mutual influence to form a feature form that can reflect the overall association relationship. For example, the statistics of the global feature distribution and the coefficients of the mutual influence are spliced to form a high-dimensional vector, and the vector is taken as the overall association feature.

[0109] Step S14310: integrating the local correlation feature, the middle-level correlation feature, and the global correlation feature to generate a multi-level network correlation feature.

[0110] In this embodiment, the local correlation feature, the middle-level correlation feature, and the global correlation feature are integrated. The integration manner can be splicing, that is, the vectors of the three features are spliced to form a new vector, which is used as the multi-level network correlation feature. For example, the vector of the local correlation feature is v_local=[l1, l2, …, lp], the vector of the middle-level correlation feature is v_middle=[m1, m2, …, mq], and the vector of the global correlation feature is v_global=[g1, g2, …, gr]. Then, the vector of the multi-level network correlation feature is v_multi=[l1, l2, …, lp, m1, m2, …, mq, g1, g2, …, gr].

[0111] Step S144: inputting the multi-level network correlation feature into a feature fusion layer of a deep learning simulation module, dynamically calculating a fusion ratio of different levels of network correlation features through an attention mechanism, weighting and fusing the multi-level network correlation feature to generate a global fusion network feature.

[0112] In this embodiment, the feature fusion layer of the deep learning simulation module includes an attention mechanism module, which is used to dynamically calculate the fusion ratio of different levels of network correlation features. First, the multi-level network correlation feature is input into the attention mechanism module, which assigns an attention weight to each level of network correlation feature. The size of the weight reflects the importance of the feature in the fusion process. The calculation method of the attention weight can be based on the importance or correlation of the feature, for example, by calculating the correlation between each level of feature and the final output result to determine the attention weight. Then, according to the above attention weight, the multi-level network correlation feature is weighted and fused. The weighted fusion manner is to multiply each level of network correlation feature by its corresponding attention weight, and then splice all the weighted features to form a global fusion network feature. For example, the attention weight of the local correlation feature is w_local, the attention weight of the middle-level correlation feature is w_middle, and the attention weight of the global correlation feature is w_global. Then, the vector of the global fusion network feature is v_fused=[w_local×v_local, w_middle×v_middle, w_global×v_global] (here, the multiplication is the multiplication of each element of the vector by the weight, and then splicing).

[0113] Step S145: input the global fusion network feature into a first simulation branch of a deep learning simulation module, perform uniformity analysis on the light distribution of each region in the room, extract the light intensity variation and light coverage range variation of each region in the room at different time periods, and generate the daylighting uniformity performance of each region in the room.

[0114] Step S1451: extract the feature part related to the light distribution of each region in the room from the global fusion network feature, wherein the feature part includes the light intensity feature content and the light coverage range feature content of each region in the room at different time periods.

[0115] In this embodiment, the global fusion network feature includes a plurality of hierarchical network correlation features, wherein the feature part related to the light distribution of each region in the room needs to be extracted. The feature part can be determined by analyzing the dimensions and meanings of the global fusion network feature. For example, the features of some dimensions correspond to the light intensity feature content of each region in the room at different time periods, and the features of some dimensions correspond to the light coverage range feature content of each region in the room at different time periods. By performing feature selection or dimension division on the global fusion network feature, these feature parts related to the light distribution of each region in the room are extracted.

[0116] Step S1452: classify the feature part according to the indoor region division and the time period division to generate the region-time period feature group corresponding to each indoor region at different time periods.

[0117] In this embodiment, the indoor region division can be performed according to the functional regions of the building, for example, the indoor regions of a commercial building can be divided into shop regions, corridor regions, public rest regions, etc. The time period division can be performed according to hours, days, months, years, etc. The extracted feature part related to the light distribution of each region in the room is classified according to the indoor region and the time period. For example, for each indoor region, the light intensity feature content and the light coverage range feature content at different time periods are classified respectively to form the region-time period feature group corresponding to the indoor region at different time periods.

[0118] Step S1453: perform time series variation analysis on the light intensity feature content in each region-time period feature group, extract the light intensity variation content of the same indoor region at adjacent time periods, and generate the region-time period light intensity variation sequence.

[0119] For the illumination intensity feature content in each region period feature group, time series change analysis is needed in this embodiment. The method of time series change analysis can be to calculate the difference or change rate of illumination intensity between adjacent time periods. For example, for the illumination intensity feature content of a certain time period t of a certain indoor region, I_t, and the illumination intensity feature content of time period t+1, I_t+1, the illumination intensity change content between adjacent time periods is ΔI_t=I_t+1-I_t (or the change rate is ΔI_t / I_t). The above-mentioned illumination intensity change content between adjacent time periods is arranged in time sequence to generate the region period illumination intensity change sequence.

[0120] Step S1454: Change trend analysis is performed on the region period illumination intensity change sequence to extract significant change content, average change content, and change frequency content in the change sequence as the illumination intensity change situation of each region in the indoor area at different time periods.

