Model training method and method and device for predicting impact resistance of reinforced concrete beam

Abstract

The invention provides a model training method and a reinforced concrete beam impact resistance prediction method and device, and belongs to the technical field of reinforced concrete.The experimental data of a reinforced concrete beam can be fitted in a Gaussian process regression mode, multiple pieces of sample data are obtained through a fitted curve, and the impact resistance of the reinforced concrete beam is predicted. Therefore, the first convolutional neural network model is trained by using the sample data to obtain the target impact resistance prediction model, and finally impact resistance prediction is performed through the target impact resistance prediction model, so that an experiment scene can be prevented from being established to carry out a freeze-thaw experiment, and the experiment efficiency is improved. Therefore, a large amount of prediction time cost and economic cost can be saved, and due to the fact that the training data utilize the advantages of the Gaussian process regression in the aspects of flexibility, uncertainty quantification capability, non-parameterization characteristics and the like, the prediction accuracy of the target impact resistance prediction model can be improved to a certain degree.

Description

Technical Field

[0001] The present application relates to the field of reinforced concrete technology, and in particular to a model training method, a method and a device for predicting the impact resistance of reinforced concrete beams. Background Art

[0002] Reinforced concrete (RC), often referred to simply as reinforced concrete in engineering, refers to a composite material composed of steel mesh, steel plates, or fibers added to concrete, which work together to improve the concrete's mechanical properties. In severely cold regions, reinforced concrete bridge structures are subject to freeze-thaw cycles during service, which can degrade the concrete's mechanical properties and, in turn, reduce the structure's bearing capacity. Furthermore, these structures may be subject to impact loads such as vehicle collisions and falling rocks, causing severe damage or even progressive collapse. Therefore, it is necessary to study the impact resistance of reinforced concrete under freeze-thaw conditions.

[0003] In related technologies, freeze-thaw experiments can be carried out on reinforced concrete, and the impact resistance data of reinforced concrete under various freeze-thaw states can be measured. Finally, the impact resistance of reinforced concrete in actual scenarios can be predicted based on experimental conditions and experimental data.

[0004] However, the above method consumes a lot of time and economic costs, and because the experimental conditions in the experimental scenario are often poorly matched with the actual scenario, the above method has poor prediction accuracy for the impact resistance of reinforced concrete under freeze-thaw conditions. Summary of the Invention

[0005] In view of the above problems, the embodiments of the present application provide a model training method, a method and device for predicting the impact resistance of reinforced concrete beams, an electronic device and a readable storage medium to overcome the above problems or at least partially solve the above problems.

[0006] In a first aspect, an embodiment of the present application provides a model training method, the method comprising: Based on the number of experimental freeze-thaw cycles, the preset proportional coefficient, and the experimental impact force platform value in the reinforced concrete beam freeze-thaw test, a first Gaussian process regression relationship is generated; wherein the first Gaussian process regression relationship includes first input data and first output data; the first input data includes the number of experimental freeze-thaw cycles and the preset proportional coefficient; the first output data includes the experimental impact force platform value; the preset proportional coefficient includes a shear span ratio or a bending-shear ratio; Determining the number of freeze-thaw cycles and the impact force platform value of the first sample based on the first Gaussian process regression relationship; Inputting the number of freeze-thaw cycles of the first sample and the preset proportional coefficient into a first convolutional neural network model to obtain a first impact force platform value output by the first convolutional neural network model; Determining a model loss value of the first convolutional neural network model based on the first impact force platform value and the sample impact force platform value; Based on the model loss value, the model parameters of the first convolutional neural network model and the hyperparameters in the first Gaussian process regression relationship are adjusted to obtain a target impact resistance performance prediction model; the target impact resistance performance prediction model is used to determine the target impact force platform value of the target reinforced concrete beam based on the first freeze-thaw cycle number and the first proportional coefficient of the target reinforced concrete beam.

[0007] Optionally, generating a first Gaussian process regression relationship based on the number of freeze-thaw cycles, a preset proportional coefficient, and an experimental impact force platform value in a freeze-thaw experiment on reinforced concrete beams includes: The dynamic Tanh function is used to normalize the number of freeze-thaw cycles, the preset proportional coefficient and the experimental impact force platform value under the freeze-thaw test of reinforced concrete beams, and the normalized values of the freeze-thaw cycle number, the proportional coefficient and the impact force platform value are obtained. Multiplying the normalized value of the number of freeze-thaw cycles and the normalized value of the proportional coefficient to obtain a first multiplied value; Gaussian process regression fitting is performed on the first multiplied value and the normalized value of the impact force platform value to obtain a first Gaussian process regression relationship.

[0008] Optionally, determining the number of freeze-thaw cycles of the first sample and the sample impact force platform value based on the first Gaussian process regression relationship includes: Determining a second multiplication value of the number of freeze-thaw cycles of the experiment and the preset proportional coefficient; In a Cartesian coordinate system, determining a first distance between an experimental coordinate point formed by each experimental impact force platform value corresponding to each of the experimental freeze-thaw cycle numbers and each second multiplied value corresponding to each of the experimental freeze-thaw cycle numbers, and a curve corresponding to the first Gaussian process regression relationship; The first experimental freeze-thaw cycle number corresponding to the coordinate point where the first distance is less than or equal to the first threshold is determined as the first sample freeze-thaw cycle number, and the first experimental impact force platform value contained in the coordinate point where the first distance is less than or equal to the first threshold is determined as the sample impact force platform value.

[0009] Optionally, determining the number of freeze-thaw cycles of the first sample and the sample impact force platform value based on the first Gaussian process regression relationship includes: In a Cartesian coordinate system, random sampling is performed on the curve corresponding to the first Gaussian process regression relationship to obtain a plurality of sampling coordinate points; Determining the number of freeze-thaw cycles of the first sample based on the abscissa value of each sampling coordinate point and a first proportional coefficient; wherein the first proportional coefficient is the same as a preset proportional coefficient of the experimental coordinate point closest to the sampling coordinate point; The ordinate value of each sampling coordinate point is determined as the sample impact force platform value.

[0010] Optionally, determining a model loss value of the first convolutional neural network model based on the first impact force platform value and the sample impact force platform value includes: Determining a mean square error loss value of the first convolutional neural network model based on the first impact force platform value and the sample impact force platform value; Determining a downward trend penalty loss value of the first convolutional neural network model based on the first impact force platform value; Based on the mean square error loss value and the downward trend penalty loss value, a model loss value of the first convolutional neural network model is determined.

[0011] Optionally, determining a downward trend penalty loss value of the first convolutional neural network model based on the first impact force platform value includes: When the first impact force platform value obtained at the first moment is less than or equal to the first impact force platform value obtained at the second moment, determining the downward trend penalty loss value of the first convolutional neural network model at the first moment to be 0; wherein the number of freeze-thaw cycles of the first sample corresponding to the first moment is greater than the number of freeze-thaw cycles of the first sample corresponding to the second moment; When the first impact force platform value obtained at the first moment is greater than the first impact force platform value obtained at the second moment, the difference between the first impact force platform value obtained at the first moment and the first impact force platform value obtained at the second moment is determined as the downward trend penalty loss value of the first convolutional neural network model at the first moment.

[0012] Optionally, determining the model loss value of the first convolutional neural network model based on the mean square error loss value and the downward trend penalty loss value includes: Determine a penalty coefficient for the downward trend penalty loss value; Based on the penalty coefficient, the downward trend penalty loss value and the mean square error loss value, a model loss value of the first convolutional neural network is determined.

[0013] In a second aspect, an embodiment of the present application provides a method for predicting the impact resistance of reinforced concrete beams, the method comprising: Obtaining a first freeze-thaw cycle number and a first proportional coefficient of a target reinforced concrete beam; the first proportional coefficient includes a shear span ratio or a bending-shear ratio; Inputting the first number of freeze-thaw cycles and the first proportional coefficient into a target impact resistance prediction model to obtain a first impact force platform value output by the target impact resistance prediction model; wherein the target impact resistance prediction model is obtained based on any of the above-described model training methods; Based on the first impact force platform value, the impact resistance performance of the target reinforced concrete beam is determined.

