Design method and system for prestressed superposed beam-column joint without groove wall

Abstract

The invention discloses a design method and system for a prestressed superposed beam-column joint without a groove wall. The invention relates to the technical field of building design. Collecting an independent variable Xi of any node i of the currently related prestressed superposed beam column without the groove wall, wherein the independent variable comprises the diameter D1 of a lower side longitudinal common stressed steel bar, the diameter D2 of an anti-seismic steel bar, the material M of the steel bar, the strength grade S of the steel bar, the type LT of the load borne by the node and the stress direction FD: Xi = [D1, D2, M, S, LT, FD]; an ideal dependent variable Yi corresponding to the independent variable Xi is regarded to comprise the length L and the wall thickness T of the connecting sleeve; according to the method, the powerful nonlinear modeling capacity of the PID neural network is utilized, the complex relation between the key parameters influencing the node stress performance and the parameters of the connecting sleeve can be quickly captured, and the design period is greatly shortened. Through a data driving mode, according to the scheme, the connecting sleeve parameters meeting the actual stress requirement can be more accurately predicted, and the design accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of architectural design, specifically to the intelligent design direction of prestressed composite beam-column joints, and particularly to a design method and system for prestressed composite beam-column joints without grooved walls. Background Art

[0002] The prestressed composite beam-column joint without a grooved wall is a technological innovation and improvement based on the traditional precast prestressed concrete assembled monolithic frame structure system (referred to as the "Shigou system" for short). In the traditional "Shigou system", through construction practice, it is found that the thickness of the grooved wall at the end of the precast prestressed composite beam is relatively thin, and it is easy to be damaged during the demolding, transportation, and on-site hoisting processes in the factory; at the same time, the grooved space is relatively small, which is not convenient for the installation of steel bars in the groove for precast beams with a relatively large beam height.

[0003] To solve these problems, the prestressed composite beam-column joint without a grooved wall cancels the grooved design at the beam end, and realizes the effective connection of the joint and the improvement of the overall performance by increasing the longitudinal stressed steel bars at the beam bottom and the U-shaped connecting steel bars to jointly bear the force of the joint. As Figure 2 It can be seen from [1] in the prior art below and its "4.2" part that the precast beam is the main part of the joint, which is precast in the factory and has standardized dimensions and shapes. Longitudinal stressed steel bars and hole positions for U-shaped connecting steel bars are reserved inside the precast beam. The post-cast concrete is used to fill the gap between the precast beam and the precast column, and the adjustable combined sleeve is used to connect the longitudinal stressed steel bars in the precast beam to form an integral beam-column joint.

[0004] (1) The prior art [1], namely the literature "Zhang Mingming, Zhang Yukai. Research on the Construction Technology of a Prestressed Composite Beam-Column Joint without a Grooved Wall [J]. Jiangsu Architecture, 2024, (04): 81-85" discloses a joint design method. The corresponding beam-column joint needs to rely on manual design, and it cannot achieve efficient data-driven and customized design for specific scenarios.

[0005] (2) The prior art [2], namely the Chinese invention patent CN202111296427.7, discloses a multi-modal input deep neural network, a beam-column design method and device for a frame structure (publication date: 2022-02-11), and the prior art [3], namely the Chinese invention patent CN202211565222.9, discloses a steel tube concrete beam-column connection joint structure and an installation method for a composite structure (publication date: 2023-06-09); both of these two prior arts generate spectral features based on node information and edge information through a graph neural network model, and can quickly complete the beam-column design of the frame structure according to key building images and design information. However, this technology can only solve the layout task of the joint and cannot complete the actual joint parameter design.

[0006] (3)Prior art [4], namely the Chinese invention patent CN202110927315.0, discloses a method for constructing an intelligent system for optimizing the design of beam-column end plate connection joints and parameter identification (publication date: 2021-11-09). By combining with the randomness of the structure, it improves the true influence of the dimension and the dispersion degree of the value range of each parameter variable on the structural sensitivity value, and realizes the design of joint parameters. However, for the prestressed composite beam-column joint without a grooved wall, its characteristic is that by increasing the longitudinal stressed reinforcement at the bottom of the beam and the U-shaped connecting reinforcement, they jointly bear the force of the joint. Therefore, the length of the connecting sleeve should be determined according to the diameter of the connected reinforcement, generally 4 to 15 times the larger value of the diameter of the lower longitudinal ordinary stressed reinforcement and the seismic reinforcement. A sufficiently long sleeve can provide a more reliable mechanical bite and increase the friction between the reinforcement and the sleeve, thus ensuring the stability of the connection and the enhancement of the load-bearing capacity. Since this prior art does not target the prestressed composite beam-column system without a grooved wall, it cannot be applied.

[0007] Therefore, the present invention proposes a design method and system for prestressed composite beam-column joints without a grooved wall. Summary of the Invention

[0008] In view of this, the embodiments of the present invention provide a design method and system for prestressed composite beam-column joints without a grooved wall to solve or alleviate the technical problems existing in the prior art, that is, how to further consider the characteristic of "increasing the longitudinal stressed reinforcement at the bottom of the beam and the U-shaped connecting reinforcement to jointly bear the force of the joint" in the prestressed composite beam-column system without a grooved wall, and to realize the intelligent design of the parameters of the joint connecting sleeve in the form of data driving for different applicable scenarios. The technical solution of the present invention is realized as follows:

[0009] In the first aspect, a design method for prestressed composite beam-column joints without a grooved wall:

[0010] (1) Overview:

[0011] The present invention aims to predict the length and wall thickness of the connecting sleeve for the prestressed composite beam-column joint without a grooved wall. First, by collecting the key independent variables of the joint, such as the diameter of the steel bar, material, strength grade, load type, and stress direction, etc., they are used as the input of the neural network. Then, using the non-linear modeling ability of the PID neural network to capture the complex relationship between the independent variables and the dependent variable, an initial prediction value is generated. Next, the D-S evidence theory algorithm is introduced. By calculating the information entropy of the initial prediction value as the confidence evaluation index and assigning the basic probability, and using the Dempster combination principle for combined calculation, a more reliable prediction result is obtained. Finally, for different joint design tasks, the above steps are repeatedly executed until the connecting sleeve parameters required for all joints are generated. This solution combines the prediction ability of the neural network with the confidence evaluation of the evidence theory, improving the accuracy and reliability of the prediction results.