[0121] In this embodiment, the method of change trend analysis on the region period illumination intensity change sequence can be based on statistical analysis or machine learning algorithm. For example, the significant change content can be extracted by calculating the extreme value in the change sequence, that is, finding the maximum and minimum values in the change sequence, and the change content corresponding to these extreme values is the significant change content. The average change content can be obtained by calculating the mean value of the change sequence, which reflects the average level of the change of illumination intensity between adjacent time periods. The change frequency content can be obtained by calculating the number of changes in the change sequence (increase or decrease), the more the number of changes, the higher the change frequency. The above-mentioned significant change content, average change content, and change frequency content are sorted as the illumination intensity change situation of each region in the indoor area at different time periods.

[0122] Step S1455: Spatial change analysis is performed on the illumination coverage range feature content in each region period feature group to extract the illumination coverage area change content and illumination coverage shape change content of the same indoor region at different time periods to generate the region period illumination coverage change sequence.

[0123] In this embodiment, for the light coverage range feature content in each region period feature group, spatial variation analysis is needed. The method of spatial variation analysis can be to calculate the light coverage area difference and the light coverage form difference of different periods. For example, for a certain period t of a certain indoor area, the light coverage area is A_t, and the light coverage area of period t+1 is A_t+1, then the light coverage area variation content is ΔA_t=A_t+1-A_t. The light coverage form variation content can be calculated by calculating the shape difference of the light coverage area of the two periods, for example, using a shape similarity index (such as Hausdorff distance) to measure. Arrange the light coverage area variation content and the light coverage form variation content of different periods in time sequence to generate a region period light coverage variation sequence.

[0124] Step S1456: Perform variation feature analysis on the region period light coverage variation sequence, and extract the maximum variation content, the average variation content, and the variation duration content in the variation sequence as the light coverage range variation of different periods of each region in the indoor area.

[0125] In this embodiment, the method of variation feature analysis on the region period light coverage variation sequence is similar to step S1454. The maximum variation content can be extracted by calculating the extreme value in the variation sequence, that is, finding the maximum value and the minimum value in the variation sequence, and the variation content corresponding to these extreme values is the maximum variation content. The average variation content can be obtained by calculating the mean value of the variation sequence, which reflects the average level of the light coverage range variation of different periods. The variation duration content can be obtained by calculating the number of consecutive periods with the same variation direction (increase or decrease) in the variation sequence. The more consecutive periods, the longer the duration of the variation. Organize the maximum variation content, the average variation content, and the variation duration content extracted above as the light coverage range variation of different periods of each region in the indoor area.

[0126] Step S1457: Integrate the light intensity variation of different periods of each region in the indoor area and the light coverage range variation of different periods of each region in the indoor area to generate the daylighting balance performance of each region in the indoor area.

[0127] In this embodiment, the light intensity variation and the light coverage variation of each region in the room at different time periods are integrated. The integration can be splicing, that is, the feature vectors of the two cases are spliced to form a new vector, which is the light balance performance of each region in the room. For example, the vector of the light intensity variation is v_intensity=[i1, i2, …, ip], and the vector of the light coverage variation is v_coverage=[c1, c2, …, cq], and the vector of the light balance performance of each region in the room is v_balance=[i1, i2, …, ip, c1, c2, …, cq].

[0128] Step S146: input the global fusion network feature into a second simulation branch of the deep learning simulation module, perform time series analysis on the light duration of each region in the room, extract the effective light duration and continuous light duration of each region in the room in different seasons and different dates, and generate the light duration performance of each region in the room.

[0129] Step S1461: extract the feature part related to the light duration of each region in the room from the global fusion network feature, and the feature part includes the light start time feature content and the light end time feature content of each region in the room in different seasons and different dates.

[0130] In this embodiment, the feature part related to the light duration of each region in the room needs to be extracted from the global fusion network feature. The above feature part can be determined by analyzing the dimensions and meanings of the global fusion network feature. For example, the features of some dimensions correspond to the light start time feature content of each region in the room in different seasons and different dates, and the features of some dimensions correspond to the light end time feature content of each region in the room in different seasons and different dates. By feature selection or dimension division of the global fusion network feature, these feature parts related to the light duration of each region in the room are extracted.

[0131] Step S1462: classify the feature part according to the indoor region division, season division, and date division to generate a region season date feature group corresponding to different seasons and different dates for each indoor region.

[0132] In this embodiment, the indoor region division, season division, and date division are similar to step S1452. The extracted feature part related to the light duration of each region in the room is classified according to the indoor region, season, and date. For example, for each indoor region, the light start time feature content and the light end time feature content of different seasons and different dates are classified respectively to form a region season date feature group corresponding to different seasons and different dates for the indoor region.

[0133] Step S1463: Time span calculation is performed on the light start time feature content and the light end time feature content in each region seasonal date feature group, and total light time content of each indoor region in the corresponding season and date is extracted.

[0134] In this embodiment, time span calculation is required for the light start time feature content and the light end time feature content in each region seasonal date feature group. The method of time span calculation is to subtract the light start time from the light end time to obtain the total light time. For example, for the light start time T_start and the light end time T_end of an indoor region in a season and a date, the total light time content is T_total=T_end-T_start.