[0014] In a third aspect, an embodiment of the present application provides a model training device, comprising: A generation module is configured to generate a first Gaussian process regression relationship based on the number of experimental freeze-thaw cycles, a preset proportional coefficient, and an experimental impact force platform value under a freeze-thaw test of reinforced concrete beams; wherein the first Gaussian process regression relationship includes first input data and first output data; the first input data includes the number of experimental freeze-thaw cycles and the preset proportional coefficient; the first output data includes the experimental impact force platform value; and the preset proportional coefficient includes a shear span ratio or a bending-shear ratio; A first determining module is configured to determine the number of freeze-thaw cycles of the first sample and a sample impact force platform value based on the first Gaussian process regression relationship; An input / output module, configured to input the number of freeze-thaw cycles of the first sample and the preset proportional coefficient into a first convolutional neural network model to obtain a first impact force platform value output by the first convolutional neural network model; a second determining module, configured to determine a model loss value of the first convolutional neural network model based on the first impact force platform value and the sample impact force platform value; An adjustment module is used to adjust the model parameters of the first convolutional neural network model and the hyperparameters in the first Gaussian process regression relationship based on the model loss value to obtain a target impact resistance prediction model; the target impact resistance prediction model is used to determine the target impact force platform value of the target reinforced concrete beam based on the first freeze-thaw cycle number and the first proportional coefficient of the target reinforced concrete beam.

[0015] Optionally, the generating module includes: The normalization submodule is used to normalize the number of freeze-thaw cycles, the preset proportional coefficient and the experimental impact force platform value under the freeze-thaw test of reinforced concrete beams using a dynamic Tanh function, and obtain the normalized value of the number of freeze-thaw cycles, the normalized value of the proportional coefficient and the normalized value of the impact force platform value; a calculation submodule, configured to multiply the normalized value of the number of freeze-thaw cycles and the normalized value of the proportional coefficient to obtain a first multiplied value; The fitting submodule is used to perform Gaussian process regression fitting on the first multiplied value and the normalized value of the impact force platform value to obtain a first Gaussian process regression relationship.

[0016] Optionally, the first determining module includes: A first determining submodule is used to determine a second multiplication value of the number of freeze-thaw cycles of the experiment and the preset proportional coefficient; A second determining submodule is configured to determine, in a Cartesian coordinate system, a first distance between an experimental impact force platform value corresponding to each of the experimental freeze-thaw cycle numbers, an experimental coordinate point formed by each second multiplied value corresponding to each of the experimental freeze-thaw cycle numbers, and a curve corresponding to the first Gaussian process regression relationship; The third determination submodule is used to determine the first experimental freeze-thaw cycle number corresponding to the coordinate point whose first distance is less than or equal to the first threshold as the first sample freeze-thaw cycle number, and determine the first experimental impact force platform value contained in the coordinate point whose first distance is less than or equal to the first threshold as the sample impact force platform value.

[0017] Optionally, the first determining module includes: a sampling submodule, configured to perform random sampling on the curve corresponding to the first Gaussian process regression relationship in a Cartesian coordinate system to obtain a plurality of sampling coordinate points; a fourth determination submodule, configured to determine the number of freeze-thaw cycles of the first sample based on the horizontal coordinate value of each sampling coordinate point and a first proportional coefficient; wherein the first proportional coefficient is the same as a preset proportional coefficient of the experimental coordinate point closest to the sampling coordinate point; The fifth determining submodule is configured to determine the ordinate value of each sampling coordinate point as a sample impact force platform value.

[0018] Optionally, the second determining module includes: a sixth determining submodule, configured to determine a mean square error loss value of the first convolutional neural network model based on the first impact force platform value and the sample impact force platform value; a seventh determining submodule, configured to determine a downward trend penalty loss value of the first convolutional neural network model based on the first impact force platform value; An eighth determination submodule is used to determine a model loss value of the first convolutional neural network model based on the mean square error loss value and the downward trend penalty loss value.

[0019] Optionally, the seventh determining submodule includes: A first determining unit is configured to determine, when a first impact force platform value obtained at a first moment is less than or equal to a first impact force platform value obtained at a second moment, that a downward trend penalty loss value of the first convolutional neural network model at the first moment is 0; wherein the number of freeze-thaw cycles of the first sample corresponding to the first moment is greater than the number of freeze-thaw cycles of the first sample corresponding to the second moment; The second determination unit is used to determine the difference between the first impact force platform value obtained at the first moment and the first impact force platform value obtained at the second moment as the downward trend penalty loss value of the first convolutional neural network model at the first moment when the first impact force platform value obtained at the first moment is greater than the first impact force platform value obtained at the second moment.

[0020] Optionally, the eighth determining submodule includes: A third determining unit is used to determine a penalty coefficient of the downward trend penalty loss value; A fourth determining unit is used to determine the model loss value of the first convolutional neural network based on the penalty coefficient, the downward trend penalty loss value and the mean square error loss value.

[0021] In a fourth aspect, an embodiment of the present application provides a device for predicting the impact resistance of reinforced concrete beams, the device comprising: An acquisition module is used to obtain the current first freeze-thaw cycle number and first proportional coefficient of the target reinforced concrete beam; the first proportional coefficient includes a shear span ratio or a bending-shear ratio; an input / output module, configured to input the first number of freeze-thaw cycles and the first proportional coefficient into a target impact resistance prediction model to obtain a first impact force platform value output by the target impact resistance prediction model; wherein the target impact resistance prediction model is obtained based on any of the above-described model training methods; A determination module is used to determine the impact resistance of the target reinforced concrete beam based on the first impact force platform value.

[0022] In a fifth aspect, an embodiment of the present application provides an electronic device comprising a memory, a processor, and a computer program stored on the memory, wherein the processor executes the computer program to implement the model training method as described in any one of the above items, or the method for predicting the impact resistance of reinforced concrete beams.

[0023] In a sixth aspect, an embodiment of the present application provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, the model training method described in any one of the above items, or the method for predicting the impact resistance of reinforced concrete beams is implemented.

[0024] The specific beneficial effects are: The embodiment of the present application generates a first Gaussian process regression relationship based on the experimental freeze-thaw cycle number, preset proportional coefficient and experimental impact force platform value under the freeze-thaw experiment of reinforced concrete beams; wherein the first Gaussian process regression relationship includes first input data and first output data; the first input data includes the experimental freeze-thaw cycle number and the preset proportional coefficient; the first output data includes the experimental impact force platform value, based on the first Gaussian process regression relationship, the first sample freeze-thaw cycle number and the sample impact force platform value are determined, the first sample freeze-thaw cycle number and the preset proportional coefficient are input into the first convolutional neural network model, and the first impact force platform value output by the first impact resistance prediction model is obtained, based on the first impact force platform value and the sample impact force platform value, the model loss value of the first convolutional neural network model is determined, based on the model loss value, the model parameters of the first convolutional neural network model and the hyperparameters in the first Gaussian process regression relationship are adjusted to obtain the target impact resistance. Impact performance prediction model; the target impact resistance performance prediction model is used to determine the target impact force platform value of the target reinforced concrete beam based on the first freeze-thaw cycle number and the first proportional coefficient of the target reinforced concrete beam. The experimental data of the reinforced concrete beam can be fitted by Gaussian process regression, and multiple sample data can be obtained by fitting the curve, so as to use the sample data to train the first convolutional neural network model to obtain the target impact resistance performance prediction model, and finally the impact resistance performance is predicted by the target impact resistance performance prediction model. When using the target impact resistance performance prediction model for prediction, it is possible to avoid setting up experimental scenarios to carry out freeze-thaw experiments, thereby saving a lot of prediction time and economic costs. Moreover, since the training data utilizes the advantages of Gaussian process regression in flexibility, uncertainty quantification ability and non-parametric characteristics, it can improve the prediction accuracy of the target impact resistance performance prediction model to a certain extent. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments of the present application. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0026] Figure 1 This is a flow chart of a model training method provided in an embodiment of the present application; Figure 2 The following are simplified structural diagrams of shear beams and cross-sectional reinforcement provided in the embodiments of the present application; (a) is a shear beam with a beam length of 1000 mm; (b) is a shear beam with a beam length of 1800 mm; (c) is a shear beam with a beam length of 2200 mm; Figure 3This is a structural diagram of the bending-resistant beam and cross-section reinforcement provided in an embodiment of the present application; Figure 4 A schematic diagram of an ultra-heavy drop weight testing machine provided in an embodiment of the present application; Figure 5 A flowchart of another model training method provided in an embodiment of the present application; Figure 6 A schematic flow chart of a method for predicting the impact resistance of reinforced concrete beams provided in an embodiment of the present application; Figure 7 This is a curve showing the effect of the number of freeze-thaw cycles on the predicted value of the model output provided in the examples of this application; Figure 8 A three-dimensional image showing the effects of shear span ratio and freeze-thaw cycle number on the predicted value of the model output provided in the embodiments of the present application; Figure 9 This is a logic block diagram of a model training device provided in an embodiment of the present application; Figure 10 This is a logic block diagram of a device for predicting the impact resistance of reinforced concrete beams provided in an embodiment of the present application; Figure 11 This is a schematic diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0027] The exemplary embodiments of the present application will be described in more detail below in conjunction with the accompanying drawings in the embodiments of the present application. Although the accompanying drawings show exemplary embodiments of the present application, it should be understood that the present application can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present application and to fully convey the scope of the present application to those skilled in the art.