[0012] (2) Technical solution:

[0013] To achieve the above objectives, the present invention selects to perform the following operating steps:

[0014] 2.1 Step S1, collect the independent variable X:

[0015] Collect the independent variable X of any node i of the prestressed composite beam-column without a grooved wall involved currently i , including the diameter D1 of the lower longitudinal ordinary stress-bearing steel bar, the diameter D2 of the seismic steel bar, the material M of the steel bar, the strength grade S of the steel bar, the load type LT and the stress direction FD borne by this node: X i = [D1, D2, M, S, LT, FD]; and the ideal dependent variable Y i corresponding to this independent variable X i is regarded as including the length L of the connecting sleeve and its wall thickness T (the material does not need to be considered because the material should be the same as that of the steel bar, that is, the material M); our goal is to predict the dependent variable Y based on the independent variable X i .

[0016] Among them:

[0017] 1) The diameter D1 of the lower longitudinal ordinary stress-bearing steel bar: a key factor affecting the sleeve length, because the sleeve needs to be long enough to provide sufficient mechanical bite;

[0018] 2) The diameter D2 of the seismic steel bar: The diameter of the seismic steel bar is larger than that of the ordinary stress-bearing steel bar, which is a decisive factor for determining the sleeve length;

[0019] 3) The material M of the steel bar: such as carbon steel, alloy steel, etc., which is digitized using the one-hot encoding technology;

[0020] 4) Strength grade S of steel bars: such as HRB400, HRB500, etc., which are also digitized using one-hot encoding technology;

[0021] 5) Load type LT: static load, dynamic load or / and seismic load;

[0022] 6) Force direction FD: tensile force, compressive force or / and shear force, which is digitized using one-hot encoding technology;

[0023] 2.2 Step S2, execute the PID neural network:

[0024] To establish the non-linear relationship between the independent variable X i and the dependent variable Y i This invention selects to use a PID neural network. By taking the independent variable X i as the input, capturing the non-linear relationship between the independent variable X i and the dependent variable Y i to generate the initial predicted value Y i ' of the dependent variable Y i ';

[0025] Among them, the hierarchical structure of the PID neural network is specifically:

[0026] (1) Input layer:

[0027] Perform encoding operations on the independent variable X i input at the current time step t for the input layer to read;

[0028] (2) Hidden layer:

[0029] Although the PID controller technology is mainly used for the calculation task of control signals, and this scheme belongs to the prediction task. However, this scheme chooses to draw on the characteristics of the PID controller, that is, the proportional term, integral term and differential term are respectively constructed into the proportional neuron P, integral neuron I and differential neuron D in the hidden layer; they all receive the independent variable X i and convert it into the proportional coefficient K p , integral coefficient K i and differential coefficient K d . For each layer in the hidden layer, it includes:

[0030] 1) Proportional neuron P: The proportional term can be analogized to the proportional factor of weight update; adjust the prediction according to the immediate relationship between the independent variable X i and the target output (initial predicted value Y i ');

[0031] 2) Integral Neuron I: The integral effect can be regarded as accumulating the influence of historical errors to ensure that long-term biases can be corrected. Similar to capturing long-term dependencies in data by adding more hidden layers or neurons to implement a memory mechanism;

[0032] 3) Differential Neuron D: The differential effect focuses on the rate of change of the error, which can be used to counter over-adjustment to better handle rapid changes on the loss function surface.

[0033] The proportional neuron P, integral neuron I, and differential neuron D are guided by the error function, repeatedly execute the prediction task, and continuously correct the bias b and weights w until the initial predicted value Y i ’;

[0034] (3) Output layer: Perform decoding operations on the initial predicted value Y i ’.

[0035] 2.2.1 Step S200, activate the hidden layer:

[0036] S2000, the proportional neuron P performs immediate relationship adjustment prediction:

[0037] ;;

[0038] S2001, the integral neuron I executes the memory mechanism:

[0039] ;

[0040] S2002, the differential neuron D calculates the rate of change of the error:

[0041] ;

[0042] Among them, w pj , w ij and w dj (j = 1, 2, 3) are the weights of the diameter D1, diameter D2, and load type LT corresponding to the proportional neuron P, integral neuron I, and differential neuron D respectively (because only these three parameters are numerical variables, and the material M, strength grade S, and force direction FD are only type variables); sign is the sign function, which determines the specific scalar corresponding to the diameter D1, diameter D2, and load type LT based on the types of the material M, strength grade S, and force direction FD; b p , b i and b d are the biases of these three neurons respectively. z p , z i and z d are the linear combination results of the proportional neuron P, integral neuron I, and differential neuron D respectively, Kp , K i and K d is the output after the non - linear transformation by the ReLU activation function.

[0043] 2.2.2 Step S201, perform prediction guidance based on the objective function F:

[0044] ;

[0045] where w i is the associated weight with the i - th output (K p , K i , K d ).

[0046] 2.2.3 Step S202, calculate the error loss:

[0047] Use a metric standard to evaluate the deviation between the current PID parameters and the desired PID parameters as the backtracking guidance. The process is as follows:

[0048] S2020, calculate the error E:

[0049] ;

[0050] where K is the number of prediction tasks executed at the current time step t; Y t-1 is the output of the previous iteration; is the output of this iteration;

[0051] S2021, based on the sum - of - squares metric of the error E, use the optimization function F’ for guidance:

[0052] ;

[0053] S2022, according to the direction of the error gradient, update the weights w pj , w ij and w dj .

[0054] 2.2.4 Step S203, iteration:

[0055] Repeat steps S200 - S202 until the error E is less than the preset threshold or the predetermined number of iterations is reached. And obtain the initial prediction value Y i ’ (i.e., the output of the last prediction task).