[0135] Step S1464: An effective light intensity standard is set, and time content in which the light intensity meets the effective light intensity standard in the total light time content of each indoor region in the corresponding season and date is extracted according to the light intensity feature content in the region seasonal date feature group.

[0136] In this embodiment, the setting of the effective light intensity standard needs to be determined according to the use demand of the indoor space and the human comfort requirement. For example, for the shop region of a commercial building, the effective light intensity standard can be set to 300 lux (lx) or more. Then, according to the light intensity feature content in the region seasonal date feature group, the time period in which the light intensity meets the effective light intensity standard is found in the total light time content. For example, for each time t in the total light time content, it is judged whether the light intensity I_t is greater than or equal to the effective light intensity standard I_standard. If yes, the time belongs to the time content meeting the standard. The above-mentioned time content meeting the standard is sorted to obtain the time content meeting the effective light intensity standard of each indoor region in the corresponding season and date.

[0137] Step S1465: The total duration of the time content meeting the effective light intensity standard is calculated as the effective light time of the indoor regions in different seasons and different dates.

[0138] In this embodiment, for each indoor area, the total duration of time that meets the valid light intensity standard in the corresponding season and corresponding date needs to be calculated. The calculation method of the total duration is to add the duration of all time periods that meet the standard. For example, the time periods that meet the standard are [t1_start, t1_end], [t2_start, t2_end], …, [tn_start, tn_end], and then the total duration is Σ(ti_end-ti_start), where i is from 1 to n. The above total duration is sorted as the valid light time of each indoor area in different seasons and different dates.

[0139] Step S1466: The continuity of the light intensity change content in the total light time content of each indoor area in the corresponding season and corresponding date is analyzed, and the time content that continuously meets the valid light intensity standard without interruption is extracted.

[0140] In this embodiment, the continuity of the light intensity change content in the total light time content of each indoor area in the corresponding season and corresponding date needs to be analyzed. The method of continuity analysis is to find the time period that continuously meets the valid light intensity standard without interruption. For example, starting from the start time of the total light time content, the light intensity at each time is sequentially judged whether it meets the standard, and when a plurality of consecutive times meet the standard, the start and end time of the time period is recorded to form a time content without interruption. The above uninterrupted time content is sorted to obtain the time content that continuously meets the valid light intensity standard without interruption for each indoor area in the corresponding season and corresponding date.

[0141] Step S1467: The longest duration and the average duration of the uninterrupted time content are calculated as the continuous light time of each indoor area in different seasons and different dates.

[0142] In this embodiment, for each indoor area, the longest duration and the average duration of the time content that continuously meets the valid light intensity standard without interruption in the corresponding season and corresponding date needs to be calculated. The calculation method of the longest duration is to find the longest duration in all uninterrupted time periods; the calculation method of the average duration is to add the duration of all uninterrupted time periods and then divide by the number of uninterrupted time periods. The longest duration and the average duration are sorted as the continuous light time of each indoor area in different seasons and different dates.

[0143] Step S1468: The valid light time of each indoor area in different seasons and different dates and the continuous light time of each indoor area in different seasons and different dates are integrated to generate the continuous performance of the daylighting of each indoor area.

[0144] In this embodiment, the effective light time and the continuous light time of each area in the room in different seasons and different dates are integrated. The integration can be splicing, that is, the feature vectors of the two cases are spliced to form a new vector, which is the light duration performance of each area in the room. For example, the vector of the effective light time is v_effective=[e1, e2, …, ep], the vector of the continuous light time is v_continuous=[c1, c2, …, cq], and the vector of the light duration performance of each area in the room is v_duration=[e1, e2, …, ep, c1, c2, …, cq].

[0145] Step S147: integrate the light balance performance of each area in the room and the light duration performance of each area in the room to generate a curtain wall lighting simulation result containing the light balance performance of each area in the room and the light duration performance of each area in the room.

[0146] In this embodiment, the light balance performance of each area in the room and the light duration performance of each area in the room are integrated. The integration can be splicing, that is, the feature vectors of the two performances are spliced to form a new vector, which is the curtain wall lighting simulation result. For example, the vector of the light balance performance of each area in the room is v_balance=[b1, b2, …, bp], the vector of the light duration performance of each area in the room is v_duration=[d1, d2, …, dq], and the vector of the curtain wall lighting simulation result is v_simulation=[b1, b2, …, bp, d1, d2, …, dq].

[0147] Step S150: generating a curtain wall design optimization guide according to the curtain wall lighting simulation result, the curtain wall design optimization guide being used to adjust the curtain wall design elements to optimize the indoor lighting performance.

[0148] Step S151: extracting the light balance performance of each area in the room from the curtain wall lighting simulation result, extracting the light intensity change and the light coverage range change of each indoor area in different time periods, classifying according to the function type of the indoor area, and generating a region function lighting balance analysis result.

[0149] In this embodiment, the indoor area function types can be divided according to the use function of the building, for example, the indoor area function types of a commercial building can be divided into shop area, office area, public rest area, etc. The indoor area lighting balance performance is extracted from the curtain wall lighting simulation results, and then the lighting intensity change and the lighting coverage range change of different time periods are classified according to the function types of the indoor areas. For example, the lighting intensity change and the lighting coverage range change of all shop areas are classified into one category, the lighting intensity change and the lighting coverage range change of all office areas are classified into another category, and so on. The classified results are sorted to generate the area function lighting balance sorting results.

