Displacement prediction method and system for glass curtain wall

Abstract

The invention relates to the technical field of constructional engineering structures, and discloses a displacement prediction method and system for a glass curtain wall, and the method comprises the steps: collecting boundary displacement response information and boundary acceleration response information of the glass curtain wall under the condition of external force application; constructing a motion equation of the glass curtain wall based on the total stiffness matrix, the total mass matrix, the boundary displacement response information and the boundary acceleration response information; and inputting the real-time sampling point parameters of the glass curtain wall into a preset full-connection neural network model to obtain a current displacement prediction result, and iteratively training the full-connection neural network model according to residual information between the motion equation and the current displacement prediction result. According to the method, iterative training is carried out on residual information obtained by combining the structural dynamics model and the deep learning model, so that the prediction model can adapt to different glass curtain walls, it is guaranteed that displacement prediction is carried out under the real-time or near-real-time condition, the prediction precision is improved, and the calculation efficiency and real-time performance are improved.

Description

Technical Field

[0001] The invention relates to the technical field of building engineering structures, and in particular to a displacement prediction method and system for a glass curtain wall. Background Art

[0002] In the current field of high-rise building design and construction, glass curtain walls are a popular external structure, which is widely popular due to its beautiful appearance, light transmission and light weight. However, glass curtain walls may be displaced and deformed when subjected to wind loads, which poses a potential threat to the safety of the building. In order to predict these displacements and deformations, two main technical means are currently used: calculation based on empirical formulas or manual finite element analysis.

[0003] First, the empirical formula method predicts displacements through a simplified mathematical model based on historical data and expert knowledge. However, the accuracy of the mathematical model depends on the quality and quantity of empirical data, and it is difficult to adapt to complex actual engineering conditions. In addition, the empirical formula method does not take into account the variability of wind loads and the complexity of curtain wall structures, resulting in obvious limitations in prediction accuracy.

[0004] On the other hand, the finite element analysis method can more accurately simulate the physical behavior of the curtain wall structure by establishing a detailed numerical model, thereby obtaining an accurate displacement prediction. Although this method can more realistically reflect the stress and deformation characteristics of the structure, its calculation process is complex and requires a lot of computing resources and time. The amount of calculation is huge in the analysis of large-scale or complex structures.

[0005] Therefore, the disadvantage of the existing technology is that it is difficult to adapt to the diversity and complexity of curtain wall structures. For curtain wall systems with new materials or innovative designs, the prediction model may not be able to accurately reflect their actual performance. For engineering practices that require rapid response, the computational efficiency of finite element analysis is often difficult to meet the needs of real-time or near real-time analysis. Summary of the invention

[0006] In view of this, the present invention provides a method and system for predicting the displacement of a glass curtain wall, so as to solve the problem of how to improve the real-time performance and accuracy of the displacement prediction of a glass curtain wall building structure.

[0007] In a first aspect, the present invention provides a method for predicting the displacement of a glass curtain wall, the method comprising: calculating the overall stiffness matrix and the overall mass matrix of the glass curtain wall based on the principle of structural dynamics;

[0008] Collect the boundary displacement response information and boundary acceleration response information of the glass curtain wall under the condition of external force;

[0009] Constructing the motion equation of the glass curtain wall based on the overall stiffness matrix, the overall mass matrix, the boundary displacement response information, and the boundary acceleration response information;

[0010] The current sampling time and sampling point position information of the glass curtain wall are used as real-time sampling point parameters and input into a preset fully connected neural network model to obtain a current displacement prediction result. According to the residual information between the motion equation and the current displacement prediction result, the fully connected neural network model is iteratively trained. When the iteration termination condition is reached, a real-time displacement prediction model of the glass curtain wall is obtained.

[0011] The next sampling time and sampling point position information of the glass curtain wall are input into the real-time displacement prediction model to obtain the real-time displacement prediction result.

[0012] The embodiment of the present invention calculates the overall stiffness matrix and mass matrix of the glass curtain wall based on the principle of structural dynamics, and constructs the motion equation in combination with the boundary displacement response information and acceleration response information. It more accurately describes the stress and deformation behavior of the curtain wall, so that the prediction model can adapt to complex actual engineering conditions. At the same time, a fully connected neural network model is introduced, and displacement prediction is performed using deep learning technology. Iterative training is performed in combination with the residual information of the motion equation, so that the prediction model can adapt to different types of curtain wall materials, thereby ensuring displacement prediction under real-time or near real-time conditions. By combining the structural dynamics model with the deep learning model, the prediction accuracy is improved, and the efficiency and real-time performance of the calculation are enhanced.

[0013] In an optional implementation, the calculation of the overall stiffness matrix and the overall mass matrix of the glass curtain wall based on the principle of structural dynamics includes:

[0014] Obtaining physical characteristic parameters and geometric structure parameters of the glass curtain wall;

[0015] Dividing the structure of the glass curtain wall into a plurality of plane units, wherein each of the plane units has a local coordinate system;

[0016] Calculate the unit stiffness matrix and unit mass matrix in the local coordinate system according to the physical characteristic parameters and geometric structure parameters corresponding to each plane unit;

[0017] The unit stiffness matrix and the unit mass matrix of each plane unit are combined into an overall stiffness matrix and an overall mass matrix of the glass curtain wall.

[0018] The embodiment of the present invention divides the curtain wall structure into multiple plane units, calculates the stiffness matrix and mass matrix of each plane unit in a local coordinate system, and then combines the matrices of each unit into an overall stiffness matrix and a mass matrix. The overall stiffness matrix and mass matrix calculated based on the principle of structural dynamics can accurately reflect the characteristics of the glass curtain wall structure under the action of external forces, thereby constructing a more accurate motion equation, further improving the prediction ability of the subsequent trained glass curtain wall model, and ensuring that the prediction result is more accurate.

[0019] In an optional implementation, the motion equation of the glass curtain wall is expressed by the following formula:

[0020]

[0021] Where M represents the overall mass matrix, K represents the overall stiffness matrix, represents the boundary acceleration response information, and v(t) represents the boundary displacement response information.

[0022] In the embodiment of the present invention, the mass matrix (M) and the stiffness matrix (K) can fully describe the physical properties of the entire glass curtain wall, and use the boundary acceleration response and boundary displacement response information to introduce the boundary conditions in the actual working conditions into the motion equation, thereby ensuring that the final motion equation more accurately simulates the actual boundary conditions and improves the accuracy of the motion equation.