[0028] Reference Figure 1 , Figure 1 A flow chart of a model training method provided in an embodiment of the present application, the method may include: Step 101: Generate a first Gaussian process regression relationship based on the number of experimental freeze-thaw cycles, a preset proportional coefficient, and an experimental impact force platform value under a freeze-thaw test of reinforced concrete beams; wherein the first Gaussian process regression relationship includes first input data and first output data; the first input data includes the number of experimental freeze-thaw cycles and the preset proportional coefficient; the first output data includes the experimental impact force platform value.

[0029] In the embodiments of the present application, the number of freeze-thaw cycles and the experimental impact force platform value under the freeze-thaw test of reinforced concrete beams can be obtained through the following experimental environment and experimental steps: Experimental environment design: 19 freeze-thaw damaged reinforced concrete beam drop hammer impact test components were tested. Among them, 12 were shear beams, with a design concrete strength grade of C30, mainly considering the influence of shear span ratio and the number of freeze-thaw cycles of large spans on the shear performance of the test beams; 7 were flexural beams, with a design concrete strength grade of C45, and the influence of the number of freeze-thaw cycles on the flexural performance of reinforced concrete beams was studied in detail. Figure 2 As shown, Figure 2 This is a simplified structural diagram of the shear beam and section reinforcement. Figure 2 (a) is a shear beam with a beam length of 1000mm, (b) is a shear beam with a beam length of 1800mm, and (c) is a shear beam with a beam length of 2200mm. The cross-sectional dimensions of the reinforced concrete shear beam design components are all 150mm×300mm. The beam lengths are 1000mm, 1800mm, and 2200mm, and the clear spans are 650mm, 1450mm, and 1850mm. The shear span ratio is The design bending-shear ratios are 1.25, 2.80, and 3.75 respectively. The longitudinal reinforcement is HRB400E grade steel bar, 2C20 at the bottom of the beam and 2C12 at the top of the beam; the stirrups are HRB400E grade steel bar, C6@120. The thickness of the steel bar cover is 25mm. Figure 3 As shown, Figure 3 The following is a simplified structural diagram of the flexural beam and cross-section reinforcement. The design dimensions of the reinforced concrete flexural beam are all 2600mm×150mm×300mm, and the design bending-shear ratio is 1.81. The longitudinal reinforcement is HRB400E grade steel bar, 2C20 at the bottom of the beam and 2C12 at the top of the beam; the stirrups are HRB400E grade steel bar, C6@120. The protective layer thickness of the steel bar is 25mm. The experimental parameters include: the shear specimens are designed with 3 shear span ratios of 1.25, 2.80, and 3.57, and 4 freeze-thaw cycles of 0, 100, 200, and 250 times; the flexural specimens are designed with 7 freeze-thaw cycles of 0, 25, 50, 75, 100, 125, and 150 times. The specimen numbers and experimental times are shown in Table 1 below: Table 1: ; Experimental process: Using Figure 4 The ultra-heavy drop hammer test machine shown in the figure is used to perform drop hammer impact loading on reinforced concrete beams damaged by freeze-thaw. Figure 4In the figure, 1 is the drop hammer, 2 is the protective frame, 3 is the pressure beam, 4 is the specimen, 5 is the bottom beam, 6 is the deflection mark, 7 is the upper and lower cutters, 8 is the fixed rod, and 9 is the rigid base. The bottom beam 5 is rigidly fixed to the rigid base 9. Specimen 4 is located in the middle of the device and is the core of the experiment. The bottom beam 5 is located at the bottom of the device and is fixed to the bottom beam 5 by a fixed rod. The pressure beam 3 is located at the top of the device, opposite to the specimen 4. The upper and lower cutters 7 are connected between the pressure beam 3 and the bottom beam 5. By adding a pair of upper and lower cutters 7 with convex corners and grooves between the pressure beam 3 and the specimen 4, the upper and lower cutters 7 can rotate up and down, thus eliminating any unnecessary constraints on the specimen 4. The protective frame surrounds the entire device, preventing the drop hammer 1 from falling and ensuring safety during the experiment. The drop hammer 1 is located at the top of the device and is the key component for applying the impact force. The specific experimental steps are as follows: 1) Adjust the position of the test beam by moving the bottom beam 5 according to the different spans; 2) Use an electric hoist to remotely control the overall raising and lowering of the drop weight 1; 3) Use a laser rangefinder to measure the height of the drop weight 1; 4) After final commissioning, manually remove the protective bolt and release the unhooker to allow the drop weight 1 to fall freely; 5) Press the high-speed camera controller before the drop weight 1 contacts the beam to record the impact process for 2.5 seconds before and after the impact. The measurement system includes an impact force sensor and a high-speed camera. Before the drop weight impact test begins, the impact force sensor is calibrated and the sensitivity is tested by striking the hammer head. The high-speed camera is installed directly opposite the mid-span of the specimen 4 to ensure a centered and clear image. During this impact test, the time history of the impact force, the time history of the mid-span displacement, the development of the component crack, and the final component failure morphology are recorded throughout. The impact force time history curve is transmitted through an impact force sensor fixed to the drop hammer 1 and transmitted to the data acquisition system, namely the Donghua testing system. The mid-span displacement time history curve and component failure process are tracked and recorded by a high-speed camera as the position changes of the specimen and its mid-span calibration observation point during impact. Ultimately, experimental data such as the impact force platform value, impact force duration, mid-span displacement peak value, and residual displacement can be obtained. Because the impact force platform value is linearly correlated with the impact force duration, mid-span displacement peak value, and residual displacement, the impact force platform value is used as the experimental output data in the embodiments of this application.

[0030] In the embodiments of the present application, the experimental environment and experimental methods described above can be used to determine the number of freeze-thaw cycles and the impact force platform value under freeze-thaw tests on reinforced concrete beams. The preset proportional coefficient can include a shear span ratio or a bending-shear ratio. For shear beams, the preset proportional coefficient is the shear span ratio; for bending beams, the preset proportional coefficient is the bending-shear ratio. For independent, actual reinforced concrete beams, the shear span ratio or bending-shear ratio is fixed.

[0031] In an embodiment of the present application, Gaussian Process Regression (GPR) is a non-parametric machine learning method based on a Bayesian framework. Gaussian process regression automatically adapts to the data distribution through the covariance function (kernel function) and can fit arbitrarily complex nonlinear relationships. The hyperparameters of the kernel function (such as length scale, signal variance, attenuation coefficient) can be optimized through maximum likelihood estimation or Bayesian method to improve the performance of the fitting model. For an objectively existing reinforced concrete beam, the greater the number of freeze-thaw cycles, the smaller the impact force platform value. However, the number of freeze-thaw cycles and the impact force platform value are not linearly correlated. Therefore, Gaussian process regression is used in the embodiment of the present application to fit the number of freeze-thaw cycles and the impact force platform value. Taking into account that the shear span ratio or bending-shear ratio may be different for different reinforced concrete beams, the preset proportional coefficient can also be used as part of the fitting data. Gaussian process regression can be expressed in the form of the following formula 1: (Formula 1); In the above formula 1, is the kernel function of Gaussian process regression, where represents the signal variance of the kernel function, represents the length scale of the kernel function, represents the attenuation coefficient of the covariance of the impact force platform value on the number of freeze-thaw cycles, Represents input data The corresponding number of freeze-thaw cycles, express The corresponding number of freeze-thaw cycles, input data It can be the product of the number of experimental freeze-thaw cycles and the preset ratio parameter. It can also be the product of the number of freeze-thaw cycles and the preset ratio parameter, but it is different from The time steps are different (i.e. the number of freeze-thaw cycles is different). Represents a Gaussian distribution, where the parameter symbols have no practical meaning. In the above formula 1, Part is the first Gaussian process regression relationship, which is used to represent the impact force platform value Correlation characteristics with the number of experimental freeze-thaw cycles and preset ratio parameters.

[0032] Step 102: Determine the number of freeze-thaw cycles of the first sample and the impact force platform value of the sample based on the first Gaussian process regression relationship.