[0056] 2.3 Step S3, perform confidence measurement:

[0057] The initial prediction value Y i’ is only a preliminary result, which still needs to be objectively evaluated and corrected before it can be officially used as a parameter. This step is based on the D-S evidence theory algorithm, using the initial predicted value Y i ’s information entropy H i ’s calculation as an evaluation index of confidence and assign the basic probability assignment (BPA). Based on the preset recognition framework O, use the Dempster combination principle for combined calculation, and finally form the dependent variable Y i .

[0058] 2.3.1 Step S300, calculate the information entropy H i :

[0059] ;

[0060] where n represents the ordinal number, and p ij is the probability of the j-th component of the probability distribution P i ’ of the initial predicted value Y i . The base of the logarithmic function is 2 or the natural logarithm.

[0061] 2.3.2 Step S301, take the reciprocal of the information entropy as the assigned value m(A) of the basic probability:

[0062] ;

[0063] where A i is a subset in the preset recognition framework O (drawn up through historical data, similar to a data set or dictionary), representing a certain category or state that the predicted value Y i ’ may belong to. ϵ is a small positive number to prevent the denominator from being zero, and V is a positive adjustment parameter used to control the influence degree of the information entropy on the basic probability.

[0064] 2.3.3 Step S302, execute the Dempster combination principle: combine the basic probabilities of different evidences to obtain the combined basic probability m′(A):

[0065] ;

[0066] where ∅ is the empty set, m1 and m2 are the basic probability assignments of different evidences, and B and C are subsets in the recognition framework O.

[0067] 2.4 Step S4, iterative execution:

[0068] Repeatedly execute Steps S1~S3 for different node design tasks until the lengths L and wall thicknesses T of the connecting sleeves required for all nodes of the prestressed composite beam-column with non-grooved walls involved currently are generated.

[0069] (III) Mechanism for solving technical problems:

[0070] This solution first identifies the key parameters affecting the mechanical properties of the node, such as the diameter of the longitudinal stressed steel bars at the bottom of the beam, the configuration of U-shaped connecting steel bars, the material and strength grade of the steel bars, as well as the type of load and the direction of force borne by the node, etc. These parameters together constitute the independent variable set for predicting the length and wall thickness of the connecting sleeve of the node. Then, by utilizing the powerful non-linear modeling ability of the PID neural network, this solution can capture the complex relationship between these independent variables and the dependent variables (i.e., the length and wall thickness of the connecting sleeve). By training the neural network to learn the optimal configuration of the connecting sleeve parameters under different combinations of independent variables. For different applicable scenarios, such as different load conditions, structural forms or construction requirements, etc., this solution can quickly generate the predicted values of the corresponding connecting sleeve parameters by adjusting the input independent variable data.

[0071] The data-driven intelligent design method not only improves the design efficiency, but also ensures that the parameter configuration of the connecting sleeve is more in line with the actual force-bearing requirements, thereby enhancing the overall mechanical properties of the beam-column joint and the safety of the structure. In addition, the D-S evidence theory algorithm is introduced to evaluate the confidence of the initial predicted value. By combining and calculating the basic probabilities of multiple evidences, this solution can further improve the accuracy and reliability of the prediction results, providing a more scientific basis for the design and construction of the prestressed composite beam-column system without a grooved wall.

[0072] Second aspect, a design system for prestressed composite beam-column joints without a grooved wall:

[0073] The system includes a processor and a memory connected to the processor. Program instructions are stored in the memory. When the program instructions are executed by the processor, the processor executes the prestressed composite beam-column joint design method as described above.

[0074] Compared with the prior art, the beneficial effects of the present invention are:

[0075] I. Improving design efficiency and accuracy: By utilizing the powerful non-linear modeling ability of the PID neural network, the present invention can quickly capture the complex relationship between the key parameters affecting the mechanical properties of the node and the parameters of the connecting sleeve, greatly shortening the design cycle. In a data-driven manner, this solution can more accurately predict the parameters of the connecting sleeve that meet the actual force-bearing requirements, improving the accuracy of the design.

[0076] II. Enhanced structural safety: The present invention fully considers the characteristic that the longitudinal stressed steel bars at the bottom of the beam and the U-shaped connecting steel bars jointly bear the force of the joint, ensuring that the parameter configuration of the connecting sleeve is more in line with the actual stress situation. Through the D-S evidence theory algorithm, the confidence of the prediction results is evaluated, further improving the reliability of the prediction results, thereby enhancing the overall mechanical properties of the beam-column joint and the structural safety.

[0077] III. Strong adaptability: The present invention can, for different application scenarios, such as different load conditions, structural forms or construction requirements, etc., quickly generate corresponding predicted values of the connecting sleeve parameters by adjusting the input independent variable data. This flexibility enables this solution to be widely applied to various prestressed composite beam-column systems without grooved walls, with strong adaptability. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0079] Figure 1 It is a schematic flowchart of the method of the present invention;

[0080] Figure 2 It is a three-dimensional structural schematic diagram of a prestressed composite beam-column joint without a grooved wall. The upper half of the figure is the main exterior, and the lower half is a 1 / 4 cross-section;

[0081] Figure 3 It is a schematic diagram of the neural network architecture of the present invention;

[0082] Figure 4 It is a schematic diagram of the hidden layer architecture of the neural network of the present invention;

[0083] Figure 5 It is a schematic flowchart of step S3 of the present invention;

[0084] Figure 6 It is a schematic diagram of the stress analysis of the control group in the test example of the present invention;

[0085] Figure 7 It is a schematic diagram of the stress analysis of the experimental group in the test example of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0086] To make the above objects, features, and advantages of the present invention more apparent and understandable, the following provides a detailed description of the specific embodiments of the present invention in conjunction with the accompanying drawings. Many specific details are set forth in the following description to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below;

[0087] It should be noted that the various embodiments in this specification are described in a progressive manner, and the key point of each embodiment is to illustrate the differences from other embodiments. For the same or similar parts among the various embodiments, reference can be made to each other. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and reference can be made to the description in the method part for the relevant parts.

[0088] Embodiment 1: As Figure 1 shown, this embodiment discloses a design method for a prestressed composite beam-column joint without a grooved wall. It includes the following steps S1 to S4.