[0023] In an optional implementation, the current sampling time and sampling point position information of the glass curtain wall are used as real-time sampling point parameters and input into a preset fully connected neural network model to obtain a current displacement prediction result. According to the residual information between the motion equation and the current displacement prediction result, the fully connected neural network model is iteratively trained. When the iteration termination condition is reached, a real-time displacement prediction model of the glass curtain wall is obtained, which specifically includes:

[0024] Constructing a fully connected neural network, and setting corresponding parameters of each neural network to initial values; wherein the neural network parameters include sampling point parameters, the number of different network layers, network width, parameters of activation function formula, parameters of loss function formula, and network learning rate; input data of the fully connected neural network includes sampling point parameters, and output data of the fully connected neural network includes displacement responses corresponding to sampling points inside the curtain wall;

[0025] In the current round of iterative training, the first sampling point parameter is input into the fully connected neural network to obtain a predicted displacement response corresponding to the first sampling point;

[0026] Performing partial derivative analysis according to the predicted displacement response using the principle of automatic differentiation to obtain a first partial derivative and a second partial derivative, wherein the first partial derivative represents a second-order partial derivative of the displacement response with respect to time, and the second partial derivative represents a second-order partial derivative of the displacement response with respect to space;

[0027] Calculating residual information about the motion equation based on the first partial derivative and the second partial derivative;

[0028] Constructing a probability density function according to the residual information;

[0029] Continue sampling based on the probability density function to obtain the second sampling point parameters, and use the gradient descent algorithm to update the slope optimization parameters in the activation function formula;

[0030] The second sampling point parameters are updated to the first sampling point parameters, and the next round of iterative training is continued; wherein the predicted displacement response in each round is also used to be brought into the loss function formula to calculate the target loss function value of each round;

[0031] When the target loss function value is lower than the preset loss function value, or the round of iterative training reaches the preset number of iterations, the iterative training is stopped to obtain the real-time displacement prediction model of the glass curtain wall.

[0032] The embodiment of the present invention is modeled by a fully connected neural network, real-time prediction is performed in combination with real-time data of sampling points, partial derivative analysis is performed using automatic differentiation, the analysis results of the partial derivatives are combined with the motion equation, and a probability density function is constructed. The sampling point parameters are adaptively updated based on residual information in each iterative training, and the sampling is focused on areas where there may be large errors in the overall sampling prediction results of the glass curtain wall, thereby improving the training efficiency and accuracy of the overall model. In each round of iteration, the activation function and network parameters of the neural network are continuously optimized by the gradient descent algorithm, and the training process is continuously updated by calculating the loss value of each round. The residual analysis with the motion equation is used to guide the training of the model, which not only enables the displacement results predicted by the neural network to conform to the physical laws of the motion of the glass curtain wall, but also obtains a real-time displacement prediction model for the glass curtain wall with higher interpretability.

[0033] In an optional embodiment, the method further includes:

[0034] Performing feature expansion on the sampling point parameters based on the feature expansion layer in the fully connected neural network to obtain feature expansion parameters of the current round;

[0035] Calculate the gradient information of the target loss function value relative to the feature expansion parameter using a back propagation algorithm;

[0036] The weight parameters and bias parameters in the activation function formula in the next round are updated according to the gradient information.

[0037] The embodiment of the present invention expands the dimension of the input data after the input layer through feature expansion, increases the high-dimensional representation of the data, and the back propagation algorithm can further optimize the weight parameters and bias of the model to ensure that all parameters of the model can work in a coordinated manner, thereby improving the prediction accuracy of the trained model.

[0038] In an optional implementation, the residual information includes a residual loss term and an integral of all residual loss terms, and the probability density function is represented by the following formula:

[0039]

[0040] Among them, p(x) represents the probability density function of the residual loss term, A represents the integral of all residual loss terms, ε(x) represents the residual loss term, T represents the residual sampling area, and the residual sampling area represents the sampling point domain where the residual is higher than the preset standard value; the k value represents the sampling concentration of the sampling points in the residual sampling area, and the c value represents the sampling uniformity of the sampling points in the residual sampling area.

[0041] The embodiment of the present invention determines the probability density function through the integral of all residual loss terms, residual loss terms, sampling concentration, sampling uniformity and residual sampling area. It should be noted that the residual information refers to the residual information calculated based on the first partial derivative and the second partial derivative and the motion equation. Since the motion equation of the glass curtain wall is combined, the constructed probability density function will be continuously updated during iterative training, so that the prediction results of the trained model are more accurate.

[0042] In an optional implementation, the loss function formula is:

[0043] L(t)=w cz loss IC +w0loss PDE +w b loss BC ;

[0044]

[0045] Among them, w cz Represents the initial loss weight coefficient, w b Represents the boundary loss weight coefficient, w o Represents the loss weight coefficient of the motion equation, loss IC Indicates the network initial value loss, loss BC Represents the network boundary loss, loss PDErepresents the motion equation loss, L(t) represents the loss function of the entire network, and N b Represents the boundary loss function sampling area, N s Represents the sampling area of ​​the initial loss function, N T represents the sampling area of ​​the motion equation loss function, t represents the sampling point, R represents the overall sampling area, T represents the residual sampling area, Represents the network parameters x at the initial sampling point c ,y c ,t c The predicted displacement response inside the curtain wall corresponding to the sampling point, u c represents the actual displacement response of the initial sampling point, Represents the parameter x at any sampling point during network training b ,y b ,t b The predicted displacement response inside the curtain wall corresponding to the sampling point, u b represents the actual displacement response of any sampling point during network training, f(t T ) represents the residual value of the network prediction by the equation of motion.

[0046] In the embodiment of the present invention, the loss function includes information on three aspects: initial value loss, boundary loss and motion equation loss. The initial value loss and boundary loss correspond to the displacement response of the curtain wall under specific conditions, while the motion equation loss corresponds to the matching degree between the network prediction and the physical equation of the glass curtain wall. Different weight coefficients are also introduced into the loss function to balance the three loss functions, so that the model prediction results are more consistent with the actual physical movement phenomenon of the glass curtain wall, thereby improving the accuracy of the final training model.

[0047] In a second aspect, the present invention provides a displacement prediction system for a glass curtain wall, the device comprising:

[0048] The overall matrix calculation module is used to calculate the overall stiffness matrix and overall mass matrix of the glass curtain wall based on the principle of structural dynamics;

[0049] The boundary response information acquisition module is used to acquire the boundary displacement response information and boundary acceleration response information of the glass curtain wall under the condition of external force application;

[0050] A motion equation construction module, used to construct the motion equation of the glass curtain wall based on the overall stiffness matrix, the overall mass matrix, the boundary displacement response information and the boundary acceleration response information;

[0051] An iterative training module is used to use the current sampling time and sampling point position information of the glass curtain wall as real-time sampling point parameters, and input them into a preset fully connected neural network model to obtain a current displacement prediction result, and iteratively train the fully connected neural network model according to the residual information between the motion equation and the current displacement prediction result, and obtain a real-time displacement prediction model of the glass curtain wall when an iteration termination condition is reached;

[0052] The real-time displacement prediction module is used to input the next sampling time and sampling point position information of the glass curtain wall into the real-time displacement prediction model to obtain the real-time displacement prediction result.

[0053] In a third aspect, the present invention provides a computer device, comprising: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the method for predicting the displacement of a glass curtain wall according to the first aspect or any corresponding embodiment thereof by executing the computer instructions.

[0054] In a fourth aspect, the present invention provides a computer-readable storage medium having computer instructions stored thereon, the computer instructions being used to enable a computer to execute the method for predicting the displacement of a glass curtain wall according to the first aspect or any corresponding embodiment thereof.

