Method, system and terminal for predicting performance of FRP confined concrete material based on summary statistics and transfer learning

Abstract

The invention discloses an FRP confined concrete material performance prediction method and system based on summary statistics and transfer learning and a terminal, and provides a transfer learning and summary statistics method to solve the problem of scarcity of data at present. The method comprises the following steps of: acquiring a plurality of experimental data to help a model to learn and improve the performance more quickly on a target domain, aggregating data from different sources or results of a plurality of experiments to reduce accidental errors of a single data point and obtain a more stable statistical index, summarizing a statistical principle, and utilizing a fixed number of output layer neurons to express a widely changing material performance rule, so as to improve the performance of the model. A Dropout method is combined to express random features on the basis of a material performance average value, a summary statistics-transfer learning algorithm can effectively utilize limited data resources, and complete statistical features of material performance are learned through more robust statistical features, so that the generalization ability and prediction precision of the model under the condition of data scarcity are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of material performance prediction, and in particular to a method, system, terminal and computer-readable storage medium for predicting FRP-confined concrete material performance based on summary statistics and transfer learning. Background Art

[0002] Currently, the mainstream research methods for analyzing composite material performance can be divided into three categories: experimental analysis, numerical models and data-driven methods.

[0003] The experimental analysis method mainly uses the test instrument to analyze the composition and microstructure information of the composite material to give the material properties of the composite material. Due to the high cost and time consumption of composite material testing and the diversity of composite material ratios, the experimental analysis method is difficult to adapt to the research needs of a large number of samples or different working conditions.

[0004] The principle of numerical models in predicting composite material performance is to discretize the geometric structure of the material into small units, establish equilibrium equations and material constitutive relations, and simulate the mechanical response of the material under different loads and environmental conditions. When data is scarce, the prediction accuracy and reliability of numerical models are limited because the model requires accurate material parameters and experimental data for calibration. Insufficient data will make it difficult for the model to capture the actual heterogeneity, microstructural characteristics and damage evolution behavior of the material, affecting the applicability of the model and the credibility of the results.

[0005] The principle of data-driven methods for predicting composite material properties is to use machine learning algorithms (such as neural networks, support vector machines, etc.) to train with a large amount of experimental data, automatically learn the complex nonlinear relationships of materials, and thus predict their mechanical responses under different conditions. Insufficient data limits the application of existing data-driven methods in three aspects: (1) It is difficult for the model to accurately learn the microstructural characteristics and mechanical behavior of the material, resulting in overfitting or underfitting, reducing the reliability and generalization ability of the prediction; (2) The fixed number of neurons in the output layer under the traditional neural network architecture is difficult to express the wide range of stress and strain of composite materials under different ratios; (3) It can only predict the average value of the range of material performance changes and cannot predict its random distribution.

[0006] Therefore, the prior art still needs to be improved and developed. Summary of the invention

[0007] The main purpose of the present invention is to provide a method, system, terminal and computer-readable storage medium for predicting the performance of FRP-confined concrete materials based on summary statistics and transfer learning, aiming to solve the problems of data scarcity and low prediction accuracy when analyzing the durability of composite materials in the prior art.

[0008] To achieve the above object, the present invention provides a method for predicting the performance of FRP-confined concrete materials based on summary statistics and transfer learning. The method for predicting the performance of FRP-confined concrete materials based on summary statistics and transfer learning comprises the following steps: Collecting a source domain data set and a target domain data set, wherein the source domain data set and the target domain data set both include multiple features and labels; Selecting a source domain deep neural network model according to the source domain, selecting a target domain deep neural network model according to the target domain, and determining parameters of the source domain deep neural network model and the target domain deep neural network model; Copying neuron signals of a lower neural perception layer in the source domain deep neural network model to the target domain deep neural network model, and keeping them unchanged during the training process of the target domain deep neural network model; Determine a performance standard for evaluating the source domain deep neural network model and the target domain deep neural network model, and select mean square error as a performance evaluation indicator for the source domain deep neural network model and the target domain deep neural network model; Determine an optimization algorithm and hyperparameters of the source domain deep neural network model, train the source domain deep neural network model using the source domain data set, and migrate the parameters of the trained source domain deep neural network model to the target domain deep neural network model; The validation data were clustered into several groups according to geometric and material characteristics, and the constitutive laws of FRP composite confined concrete predicted by the model were compared with the experimental results. The FRP composite material to be predicted is obtained, the performance of the predicted FRP composite material is predicted by using the target domain deep neural network model, and the FRP material performance prediction result is output.

[0009] Optionally, in the FRP-confined concrete material performance prediction method based on summary statistics and transfer learning, the first layer of the source domain deep neural network model and the target domain deep neural network model is an input layer, the number of neurons in the input layer is the same as the number of features, the last layer is an output layer, the number of neurons in the output layer is the same as the number of labels, the middle layer is a hidden layer, and the number of neurons in the hidden layer is a hyperparameter.