[0089] In this embodiment, regarding step S1, collect the independent variable X i : For any node i of the prestressed composite beam-column without a grooved wall, collect its independent variable X i , and this independent variable includes six key elements: the diameter D1 of the lower longitudinal ordinary stressed reinforcement, the diameter D2 of the seismic reinforcement, the material M of the reinforcement, the strength grade S of the reinforcement, the load type LT borne by this node, and the stress direction FD.

[0090] The expression of the independent variable X i is: Xi = [D1, D2, M, S, LT, FD]. Wherein:

[0091] (1) The diameter D1 of the lower longitudinal ordinary stressed reinforcement: This is a key factor affecting the length of the sleeve. The sleeve needs to be long enough to provide sufficient mechanical bite, so the size of D1 is directly related to the designed length of the sleeve.

[0092] (2) The diameter D2 of the seismic reinforcement: The diameter of the seismic reinforcement is usually larger than that of the ordinary stressed reinforcement, and it is one of the decisive factors for determining the length of the sleeve. Under dynamic loads such as earthquakes, the diameter of the seismic reinforcement is crucial for the bearing capacity and stability of the sleeve.

[0093] (3) The material M of the reinforcement: Such as carbon steel, alloy steel, etc. Since the material type is non-numerical data, we use the one-hot encoding technology for digital processing to convert each material into a unique binary vector.

[0094] (4)Strength grade S of steel bars: such as HRB400, HRB500, etc. Similarly, the strength grade is also non-numerical data, and we use one-hot encoding technology for digitization for subsequent model processing.

[0095] (5)Load type LT: including static load, dynamic load or / and seismic load. These load types have different design requirements for the sleeve, so they need to be taken as part of the independent variables.

[0096] (6)Force direction FD: including tension, compression or / and shear force. Different force directions will affect the force and deformation of the sleeve, so digital processing is also required. Here, one-hot encoding technology is also used.

[0097] The ideal dependent variables Yi corresponding to the independent variable Xi include the length L and wall thickness T of the connecting sleeve. Our goal is to predict these two dependent variables based on the independent variable X.

[0098] It should be noted that the material of the sleeve is usually the same as that of the steel bar, that is, the material M. Therefore, the material factor is not considered separately in the dependent variable.

[0099] In this embodiment, as Figures 3 - 4 shown, regarding step S2, execute the PID neural network: To establish the non-linear relationship between the independent variable X i and the dependent variable Y i of the present invention, a PID neural network is selected. By taking the independent variable X i as the input, capture the non-linear relationship between the independent variable X i and the dependent variable Y i to generate the initial predicted value Y i ' of the dependent variable Y i ';

[0100] Among them, the hierarchical structure of the PID neural network is specifically:

[0101] (1)Input layer: Perform encoding operations on the independent variable X i input at the current time step t for the input layer to read; the encoding operation is a traditional numerical conversion or feature extraction technology to ensure that the input data is suitable for neural network processing.

[0102] (2)Hidden layer: Although the PID controller technology is mainly used for the calculation task of control signals, and this solution belongs to the prediction task. However, this solution chooses to draw on the characteristics of the PID controller, that is, to construct the proportional term, integral term, and differential term into the proportional neuron P, integral neuron I, and differential neuron D in the hidden layer respectively; they all receive the independent variable X i and convert it into the proportional coefficient K p and the integral coefficient Ki and the differential coefficient K d . For each layer in the hidden layer, it includes:

[0103] 1) Proportional neuron P: The proportional term can be analogous to the proportional factor of weight update; adjust the prediction according to the immediate relationship between the independent variable X i and the target output (the initial predicted value Y i ').

[0104] 2) Integral neuron I: The integral effect can be regarded as accumulating the influence of historical errors to ensure that long-term biases can be corrected. Similar to capturing long-term dependencies in data by adding more hidden layers or neurons, a memory mechanism is implemented;

[0105] 3) Differential neuron D: The differential effect focuses on the rate of change of the error, which can be used to counteract over-adjustment in order to better handle the rapid changes on the loss function surface.

[0106] The proportional neuron P, integral neuron I, and differential neuron D are guided by the error function, perform the prediction task in a loop, and continuously correct the bias b and weights w until the initial predicted value Y i ' is calculated; among them, the proportional neuron P is responsible for adjusting the prediction according to the current error, the integral neuron I is responsible for accumulating historical errors to correct long-term biases, and the differential neuron D is responsible for paying attention to the rate of change of the error to prevent over-adjustment. By continuously performing the prediction task in a loop and correcting the bias b and weights w, the PID neural network can gradually approach the true dependent variable Y i , thereby generating the initial predicted value Y i '.

[0107] (3) Output layer: Perform a decoding operation on the initial predicted value Y i '.

[0108] Specifically, in step S200, activate the hidden layer:

[0109] S2000, the proportional neuron P performs an immediate relationship adjustment prediction:

[0110] ;;

[0111] S2001, the integral neuron I performs a memory mechanism:

[0112] ;

[0113] S2002, the differential neuron D calculates the rate of change of the error:

[0114] ;

[0115] Among them, w pj , w ij and w dj (j = 1, 2, 3) are the weights of the diameter D1, diameter D2, and load type LT corresponding to the proportional neuron P, integral neuron I, and derivative neuron D respectively (since only these three parameters are numerical variables, the material M, strength grade S, and force direction FD are only type variables); sign is the sign function, which determines the specific scalar corresponding to the diameter D1, diameter D2, and load type LT based on the types of the material M, strength grade S, and force direction FD; b p , b i and b d are the biases of these three neurons respectively. z p , z i and z d are the linear combination results of the proportional neuron P, integral neuron I, and derivative neuron D respectively, and K p , K i and K d are the outputs after non-linear transformation through the ReLU activation function.

[0116] Given that neural networks mainly process numerical data, non-numerical variables (materials, strength grades) must be converted into numerical forms. A scalar is assigned to each category through the sign function sign.