[0055] In a fifth aspect, the present invention provides a computer program product, comprising computer instructions, wherein the computer instructions are used to enable a computer to execute the method for predicting the displacement of a glass curtain wall according to the first aspect or any corresponding embodiment thereof. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0057] Figure 1 is a schematic flow chart of a method for predicting displacement of a glass curtain wall according to an embodiment of the present invention;

[0058] Figure 2 is a schematic flow chart of another method for predicting displacement of a glass curtain wall according to an embodiment of the present invention;

[0059] Figure 3 is a schematic diagram of structural division of a method for predicting displacement of a glass curtain wall according to an embodiment of the present invention;

[0060] Figure 4is a schematic flow chart of another method for predicting displacement of a glass curtain wall according to an embodiment of the present invention;

[0061] Figure 5 is a schematic diagram of a motion equation loss function of a displacement prediction method for a glass curtain wall according to an embodiment of the present invention;

[0062] Figure 6 is a schematic diagram of an overall loss function of a method for predicting displacement of a glass curtain wall according to an embodiment of the present invention;

[0063] Figure 7 is a structural block diagram of a displacement prediction system for a glass curtain wall according to an embodiment of the present invention;

[0064] Figure 8 It is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0065] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0066] In the current field of high-rise building design and construction, glass curtain walls are favored for their beauty, light transmission and light weight. However, the displacement and deformation of glass curtain walls under wind loads may pose a potential threat to building safety. Existing displacement prediction methods mainly rely on two technical means: empirical formulas and finite element analysis, but each of them has certain disadvantages.

[0067] The empirical formula method is based on historical data and expert knowledge and uses a simplified mathematical model to predict displacement. Although this method is easy to operate and has a fast calculation speed, its prediction accuracy is limited by the quality and quantity of empirical data and is difficult to adapt to complex actual engineering conditions. In addition, due to insufficient consideration of the variability of wind loads and the complexity of curtain wall structures, the prediction accuracy of the empirical formula method has obvious defects.

[0068] The finite element analysis method simulates the physical behavior of the curtain wall structure by establishing a detailed numerical model to obtain a more accurate displacement prediction. Even though this method can more realistically reflect the structural characteristics, its calculation process is complicated and requires a lot of computing resources and time. The efficiency and practicality of finite element analysis are not high.

[0069] Based on the existing technology, there are mainly the following disadvantages:

[0070] (1) Insufficient adaptability. Existing methods are difficult to adapt to the diversity and complexity of curtain wall structures. For curtain wall systems made of new materials or innovative designs, the prediction model may not accurately reflect their actual performance.

[0071] (2) Low computational efficiency: When faced with engineering practices that require rapid response, the computational efficiency of finite element analysis is often unable to meet the needs of real-time or near real-time analysis.

[0072] (3) Limited generalization ability: The prediction results of the empirical formula method are usually limited to specific application scenarios and lack sufficient generalization ability, making it difficult to extend to a wider range of engineering applications.

[0073] (4) Strong data dependence: Existing methods rely heavily on the quality and completeness of historical data. Data deficiencies or deviations will directly affect the accuracy of prediction results.

[0074] (5) Lack of real-time update mechanism. As the building is used for a long time and the environmental conditions change, the existing methods fail to achieve real-time update and self-optimization of the prediction model.

[0075] The embodiment of the present invention provides a method for predicting the displacement of a glass curtain wall, which is applied to the design and construction of glass curtain walls of high-rise buildings, such as in scenarios where the stress and displacement changes of the curtain wall structure need to be monitored in real time. By integrating the unit stiffness matrix and the unit mass matrix in the local coordinate system, and combining the known displacement response of the curtain wall boundary, the displacement of a specific point inside the glass curtain wall can be accurately inverted and predicted, which not only improves the accuracy and efficiency of the prediction, but also enhances the generalization ability and adaptability of the model through embedded physical laws, and has the ability of real-time monitoring and self-optimization, effectively solving the limitations of traditional displacement prediction methods in terms of computational complexity, data dependence, and real-time updating.

[0076] According to an embodiment of the present invention, an embodiment of a method for predicting displacement of a glass curtain wall is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0077] In this embodiment, a method for predicting the displacement of a glass curtain wall is provided, which can be used in the above-mentioned computer. Figure 1 is a flow chart of a method for predicting displacement of a glass curtain wall according to an embodiment of the present invention. Figure 1 As shown, the process includes the following steps:

[0078] Step S101, calculating the overall stiffness matrix and overall mass matrix of the glass curtain wall based on the principle of structural dynamics.

[0079] It should be noted that the principle of structural dynamics refers to the response law of the structure under the action of external force, including displacement, velocity, acceleration, etc. The stiffness matrix refers to the displacement deformation of the structure under the action of unit external force, and the mass matrix refers to the mass distribution of the structure.

[0080] Specifically, step S101 is implemented by using the finite element method (FEM), the glass curtain wall is discretized and divided into a plurality of finite element units, the mass matrix and stiffness matrix of the units are calculated respectively, and the overall mass matrix and stiffness matrix are further calculated for the subsequent process of constructing the motion equation.

[0081] Step S102: collecting boundary displacement response information and boundary acceleration response information of the glass curtain wall under the condition of external force being applied.

[0082] It should be noted that the boundary displacement response information refers to the displacement change of the boundary position of the glass curtain wall under the action of external force, and the boundary acceleration response information refers to the acceleration change of the boundary position of the glass curtain wall under the action of external force.

[0083] In actual operation, displacement sensors and acceleration sensors installed at key locations of the glass curtain wall are used to monitor its response information in real time. In essence, the boundary displacement response information and boundary acceleration response information characterize the actual dynamic response of the glass curtain wall under the action of external forces, providing a basis for building an accurate physical model.

[0084] Step S103, constructing the motion equation of the glass curtain wall based on the overall stiffness matrix, the overall mass matrix, the boundary displacement response information and the boundary acceleration response information.

[0085] It should be noted that the motion equation describes the dynamic behavior of the glass curtain wall under the action of external forces, and this embodiment adopts the form of a second-order partial differential equation.

[0086] Specifically, according to the finite element method and dynamics theory, the motion equation of the glass curtain wall is constructed using the overall stiffness matrix and mass matrix calculated in step S101 and the boundary response information collected in step S102.

[0087] Step S104, taking the current sampling time and sampling point position information of the glass curtain wall as real-time sampling point parameters, and inputting them into a preset fully connected neural network model to obtain a current displacement prediction result, and iteratively training the fully connected neural network model according to the residual information between the motion equation and the current displacement prediction result, and when the iteration termination condition is reached, a real-time displacement prediction model of the glass curtain wall is obtained.

[0088] It should be noted that the fully connected neural network model refers to an artificial neural network architecture in which the neurons in each layer are connected to each neuron in the next layer, and the sampling point parameters include the current time and the location of the sampling point.

[0089] Specifically, the current sampling time of the glass curtain wall and the sampling point position are used as inputs and passed into the preset fully connected neural network model. In this embodiment, the FEM-PINN network architecture is adopted. The preset neural network model has been pre-trained with a large amount of historical data, and can predict the displacement response of the glass curtain wall based on the input spatiotemporal information. Through back propagation and loss function optimization, the weights and biases of the network are continuously adjusted until the residual is minimized to obtain accurate displacement prediction.

[0090] It can be understood that compared with the traditional finite element method, FEM-PINN reduces the demand for computing resources by learning data-driven models, greatly improving the solution speed, which is especially suitable for large-scale or multi-parameter problems. In addition, FEM-PINN directly embeds the physical laws corresponding to the motion equations of the glass curtain wall into the neural network model, ensuring the accuracy of the prediction results and the consistency with the physical laws. It is especially suitable for complex boundary conditions and load conditions. Even under limited data samples, FEM-PINN can capture the basic laws of the structural response of the glass curtain wall and effectively generalize it, providing flexibility and adaptability for structural design and analysis.