[0033] In an embodiment of the present application, the number of freeze-thaw cycles of a first sample and the sample impact force plateau value can be determined based on a first Gaussian process regression relationship. Random sampling can be performed within the first Gaussian process regression relationship to obtain multiple first number of freeze-thaw cycles of a sample and sample impact force plateau values. Alternatively, the first Gaussian process regression relationship can be used to screen the experimental number of freeze-thaw cycles and the experimental impact force plateau values to obtain the first number of freeze-thaw cycles of a sample and the sample impact force plateau value.

[0034] Step 103: Input the number of freeze-thaw cycles of the first sample and the preset proportional coefficient into a first convolutional neural network model to obtain a first impact force platform value output by the first convolutional neural network model.

[0035] In an embodiment of the present application, a convolutional neural network model can be used as a training model. The number of freeze-thaw cycles of the first sample and a preset proportional coefficient can be input into the first convolutional neural network model, so that the first impact force platform value output by the first convolutional neural network model can be obtained. Among them, the first convolutional neural network may include a convolution layer, a fully connected layer and an output layer. The convolution layer can have 8 convolution kernels, and each convolution kernel can have its own corresponding bias. The number of layers of the fully connected layer can be set to a larger value so that the first convolutional neural network can fully learn the Gaussian process regression relationship. The convolution layer and the fully connected layer are directly connected. The activation function of the first layer of the fully connected layer is the ReLU function. The activation function of the output layer is the sigmoid function.

[0036] Step 104: Determine a model loss value of the first convolutional neural network model based on the first impact force platform value and the sample impact force platform value.

[0037] In an embodiment of the present application, a model loss value of the first convolutional neural network model can be calculated based on the first impact force platform value and the sample impact force platform value, wherein the model loss value can be a mean square error loss value or a mean absolute error loss value.

[0038] Step 105: Based on the model loss value, adjust the model parameters of the first convolutional neural network model and the hyperparameters in the first Gaussian process regression relationship to obtain a target impact resistance prediction model; the target impact resistance prediction model is used to determine the target impact force platform value of the target reinforced concrete beam based on the first freeze-thaw cycle number and the first proportional coefficient of the target reinforced concrete beam.

[0039] In the embodiment of the present application, the model parameters of the first convolutional neural network model and the hyperparameters in the first Gaussian process regression relationship can be adjusted according to the model loss value, so as to obtain the target impact resistance performance prediction model. Wherein, referring to Formula 1, the hyperparameters in the first Gaussian process regression relationship may include (signal variance of the kernel function), (the length scale of the kernel function) and (Attenuation coefficient of the covariance of the impact force platform value to the number of freeze-thaw cycles). The direction of adjusting the above-mentioned model parameters and hyperparameters can be the direction of reducing the model loss value. In addition, the above-mentioned model parameters and hyperparameters can be adjusted multiple times. After each adjustment is completed, the process of steps 101 to 105 can be repeated to continuously train the first convolutional neural network and continuously optimize the Gaussian process regression relationship. When the number of training times of the first convolutional neural network model reaches a preset number of times, or the model loss value meets the preset convergence conditions, the training of the first convolutional neural network model can be stopped, and the first convolutional neural network model obtained after the last model parameter adjustment is determined as the target impact resistance performance prediction model. The target impact resistance performance prediction model can be used to determine the target impact force platform value of the target reinforced concrete beam based on the first number of freeze-thaw cycles and the first proportional coefficient of the target reinforced concrete beam.

[0040] In an embodiment of the present application, a first Gaussian process regression relationship is generated based on the experimental freeze-thaw cycle number, preset proportional coefficient and experimental impact force platform value under the freeze-thaw experiment of reinforced concrete beams; wherein, the first Gaussian process regression relationship includes first input data and first output data; the first input data includes the experimental freeze-thaw cycle number and the preset proportional coefficient; the first output data includes the experimental impact force platform value, based on the first Gaussian process regression relationship, the first sample freeze-thaw cycle number and the sample impact force platform value are determined, the first sample freeze-thaw cycle number and the preset proportional coefficient are input into the first convolutional neural network model, and the first impact force platform value output by the first impact resistance prediction model is obtained, based on the first impact force platform value and the sample impact force platform value, the model loss value of the first convolutional neural network model is determined, based on the model loss value, the model parameters of the first convolutional neural network model and the hyperparameters in the first Gaussian process regression relationship are adjusted to obtain the target The target impact resistance performance prediction model is used to determine the target impact force platform value of the target reinforced concrete beam based on the first freeze-thaw cycle number and the first proportional coefficient of the target reinforced concrete beam. The experimental data of the reinforced concrete beam can be fitted by Gaussian process regression, and multiple sample data can be obtained by fitting the curve, so as to use the sample data to train the first convolutional neural network model to obtain the target impact resistance performance prediction model, and finally the impact resistance prediction is performed by the target impact resistance performance prediction model. When using the target impact resistance performance prediction model for prediction, it is possible to avoid setting up experimental scenarios for freeze-thaw experiments, thereby saving a lot of prediction time and economic costs. Moreover, since the training data utilizes the advantages of Gaussian process regression in flexibility, uncertainty quantification ability and non-parametric characteristics, it can improve the prediction accuracy of the target impact resistance performance prediction model to a certain extent.

[0041] Reference Figure 5 , Figure 5 A flowchart of another model training method provided in an embodiment of the present application, wherein the method may include: Step 201: Use a dynamic Tanh function to normalize the number of freeze-thaw cycles, the preset proportional coefficient, and the experimental impact force platform value under the freeze-thaw test of reinforced concrete beams, respectively, to obtain a normalized value of the number of freeze-thaw cycles, a normalized value of the proportional coefficient, and a normalized value of the impact force platform value.

[0042] In the embodiments of the present application, Dynamic Tanh (DyT) is a simple and efficient method for replacing the normalization layer (such as Layer Normalization, LN) in deep learning models such as Transformer. Its expression is shown in the following formula 2: (Formula 2); In the above formula 2, is a learnable scalar parameter used to dynamically adjust the scaling of the input. and Tanh is a channel-by-channel learnable parameter, similar to the affine transformation parameter in the normalization layer. The hyperbolic function Tanh can compress the input to the range of [-1, 1] to simulate the compression effect of the normalization layer on extreme values.

[0043] In an embodiment of the present application, the dynamic Tanh function of the above form can be used to normalize the experimental freeze-thaw cycle number, the preset proportional coefficient and the experimental impact force platform value under the freeze-thaw test of reinforced concrete beams, respectively, so as to obtain the normalized value of the freeze-thaw cycle number, the normalized value of the proportional coefficient and the normalized value of the impact force platform value. Among them, the experimental freeze-thaw cycle number can correspond to the normalized value of the freeze-thaw cycle number, the preset proportional coefficient can correspond to the normalized value of the proportional coefficient, and the experimental impact force platform value can correspond to the normalized value of the impact force platform value. For other implementation contents of this step, please refer to the embodiment content of step 101 and will not be repeated here.

[0044] Step 202: multiply the normalized value of the number of freeze-thaw cycles and the normalized value of the proportional coefficient to obtain a first multiplied value.

[0045] In an embodiment of the present application, the normalized value of the number of freeze-thaw cycles and the normalized value of the proportional coefficient may be multiplied together to obtain a first multiplied value.

[0046] Step 203 : Perform Gaussian process regression fitting on the first multiplied value and the normalized value of the impact force platform value to obtain a first Gaussian process regression relationship.

[0047] In the embodiment of the present application, the first multiplication value and the normalized value of the impact force platform value can be fitted in a Gaussian process regression manner, thereby obtaining a first Gaussian process regression relationship. The first Gaussian process regression relationship can refer to the form of formula 1, but the kernel function middle, and Represents the two first multiplication values at different time steps, Indicates the normalized value of the impact force platform value.

[0048] In an embodiment of the present application, by adopting the dynamic Tanh function, the experimental freeze-thaw cycle number, the preset proportional coefficient and the experimental impact force platform value under the freeze-thaw test of reinforced concrete beams are normalized respectively to obtain the normalized value of the freeze-thaw cycle number, the normalized value of the proportional coefficient and the normalized value of the impact force platform value. The normalized value of the freeze-thaw cycle number and the normalized value of the proportional coefficient are multiplied to obtain a first multiplied value. The first multiplied value and the normalized value of the impact force platform value are subjected to Gaussian process regression fitting to obtain a first Gaussian process regression relationship. The dynamic Tanh function can be used to normalize the data, and the normalized data can be fitted to obtain the first Gaussian process regression relationship. It combines the efficiency, robustness and ease of use of the dynamic Tanh function, and can improve the accuracy of the fitted first Gaussian process regression relationship to a certain extent.

[0049] Step 204 : Determine the number of freeze-thaw cycles of the first sample and the impact force platform value of the sample based on the first Gaussian process regression relationship.