[0150] Step S152: Extract the indoor area lighting duration performance from the curtain wall lighting simulation results, extract the effective lighting time and continuous lighting time of different seasons and different dates of each indoor area, classify them according to the indoor area use frequency, and generate the area use lighting duration sorting results.

[0151] In this embodiment, the indoor area use frequency can be divided according to the use of the building, for example, the indoor area use frequency of a commercial building can be divided into high-frequency use area (such as shop area), medium-frequency use area (such as corridor area), low-frequency use area (such as equipment room area), etc. The indoor area lighting duration performance is extracted from the curtain wall lighting simulation results, and then the effective lighting time and the continuous lighting time of different seasons and different dates are classified according to the use frequency of the indoor areas. For example, the effective lighting time and the continuous lighting time of all high-frequency use areas are classified into one category, the effective lighting time and the continuous lighting time of all medium-frequency use areas are classified into another category, and so on. The classified results are sorted to generate the area use lighting duration sorting results.

[0152] Step S153: Compare the area function lighting balance sorting results with the preset area function lighting balance standard, extract the indoor areas that do not meet the standard and the corresponding feature content that does not meet the standard, and generate the balance optimization direction.

[0153] In this embodiment, the preset area function lighting balance standard needs to be determined according to the function demand of the indoor area and the human comfort requirement. For example, for the shop area, the preset standard deviation of the lighting intensity change is not more than a certain value, the maximum value of the lighting coverage range change is not more than a certain value, etc. The area function lighting balance sorting results are compared with the preset standard to find the indoor areas that do not meet the standard. Then, the feature content corresponding to these indoor areas that do not meet the standard is analyzed, for example, the lighting intensity change is too large, the lighting coverage range change is too large, etc. According to the above feature content, the balance optimization direction is determined, for example, the curtain wall structure needs to be adjusted to reduce the lighting intensity change, or the curtain wall material needs to be adjusted to optimize the lighting coverage range change, etc.

[0154] Step S154: Comparing the region use lighting duration combing result with the preset region use lighting duration standard, extracting indoor regions which do not meet the standard and corresponding feature content which do not meet the standard, and generating a persistence optimization direction.

[0155] In this embodiment, the preset region use lighting duration standard needs to be determined according to the use frequency and use demand of the indoor region. For example, for high-frequency use regions, the preset effective lighting time is not less than a certain time length, and the average length of continuous lighting time is not less than a certain time length. Comparing the region use lighting duration combing result with the preset standard, finding out the indoor regions which do not meet the standard. Then, analyzing the corresponding feature content of these indoor regions which do not meet the standard, such as too short effective lighting time, too short continuous lighting time, etc. According to the above feature content, determine the persistence optimization direction, such as need to adjust the curtain wall structure to increase the effective lighting time, or adjust the curtain wall material to prolong the continuous lighting time, etc.

[0156] Step S155: Extracting the curtain wall design element content that needs to be adjusted according to the balance optimization direction, determining the specific direction of adjusting the curtain wall structure arrangement or the light transmission performance of the curtain wall material, and generating a structure material adjustment suggestion.

[0157] In this embodiment, according to the balance optimization direction, the curtain wall design element content that needs to be adjusted is analyzed. For example, if the balance optimization direction is to reduce the change of lighting intensity, then the curtain wall structure arrangement may need to be adjusted, such as changing the angle or size of the curtain wall panel to optimize the incidence and reflection of light; or adjusting the light transmission performance of the curtain wall material, such as selecting a material with more stable light transmission rate to reduce the change of lighting intensity. Determine the specific direction of adjustment, for example, the adjustment direction of the curtain wall structure arrangement can be to increase or decrease the inclination angle of the curtain wall panel, and the adjustment direction of the light transmission performance of the curtain wall material can be to increase or decrease the light transmission rate of the material, etc. Organize the above adjustment suggestions to generate a structure material adjustment suggestion.

[0158] Step S156: Extracting the curtain wall design element content that needs to be adjusted according to the persistence optimization direction, determining the specific direction of adjusting the curtain wall structure arrangement or the light transmission performance of the curtain wall material, and generating a supplementary structure material adjustment suggestion.

[0159] In this embodiment, according to the persistence optimization direction, the curtain wall design element content that needs to be adjusted is analyzed. For example, if the persistence optimization direction is to increase the effective illumination time, the curtain wall structure arrangement may need to be adjusted, such as changing the orientation or area of the curtain wall to increase the incident time of light; or adjusting the light transmission performance of the curtain wall material, such as selecting a material with higher light transmittance to increase the amount of light transmission. Determine the specific direction of adjustment, for example, the adjustment direction of the curtain wall structure arrangement can be to adjust the orientation of the curtain wall to the south, and the adjustment direction of the light transmission performance of the curtain wall material can be to replace it with a glass with higher light transmittance. Organize the above adjustment suggestions to generate supplementary structural material adjustment suggestions.