[0117] It can be understood that through the collaborative action of the proportional, integral, and derivative neurons, the PID neural network can more accurately capture the non-linear relationship between the independent variable X i and the dependent variable Y i , improving the accuracy of prediction. The memory mechanism of the integral neuron helps accumulate historical error information, correct long-term biases, and enhance the stability of the model. The derivative neuron focuses on the rate of change of the error and can quickly respond to changes in the independent variable X i , improving the response speed and adaptability of the model.

[0118] Specifically, in step S201, perform prediction guidance based on the objective function F:

[0119] ;

[0120] Among them, w i is the associated weight with the i-th output (K p , K i , K d ). Note that i here does not refer to the i-th node.

[0121] It should be noted that in steps S200 - 201:

[0122] (1)The proportional neuron P receives multiple input signals, which usually include various parameters related to the node design, such as numerical representations of the diameter D1, diameter D2, and load type LT. Through weighted summation, the proportional neuron P linearly combines these input signals into a single output value z p . In this process, each input signal is assigned a weight (such as w p1 , w p2 , w p3 ) according to its importance, and these weights reflect the influence degree of the input signals on the output value zp. Additionally, the proportional neuron P also includes a bias term b p for adjusting the baseline level of the output value z p . The output value z p obtained from the linear combination is then fed into the ReLU activation function for non-linear transformation, which sets all negative values to zero while keeping positive values unchanged. This property makes the output K p (i.e., zp after ReLU transformation) of the proportional neuron P always non-negative, conforming to the physical meaning of the proportional coefficient, because the proportional coefficient usually represents a positive gain or amplification effect. Through the ReLU activation function, the proportional neuron P realizes a non-linear mapping of the input signals, thus being able to capture the complex non-linear relationship between the independent variable and the dependent variable.

[0123] Regarding its immediate relationship adjustment prediction: The output K p of the proportional neuron P represents the result of the immediate relationship adjustment prediction. Since K p is calculated based on the current input signals, it can reflect the immediate relationships between the input signals and adjust the prediction value according to these relationships. In the design method of prestressed composite beam-column joints, the output K p of the proportional neuron P can be used to adjust the stress state of the joint, ensuring that the longitudinal stressed steel bars at the bottom of the beam and the U-shaped connecting steel bars can jointly bear the joint stress, thereby improving the bearing capacity and stability of the joint.

[0124] (2)The integral neuron I: In the design method of prestressed composite beam-column joints without grooved walls, the embodiment and role of the execution of the memory mechanism are mainly based on its accumulation and processing of historical error information. The integral neuron I receives the same input signals as the proportional neuron P and calculates the output value K i through linear combination and non-linear transformation. Different from the proportional neuron P, the output value K i of the integral neuron I depends not only on the current input signals but also on the past input signals and error information. This is because the integral neuron I calculates the output value K iWhen it does, it accumulates and processes the historical error information. This accumulation and processing process enables the integrating neuron I to remember past input signals and error information, thus achieving long-term memory of the system's dynamic characteristics. The method is to calculate an error e(t) at each time step t, and this error is the error E of the previous time step (see S202 for details). A bias b i (t) at the next time step t + 1 is a proportion of the current bias plus the current error:

[0125] b i (t + 1) = b i (t) + k e * e(t)

[0126] where k e is a constant that determines the degree of influence of the error on the bias update.

[0127] Then, the output K of the integrating neuron I i is expressed as:

[0128] ;

[0129] where f(input signal, t) is the partial output (excluding the influence of the accumulated error) calculated by the integrating neuron I at time t based on the current input signal.

[0130] By remembering past input signals and error information, the integrating neuron I can correct the long-term deviation of the system, improving the stability and accuracy of the system. In the design method of prestressed composite beam-column joints, the output K of the integrating neuron I i can be used to adjust the stress state of the joint, ensuring that the joint can maintain stable performance during long-term use. In addition, the memory mechanism of the integrating neuron I helps the system adapt to changes in the external environment, improving the robustness and adaptability of the system.

[0131] (3) Differentiating neuron D: In the design method of prestressed composite beam-column joints without grooved walls, the manifestation and role of calculating the error change rate are mainly based on its sensitive capture and rapid response to the system's dynamic characteristics. The differentiating neuron D receives the same input signals as the proportional neuron P and the integrating neuron I, and calculates the output value K d . Different from the proportional neuron P and the integrating neuron I, the output value K of the differentiating neuron D d represents the change rate of the error, that is, the rate of change of the error over time. The method is: K d (t)=k d ⋅(e(t)−e(t−1));

[0132] where: Kd (t) is the output value of the differential neuron D at time step t. k d is the differential gain, which is a constant used to adjust the strength of the differential action. e(t) is the error signal at the current time step. e(t−1) is the error signal at the previous time step. Δt is the interval of the time step, which is a fixed value in a discrete-time system.

[0133] By calculating the rate of change of the error, the differential neuron D can capture the small changes in the dynamic characteristics of the system, providing a basis for the rapid response of the system. The rate of change of the error reflects the dynamic changes of the system state and is an important indicator of the system stability and control performance. In the design method of prestressed composite beam-column joints, the output K d (t) (Note: This output can be regarded as or replace K d above) can be used to adjust the stress state of the joint, ensuring that the joint can maintain dynamic balance during the stress process. By introducing the differential neuron D, the system can quickly respond to the changes in external loads, improving the bearing capacity and stability of the joint. In addition, the differential neuron D helps the system suppress high-frequency disturbances and improve the anti-interference ability of the system.

[0134] Specifically, in step S202, calculate the error loss: Use a metric standard to evaluate the deviation between the current PID parameters and the desired PID parameters as the backtracking guidance. The process is as follows:

[0135] S2020, calculate the error E: ;

[0136] Among them, K is the number of prediction tasks executed at the current time step t, reflecting the comprehensive deviation on multiple tasks; Y t-1 is the output of the previous iteration; is the output of this iteration;

[0137] S2021, based on the sum of squares metric of the error E, use the optimization function F’ for guidance:

[0138] ;

[0139] S2022, according to the direction of the error gradient, update the weights w pj 、w ij and w dj . In other words, by calculating the error gradient (i.e., the derivative of the error with respect to the weights), the direction and magnitude of the weight adjustment can be determined. Updating the weights along the opposite direction of the error gradient can reduce the error loss and make the predicted output closer to the actual output.