[0050] In the embodiment of the present application, the implementation content of this step can refer to the embodiment content of step 102 and will not be repeated here.

[0051] Optionally, step 204 may include the following sub-steps: Sub-step 2041: determining a second multiplication value of the number of freeze-thaw cycles in the experiment and the preset proportional coefficient.

[0052] In an embodiment of the present application, the second multiplication value of the number of experimental freeze-thaw cycles and the preset proportional coefficient can be calculated. In order to match the second multiplication value with the first Gaussian process regression relationship, if the data normalized by the dynamic Tanh function is used in the first Gaussian process regression relationship, when calculating the second multiplication value here, the dynamic Tanh function can also be used to normalize the number of experimental freeze-thaw cycles and the preset proportional coefficient respectively, and then the numerical values obtained after normalization are multiplied to obtain the second multiplication value. If the original data is used in the first Gaussian process regression relationship, when calculating the second multiplication value, it is not necessary to normalize and can be directly multiplied.

[0053] Sub-step 2042: Determine, in a Cartesian coordinate system, the first distance between the experimental impact force platform value corresponding to each of the experimental freeze-thaw cycle numbers, the experimental coordinate point formed by each second multiplication value corresponding to each of the experimental freeze-thaw cycle numbers, and the curve corresponding to the first Gaussian process regression relationship.

[0054] In an embodiment of the present application, after obtaining the first Gaussian process regression relationship, the curve corresponding to the first Gaussian process regression relationship can be generated in a Cartesian coordinate system. At the same time, with the experimental freeze-thaw cycle number as a connecting link, the corresponding relationship between the experimental impact force platform value and the second multiplication value can be constructed, and after the experimental impact force platform value is normalized using a dynamic Tanh function, a coordinate point is constructed in a Cartesian coordinate system with the second multiplication value. For each coordinate point, the second multiplication value and the experimental impact force platform value it contains can correspond to the same experimental freeze-thaw cycle number. Afterwards, the first distance between each coordinate point and the curve of the first Gaussian process regression relationship can be calculated. The calculation method of the first distance can refer to the calculation method of the distance from the mathematical midpoint to the curve, which will not be repeated here.

[0055] Sub-step 2043: Determine the first experimental freeze-thaw cycle number corresponding to the coordinate point whose first distance is less than or equal to the first threshold as the first sample freeze-thaw cycle number, and determine the first experimental impact force platform value contained in the coordinate point whose first distance is less than or equal to the first threshold as the sample impact force platform value.

[0056] In an embodiment of the present application, a first threshold value can be set to measure the first distance from the coordinate point to the curve corresponding to the first Gaussian process regression relationship. If the first distance is greater than the first threshold value, it indicates that the data error corresponding to the coordinate point is too large and can be discarded. If the first distance is less than or equal to the first threshold value, it can be considered that the data error corresponding to the coordinate point is small, and the first experimental freeze-thaw cycle number corresponding to the coordinate point can be determined as the first sample freeze-thaw cycle number, and the first experimental impact force platform value in the coordinate point can be determined as the sample impact force platform value. Among them, if the data normalized by the dynamic Tanh function is used in the first Gaussian process regression relationship, the first sample freeze-thaw cycle number and the sample impact force platform value can be the values normalized by the dynamic Tanh function, otherwise the first sample freeze-thaw cycle number and the sample impact force platform value can be actual values.

[0057] In an embodiment of the present application, by determining the second multiplication value of the experimental freeze-thaw cycle number and the preset proportional coefficient, in the Cartesian coordinate system, the experimental impact force platform value corresponding to each experimental freeze-thaw cycle number, and the experimental coordinate point composed of each second multiplication value corresponding to each experimental freeze-thaw cycle number, and the first distance between the curve corresponding to the first Gaussian process regression relationship are determined, the first experimental freeze-thaw cycle number corresponding to the coordinate point whose first distance is less than or equal to the first threshold is determined as the first sample freeze-thaw cycle number, and the first experimental impact force platform value contained in the coordinate point whose first distance is less than or equal to the first threshold is determined as the sample impact force platform value. The experimental data can be screened to obtain the first sample freeze-thaw cycle number and the sample impact force platform value for training, which can improve the accuracy and reliability of the first sample freeze-thaw cycle number and the sample impact force platform value to a certain extent.

[0058] Sub-step 2044 , performing random sampling on the curve corresponding to the first Gaussian process regression relationship in a Cartesian coordinate system to obtain a plurality of sampling coordinate points.

[0059] In an embodiment of the present application, in a Cartesian coordinate system, random sampling may be performed on a curve corresponding to the first Gaussian process regression relationship, thereby obtaining a plurality of sampling coordinate points.

[0060] Sub-step 2045 , determining the number of freeze-thaw cycles of the first sample based on the horizontal coordinate value of each sampling coordinate point and a first proportional coefficient; wherein the first proportional coefficient is the same as the preset proportional coefficient of the experimental coordinate point closest to the sampling coordinate point.

[0061] In an embodiment of the present application, the number of freeze-thaw cycles of the first sample can be calculated based on the horizontal coordinate value of each sampling coordinate point and a preset proportional coefficient, wherein the value of the preset proportional coefficient can be the same as the preset proportional coefficient of the experimental coordinate point closest to the sampling coordinate point.

[0062] Sub-step 2046: determining the ordinate value of each sampling coordinate point as the sample impact force platform value.

[0063] In the embodiment of the present application, the vertical coordinate value of each sampling point can be determined as the sample impact force platform value.

[0064] In an embodiment of the present application, random sampling is performed on the curve corresponding to the first Gaussian process regression relationship in a Cartesian coordinate system to obtain multiple sampling coordinate points, and the number of freeze-thaw cycles of the first sample is determined based on the horizontal coordinate value of each sampling coordinate point and the first proportional coefficient; wherein the first proportional coefficient is the same as the preset proportional coefficient of the experimental coordinate point closest to the sampling coordinate point, and the vertical coordinate value of each sampling coordinate point is determined as the sample impact force platform value. The first sample freeze-thaw cycle number and the sample impact force platform value can be obtained by sampling the first Gaussian process regression relationship curve, which can improve the accuracy of the first sample freeze-thaw cycle number and the sample impact force platform value to a certain extent.

[0065] Step 205: Input the number of freeze-thaw cycles of the first sample and the preset proportional coefficient into a first convolutional neural network model to obtain a first impact force platform value output by the first convolutional neural network model.

[0066] In an embodiment of the present application, when the number of freeze-thaw cycles of the first sample is input into the first convolutional neural network model, it can be input in ascending order. For other implementation details of this step, please refer to the embodiment details of step 103 and will not be repeated here.

[0067] Step 206: Determine the mean square error loss value of the first convolutional neural network model based on the first impact force platform value and the sample impact force platform value.

[0068] In an embodiment of the present application, the model loss value can be composed of a mean square error loss value and a downward trend penalty loss value. The mean square error loss value of the first convolutional neural network model can be calculated based on the first impact force platform value and the sample impact force platform value, and the calculation method is shown in the following formula 3: (Formula 3); In the above formula 3, represents the mean square error loss value, Indicates the total number of first impact force plateau values. Indicates the sample impact force platform value, Indicates the first impact force platform value.

[0069] Step 207: Determine a downward trend penalty loss value of the first convolutional neural network model based on the first impact force platform value.

[0070] In the embodiment of the present application, the first impact force plateau value output by the model may increase with the number of freeze-thaw cycles, which is inconsistent with the actual situation. Therefore, a downward trend penalty loss value can be added to the model loss value to accelerate the convergence of the model. The downward trend penalty loss value of the first convolutional neural network model can be calculated based on the first impact force plateau value, and its calculation method is shown in the following formula 4: (Formula 4); In the above formula 4, Indicates the downward trend penalty loss value, express The total number of calculations. Indicates The first impact force platform value outputted later is Indicates the The first impact force platform value output.

[0071] Optionally, step 207 may include the following sub-steps: Sub-step 2071, when the first impact force platform value obtained at the first moment is less than or equal to the first impact force platform value obtained at the second moment, determining that the downward trend penalty loss value of the first convolutional neural network model at the first moment is 0; wherein the number of freeze-thaw cycles of the first sample corresponding to the first moment is greater than the number of freeze-thaw cycles of the first sample corresponding to the second moment.