[0010] Optionally, in the method for predicting FRP-confined concrete material properties based on summary statistics and transfer learning, the mean square error is: ; ; ; in, represents the mean square error of the total prediction, represents the mean square error of the predicted ultimate strain, represents the weight of stress prediction, represents the mean square error of the predicted ultimate stress, represents the ultimate strain of the experiment, represents the predicted ultimate strain, represents the number of selected stress points, Indicates The experimental stress, Indicates Predicted stress.

[0011] Optionally, in the method for predicting the properties of FRP-confined concrete materials based on summary statistics and transfer learning, the optimization algorithm of the source domain deep neural network model uses a gradient descent algorithm; the hyperparameters of the source domain deep neural network model are determined by cross-validation, and the hyperparameters of the source domain deep neural network model include the number of hidden layers of the neural network, the number of neurons in each hidden layer, and the learning rate.

[0012] Optionally, the method for predicting FRP-confined concrete material properties based on summary statistics and transfer learning, wherein the step of using the source domain data set to train the source domain deep neural network model specifically includes: Dividing the source domain data set into a training set and a validation set according to a preset ratio; In each iteration of the cross-validation process, the training set is used to train the source domain deep neural network model, and the validation set is used to validate the source domain deep neural network model; The loss value is calculated according to the loss function, and when the loss value is minimized, the trained source domain deep neural network model is obtained.

[0013] Optionally, in the method for predicting the performance of FRP-confined concrete materials based on summary statistics and transfer learning, the geometric features include the width, height, length, corner radius of the cross section and FRP thickness of the concrete specimen; The material characteristics include the ultimate strength of concrete, the ultimate strain of concrete, the fracture strain of FRP, the elastic modulus of FRP and the elastic modulus of FRP.

[0014] Optionally, in the method for predicting FRP-confined concrete material properties based on summary statistics and transfer learning, the source domain deep neural network model and the target domain deep neural network model have the same number of neuron layers, the number of neurons in each layer, the activation function of the nonlinear layer, and the dropout rate of the dropout layer.

[0015] In addition, to achieve the above-mentioned purpose, the present invention also provides a FRP-confined concrete material performance prediction system based on summary statistics and transfer learning, wherein the FRP-confined concrete material performance prediction system based on summary statistics and transfer learning includes: A data collection module, used to collect a source domain data set and a target domain data set, wherein the source domain data set and the target domain data set both include multiple features and labels; A model selection module, used to select a source domain deep neural network model according to the source domain, select a target domain deep neural network model according to the target domain, and determine parameters of the source domain deep neural network model and the target domain deep neural network model; A transfer learning module, used to copy the neuron signals of the lower neural perception layer in the source domain deep neural network model to the target domain deep neural network model, and keep them unchanged during the training process of the target domain deep neural network model; A model evaluation module, used to determine a performance standard for evaluating the source domain deep neural network model and the target domain deep neural network model, and select mean square error as a performance evaluation indicator for the source domain deep neural network model and the target domain deep neural network model; A model training module, used to determine the optimization algorithm and hyperparameters of the source domain deep neural network model, and use the source domain data set to train the source domain deep neural network model, and migrate the parameters of the trained source domain deep neural network model to the target domain deep neural network model; Effect comparison module, which is used to cluster the validation data into several groups according to geometric and material characteristics, and compare the constitutive law of FRP composite confined concrete predicted by the model with the test results; The performance prediction module is used to obtain the FRP composite material to be predicted, perform performance prediction on the predicted FRP composite material through the target domain deep neural network model, and output the FRP material performance prediction result.

[0016] In addition, to achieve the above-mentioned purpose, the present invention also provides a terminal, wherein the terminal includes: a memory, a processor, and a FRP-constrained concrete material performance prediction program based on summary statistics and transfer learning stored in the memory and executable on the processor, wherein the FRP-constrained concrete material performance prediction program based on summary statistics and transfer learning implements the steps of the FRP-constrained concrete material performance prediction method based on summary statistics and transfer learning as described above when executed by the processor.

[0017] In addition, to achieve the above-mentioned purpose, the present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a FRP-constrained concrete material performance prediction program based on summary statistics and transfer learning, and when the FRP-constrained concrete material performance prediction program based on summary statistics and transfer learning is executed by a processor, the steps of the FRP-constrained concrete material performance prediction method based on summary statistics and transfer learning as described above are implemented.