[0160] Step S157: Integrate the structural material adjustment suggestion and the supplementary structural material adjustment suggestion to determine the priority order of adjustment and the specific range of adjustment, and generate a curtain wall design optimization guide.

[0161] In this embodiment, the structural material adjustment suggestion and the supplementary structural material adjustment suggestion are integrated, first the importance and urgency of each adjustment suggestion is analyzed to determine the priority order of adjustment. For example, the adjustment suggestion that has a greater impact on human comfort has a higher priority; the adjustment suggestion that is easy to implement can also be appropriately improved in priority. Then, the specific range of each adjustment suggestion is determined, for example, the adjustment range of the curtain wall structure arrangement can be the angle adjustment of some curtain wall panels, and the adjustment range of the light transmission performance of the curtain wall material can be the replacement of the curtain wall material in some areas. Organize the priority order and specific range of adjustment to generate a curtain wall design optimization guide, which can provide specific adjustment direction and implementation range for curtain wall designers to optimize indoor lighting performance.

[0162] Based on the same inventive concept, please refer to Figure 2 , shows the structure schematic block diagram of the deep learning-based curtain wall lighting performance simulation system 100 provided by the embodiment of the present application for executing the above-mentioned deep learning-based curtain wall lighting performance simulation method, which can include a communication unit 110, a machine-readable storage medium 120 and a processor 130.

[0163] In this embodiment, the machine-readable storage medium 120 and the processor 130 are located in the deep learning-based curtain wall lighting performance simulation system 100 and are separately arranged. However, it should be understood that the machine-readable storage medium 120 can also be independent of the deep learning-based curtain wall lighting performance simulation system 100, and can be accessed by the processor 130 through a bus interface. Alternatively, the machine-readable storage medium 120 can also be integrated into the processor 130, and can communicate and interact with external systems through the communication unit 110.

[0164] The processor 130 is the control center of the deep learning-based curtain wall lighting performance simulation system 100, connects various parts of the deep learning-based curtain wall lighting performance simulation system 100 through various interfaces and lines, executes the software programs and / or modules stored in the machine readable storage medium 120 and calls the data stored in the machine readable storage medium 120, executes various functions of the deep learning-based curtain wall lighting performance simulation system 100 and processes data, thereby monitoring the deep learning-based curtain wall lighting performance simulation system 100 as a whole. Optionally, the processor 130 can include one or more processing cores; for example, the processor 130 can integrate an application processor and a modem processor, wherein the application processor mainly processes the operating system, user interface and application program, etc., and the modem processor mainly processes wireless communication. It can be understood that the above-mentioned modem processor can also not be integrated into the processor. Among them, the machine readable storage medium 120 is used to store machine executable instructions for executing the scheme of the present application, and the processor 130 is used to execute the machine executable instructions stored in the machine readable storage medium 120 to realize the deep learning-based curtain wall lighting performance simulation method provided by the foregoing method embodiment.

[0165] It should be noted that, in order to simplify the expression of the present disclosure and help to understand one or more embodiments of the present application, in the foregoing description of the embodiments of the present application, various features are sometimes combined into one embodiment, figure or description thereof.