[0072] In an embodiment of the present application, if the downward trend penalty loss value contains a negative value, the convergence speed of the model may be relatively slow. Therefore, the downward trend penalty loss value can be improved. When the first impact force platform value obtained at the first moment is less than or equal to the first impact force platform value obtained at the second moment, it can be determined that the downward trend penalty loss value of the first convolutional neural network model at the first moment is 0. Among them, the number of freeze-thaw cycles of the first sample corresponding to the first moment is greater than the number of freeze-thaw cycles of the first sample corresponding to the second moment, that is, on the time axis, the first moment is located behind the second moment.

[0073] In sub-step 2072, when the first impact force platform value obtained at the first moment is greater than the first impact force platform value obtained at the second moment, the difference between the first impact force platform value obtained at the first moment and the first impact force platform value obtained at the second moment is determined as the downward trend penalty loss value of the first convolutional neural network model at the first moment.

[0074] In an embodiment of the present application, if the first impact force platform value obtained at the first moment is greater than the first impact force platform value obtained at the second moment, the difference between the first impact force platform value obtained at the first moment and the first impact force platform value obtained at the second moment can be calculated, and the difference can be determined as the downward trend penalty loss value of the first convolutional neural network model at the first moment. Combining sub-steps 2071 and 2072, the downward trend penalty loss value can be changed to the form of the following formula 5: (Formula 5); By calculating the downward trend penalty loss value in the form of the above formula 5, the method of sub-step 2071 and sub-step 2072 can be implemented. The meaning of each parameter symbol in the above formula 5 can refer to the meaning of each parameter symbol in formula 4, and will not be repeated here.

[0075] In an embodiment of the present application, by determining that the downward trend penalty loss value of the first convolutional neural network model at the first moment is 0 when the first impact force platform value obtained at the first moment is less than or equal to the first impact force platform value obtained at the second moment; wherein, the number of freeze-thaw cycles of the first sample corresponding to the first moment is greater than the number of freeze-thaw cycles of the first sample corresponding to the second moment, and when the first impact force platform value obtained at the first moment is greater than the first impact force platform value obtained at the second moment, the difference between the first impact force platform value obtained at the first moment and the first impact force platform value obtained at the second moment is determined as the downward trend penalty loss value of the first convolutional neural network model at the first moment, and the downward trend penalty loss value can be calculated only when the output rises, which can improve the reliability of the downward trend penalty loss value to a certain extent and improve the convergence efficiency of the first convolutional neural network model.

[0076] Step 208: Determine a model loss value of the first convolutional neural network model based on the mean square error loss value and the downward trend penalty loss value.

[0077] In an embodiment of the present application, the model loss value of the first convolutional neural network model can be calculated based on the mean square error loss value and the downtrend penalty loss value. For example, the model loss value can be directly obtained by adding the mean square error loss value and the downtrend penalty loss value, or a weight value can be assigned to the mean square error loss value and / or the downtrend penalty loss value, and then the mean square error loss value and the downtrend penalty value are weighted and summed to obtain the model loss value.

[0078] Optionally, step 208 may include the following sub-steps: Sub-step 2081: determining a penalty coefficient for the downward trend penalty loss value.

[0079] In an embodiment of the present application, a penalty coefficient of the downward trend penalty loss value may be determined to limit the role of the downward trend penalty loss value in the model loss value.

[0080] Sub-step 2082: Determine the model loss value of the first convolutional neural network based on the penalty coefficient, the downward trend penalty loss value, and the mean square error loss value.

[0081] In an embodiment of the present application, the model loss value of the first convolutional neural network model can be calculated based on the penalty coefficient, the downward trend penalty loss value, and the mean square error loss value, and the calculation method is shown in the following formula 6: (Formula 6); In the above formula 6, Represents the model loss value, represents the mean square error loss value, represents the penalty coefficient, Indicates the downward trend penalty loss value.

[0082] In an embodiment of the present application, by determining the penalty coefficient of the downward trend penalty loss value, and determining the model loss value of the first convolutional neural network based on the penalty coefficient, the downward trend penalty loss value and the mean square error loss value, the weight ratio of the downward trend penalty loss value in the model loss value can be flexibly adjusted, thereby improving the flexibility and reliability of the model loss value to a certain extent.

[0083] In an embodiment of the present application, the mean square error loss value of the first convolutional neural network model is determined based on the first impact force platform value and the sample impact force platform value, the downward trend penalty loss value of the first convolutional neural network model is determined based on the first impact force platform value, and the model loss value of the first convolutional neural network model is determined based on the mean square error loss value and the downward trend penalty loss value. The downward trend penalty loss value can be added to the model loss value, which can reduce the output increase during model training and improve the accuracy and availability of the model loss value to a certain extent.

[0084] Step 209: Based on the model loss value, adjust the model parameters of the first convolutional neural network model and the hyperparameters in the first Gaussian process regression relationship to obtain a target impact resistance performance prediction model.

[0085] In the embodiment of the present application, since a downward trend penalty loss value is added to the model loss value, during the model training process, not only can the model convergence be accelerated, but the Gaussian process regression fitting can also be made more accurate and more in line with the actual situation. For other implementation details of this step, please refer to the embodiment details of step 105 and will not be repeated here.

[0086] Reference Figure 6 , Figure 6 A schematic flow chart of a method for predicting the impact resistance of reinforced concrete beams provided in an embodiment of the present application, the method may include: Step 301: Obtain the current first freeze-thaw cycle number and first proportional coefficient of the target reinforced concrete beam; the first proportional coefficient includes a shear span ratio or a bending-shear ratio.

[0087] In an embodiment of the present application, the current first freeze-thaw cycle number and first proportional coefficient of the target reinforced concrete beam can be obtained, where the first proportional coefficient can include a shear span ratio or a bending-shear ratio. If the target reinforced concrete beam is a shear beam, the first proportional coefficient is the shear span ratio; if the target reinforced concrete beam is a bending beam, the first proportional coefficient is the bending-shear ratio.

[0088] Step 302: Input the first number of freeze-thaw cycles and the first proportional coefficient into a target impact resistance prediction model to obtain a first impact force platform value output by the target impact resistance prediction model; wherein the target impact resistance prediction model is obtained based on any of the model training methods described above.

[0089] In an embodiment of the present application, the first number of freeze-thaw cycles and the first proportional coefficient can be input into a target impact resistance prediction model to obtain a first impact force platform value output by the target impact resistance prediction model. The target impact resistance prediction model can be obtained using any of the above-described model training methods.

[0090] Step 303: Determine the impact resistance of the target reinforced concrete beam based on the first impact force platform value.

[0091] In an embodiment of the present application, the impact resistance of a target reinforced concrete beam can be determined based on a first impact force platform value. The impact resistance can be graded based on the impact force platform value. If the first impact force platform value is less than a first preset threshold, the impact resistance of the target reinforced concrete beam can be considered poor; if the first impact force platform value is greater than or equal to the first preset threshold and less than a second preset threshold, the impact resistance of the target reinforced concrete beam can be considered moderate; and if the first impact force platform value is greater than or equal to a second preset threshold, the impact resistance of the target reinforced concrete beam can be considered good. The second preset threshold can be greater than the first preset threshold.

[0092] In an embodiment of the present application, by obtaining the current first freeze-thaw cycle number and the first proportional coefficient of the target reinforced concrete beam; the first proportional coefficient includes the shear span ratio or the bending-shear ratio, the first freeze-thaw cycle number and the first proportional coefficient are input into the target impact resistance prediction model, and the first impact force platform value output by the target impact resistance prediction model is obtained; wherein, the target impact resistance prediction model is obtained based on any of the model training methods described above, and the impact resistance of the target reinforced concrete beam is determined based on the first impact force platform value. The target impact resistance prediction model can be used to predict the first impact force platform value of the target reinforced concrete beam, and the impact resistance of the target reinforced concrete beam is determined according to the first impact force platform value. This can avoid a lot of time and economic costs caused by repeated experiments, and can improve the accuracy of the impact resistance prediction to a certain extent.

[0093] In the examples of the present application, the following Tables 2 and 3 respectively show the impact test results of freeze-thaw damaged reinforced concrete shear beams and freeze-thaw damaged reinforced concrete bending beams: Table 2: ; Table 3: ; The specimen numbers in Tables 2 and 3 are the same as those in Table 1. Figure 7 The influence curve of the number of freeze-thaw cycles on the predicted value of the model output is given. Figure 8 The three-dimensional graph of the influence of shear span ratio and freeze-thaw cycle number on the predicted value of model output is given. Figure 7 and Figure 8 It can be seen that for shear beams, relatively accurate prediction results can be achieved when the shear span is relatively small, and inaccurate prediction results may occasionally occur when the shear span is relatively large. Combined with Table 3, it can be seen that for bending beams, the deviation between the predicted value and the true value is irregular, but overall close to the true value. Table 2, Table 3, Figure 7 and Figure 8 It can be shown that the technical solution of this application is effective as a whole.