[0091] Step S105, inputting the next sampling time and sampling point position information of the glass curtain wall into the real-time displacement prediction model to obtain a real-time displacement prediction result.

[0092] Specifically, using the trained FEM-PINN model, the sampling point information of the next moment is input into the model. The model will predict the displacement of the glass curtain wall at that moment based on the prediction law and motion equation learned internally. Since the trained model is directly used in the prediction process of the next moment, the prediction result of the model is more real-time.

[0093] The embodiment of the present invention calculates the overall stiffness matrix and mass matrix of the glass curtain wall based on the principle of structural dynamics, and constructs the motion equation in combination with the boundary displacement response information and acceleration response information. It more accurately describes the stress and deformation behavior of the curtain wall, so that the prediction model can adapt to complex actual engineering conditions. At the same time, a fully connected neural network model is introduced, and displacement prediction is performed using deep learning technology. Iterative training is performed in combination with the residual information of the motion equation, so that the prediction system can adapt to different types of curtain wall materials, thereby ensuring displacement prediction under real-time or near real-time conditions. By combining the structural dynamics model with the deep learning model, the prediction accuracy is improved, and the efficiency and real-time performance of the calculation are enhanced.

[0094] In this embodiment, a method for predicting the displacement of a glass curtain wall is provided, which can be used in the above-mentioned computer. Figure 2 is a flow chart of a method for predicting displacement of a glass curtain wall according to an embodiment of the present invention. Figure 2 As shown, the process includes the following steps:

[0095] Step S201, calculating the overall stiffness matrix and overall mass matrix of the glass curtain wall based on the principle of structural dynamics.

[0096] Specifically, the above step S201 includes:

[0097] Step S2011, obtaining physical characteristic parameters and geometric structure parameters of the glass curtain wall.

[0098] Step S2012: dividing the structure of the glass curtain wall into a plurality of plane units, wherein each of the plane units has a local coordinate system.

[0099] Step S2013, calculating the unit stiffness matrix and the unit mass matrix in the local coordinate system according to the physical characteristic parameters and geometric structure parameters corresponding to each planar unit.

[0100] Step S2014: Combining the unit stiffness matrix and unit mass matrix of each plane unit into the overall stiffness matrix and overall mass matrix of the glass curtain wall.

[0101] It should be noted that physical property parameters refer to the material properties of the glass curtain wall, such as density, elastic modulus, Poisson's ratio, etc., which represent the response characteristics of the material when subjected to external forces. Geometric structural parameters refer to the shape and size of the glass curtain wall, including the thickness of the panel, boundary conditions, span, etc. Plane unit refers to the basic unit in finite element analysis, which can be referred to Figure 3 , Figure 3 The entire glass curtain wall is divided into 16 plane units. The local coordinate system refers to an independent coordinate system set relative to each plane unit, which is used to determine the stiffness and mass matrix of each unit in its own coordinate system.

[0102] Specifically, the physical characteristic parameters of the glass curtain wall are obtained by consulting the material manual or conducting experimental tests, and the geometric structure parameters are obtained by using design drawings or measuring tools. Through finite element analysis, mesh division is performed to decompose the glass curtain wall structure into multiple finite element units, and when dividing, the appropriate unit type is selected to reasonably set the density of the mesh to ensure the calculation accuracy. The stiffness and mass matrix of each unit are first calculated in its local coordinate system. Through the physical characteristic parameters and geometric parameters, the expressions of the expected stiffness matrix and unit mass matrix are derived based on the finite element theory. The stiffness matrix is ​​calculated by combining the geometric parameters through the structural mechanics formula, and the unit mass matrix is ​​calculated by the mass integral formula according to the mass density distribution. Finally, the stiffness matrix and mass matrix of all units are superimposed by recombination to obtain the overall stiffness matrix and mass matrix of the entire glass curtain wall.

[0103] For example, the element stiffness matrix in the local coordinate system is calculated as follows:

[0104]

[0105] For example, the unit mass matrix in the local coordinate system is calculated as follows:

[0106]

[0107] The element stiffness matrix and consistent mass matrix in the above local coordinate system need to be converted into the expression form in the global coordinate system. The conversion method is achieved through the following formula:

[0108]

[0109] The transformation matrix is:

[0110]

[0111] Finally, after coordinate transformation, the stiffness matrix and mass matrix of each unit of the curtain wall in the global coordinate system are obtained, and they are assembled to obtain the stiffness matrix K and mass matrix M of the curtain wall.

[0112] The embodiment of the present invention divides the curtain wall structure into multiple plane units, calculates the stiffness matrix and mass matrix of each plane unit in a local coordinate system, and then combines the matrices of each unit into an overall stiffness matrix and a mass matrix. The overall stiffness matrix and mass matrix calculated based on the principle of structural dynamics can accurately reflect the characteristics of the glass curtain wall structure under the action of external forces, thereby constructing a more accurate motion equation, further improving the prediction ability of the subsequent trained glass curtain wall model, and ensuring that the prediction result is more accurate.

[0113] Step S202: Collect the boundary displacement response information and boundary acceleration response information of the glass curtain wall under the condition of external force. Figure 1 Step S102 of the illustrated embodiment will not be described in detail here.

[0114] Step S203: construct the motion equation of the glass curtain wall based on the overall stiffness matrix, overall mass matrix, boundary displacement response information, and boundary acceleration response information. Figure 1 Step S103 of the illustrated embodiment will not be described in detail here.

[0115] In an optional implementation, the motion equation of the glass curtain wall is expressed by the following formula:

[0116]

[0117] Where M represents the overall mass matrix, K represents the overall stiffness matrix, represents the boundary acceleration response information, and v(t) represents the boundary displacement response information.

[0118] In the embodiment of the present invention, the mass matrix (M) and the stiffness matrix (K) can fully describe the physical properties of the entire glass curtain wall, and use the boundary acceleration response and boundary displacement response information to introduce the boundary conditions in the actual working conditions into the motion equation, thereby ensuring that the final motion equation more accurately simulates the actual boundary conditions and improves the accuracy of the motion equation.

[0119] Step S204: The current sampling time and sampling point location information of the glass curtain wall are used as real-time sampling point parameters and input into a preset fully connected neural network model to obtain the current displacement prediction result. The fully connected neural network model is iteratively trained according to the residual information between the motion equation and the current displacement prediction result. When the iteration termination condition is reached, a real-time displacement prediction model of the glass curtain wall is obtained. For details, please refer to Figure 1 Step S104 of the illustrated embodiment will not be described in detail here.

[0120] Step S205: input the next sampling time and sampling point location information of the glass curtain wall into the real-time displacement prediction model to obtain the real-time displacement prediction result. Figure 1 Step S105 of the illustrated embodiment will not be described in detail here.

[0121] In this embodiment, a method for predicting the displacement of a glass curtain wall is provided, which can be used in the above-mentioned computer. The process includes the following steps:

[0122] Step S401, calculate the overall stiffness matrix and overall mass matrix of the glass curtain wall based on the principle of structural dynamics. Figure 1Step S101 of the illustrated embodiment will not be described in detail here.

[0123] Step S402: Collect the boundary displacement response information and boundary acceleration response information of the glass curtain wall under the condition of external force. Figure 1 Step S102 of the illustrated embodiment will not be described in detail here.

[0124] Step S403: construct the motion equation of the glass curtain wall based on the overall stiffness matrix, overall mass matrix, boundary displacement response information, and boundary acceleration response information. Figure 1 Step S103 of the illustrated embodiment will not be described in detail here.