Claims

1. A method for simulating the performance of a curtain wall based on deep learning, characterized in that, The method comprises: integrate curtain wall design elements and target area lighting elements, the curtain wall design elements include curtain wall structure arrangement, curtain wall material light transmission performance, the lighting elements include different time period light direction, different time period light intensity change; input the curtain wall design elements and the lighting elements into the pre-trained daylighting correlation network construction model to generate the curtain wall part light receiving performance distribution and the indoor space light diffusion performance distribution; construct a curtain wall daylighting dynamic correlation network according to the light receiving performance distribution and the light diffusion performance distribution, the curtain wall daylighting dynamic correlation network is used for presenting the dynamic action relationship between the curtain wall design elements, the lighting elements and the indoor daylighting performance; start the deep learning simulation module to interactively simulate and calculate the curtain wall daylighting dynamic correlation network to generate the curtain wall daylighting simulation result including indoor area daylighting balance performance and indoor area daylighting persistence performance; generate curtain wall design optimization guidance according to the curtain wall daylighting simulation result, the curtain wall design optimization guidance is used for adjusting the curtain wall design elements to optimize the indoor daylighting performance; the curtain wall design elements and the lighting elements into the pre-trained daylighting correlation network construction model to generate the curtain wall part light receiving performance distribution and the indoor space light diffusion performance distribution, comprising: perform spatial digital conversion on the curtain wall structure arrangement in the curtain wall design elements to form a curtain wall structure digital form with coordinate attributes, each structure part in the curtain wall structure digital form has corresponding position parameters and size parameters; perform feature conversion on the curtain wall material light transmission performance in the curtain wall design elements to form a material light transmission feature form recognizable by the daylighting correlation network construction model, the dimension of the material light transmission feature form matches the input layer dimension of the daylighting correlation network construction model; perform direction conversion on the different time period light direction in the lighting elements to form a light action direction set, each direction information in the light action direction set corresponds to the light incidence direction of a specific time period; perform time sequence conversion on the different time period light intensity change in the lighting elements to form a light intensity time sequence feature, the time interval of the light intensity time sequence feature is consistent with the time period division of the light action direction set; input the curtain wall structure digital form, the material light transmission feature form, the light action direction set and the light intensity time sequence feature into the input layer of the pre-trained daylighting correlation network construction model; perform spatial analysis on the curtain wall structure digital form through the structure feature processing layer of the daylighting correlation network construction model to generate the spatial position features of the curtain wall parts; perform attribute analysis on the material light transmission feature form through the material feature processing layer of the daylighting correlation network construction model to generate the material light transmission attribute features of the curtain wall parts; perform space-time correlation on the light action direction set and the light intensity time sequence feature through the light feature processing layer of the daylighting correlation network construction model to generate the light action features of each time period; The space position feature, the material light transmission property feature and the light action feature are cross-element fused by the element interaction layer of the daylighting correlation network construction model to generate light receiving comprehensive features of each part of the curtain wall; A light receiving performance distribution of each part of the curtain wall is generated according to the spatial distribution of the light receiving comprehensive features of each part of the curtain wall; The indoor diffusion simulation layer of the daylighting correlation network construction model simulates the indoor diffusion path and energy change process of light after penetrating through the curtain wall according to the light receiving performance distribution and the spatial parameters of the digital form of the curtain wall structure to generate an indoor space light diffusion performance distribution. 2.The deep learning-based curtain wall daylighting performance simulation method according to claim 1, characterized in that, The curtain wall daylighting dynamic correlation network is constructed according to the light receiving performance distribution and the light diffusion performance distribution, including: The light receiving intensity performance, light receiving time performance and light incidence angle performance of each part of the curtain wall are extracted from the light receiving performance distribution as the first type of node information of the curtain wall daylighting dynamic correlation network; The light reaching intensity performance, light coverage range performance and light balance degree performance of each indoor area are extracted from the light diffusion performance distribution as the second type of node information of the curtain wall daylighting dynamic correlation network; The curtain wall panel size performance, panel splicing gap performance and support frame shielding performance are extracted from the curtain wall structure arrangement of the curtain wall design elements as the third type of node information of the curtain wall daylighting dynamic correlation network; The material light transmission rate performance, material reflectivity performance and material absorption rate performance are extracted from the material light transmission performance of the curtain wall design elements as the fourth type of node information of the curtain wall daylighting dynamic correlation network; The light intensity peak value performance at different time periods and the light direction change performance at different time periods are extracted from the light action elements as the fifth type of node information of the curtain wall daylighting dynamic correlation network; A node group of the curtain wall daylighting dynamic correlation network is established, which includes curtain wall part nodes, indoor area nodes, curtain wall structure nodes, curtain wall material nodes and light action nodes, each type of node corresponding to the above different types of node information; The action relationship between the curtain wall part nodes and the indoor area nodes is analyzed, and a first type of connection relationship between the curtain wall part nodes and the indoor area nodes is established according to the diffusion path of light from the curtain wall part to the indoor area, the action degree of the first type of connection relationship being determined according to the energy transmission in the light diffusion process; The action relationship between the curtain wall part nodes and the curtain wall structure nodes is analyzed, and a second type of connection relationship between the curtain wall part nodes and the curtain wall structure nodes is established according to the influence of the curtain wall structure on the light receiving of the curtain wall part, the action degree of the second type of connection relationship being determined according to the influence of the structure shielding on the light receiving; The action relationship between the curtain wall part nodes and the curtain wall material nodes is analyzed, and a third type of connection relationship between the curtain wall part nodes and the curtain wall material nodes is established according to the influence of the curtain wall material on the light absorption and reflection of the curtain wall part, the action degree of the third type of connection relationship being determined according to the influence of the material light transmission rate on the light transmission; analyze the action relationship between the curtain wall part node and the light action node, establish a fourth type of connection relationship between the curtain wall part node and the light action node according to the influence of the light action on the light bearing of the curtain wall part, and the action degree of the fourth type of connection relationship is determined according to the influence of the light intensity and the light direction on the light bearing of the curtain wall part; analyze the action relationship between the indoor area node and the curtain wall structure node, establish a fifth type of connection relationship between the indoor area node and the curtain wall structure node according to the influence of the curtain wall structure on the light diffusion of the indoor area, and the action degree of the fifth type of connection relationship is determined according to the influence of the structure arrangement on the indoor light distribution; analyze the action relationship between the indoor area node and the curtain wall material node, establish a sixth type of connection relationship between the indoor area node and the curtain wall material node according to the influence of the curtain wall material on the light intensity of the indoor area, and the action degree of the sixth type of connection relationship is determined according to the influence of the light transmittance of the material on the indoor light intensity; Integrate the above various types of nodes, various types of node information and various types of connection relationships to form a complete curtain wall daylighting dynamic correlation network, which is used to present the dynamic action relationship between curtain wall design elements, light action elements and indoor daylighting performance. 3.The deep learning-based curtain wall daylighting performance simulation method of claim 1, wherein, The starting deep learning simulation module interacts with the curtain wall daylighting dynamic correlation network to generate a curtain wall daylighting simulation result containing indoor area daylighting balance performance and indoor area daylighting persistence performance, including: The network structure of the curtain wall daylighting dynamic correlation network is transformed, the node information and connection relationship information in the curtain wall daylighting dynamic correlation network are transformed into a network feature form recognizable by the deep learning simulation module, the number of rows of the network feature form corresponds to the number of nodes in the curtain wall daylighting dynamic correlation network, and the number of columns corresponds to the sum of the node information dimension and the connection relationship feature dimension; The network feature form is input into the network feature processing layer of the deep learning simulation module, the neighborhood feature of the network feature form is integrated through network convolution operation, and the neighborhood integration feature of each node is generated; The neighborhood integration feature is analyzed by the multi-level network analysis layer of the deep learning simulation module to extract local correlation features and overall correlation features of the curtain wall daylighting dynamic correlation network, and generate multi-level network correlation features; The multi-level network correlation features are input into the feature fusion layer of the deep learning simulation module, the fusion proportion of different level network correlation features is dynamically calculated through the attention mechanism, the multi-level network correlation features are weighted and fused to generate global fusion network features; The global fusion network features are input into the first simulation branch of the deep learning simulation module, the balance of the light distribution of the indoor area is analyzed, the light intensity change and the light coverage change of the indoor area at different time periods are extracted, and the indoor area daylighting balance performance is generated. input the global fusion network features into a second simulation branch of the deep learning simulation module, perform time sequence analysis on the light duration of each region in the room, extract the effective light duration and continuous light duration of each region in the room in different seasons and on different dates, and generate the light duration performance of each region in the room; integrate the light balance performance of each region in the room and the light duration performance of each region in the room to generate the curtain wall daylighting simulation result containing the light balance performance of each region in the room and the light duration performance of each region in the room.