[0140] Specifically, in step S203, iteration: repeat steps S200 - S202 until the error E is less than the preset threshold or the predetermined number of iterations is reached. And obtain the initial predicted value Y i ’ (i.e., the output of the last prediction task). Through the iterative optimization process, the PID parameters can be continuously adjusted to make the predicted output closer to the actual output, improving the accuracy of the parameter design of the node connection sleeve. For different application scenarios, the intelligent parameter design of the node connection sleeve is realized in a data-driven form, enhancing the adaptability and robustness of the system. The automated iterative optimization process reduces the time and difficulty of manual debugging and improves the design efficiency.

[0141] In this embodiment, as Figure 5 shown, regarding step S3, perform confidence measurement: the initial predicted value Y i ’ is only a preliminary result, which still needs to be objectively evaluated and corrected before it can be officially used as a parameter. This step is based on the D-S evidence theory algorithm. Using the information entropy H i of the initial predicted value Y i as the evaluation index of confidence and assigning the basic probability assignment (BPA), and performing combined calculation based on the preset recognition framework O using the Dempster combination principle, and finally forming the dependent variable Y i .

[0142] Specifically, in step S300, information entropy is an index to measure the uncertainty or chaos degree of information. In this method, we use information entropy to evaluate the confidence of the initial predicted value Y i’ . Calculate the information entropy ;

[0143] where n represents the ordinal number, and p ij is the probability of the j-th component of the probability distribution P i of the initial predicted value Y i ’, reflecting the relative importance of this component in the overall prediction; the base of the logarithmic function is 2 or the natural logarithm; the logarithmic function is used to measure the “surprise” degree of the probability value. By summing and taking the negative value, the information entropy Hi reflects the overall uncertainty or chaos degree of the predicted value Yi’. The larger Hi is, the higher the uncertainty of the predicted value.

[0144] Specifically, in step S301, in order to convert the information entropy into a measure of confidence, we take the reciprocal of the information entropy as the assigned value of the basic probability. Take the reciprocal of the information entropy as the assigned value of the basic probability m(A):

[0145] ;

[0146] where A iis a subset within the preset recognition framework O (formulated through historical data, similar to a dataset or dictionary), representing the predicted value Y i ’may belong to a certain category or state. ϵ is a small positive number to prevent the denominator from being zero, and V is a positive adjustment parameter used to control the influence degree of information entropy on the basic probability.

[0147] Specifically, in step S302, when multiple pieces of evidence or sources provide predicted values, we need to combine these predicted values to obtain a more accurate confidence assessment. The Dempster combination principle provides an effective combination method: combining the basic probabilities of different pieces of evidence to obtain the combined basic probability m′(A):

[0148] ;

[0149] where m1 and m2 are the basic probability assignments of different pieces of evidence, B and C are two other different subsets within the recognition framework O. ∅ is the empty set; by summing the products of the basic probabilities that satisfy B∩C = A and dividing by a normalization factor (1 minus the sum of the products of the basic probabilities where B∩C = ∅), we obtain the combined basic probability m′(A). The Dempster combination principle can comprehensively consider the information of multiple pieces of evidence, making the combined confidence more accurate and reliable.

[0150] It can be understood that through the confidence measure, we can objectively evaluate and correct the initial predicted value, thereby improving the prediction accuracy. Utilizing information entropy and the Dempster combination principle, we can effectively handle the uncertainty and conflicting information in the predicted value and enhance the robustness of the method.

[0151] In this embodiment, regarding step S4, iterate and execute: Repeat steps S1~S3 for different node design tasks until the lengths L and wall thicknesses T of the connection sleeves required for all nodes of the prestressed composite beam-column with a non-grooved wall involved currently are generated.

[0152] Embodiment 2: In Embodiment 1, w i is the associated weight with the i-th output (K p , K i , K d ); This embodiment further provides a dynamic intelligent assignment strategy for the associated weight w i , that is, using the least squares method to perform residual backtracking on the output of each K p ,K i ,K d and its correlation, and imposing rate control to achieve the dynamic intelligent assignment of the associated weight w iAssignment. The least squares method is a commonly used optimization method for minimizing the sum of squared residuals between predicted values and actual values, thereby finding the optimal parameter estimates.

[0153] P1. Data collection: Collect a series of inputs, actual outputs, and K at a series of time steps p ,K i ,K d of the calculated outputs. Let the input signal be x(t) and the actual output be y(t) at time step t, and the outputs of K p ,K i ,K d be K p (t),K i (t),K d (t) respectively.

[0154] P2. Calculate the linear combination: ; where is the result of the linear combination, and w pi ,w ii ,w di are the correlation weights to be determined.

[0155] P3. Calculate the residual between the actual output and the predicted output ; P4. Use the least squares method with the goal of minimizing the sum of squared residuals ; where N is the total number of time steps.

[0156] P5. Solve for the weights: Minimize the objective function J by solving a system of linear equations or using a numerical optimization method (such as the gradient descent method) to obtain the optimal weights w pi ,w ii ,w di .

[0157] This embodiment uses matrix form to represent and solve this problem, i.e.: W = (K T K) −1 K T Y’’; where w i = W = [w pi ,w ii ,w di T , K is a matrix containing K p (t),K i (t),K d (t), and Y’’ is a vector containing the actual output y(t). T is the matrix transpose operation.

[0158] ​P6. Rate Control: To avoid system instability caused by overly rapid weight updates, a learning rate α is used to adjust the weight update step: w i (new) = w i (old) + α ⋅ Δw i ; where Δw i is the weight update amount calculated by the least squares method. w i (new) represents the new correlation weight, and w i (old) represents the correlation weight of the previous iteration.

[0159] P7. In practical applications, multiple iterations are required to gradually optimize the weights.

[0160] By minimizing the sum of squared residuals between the predicted output and the actual output, the least squares method can find the optimal correlation weight w i , making the predicted output of the model closer to the actual output. This helps improve the accuracy and precision of the control system. The adjustment of the correlation weight w i takes into account the dynamic characteristics of the system. Through reasonable weight allocation, the system can be made more stable, reducing overshoot and oscillation phenomena.