[0125] Step S404, taking the current sampling time and sampling point position information of the glass curtain wall as real-time sampling point parameters, and inputting them into a preset fully connected neural network model to obtain a current displacement prediction result, iteratively training the fully connected neural network model according to the residual information between the motion equation and the current displacement prediction result, and obtaining a real-time displacement prediction model of the glass curtain wall when the iteration termination condition is reached.

[0126] Specifically, Figure 4 As shown, the above step S404 includes:

[0127] Step S4041, constructing a fully connected neural network, and setting the corresponding neural network parameters to initial values; wherein the neural network parameters include sampling point parameters, the number of different network layers, network width, parameters of the activation function formula, parameters of the loss function formula, and network learning rate; the input data of the fully connected neural network includes the sampling point parameters, and the output data of the fully connected neural network includes the displacement response corresponding to the sampling point inside the curtain wall.

[0128] Step S4042: During the current round of iterative training, the first sampling point parameter is input into the fully connected neural network to obtain a predicted displacement response corresponding to the first sampling point.

[0129] Step S4043, using the principle of automatic differentiation to perform partial derivative analysis based on the predicted displacement response to obtain a first partial derivative and a second partial derivative, wherein the first partial derivative represents the second-order partial derivative of the displacement response with respect to time, and the second partial derivative represents the second-order partial derivative of the displacement response with respect to space.

[0130] Step S4044, calculating residual information of the motion equation based on the first partial derivative and the second partial derivative.

[0131] Step S4045: construct a probability density function according to the residual information.

[0132] Step S4046: Continue sampling based on the probability density function to obtain the second sampling point parameters, and use the gradient descent algorithm to update the slope optimization parameters in the activation function formula.

[0133] Step S4047, updating the second sampling point parameters to the first sampling point parameters, and continuing the next round of iterative training; wherein the predicted displacement response in each round is also used to be brought into the loss function formula to calculate the target loss function value of each round.

[0134] Step S4048, when the target loss function value is lower than the preset loss function value, or the round of iterative training reaches the preset number of iterations, the iterative training is stopped to obtain the real-time displacement prediction model of the glass curtain wall.

[0135] Exemplarily, to implement the above steps S4041 to S4048, it is necessary to first construct a fully connected neural network (FEM-PINN) and set initial parameters for it. The input of the network is the location information (x, y) and time information t of the sampling point, and the output is the displacement response of the sampling point inside the curtain wall. Use Pytorch to write the FEM-PINN network code and build a "1+1+8+1" network architecture, that is, 1 input layer, 1 feature expansion layer, 8 hidden layers and 1 output layer, each hidden layer has 50 neurons. Use the random initialization method to assign initial values ​​to the network weights and biases. During initialization, the dimension of the input layer is 3 (corresponding to x, y, t), and the dimension of the output layer is 1 (corresponding to the displacement response). The input data is transformed through the linear transformation and nonlinear activation function of each layer to obtain the predicted value of the output layer. The displacement value finally output is the predicted result of a specific sampling point at a specific time. Automatic differentiation technology is used to calculate the second-order partial derivatives of the displacement response with respect to time and space. Automatic differentiation technology can calculate the derivative of the neural network output to the input through the chain rule, calculate the second-order partial derivative of the displacement with respect to time (acceleration) and the second-order partial derivative of the displacement with respect to the spatial coordinate (stress), and calculate the residual based on the above calculated partial derivatives. A probability density function is constructed based on the residual information, and the residual is converted into a probability distribution to guide the next sampling. The slope in the activation function is updated using the gradient descent algorithm, and the new sampling point is used as input. The previous steps are repeated. After each round of iteration, the loss function is calculated, and the fitting effect of the model is judged based on its value.

[0136] In a specific embodiment, in the above step S4046, the slope optimization parameter in the activation function formula is updated using the gradient descent algorithm, which specifically includes: according to the slope optimization parameter of the current round, and using the gradient descent algorithm to update the slope optimization parameter of the next round; the gradient descent algorithm is represented by the following formula:

[0137] a k+1 =ak -η a J K (a);

[0138] Among them, η represents the network learning rate, a k represents the slope optimization parameter of the Kth iteration, a k+1 represents the slope optimization parameter of the K+1th iteration, η a J K (a) represents the gradient of the loss function along direction a.

[0139] It is understandable that a is found by minimizing the loss function through the gradient descent method. For example, the slope transformation of the adaptive Sigmoid function under different a is as follows:

[0140]

[0141] The embodiment of the present invention is modeled by a fully connected neural network, real-time prediction is performed in combination with real-time data of sampling points, partial derivative analysis is performed using automatic differentiation, the analysis results of the partial derivatives are combined with the motion equation, and a probability density function is constructed. The sampling point parameters are adaptively updated based on residual information in each iterative training, and the sampling is focused on areas where there may be large errors in the overall sampling prediction results of the glass curtain wall, thereby improving the training efficiency and accuracy of the overall model. In each round of iteration, the activation function and network parameters of the neural network are continuously optimized by the gradient descent algorithm, and the training process is continuously updated by calculating the loss value of each round. The residual analysis with the motion equation is used to guide the training of the model, which not only enables the displacement results predicted by the neural network to conform to the physical laws of the motion of the glass curtain wall, but also obtains a real-time displacement prediction model for the glass curtain wall with higher interpretability.

[0142] In an optional implementation, a displacement prediction method for a glass curtain wall provided in this embodiment further includes: performing feature expansion on the sampling point parameters based on the feature expansion layer in the fully connected neural network to obtain the feature expansion parameters of the current round; calculating the gradient information of the target loss function value relative to the feature expansion parameters using a back propagation algorithm; and updating the weight parameters and bias parameters in the activation function formula in the next round according to the gradient information.

[0143] It should be noted that in a fully connected neural network, the input parameters of the sampling points are first processed through the feature expansion layer. The input spatial response variable is directly expanded into a high-dimensional form, such as ln sin(λn x), where λn and ln are parameters that can be automatically updated during network training. The feature expansion method increases the dimension of the input data, allowing the network to capture more complex spatial features. During the training process, the back propagation algorithm is used to calculate the gradient information of the target loss function relative to the feature expansion parameters (including parameters such as λn and ln). According to the gradient information calculated in the previous step, the weight parameters and bias parameters in the activation function formula in the next round of training are updated, and these parameters are optimized to update the network weights, minimize the prediction error, and improve the prediction performance of the model.

[0144] It is understandable that the values ​​of λn and ln are automatically adjusted by the gradient descent algorithm to better fit the training data and capture the complex patterns in the input data. The introduction of the parameter λn provides flexibility for the network, allowing the network to learn the spatial frequency characteristics in the data, while the ln parameter further enhances the model's sensitivity to data changes.

[0145] The embodiment of the present invention expands the dimension of the input data after the input layer through feature expansion, increases the high-dimensional representation of the data, and the back propagation algorithm can further optimize the weight parameters and bias of the model to ensure that all parameters of the model can work in a coordinated manner, thereby improving the prediction accuracy of the trained model.