[0094] refer to Figure 9 , Figure 9 A logical block diagram of a model training device provided in an embodiment of the present application, wherein the device 700 may include: A generation module 701 is configured to generate a first Gaussian process regression relationship based on the number of freeze-thaw cycles, a preset proportional coefficient, and an experimental impact force platform value in a freeze-thaw experiment on reinforced concrete beams; wherein the first Gaussian process regression relationship includes first input data and first output data; the first input data includes the number of freeze-thaw cycles and the preset proportional coefficient; the first output data includes the experimental impact force platform value; and the preset proportional coefficient includes a shear span ratio or a bending-shear ratio. A first determining module 702 is configured to determine the number of freeze-thaw cycles of the first sample and the impact force platform value of the sample based on the first Gaussian process regression relationship; An input / output module 703 is configured to input the number of freeze-thaw cycles of the first sample and the preset proportional coefficient into a first convolutional neural network model to obtain a first impact force platform value output by the first convolutional neural network model; a second determining module 704, configured to determine a model loss value of the first convolutional neural network model based on the first impact force platform value and the sample impact force platform value; Adjustment module 705 is used to adjust the model parameters of the first convolutional neural network model and the hyperparameters in the first Gaussian process regression relationship based on the model loss value to obtain a target impact resistance performance prediction model; the target impact resistance performance prediction model is used to determine the target impact force platform value of the target reinforced concrete beam based on the first freeze-thaw cycle number and the first proportional coefficient of the target reinforced concrete beam.

[0095] Optionally, the generating module 701 includes: The normalization submodule is used to normalize the number of freeze-thaw cycles, the preset proportional coefficient and the experimental impact force platform value under the freeze-thaw test of reinforced concrete beams using a dynamic Tanh function, and obtain the normalized value of the number of freeze-thaw cycles, the normalized value of the proportional coefficient and the normalized value of the impact force platform value; a calculation submodule, configured to multiply the normalized value of the number of freeze-thaw cycles and the normalized value of the proportional coefficient to obtain a first multiplied value; The fitting submodule is used to perform Gaussian process regression fitting on the first multiplied value and the normalized value of the impact force platform value to obtain a first Gaussian process regression relationship.

[0096] Optionally, the first determining module 702 includes: A first determining submodule is used to determine a second multiplication value of the number of freeze-thaw cycles of the experiment and the preset proportional coefficient; A second determining submodule is configured to determine, in a Cartesian coordinate system, a first distance between an experimental impact force platform value corresponding to each of the experimental freeze-thaw cycle numbers, an experimental coordinate point formed by each second multiplied value corresponding to each of the experimental freeze-thaw cycle numbers, and a curve corresponding to the first Gaussian process regression relationship; The third determination submodule is used to determine the first experimental freeze-thaw cycle number corresponding to the coordinate point whose first distance is less than or equal to the first threshold as the first sample freeze-thaw cycle number, and determine the first experimental impact force platform value contained in the coordinate point whose first distance is less than or equal to the first threshold as the sample impact force platform value.

[0097] Optionally, the first determining module 702 includes: a sampling submodule, configured to perform random sampling on the curve corresponding to the first Gaussian process regression relationship in a Cartesian coordinate system to obtain a plurality of sampling coordinate points; a fourth determination submodule, configured to determine the number of freeze-thaw cycles of the first sample based on the horizontal coordinate value of each sampling coordinate point and a first proportional coefficient; wherein the first proportional coefficient is the same as a preset proportional coefficient of the experimental coordinate point closest to the sampling coordinate point; The fifth determining submodule is configured to determine the ordinate value of each sampling coordinate point as a sample impact force platform value.

[0098] Optionally, the second determining module 704 includes: a sixth determining submodule, configured to determine a mean square error loss value of the first convolutional neural network model based on the first impact force platform value and the sample impact force platform value; a seventh determining submodule, configured to determine a downward trend penalty loss value of the first convolutional neural network model based on the first impact force platform value; An eighth determination submodule is used to determine a model loss value of the first convolutional neural network model based on the mean square error loss value and the downward trend penalty loss value.

[0099] Optionally, the seventh determining submodule includes: A first determining unit is configured to determine, when a first impact force platform value obtained at a first moment is less than or equal to a first impact force platform value obtained at a second moment, that a downward trend penalty loss value of the first convolutional neural network model at the first moment is 0; wherein the number of freeze-thaw cycles of the first sample corresponding to the first moment is greater than the number of freeze-thaw cycles of the first sample corresponding to the second moment; The second determination unit is used to determine the difference between the first impact force platform value obtained at the first moment and the first impact force platform value obtained at the second moment as the downward trend penalty loss value of the first convolutional neural network model at the first moment when the first impact force platform value obtained at the first moment is greater than the first impact force platform value obtained at the second moment.

[0100] Optionally, the eighth determining submodule includes: A third determining unit is used to determine a penalty coefficient of the downward trend penalty loss value; A fourth determining unit is used to determine the model loss value of the first convolutional neural network based on the penalty coefficient, the downward trend penalty loss value and the mean square error loss value.

[0101] The model training device in the embodiments of the present application can be an electronic device or a component of an electronic device, such as an integrated circuit or chip. The electronic device can be a terminal or other device other than a terminal. For example, the electronic device can be a GPU box, a tablet computer, a laptop computer, a robot, an ultra-mobile personal computer (UMPC), or a personal digital assistant (PDA), etc. It can also be a server, a personal computer (PC), etc., and the embodiments of the present application do not specifically limit this.

[0102] The model training device in the embodiments of the present application may be a device having an operating system. The operating system may be an Android operating system, a Linux operating system, a Windows operating system, or other possible operating systems, which are not specifically limited in the embodiments of the present application.

[0103] The model training device provided in the embodiment of the present application can achieve Figure 1 and Figure 5 To avoid repetition, the various processes implemented in the method embodiment are not described here.

[0104] Reference Figure 10 , Figure 10 This is a logic block diagram of a device for predicting the impact resistance of reinforced concrete beams provided in an embodiment of the present application. The device 800 may include: An acquisition module 801 is configured to acquire a first freeze-thaw cycle number and a first proportional coefficient of a target reinforced concrete beam; the first proportional coefficient includes a shear span ratio or a bending-shear ratio; An input / output module 802 is configured to input the first number of freeze-thaw cycles and the first proportional coefficient into a target impact resistance prediction model to obtain a first impact force platform value output by the target impact resistance prediction model; wherein the target impact resistance prediction model is obtained based on any of the above-described model training methods; The determination module 803 is configured to determine the impact resistance of the target reinforced concrete beam based on the first impact force platform value.

[0105] The device for predicting the impact resistance of reinforced concrete beams in the embodiments of the present application can be an electronic device or a component of an electronic device, such as an integrated circuit or chip. The electronic device can be a terminal or other device other than a terminal. Exemplary electronic devices include GPU boxes, tablet computers, laptop computers, robots, ultra-mobile personal computers (UMPCs), personal digital assistants (PDAs), servers, personal computers (PCs), and the like, but are not specifically limited in the embodiments of the present application.

[0106] The device for predicting the impact resistance of reinforced concrete beams in the embodiments of the present application can be a device having an operating system. The operating system can be an Android operating system, a Linux operating system, a Windows operating system, or other possible operating systems, which are not specifically limited in the embodiments of the present application.

[0107] The device for predicting the impact resistance of reinforced concrete beams provided in the embodiment of the present application can achieve Figure 6 To avoid repetition, the various processes implemented in the method embodiment are not described here.

[0108] The present application provides an electronic device. Figure 11 The electronic device 90 includes: a processor 901, a memory 902, and a computer program 9021 stored in the memory 902 and executable on the processor 901. When the processor 901 executes the program, the model training method of the aforementioned embodiment or the method for predicting the impact resistance of reinforced concrete beams is implemented.

[0109] An embodiment of the present application also provides a computer-readable storage medium having a computer program / instruction stored thereon. When the computer program / instruction is executed by a processor, the model training method disclosed in the embodiment of the present application, or the steps in the method for predicting the impact resistance of reinforced concrete beams, are implemented.

[0110] An embodiment of the present application also provides a computer program product, which, when executed on an electronic device, enables a processor to implement the model training method disclosed in the embodiment of the present application, or the steps in the method for predicting the impact resistance of reinforced concrete beams.