[0161] By introducing the differential term K d , the system can respond quickly to the rate of change of the error, improving the dynamic performance of the system. And the least squares method can optimize the weight of w d , making the differential action more effective. Moreover, this scheme can dynamically adjust the correlation weight w i according to the residual between the actual output and the predicted output of the system, making the system have a certain self - adaptability. This means that the system can automatically adjust the control strategy according to different working conditions and external disturbances, improving the robustness of the system. By automatically calculating the correlation weight w i using the least squares method, the parameter adjustment process can be simplified, reducing the time and difficulty of manual debugging. At the same time, since the correlation weight w i is automatically calculated based on system data, it is more objective and accurate.

[0162] Example 3: In Example 1, p ij represents the probability of the j - th component of the probability distribution P i of the initial predicted value Y i ’. To clarify the assignment of p ij instead of the subjective assignment in Example 1, this example provides a data - driven method to determine this probability value:

[0163] P1. Let the probability p ijSubject to the k-dimensional random vector XX = (XX1, XX2, …, XXk) T , according to the theory of multivariate normal distribution, its probability density function is formulated as follows:

[0164] ;

[0165] where: μ = (μ1, μ2, …, μ k ) T is the mean vector, representing the expected value of each random variable. Σ is the covariance matrix, representing the covariance between the random variables. |Σ| is the determinant of the covariance matrix. Σ−1 is the inverse matrix of the covariance matrix.

[0166] P2. Calculate the Mahalanobis distance: The Mahalanobis distance is a statistic that describes the distance between a sample point and the mean vector, taking into account the covariance between the variables. For the j-th component of the initial predicted value Y i ’, we can calculate its Mahalanobis distance d j as follows: ;

[0167] where, y j ’ is the j-th component of the initial predicted value Y i ’.

[0168] P3. Substitute the Mahalanobis distance into the probability density function of the multivariate normal distribution, and the probability density of the j-th component of the initial predicted value Y i ’ can be obtained: ;

[0169] The multivariate normal distribution can accurately model the joint probability distribution between multiple random variables, taking into account their correlation. This accuracy enables the prediction model based on the multivariate normal distribution to more accurately reflect the actual situation and improve the accuracy of prediction. By assigning probabilities to each component of the initial predicted value Y i ’, the uncertainty of the prediction can be evaluated more comprehensively. It helps to reduce the errors that may be caused by a single predicted value and improve the reliability of the prediction.

[0170] Test example:

[0171] (I) Test purpose:

[0172] This example aims to comparatively evaluate the effects of two design methods for prestressed composite beam-column joints without grooved walls, that is, the experimental group uses the method of combining PID neural network with D-S evidence theory algorithm, and the control group uses the method of combining the neural network model and structural randomness disclosed in Chinese invention patent CN202110927315.0. Through finite element analysis, the superiority and practicability of the experimental group method in designing joint parameters are verified.

[0173] (2) Experimental group and control group:

[0174] 2.1 Experimental group:

[0175] Method: The PID neural network combined with the D-S evidence theory algorithm described in Example 1 is adopted to realize the design of node parameters.

[0176] Feature: It can identify the key parameters affecting the mechanical properties of the node, and capture the complex relationship between the independent variable and the dependent variable through the nonlinear modeling ability of the PID neural network, and generate the optimal predicted value of the connection sleeve parameters.

[0177] 2.2 Control group:

[0178] Method: The neural network model disclosed in Chinese invention patent CN202110927315.0 is adopted to realize the design of node parameters by combining with the randomness of the structure.

[0179] Feature: Based on the existing neural network model, no specific optimization is carried out for the prestressed composite beam-column joints without grooved walls.

[0180] (3) Test method:

[0181] A prestressed composite beam-column environment without grooved walls is proposed, which is a precast prestressed assembled integral frame - cast-in-place shear wall structure. The maximum number of bottom bars of the precast beam is 12C25. Both the experimental group and the control group reasonably optimize the arrangement of the bottom bars of the precast beam with the parameters given by the BIM model. The single-row bottom bars of the beam are adjusted to five rows, with no more than 3 bars in each row, and the positions between the bars are reasonably adjusted.

[0182] The AUTODESK INVENTOR finite element analysis module is used to import the proposed BIM model. In order to avoid the bottom bars of the precast beam in different directions, vertical loads are applied asymmetrically at the beam ends. Before the specimen yields, it is controlled by load, loaded once every cycle, and gradually loaded with a small amplitude to the calculated cracking load.

[0183] The finite element visualization stress diagrams of the experimental group and the control group are shown to visually compare the stress effects of the two.

[0184] (4) Test results:

[0185] 4.1 Comparison of stress effects:

[0186] From Figures 6 - 7 It can be clearly seen that the stress effect of the experimental group is significantly better than that of the control group. The experimental group shows more uniform performance in terms of node force, with less stress concentration, indicating better overall mechanical properties.

[0187] 4.2 Cause analysis:

[0188] The model of the control group is not designed for the prestressed composite beam-column joints with non-grooved walls, and it is impossible to optimize the joint force by increasing the longitudinal reinforcement at the bottom of the beam and the U-shaped connecting reinforcement. The parameter prediction of the length of the connecting sleeve in the control group is not clear, resulting in the friction module between the reinforcement and the sleeve, which cannot ensure the connection stability and load capacity.

[0189] 4.3 Advantages of the experimental group method:

[0190] Identify the key parameters affecting the joint force performance, such as the diameter of the longitudinal reinforcement at the bottom of the beam, the configuration of the U-shaped connecting reinforcement, etc. Utilize the non-linear modeling ability of the PID neural network to capture the complex relationship between the independent variable and the dependent variable, and generate the optimal predicted value of the connecting sleeve parameters. Introduce the D-S evidence theory algorithm to evaluate and correct the confidence of the initial predicted value, which can improve the accuracy and reliability of the prediction results.