[0146] In an optional implementation, the residual information includes a residual loss term and an integral of all residual loss terms, and the probability density function is represented by the following formula:

[0147]

[0148] Among them, p(x) represents the probability density function of the residual loss term, A represents the integral of all residual loss terms, ε(x) represents the residual loss term, T represents the residual sampling area, and the residual sampling area represents the sampling point domain where the residual is higher than the preset standard value; the k value represents the sampling concentration of the sampling points in the residual sampling area, and the c value represents the sampling uniformity of the sampling points in the residual sampling area.

[0149] It should be noted that the residual information includes the residual loss term and the integral of all residual loss terms. The residual information is calculated based on the first partial derivative and the second partial derivative. The residual sampling area T is a specific area that represents the sampling point domain where the residual is higher than the preset standard value. In this area, the location of the sampling point is determined by the probability density function to ensure that there are more sampling points in the area with larger residuals. Two important hyperparameters are introduced in the sampling process to adjust the density and uniformity of the sampling.

[0150] Among them, the k value represents the concentration of sampling points in the residual sampling area. A larger k value means that more dense sampling will be performed in areas with larger residuals. The c value represents the uniformity of the sampling points. A larger c value means that the sampling points will be more evenly distributed in the sampling area.

[0151] It is understandable that as the model is trained, the probability density function will be updated in each iteration, thereby continuously optimizing the distribution of sampling points and making model training more efficient. Among them, the residual sampling points are used to train the FEM-PINN model. Through this residual-based sampling method, the RAD method can optimize the network weights, minimize the residual loss, and ultimately improve the prediction accuracy of the model. The combination of the RAD method and FEM-PINN improves the efficiency of data sampling and ensures the physical consistency of the model output, so that the network can not only converge faster, but also better conform to the laws of physics.

[0152] The embodiment of the present invention determines the probability density function through the integral of all residual loss terms, residual loss terms, sampling concentration, sampling uniformity and residual sampling area. It should be noted that the residual information refers to the residual information calculated based on the first partial derivative and the second partial derivative and the motion equation. Since the motion equation of the glass curtain wall is combined, the constructed probability density function will be continuously updated during iterative training, so that the prediction results of the trained model are more accurate.

[0153] Step S405: input the next sampling time and sampling point location information of the glass curtain wall into the real-time displacement prediction model to obtain the real-time displacement prediction result. Figure 1 Step S105 of the illustrated embodiment will not be described in detail here.

[0154] In an optional implementation, the loss function formula is:

[0155] L(t)=w cz loss IC +w0loss PDE +w b loss BC ;

[0156]

[0157] Among them, w cz Represents the initial loss weight coefficient, w b Represents the boundary loss weight coefficient, w o Represents the loss weight coefficient of the motion equation, loss IC Indicates the network initial value loss, loss BC Represents the network boundary loss, loss PDErepresents the motion equation loss, L(t) represents the loss function of the entire network, and N b Represents the boundary loss function sampling area, N s Represents the sampling area of ​​the initial loss function, N T represents the sampling area of ​​the motion equation loss function, t represents the sampling point, r represents the overall sampling area, T represents the residual sampling area, Represents the network parameters x at the initial sampling point c ,y c ,t c The predicted displacement response inside the curtain wall corresponding to the sampling point, u c represents the actual displacement response of the initial sampling point, Represents the parameter x at any sampling point during network training b ,y b ,t b The predicted displacement response inside the curtain wall corresponding to the sampling point, u b represents the actual displacement response of any sampling point during network training, f(t T ) represents the residual value of the network prediction by the equation of motion.

[0158] In the actual training process, the network architecture of the FEM-PINN algorithm for solving PDE (motion equation loss) considering data features is as follows: the input of the network is the coordinate information sampled from the residual region T, that is, (x, y, t), where x and y are spatial coordinates and t is the time coordinate. The network processes the input x and y through the feature expansion layer and expands them into a feature form suitable for the network. The number of layers of the network is L, the number of neurons in the hidden layer is m, the activation function selects the Tanh function, and the learning rate is a. The output of the network is the approximate predicted value of the displacement response of certain specific points inside the curtain wall.

[0159] The algorithm steps of FEM-PINN for solving PDE considering data characteristics are as follows: Step 1: Build a fully connected neural network and initialize the parameters. Step 2: Define the loss function, which considers the residual, initial conditions and boundary conditions of the partial differential equation. Step 3: Start training the model. Step 4: Perform a certain number of iterative training (K increases by 1 in each iteration) until the network training requirements are met.

[0160] After training with the above algorithm, the motion equation loss function of FEM-PINN is as follows: Figure 5 As shown, the total loss function of the training process is as follows Figure 6 As the number of training rounds decreases, the loss training threshold is about 10 6 The decline becomes gentle, at which point both the motion equation and the overall loss function have dropped below a certain threshold, and the model iterative training is completed.

[0161] In the embodiment of the present invention, the loss function includes information on three aspects: initial value loss, boundary loss and motion equation loss. The initial value loss and boundary loss correspond to the displacement response of the curtain wall under specific conditions, while the motion equation loss corresponds to the matching degree between the network prediction and the physical equation of the glass curtain wall. Different weight coefficients are also introduced into the loss function to balance the three loss functions, so that the model prediction results are more consistent with the actual physical movement phenomenon of the glass curtain wall, thereby improving the accuracy of the final training model.

[0162] In this embodiment, a displacement prediction system for a glass curtain wall is also provided, and the device is used to implement the above-mentioned embodiments and preferred implementation modes, and the descriptions that have been made are omitted. As used below, the term "module" can implement a combination of software and / or hardware of a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceivable.

[0163] This embodiment provides a displacement prediction system for a glass curtain wall. Figure 7 As shown, including:

[0164] The overall matrix calculation module 701 is used to calculate the overall stiffness matrix and overall mass matrix of the glass curtain wall based on the principle of structural dynamics;

[0165] The boundary response information acquisition module 702 is used to acquire boundary displacement response information and boundary acceleration response information of the glass curtain wall under the condition of external force application;

[0166] A motion equation construction module 703 is used to construct the motion equation of the glass curtain wall based on the overall stiffness matrix, the overall mass matrix, the boundary displacement response information and the boundary acceleration response information;

[0167] Iterative training module 704, used to use the current sampling time and sampling point position information of the glass curtain wall as real-time sampling point parameters, and input them into a preset fully connected neural network model to obtain a current displacement prediction result, and iteratively train the fully connected neural network model according to the residual information between the motion equation and the current displacement prediction result, and obtain a real-time displacement prediction model of the glass curtain wall when the iteration termination condition is met;

[0168] The real-time displacement prediction module 705 is used to input the next sampling time and sampling point position information of the glass curtain wall into the real-time displacement prediction model to obtain the real-time displacement prediction result.

[0169] In some optional implementations, the overall matrix calculation module 701 includes:

[0170] A physical characteristic parameter unit, used to obtain the physical characteristic parameters and geometric structure parameters of the glass curtain wall;

[0171] A plane unit division unit, used to divide the structure of the glass curtain wall into a plurality of plane units, wherein each of the plane units has a local coordinate system;

[0172] A local unit stiffness matrix calculation unit is used to calculate the unit stiffness matrix and unit mass matrix in the local coordinate system according to the physical characteristic parameters and geometric structure parameters corresponding to each plane unit;

[0173] The overall stiffness and mass matrix assembly unit is used to combine the unit stiffness matrix and unit mass matrix of each plane unit into the overall stiffness matrix and overall mass matrix of the glass curtain wall.