4. The deep learning-based curtain wall daylighting performance simulation method according to claim 3, characterized in that, The network feature processing layer of the deep learning simulation module integrates the network feature form to generate the neighborhood integrated features of each node, including: identify the direct neighboring node group of each node in the curtain wall daylighting dynamic correlation network, which includes all nodes directly correlated with the node through the connection relationship; calculate the connection relationship action degree between each node and each node in its direct neighboring node group, which is determined according to the action degree parameter of the connection relationship; standardize the feature row information corresponding to each node in the network feature form and the feature row information corresponding to each node in the direct neighboring node group of the node to eliminate the dimension difference of the feature quantity and generate the standardized feature row information; perform element-level operation on the standardized feature row information to generate a set of adjacent node feature operation results; perform weighted operation on each result in the set of adjacent node feature operation results and the corresponding connection relationship action degree to generate a set of weighted adjacent node features; perform element-level summation operation on all features in the set of weighted adjacent node features to generate a neighborhood feature summation result; perform regularization processing on the neighborhood feature summation result to eliminate the feature value difference caused by the difference in the number of adjacent nodes of different nodes, and generate a regularized neighborhood feature; combine the original feature row information corresponding to each node with the regularized neighborhood feature of the node to generate the neighborhood integrated features of each node.

5. The deep learning-based curtain wall daylighting performance simulation method according to claim 3, characterized in that, The multi-level network analysis layer of the deep learning simulation module performs network structure analysis of different levels on the neighborhood integrated features to extract local correlation features and overall correlation features of the curtain wall daylighting dynamic correlation network and generate multi-level network correlation features, including: set the first level analysis range, which corresponds to the direct neighboring range of nodes in the curtain wall daylighting dynamic correlation network, for extracting local correlation features; perform standardization processing on the neighborhood integrated features of each node within the first level analysis range, perform adjacent node feature difference analysis on the standardized neighborhood integrated features, extract the feature difference between the node and its direct neighboring nodes, and generate the local feature difference situation; perform feature screening on the local feature difference situation to retain key feature content that can reflect the local correlation relationship, and generate the local correlation feature; set the second level analysis range, which corresponds to the indirect neighboring range of nodes in the curtain wall daylighting dynamic correlation network, i.e., the direct neighboring range of direct neighboring nodes, for extracting intermediate level correlation features; In the second level analysis range, the neighborhood integration features of each node are standardized, the standardized neighborhood integration features are analyzed for indirect adjacent node feature transmission, the influence of indirect adjacent node features transmitted by direct adjacent nodes is extracted, and an intermediate level feature transmission situation is generated; The intermediate level feature transmission situation is feature extracted, key feature content reflecting the intermediate level correlation is retained, and an intermediate level correlation feature is generated; A third level analysis range is set, the third level analysis range corresponds to the whole node range of the curtain wall lighting dynamic correlation network, and overall correlation features are extracted; In the third level analysis range, the neighborhood integration features of all nodes are standardized, the standardized neighborhood integration features are globally statistically analyzed, the overall distribution and mutual influence of global node features are extracted, and a global feature sorting result is generated; The global feature sorting result is feature integrated, the global feature distribution and mutual influence are converted into a feature form reflecting the overall correlation, and an overall correlation feature is generated; The local correlation feature, the intermediate level correlation feature, and the overall correlation feature are integrated to generate a multi-level network correlation feature.