[0191] (V) Conclusion:

[0192] The method of combining the PID neural network with the D-S evidence theory algorithm adopted by the experimental group shows obvious advantages in the design of joint parameters, and the stress effect is better than that of the control group. The method of the experimental group can identify the key parameters, and generate the optimal predicted value of the connecting sleeve parameters through the non-linear modeling ability of the PID neural network, improving the design efficiency and accuracy. Introducing the D-S evidence theory algorithm to evaluate the confidence of the initial predicted value further improves the reliability and scientific nature of the prediction results.

[0193] For those skilled in the art, it can be further realized that the units and algorithm steps of each example described in combination with the embodiments disclosed in this article can be implemented by electronic hardware, computer software, or a combination of the two. To clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described according to functions in the above description. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

[0194] Meanwhile, those skilled in the art can understand that all or part of the processes in the methods of implementing the above-mentioned all embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned various methods. Among them, any reference to a memory, storage, database or other medium provided in the present application and used in the embodiments can include non-volatile and / or volatile memories. The non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. The volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and Rambus dynamic RAM (RDRAM), etc.

[0195] The above embodiments only express the implementation manners of the relevant practical applications of the present invention. The descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the invention patent. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several deformations and improvements can still be made, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the present invention patent shall be subject to the appended claims.

Claims

1. Design method for prestressed composite beam-column joints with non-grooved walls, characterized in that, It includes the following execution steps: S1. Collect the independent variable X of any node i of the prestressed composite beam-column with a non-grooved wall involved currently i , including the diameter D1 of the lower longitudinal ordinary stressed reinforcement, the diameter D2 of the seismic reinforcement, the material M of the reinforcement, the strength grade S of the reinforcement, the load type LT and the stress direction FD borne by this node: X i = [D1, D2, M, S, LT, FD]; and take the ideal dependent variable Y i corresponding to this independent variable X i as including the length L and wall thickness T of the connecting sleeve; S2, the PID neural network captures the independent variable X i as input, captures the independent variable X i and the dependent variable Y i to generate the initial predicted value Y of the dependent variable Y i '; i '; S3, using the initial predicted value Y i 's information entropy H i 's calculation as an evaluation index of confidence and assign basic probabilities, and perform combined calculation using the Dempster combination principle based on the preset identification framework O, and finally form the dependent variable Y i .

2. The design method of the prestressed composite beam-column joint according to claim 1, characterized in that: In the S2, the PID neural network includes: Input layer: input the independent variable X at the current time step t i Perform an encoding operation for reading by the input layer; Hidden layer: It includes proportional neuron P, integral neuron I, and derivative neuron D. All are guided by the error function, repeatedly execute the prediction task, and continuously correct the bias b and weight w until the initial predicted value Y is calculated. i ’; Output layer: Perform a decoding operation on the initial predicted value Y i ’.

3. The design method of the prestressed composite beam-column joint according to claim 2, characterized in that: In the S2, the execution steps of the hidden layer include: S200, activate the hidden layer: S2000, the proportional neuron P performs immediate relationship adjustment prediction; S2001, the integral neuron I performs a memory mechanism; S2002, the derivative neuron D calculates the change rate of the error; S201, perform prediction guidance based on the objective function F: ; where w i is the associated weight with the i-th output (K p , K i , K d ); S202, a metric is used to evaluate the deviation between the current PID parameters and the desired PID parameters as a backtracking guide.

4. The design method of the prestressed composite beam-column joint according to claim 3, wherein: In the S200: S2000， ；； S2001， ； S2002， ； Among them, w pj 、w ij and w dj are the weights of the diameter D1, diameter D2, and load type LT corresponding to the proportional neuron P, integral neuron I, and derivative neuron D, respectively; sign is the sign function, which determines the scalar corresponding to the diameter D1, diameter D2, and load type LT based on the type of material M, strength grade S, and force direction FD; b p 、b i and b d are the biases of these three neurons respectively; z p 、z i and z d are the linear combination results of these three neurons respectively, and K p 、K i and K d are the outputs after non-linear transformation by the ReLU activation function.

5. The design method of the prestressed composite beam-column joint according to claim 4, characterized in that: In the S202, it includes: S2020, calculate the error E: ； where K is the number of prediction tasks executed at the current time step t; Y t-1 is the output of the previous iteration; is the output of this iteration; S2021, based on the sum of squares metric of the error E, be guided by the optimization function F'; ； In S2022, update the weights w again according to the direction of the error gradient pj , w ij and w dj .

6. The prestressed composite beam-column joint design method according to claim 5, wherein: In S2, it further includes: repeatedly executing steps S200 to S202 until the error E is less than a preset threshold or a predetermined number of iterations is reached; and obtaining an initial predicted value Y i ’.

7. The design method of the prestressed composite beam-column joint according to claim 2, characterized in that: The execution steps of the S3 include: S300, calculate the information entropy H i : ; where n represents an ordinal number, p ij is the probability distribution P i of the initial predicted value Y i for the j-th component; the base of the logarithmic function is 2 or the natural logarithm; S301, use the reciprocal of the information entropy as the assigned value of the basic probability ; Among them, A i is a subset in the preset recognition framework O, representing a certain category to which the predicted value Y i ’ belongs, and ϵ is a positive number; V is a positive number adjustment parameter.

8. The design method of the prestressed composite beam-column joint according to claim 7, characterized in that: The execution steps of the S3 also include: ； Where ∅ is the empty set, m1 and m2 are the basic probability assignments of different evidences, and B and C are subsets in the recognition frame O.

9. The design method of the prestressed composite beam-column joint according to any one of claims 1 to 8, characterized in that: It also includes S4, which is iteratively executed: Steps S1 to S3 are repeatedly executed for different node design tasks until the lengths L and wall thicknesses T of the connection sleeves required for all nodes of the prestressed composite beam-column with a non-grooved wall involved currently are generated.

10. Design method and system for prestressed composite beam-column joint without grooved wall, characterized in that: The system includes a processor and a memory connected to the processor. Program instructions are stored in the memory. When the program instructions are executed by the processor, the processor executes the prestressed composite beam-column node design method according to any one of claims 1-9.

Images