[0174] In some optional implementations, the motion equation of the glass curtain wall is expressed by the following formula:

[0175]

[0176] Where M represents the overall mass matrix, K represents the overall stiffness matrix, represents the boundary acceleration response information, and v(t) represents the boundary displacement response information.

[0177] In some optional implementations, the iterative training module 704 specifically includes:

[0178] A neural network construction unit is used to construct a fully connected neural network and set the corresponding neural network parameters to initial values; wherein the neural network parameters include sampling point parameters, the number of different network layers, network width, parameters of activation function formula, parameters of loss function formula, and network learning rate; the input data of the fully connected neural network includes sampling point parameters, and the output data of the fully connected neural network includes displacement responses corresponding to the sampling points inside the curtain wall;

[0179] A predicted displacement response unit, used for inputting the first sampling point parameter into the fully connected neural network to obtain a predicted displacement response corresponding to the first sampling point during the current round of iterative training;

[0180] A partial derivative calculation unit, configured to perform partial derivative analysis according to the predicted displacement response using an automatic differentiation principle to obtain a first partial derivative and a second partial derivative, wherein the first partial derivative represents a second-order partial derivative of the displacement response with respect to time, and the second partial derivative represents a second-order partial derivative of the displacement response with respect to space;

[0181] A residual calculation unit, used for calculating residual information with respect to the motion equation based on the first partial derivative and the second partial derivative;

[0182] A probability density function construction unit, used to construct a probability density function according to the residual information;

[0183] A sampling parameter updating unit, used to continue sampling based on the probability density function to obtain the second sampling point parameter, and to update the slope optimization parameter in the activation function formula using a gradient descent algorithm;

[0184] An iterative training unit, used to update the second sampling point parameter to the first sampling point parameter, and continue the next round of iterative training; wherein the predicted displacement response in each round is also used to be brought into the loss function formula to calculate the target loss function value of each round;

[0185] The loss function calculation unit is used to stop iterative training when the target loss function value is lower than a preset loss function value or the number of iterative training rounds reaches a preset number of iterations to obtain a real-time displacement prediction model for the glass curtain wall.

[0186] In some optional embodiments, the above device further includes:

[0187] A feature expansion module, used to perform feature expansion on the sampling point parameters based on the feature expansion layer in the fully connected neural network to obtain feature expansion parameters of the current round;

[0188] A gradient calculation module, used to calculate the gradient information of the target loss function value relative to the feature expansion parameter using a back propagation algorithm;

[0189] A parameter updating module is used to update the weight parameters and bias parameters in the activation function formula in the next round according to the gradient information.

[0190] In some optional implementations, the residual information includes a residual loss term and an integral of all residual loss terms, and the probability density function is represented by the following formula:

[0191]

[0192] Among them, p(x) represents the probability density function of the residual loss term, A represents the integral of all residual loss terms, ε(x) represents the residual loss term, T represents the residual sampling area, and the residual sampling area represents the sampling point domain where the residual is higher than the preset standard value; the k value represents the sampling concentration of the sampling points in the residual sampling area, and the c value represents the sampling uniformity of the sampling points in the residual sampling area.

[0193] In some optional implementations, the loss function formula is:

[0194] L(t)=w cz loss IC +w0loss PDE+w b loss BC ;

[0195]

[0196] Among them, w cz Represents the initial loss weight coefficient, w b Represents the boundary loss weight coefficient, w o Represents the loss weight coefficient of the motion equation, loss IC Indicates the network initial value loss, loss BC Represents the network boundary loss, loss PDE represents the motion equation loss, L(t) represents the loss function of the entire network, and N b Represents the boundary loss function sampling area, N s Represents the sampling area of ​​the initial loss function, N T represents the sampling area of ​​the motion equation loss function, t represents the sampling point, R represents the overall sampling area, T represents the residual sampling area, Represents the network parameters x at the initial sampling point c ,y c ,t c The predicted displacement response inside the curtain wall corresponding to the sampling point, u c represents the actual displacement response of the initial sampling point, Represents the parameter x at any sampling point during network training b ,y b ,t b The predicted displacement response inside the curtain wall corresponding to the sampling point, u b represents the actual displacement response of any sampling point during network training, f(t T ) represents the residual value of the network prediction by the equation of motion.

[0197] The further functional description of each of the above modules and units is the same as that of the above corresponding embodiments and will not be repeated here.

[0198] The displacement prediction system of the glass curtain wall in this embodiment is presented in the form of a functional unit, where the unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that executes one or more software or fixed programs, and / or other devices that can provide the above functions.

[0199] The embodiment of the present invention also provides a computer device having the above Figure 7 The displacement prediction system of the glass curtain wall is shown.

[0200] See also Figure 8, Figure 8 is a schematic diagram of the structure of a computer device provided by an optional embodiment of the present invention, such as Figure 8 As shown, the computer device includes: one or more processors 10, a memory 20, and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. Various components are connected to each other using different buses for communication, and can be installed on a common mainboard or installed in other ways as needed. The processor can process the instructions executed in the computer device, including instructions stored in or on the memory to display the graphical information of the GUI on an external input / output device (such as, a display device coupled to the interface). In some optional embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories and multiple memories. Similarly, multiple computer devices can be connected, and each device provides some necessary operations (for example, as a server array, a group of blade servers, or a multi-processor system). Figure 8 A processor 10 is taken as an example.

[0201] The processor 10 may be a central processing unit, a network processor or a combination thereof. The processor 10 may further include a hardware chip. The hardware chip may be a dedicated integrated circuit, a programmable logic device or a combination thereof. The programmable logic device may be a complex programmable logic device, a field programmable gate array, a general purpose array logic or any combination thereof.

[0202] The memory 20 stores instructions executable by at least one processor 10, so that at least one processor 10 executes the method shown in the above embodiment.

[0203] The memory 20 may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created according to the use of the computer device, etc. In addition, the memory 20 may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some optional embodiments, the memory 20 may optionally include a memory remotely arranged relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0204] The memory 20 may include a volatile memory, such as a random access memory; the memory may also include a non-volatile memory, such as a flash memory, a hard disk or a solid state drive; the memory 20 may also include a combination of the above types of memory.

[0205] The computer device further comprises a communication interface 30 for the computer device to communicate with other devices or a communication network.

[0206] The embodiment of the present invention also provides a computer-readable storage medium. The method according to the embodiment of the present invention can be implemented in hardware, firmware, or can be implemented as a computer code that can be recorded in a storage medium, or can be implemented as a computer code that is originally stored in a remote storage medium or a non-temporary machine-readable storage medium and will be stored in a local storage medium through a network download, so that the method described herein can be stored in such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only storage memory, a random access memory, a flash memory, a hard disk or a solid-state hard disk, etc.; further, the storage medium can also include a combination of the above types of memories. It can be understood that a computer, a processor, a microprocessor controller, or programmable hardware includes a storage component that can store or receive software or computer code. When the software or computer code is accessed and executed by a computer, a processor, or hardware, the method shown in the above embodiment is implemented.

[0207] A part of the present invention may be applied as a computer program product, such as a computer program instruction, which, when executed by a computer, can call or provide the method and / or technical solution according to the present invention through the operation of the computer. Those skilled in the art should understand that the existence of the computer program instruction in a computer-readable medium includes, but is not limited to, a source file, an executable file, an installation package file, etc., and accordingly, the way in which the computer program instruction is executed by the computer includes, but is not limited to: the computer directly executes the instruction, or the computer compiles the instruction and then executes the corresponding compiled program, or the computer reads and executes the instruction, or the computer reads and installs the instruction and then executes the corresponding installed program. Here, the computer-readable medium may be any available computer-readable storage medium or communication medium accessible to the computer.