6. The deep learning-based curtain wall daylighting performance simulation method according to claim 3, characterized in that, The global fusion network feature is input into a first simulation branch of the deep learning simulation module, the light distribution of each indoor area is balanced, the light intensity change and the light coverage range change of each indoor area at different times are extracted, and the indoor area lighting balance performance is generated, including: The feature part related to the light distribution of each indoor area is extracted from the global fusion network feature, and the feature part includes the light intensity feature content and the light coverage range feature content of each indoor area at different times; The feature part is classified according to indoor area division and time division to generate a region-time feature group corresponding to each indoor area at different times; The light intensity feature content in each region-time feature group is analyzed for time series change, the light intensity change content of the same indoor area at adjacent times is extracted, and a region-time light intensity change sequence is generated; The region-time light intensity change sequence is analyzed for change trend, and the significant change content, the average change content, and the change frequency content in the change sequence are extracted as the light intensity change of each indoor area at different times; The light coverage range feature content in each region-time feature group is analyzed for spatial change, the light coverage area change content and the light coverage form change content of the same indoor area at different times are extracted, and a region-time light coverage change sequence is generated; The region-time light coverage change sequence is analyzed for change feature, and the maximum change content, the average change content, and the change duration content in the change sequence are extracted as the light coverage range change of each indoor area at different times; The light intensity change of each indoor area at different times and the light coverage range change of each indoor area at different times are integrated to generate the indoor area lighting balance performance.

7. The deep learning-based curtain wall daylighting performance simulation method according to claim 3, characterized in that, The global fusion network feature is input into a second simulation branch of the deep learning simulation module, and time sequence analysis is performed on the light duration of each indoor area to extract the effective light duration of each indoor area in different seasons and on different dates, and generate the indoor area lighting duration performance, including: A feature part related to the light duration of each indoor area is extracted from the global fusion network feature, and the feature part includes light start time feature content and light end time feature content of each indoor area in different seasons and on different dates; The feature part is classified according to indoor area division, seasonal division, and date division to generate a region-season-date feature group corresponding to each indoor area in different seasons and on different dates; The light start time feature content and the light end time feature content in each region-season-date feature group are calculated to extract the total light duration of each indoor area in the corresponding season and on the corresponding date; An effective light intensity standard is set, and the total light duration of each indoor area in the corresponding season and on the corresponding date is extracted according to the light intensity feature content in the region-season-date feature group; The total duration of the time content that meets the effective light intensity standard is calculated as the effective light duration of each indoor area in different seasons and on different dates; The light intensity change content in the total light duration of each indoor area in the corresponding season and on the corresponding date is analyzed for continuity to extract the time content that continuously meets the effective light intensity standard without interruption; The longest duration and the average duration of the uninterrupted time content are calculated as the continuous light duration of each indoor area in different seasons and on different dates; The effective light duration of each indoor area in different seasons and on different dates and the continuous light duration of each indoor area in different seasons and on different dates are integrated to generate the indoor area lighting duration performance. 8.The deep learning based curtain wall daylighting performance simulation method of claim 1, wherein, The curtain wall lighting simulation result is used to generate curtain wall design optimization guidance, including: The indoor area lighting balance performance is extracted from the curtain wall lighting simulation result to extract the light intensity change and the light coverage range change of each indoor area at different times, and the performance is classified according to the indoor area function type to generate a region function lighting balance sorting result; The indoor area lighting duration performance is extracted from the curtain wall lighting simulation result to extract the effective light duration and the continuous light duration of each indoor area in different seasons and on different dates, and the performance is classified according to the indoor area usage frequency to generate a region usage lighting duration sorting result; The region function lighting balance sorting result is compared with a preset region function lighting balance standard to extract indoor areas that do not meet the standard and corresponding feature content that does not meet the standard, and a balance optimization direction is generated; The region usage lighting duration sorting result is compared with a preset region usage lighting duration standard to extract indoor areas that do not meet the standard and corresponding feature content that does not meet the standard, and a duration optimization direction is generated; According to the uniformity optimization direction, extract the curtain wall design element content that needs to be adjusted, determine the specific direction of adjusting the curtain wall structure arrangement or the light transmission performance of the curtain wall material, and generate a structure material adjustment suggestion; According to the continuity optimization direction, extract the curtain wall design element content that needs to be adjusted, determine the specific direction of adjusting the curtain wall structure arrangement or the light transmission performance of the curtain wall material, and generate a supplementary structure material adjustment suggestion; Integrate the structure material adjustment suggestion and the supplementary structure material adjustment suggestion, determine the priority order of adjustment and the specific range of adjustment, and generate a curtain wall design optimization guide. 9.A deep learning-based curtain wall daylighting performance simulation system, characterized in that, It comprises: a processor; a machine-readable storage medium for storing machine-executable instructions of the processor; wherein the processor is configured to execute the machine-executable instructions to perform the curtain wall daylighting performance simulation method based on deep learning in any one of claims 1 to 8.

Images