[0111] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0112] The embodiments of the present application are described with reference to the flowcharts and / or block diagrams of the methods, apparatuses, electronic devices, and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing terminal device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing terminal device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0113] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing terminal device to operate in a specific manner, so that the instructions stored in the computer readable memory produce a manufactured product including an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0114] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal device so that a series of operating steps are executed on the computer or other programmable terminal device to produce a computer-implemented process, thereby providing instructions for executing on the computer or other programmable terminal device to implement the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0115] Although preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they become aware of the basic inventive concepts. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the embodiments of the present invention.

[0116] Finally, it should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "includes," or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article, or terminal device that includes a series of elements includes not only those elements, but also other elements not explicitly listed, or elements that are inherent to such process, method, article, or terminal device. In the absence of further restrictions, an element defined by the phrase "comprises a ..." does not exclude the presence of additional identical elements in the process, method, article, or terminal device that includes the element.

[0117] The above is a detailed introduction to a model training method, a reinforced concrete beam impact resistance prediction method and a device provided by this application. This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method of this application and its core idea; at the same time, for general technical personnel in this field, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on this application.

Claims

1. A model training method, characterized in that: The method comprises: Based on the number of experimental freeze-thaw cycles, the preset proportional coefficient, and the experimental impact force platform value in the reinforced concrete beam freeze-thaw test, a first Gaussian process regression relationship is generated; wherein the first Gaussian process regression relationship includes first input data and first output data; the first input data includes the number of experimental freeze-thaw cycles and the preset proportional coefficient; the first output data includes the experimental impact force platform value; the preset proportional coefficient includes a shear span ratio or a bending-shear ratio; Determining the number of freeze-thaw cycles and the impact force platform value of the first sample based on the first Gaussian process regression relationship; Inputting the number of freeze-thaw cycles of the first sample and the preset proportional coefficient into a first convolutional neural network model to obtain a first impact force platform value output by the first convolutional neural network model; Determining a model loss value of the first convolutional neural network model based on the first impact force platform value and the sample impact force platform value; Based on the model loss value, the model parameters of the first convolutional neural network model and the hyperparameters in the first Gaussian process regression relationship are adjusted to obtain a target impact resistance performance prediction model; the target impact resistance performance prediction model is used to determine the target impact force platform value of the target reinforced concrete beam based on the first freeze-thaw cycle number and the first proportional coefficient of the target reinforced concrete beam.

2. The method according to claim 1, characterized in that The first Gaussian process regression relationship is generated based on the number of freeze-thaw cycles, the preset proportional coefficient, and the experimental impact force platform value under the freeze-thaw test of reinforced concrete beams, including: The dynamic Tanh function is used to normalize the number of freeze-thaw cycles, the preset proportional coefficient and the experimental impact force platform value under the freeze-thaw test of reinforced concrete beams, and the normalized values of the freeze-thaw cycle number, the proportional coefficient and the impact force platform value are obtained. Multiplying the normalized value of the number of freeze-thaw cycles and the normalized value of the proportional coefficient to obtain a first multiplied value; Gaussian process regression fitting is performed on the first multiplied value and the normalized value of the impact force platform value to obtain a first Gaussian process regression relationship.

3. The method according to claim 1, characterized in that Determining the number of freeze-thaw cycles and the impact force platform value of the first sample based on the first Gaussian process regression relationship includes: Determining a second multiplication value of the number of freeze-thaw cycles of the experiment and the preset proportional coefficient; In a Cartesian coordinate system, determining a first distance between an experimental coordinate point formed by each experimental impact force platform value corresponding to each of the experimental freeze-thaw cycle numbers and each second multiplied value corresponding to each of the experimental freeze-thaw cycle numbers, and a curve corresponding to the first Gaussian process regression relationship; The first experimental freeze-thaw cycle number corresponding to the coordinate point where the first distance is less than or equal to the first threshold is determined as the first sample freeze-thaw cycle number, and the first experimental impact force platform value contained in the coordinate point where the first distance is less than or equal to the first threshold is determined as the sample impact force platform value.

4. The method according to claim 1 or 3, characterized in that Determining the number of freeze-thaw cycles and the impact force platform value of the first sample based on the first Gaussian process regression relationship includes: In a Cartesian coordinate system, random sampling is performed on the curve corresponding to the first Gaussian process regression relationship to obtain a plurality of sampling coordinate points; Determining the number of freeze-thaw cycles of the first sample based on the abscissa value of each sampling coordinate point and a first proportional coefficient; wherein the first proportional coefficient is the same as a preset proportional coefficient of the experimental coordinate point closest to the sampling coordinate point; The ordinate value of each sampling coordinate point is determined as the sample impact force platform value.

5. The method according to claim 2, characterized in that The determining, based on the first impact force platform value and the sample impact force platform value, a model loss value of the first convolutional neural network model includes: Determining a mean square error loss value of the first convolutional neural network model based on the first impact force platform value and the sample impact force platform value; Determining a downward trend penalty loss value of the first convolutional neural network model based on the first impact force platform value; Based on the mean square error loss value and the downward trend penalty loss value, a model loss value of the first convolutional neural network model is determined.

6. The method according to claim 5, characterized in that The determining, based on the first impact force platform value, a downward trend penalty loss value of the first convolutional neural network model includes: When the first impact force platform value obtained at the first moment is less than or equal to the first impact force platform value obtained at the second moment, determining the downward trend penalty loss value of the first convolutional neural network model at the first moment to be 0; wherein the number of freeze-thaw cycles of the first sample corresponding to the first moment is greater than the number of freeze-thaw cycles of the first sample corresponding to the second moment; When the first impact force platform value obtained at the first moment is greater than the first impact force platform value obtained at the second moment, the difference between the first impact force platform value obtained at the first moment and the first impact force platform value obtained at the second moment is determined as the downward trend penalty loss value of the first convolutional neural network model at the first moment.

7. The method according to claim 5, characterized in that The determining, based on the mean square error loss value and the downward trend penalty loss value, a model loss value of the first convolutional neural network model includes: Determine a penalty coefficient for the downward trend penalty loss value; Based on the penalty coefficient, the downward trend penalty loss value and the mean square error loss value, a model loss value of the first convolutional neural network is determined.

8. A method for predicting the impact resistance of reinforced concrete beams, characterized in that: The method comprises: Obtaining a first freeze-thaw cycle number and a first proportional coefficient of a target reinforced concrete beam; the first proportional coefficient includes a shear span ratio or a bending-shear ratio; Inputting the first number of freeze-thaw cycles and the first proportional coefficient into a target impact resistance prediction model to obtain a first impact force platform value output by the target impact resistance prediction model; wherein the target impact resistance prediction model is obtained based on the model training method according to any one of claims 1 to 7; Based on the first impact force platform value, the impact resistance performance of the target reinforced concrete beam is determined.

9. A model training device, characterized in that: The device comprises: A generation module is configured to generate a first Gaussian process regression relationship based on the number of experimental freeze-thaw cycles, a preset proportional coefficient, and an experimental impact force platform value under a freeze-thaw test of reinforced concrete beams; wherein the first Gaussian process regression relationship includes first input data and first output data; the first input data includes the number of experimental freeze-thaw cycles and the preset proportional coefficient; the first output data includes the experimental impact force platform value; and the preset proportional coefficient includes a shear span ratio or a bending-shear ratio; A first determining module is configured to determine the number of freeze-thaw cycles of the first sample and a sample impact force platform value based on the first Gaussian process regression relationship; An input / output module, configured to input the number of freeze-thaw cycles of the first sample and the preset proportional coefficient into a first convolutional neural network model to obtain a first impact force platform value output by the first convolutional neural network model; a second determining module, configured to determine a model loss value of the first convolutional neural network model based on the first impact force platform value and the sample impact force platform value; An adjustment module is used to adjust the model parameters of the first convolutional neural network model and the hyperparameters in the first Gaussian process regression relationship based on the model loss value to obtain a target impact resistance prediction model; the target impact resistance prediction model is used to determine the target impact force platform value of the target reinforced concrete beam based on the first freeze-thaw cycle number and the first proportional coefficient of the target reinforced concrete beam.

10. A device for predicting the impact resistance of reinforced concrete beams, characterized in that: The device comprises: An acquisition module is used to obtain the current first freeze-thaw cycle number and first proportional coefficient of the target reinforced concrete beam; the first proportional coefficient includes a shear span ratio or a bending-shear ratio; an input / output module, configured to input the first number of freeze-thaw cycles and the first proportional coefficient into a target impact resistance prediction model to obtain a first impact force platform value output by the target impact resistance prediction model; wherein the target impact resistance prediction model is obtained based on the model training method according to any one of claims 1 to 7; A determination module is used to determine the impact resistance of the target reinforced concrete beam based on the first impact force platform value.

Images