[0208] Although the embodiments of the present invention have been described in conjunction with the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present invention, and such modifications and variations are all within the defined scope.

Claims

1. A method for predicting the displacement of a glass curtain wall, characterized in that: The method comprises: Calculate the overall stiffness matrix and overall mass matrix of the glass curtain wall based on the principles of structural dynamics; Collect the boundary displacement response information and boundary acceleration response information of the glass curtain wall under the condition of external force; Constructing the motion equation of the glass curtain wall based on the overall stiffness matrix, the overall mass matrix, the boundary displacement response information, and the boundary acceleration response information; The current sampling time and sampling point position information of the glass curtain wall are used as real-time sampling point parameters and input into a preset fully connected neural network model to obtain a current displacement prediction result. According to the residual information between the motion equation and the current displacement prediction result, the fully connected neural network model is iteratively trained. When the iteration termination condition is reached, a real-time displacement prediction model of the glass curtain wall is obtained. The next sampling time and sampling point position information of the glass curtain wall are input into the real-time displacement prediction model to obtain the real-time displacement prediction result.

2. The method according to claim 1, characterized in that The calculation of the overall stiffness matrix and overall mass matrix of the glass curtain wall based on the principle of structural dynamics includes: Obtaining physical characteristic parameters and geometric structure parameters of the glass curtain wall; Dividing the structure of the glass curtain wall into a plurality of plane units, wherein each of the plane units has a local coordinate system; Calculate the unit stiffness matrix and unit mass matrix in the local coordinate system according to the physical characteristic parameters and geometric structure parameters corresponding to each plane unit; The unit stiffness matrix and the unit mass matrix of each plane unit are combined into an overall stiffness matrix and an overall mass matrix of the glass curtain wall.

3. The method according to claim 1 or 2, characterized in that: The motion equation of the glass curtain wall is expressed by the following formula: Where M represents the overall mass matrix, K represents the overall stiffness matrix, represents the boundary acceleration response information, and v(t) represents the boundary displacement response information.

4. The method according to claim 3, characterized in that The current sampling time and sampling point position information of the glass curtain wall are used as real-time sampling point parameters and input into a preset fully connected neural network model to obtain a current displacement prediction result. According to the residual information between the motion equation and the current displacement prediction result, the fully connected neural network model is iteratively trained. When the iteration termination condition is reached, a real-time displacement prediction model of the glass curtain wall is obtained, which specifically includes: Constructing a fully connected neural network, and setting corresponding parameters of each neural network to initial values; wherein the neural network parameters include sampling point parameters, the number of different network layers, network width, parameters of activation function formula, parameters of loss function formula, and network learning rate; input data of the fully connected neural network includes sampling point parameters, and output data of the fully connected neural network includes displacement responses corresponding to sampling points inside the curtain wall; In the current round of iterative training, the first sampling point parameter is input into the fully connected neural network to obtain a predicted displacement response corresponding to the first sampling point; Performing partial derivative analysis according to the predicted displacement response using the principle of automatic differentiation to obtain a first partial derivative and a second partial derivative, wherein the first partial derivative represents a second-order partial derivative of the displacement response with respect to time, and the second partial derivative represents a second-order partial derivative of the displacement response with respect to space; Calculating residual information about the motion equation based on the first partial derivative and the second partial derivative; Constructing a probability density function according to the residual information; Continue sampling based on the probability density function to obtain the second sampling point parameters, and use the gradient descent algorithm to update the slope optimization parameters in the activation function formula; The second sampling point parameters are updated to the first sampling point parameters, and the next round of iterative training is continued; wherein the predicted displacement response in each round is also used to be brought into the loss function formula to calculate the target loss function value of each round; When the target loss function value is lower than the preset loss function value, or the round of iterative training reaches the preset number of iterations, the iterative training is stopped to obtain the real-time displacement prediction model of the glass curtain wall.

5. The method according to claim 4, characterized in that The method further comprises: Performing feature expansion on the sampling point parameters based on the feature expansion layer in the fully connected neural network to obtain feature expansion parameters of the current round; Calculate the gradient information of the target loss function value relative to the feature expansion parameter using a back propagation algorithm; The weight parameters and bias parameters in the activation function formula in the next round are updated according to the gradient information.

6. The method according to claim 4, characterized in that The residual information includes the residual loss term and the integral of all residual loss terms, and the probability density function is expressed by the following formula: Among them, p(x) represents the probability density function of the residual loss term, A represents the integral of all residual loss terms, ε(x) represents the residual loss term, T represents the residual sampling area, and the residual sampling area represents the sampling point domain where the residual is higher than the preset standard value; the k value represents the sampling concentration of the sampling points in the residual sampling area, and the c value represents the sampling uniformity of the sampling points in the residual sampling area.

7. The method according to claim 4, characterized in that The loss function formula is: L(t)=w cz loss IC +w0loss PDE +w b loss BC ; Among them, w cz Represents the initial loss weight coefficient, w b Represents the boundary loss weight coefficient, w o Represents the loss weight coefficient of the motion equation, loss IC Indicates the network initial value loss, loss BC Represents the network boundary loss, loss PDE represents the motion equation loss, L(t) represents the loss function of the entire network, and N b Represents the boundary loss function sampling area, N s Represents the initial value loss function sampling area, N T represents the sampling area of ​​the motion equation loss function, t represents the sampling point, R represents the overall sampling area, T represents the residual sampling area, Represents the network parameters x at the initial sampling point c ,y c ,t c The predicted displacement response inside the curtain wall corresponding to the sampling point, u c represents the actual displacement response of the initial sampling point, Represents the parameter x at any sampling point during network training b ,y b ,t b The predicted displacement response inside the curtain wall corresponding to the sampling point, u b represents the actual displacement response of any sampling point during network training, f(t T ) represents the residual value of the network prediction by the equation of motion.

8. A displacement prediction system for a glass curtain wall, characterized in that: The system comprises: The overall matrix calculation module is used to calculate the overall stiffness matrix and overall mass matrix of the glass curtain wall based on the principle of structural dynamics; The boundary response information acquisition module is used to acquire the boundary displacement response information and boundary acceleration response information of the glass curtain wall under the condition of external force application; A motion equation construction module, used to construct the motion equation of the glass curtain wall based on the overall stiffness matrix, the overall mass matrix, the boundary displacement response information and the boundary acceleration response information; An iterative training module is used to use the current sampling time and sampling point position information of the glass curtain wall as real-time sampling point parameters, and input them into a preset fully connected neural network model to obtain a current displacement prediction result, and iteratively train the fully connected neural network model according to the residual information between the motion equation and the current displacement prediction result, and obtain a real-time displacement prediction model of the glass curtain wall when an iteration termination condition is reached; The real-time displacement prediction module is used to input the next sampling time and sampling point position information of the glass curtain wall into the real-time displacement prediction model to obtain the real-time displacement prediction result.

9. A computer device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the displacement prediction method of the glass curtain wall according to any one of claims 1 to 7 by executing the computer instructions.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the method for predicting the displacement of a glass curtain wall according to any one of claims 1 to 7.

Images