[0018] In the present invention, a source domain data set and a target domain data set are collected, and the source domain data set and the target domain data set both include multiple features and labels; a source domain deep neural network model is selected according to the source domain, and a target domain deep neural network model is selected according to the target domain, and parameters of the source domain deep neural network model and the target domain deep neural network model are determined; neuron signals of a low-level neural perception layer in the source domain deep neural network model are copied to the target domain deep neural network model, and remain unchanged during the training process of the target domain deep neural network model; performance criteria for evaluating the source domain deep neural network model and the target domain deep neural network model are determined, and mean square error is selected. The method comprises the following steps: obtaining a performance evaluation index of the source domain deep neural network model and the target domain deep neural network model; determining an optimization algorithm and hyperparameters of the source domain deep neural network model, training the source domain deep neural network model using the source domain data set, and migrating the parameters of the trained source domain deep neural network model to the target domain deep neural network model; clustering the verification data into several groups according to geometric features and material features, and comparing the constitutive law of FRP composite material confined concrete predicted by the model with the test results; obtaining the FRP composite material to be predicted, predicting the performance of the predicted FRP composite material through the target domain deep neural network model, and outputting the FRP material performance prediction result. The present invention proposes a method of transfer learning and summary statistics to solve the current problem of data scarcity. Transfer learning can enable the model to obtain good training results under limited data conditions. Transfer learning helps the model learn faster and improve performance in the target domain by transferring knowledge learned in the source domain to the target domain. It reduces the accidental errors of a single data point by aggregating data from different sources or the results of multiple experiments, and obtains more stable statistical indicators. The summary statistics principle uses a fixed number of output layer neurons to express widely varying material performance laws, and combines the Dropout method to express random characteristics based on the average value of material performance. The summary statistics-transfer learning algorithm can effectively utilize limited data resources, so that the model does not rely on individual data points and learns the complete statistical characteristics of material performance through more robust statistical characteristics, thereby improving the generalization ability and prediction accuracy of the model under data scarcity conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1It is a structural schematic diagram of transfer learning in a preferred embodiment of the FRP-confined concrete material performance prediction method based on summary statistics and transfer learning of the present invention; Figure 2 It is a schematic diagram of a feature representation module in a preferred embodiment of the method for predicting the performance of FRP-confined concrete materials based on summary statistics and transfer learning of the present invention; Figure 3 It is a schematic diagram of a target representation module in a preferred embodiment of the method for predicting the performance of FRP-confined concrete materials based on summary statistics and transfer learning of the present invention; Figure 4 It is a schematic diagram of a pre-training module in a preferred embodiment of the method for predicting the performance of FRP-confined concrete materials based on summary statistics and transfer learning of the present invention; Figure 5 It is a schematic diagram of a transfer learning module in a preferred embodiment of the method for predicting the performance of FRP-confined concrete materials based on summary statistics and transfer learning of the present invention; Figure 6 It is a flow chart of a preferred embodiment of the method for predicting the performance of FRP-confined concrete materials based on summary statistics and transfer learning of the present invention; Figure 7 It is a structural diagram of a preferred embodiment of the FRP-confined concrete material performance prediction system based on summary statistics and transfer learning of the present invention; Figure 8 It is a structural diagram of a preferred embodiment of the terminal of the present invention. DETAILED DESCRIPTION

[0020] In order to make the purpose, technical solution and advantages of the present invention clearer and more specific, the present invention is further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0021] like Figure 1 As shown, it represents the structure diagram of the transfer learning model. The method of deep learning network based on summary statistics and transfer learning described in the present invention consists of four modules: feature representation module, target representation module, pre-training module and transfer learning module.

[0022] like Figure 2 As shown in the figure, the feature representation module: the performance of building materials is usually determined according to specific test criteria, and the input characteristics of the specimen can be represented by geometric characteristics and material characteristics. Therefore, the specimen of FRP (Fiber Reinforced Polymer, fiber reinforced composite material) composite material confined concrete is described from the two aspects of geometric shape and material characteristics, and adaptive design is performed on it.

[0023] like Figure 3As shown, the target representation module: the constitutive law of FRP composite material confined concrete is summarized and statistically processed to extract sufficient information. The present invention uses a summary statistical method to extract key information about the constitutive law, including characteristic points ((strength, ultimate strain, etc. when the material is destroyed) and the corresponding relationship between the stress-strain process before the material is destroyed. The test data is divided into several groups according to some characteristics using a clustering method. Each group of data uses a dropout layer in the last hidden layer to randomly inactivate neurons for Monte Carlo simulation to generate 1 ultimate strain and n stresses. Among them, the n stresses correspond to n fixed percentage ultimate strains, for example, when n=10, the 10 stresses correspond to 10%, 20%, 30%, ... 100% ultimate strains, and the maximum value is the ultimate strength. The target representation module completely extracts the mean and standard deviation of the constitutive law of FRP composite material confined concrete, and uses 5 curves to represent the results: one curve corresponds to the mean, two curves correspond to the mean ±1 times the standard deviation, and two curves correspond to the mean ±2 times the standard deviation.

[0024] like Figure 4 As shown, pre-training module (source domain task): The pre-training module adopts the deep neural network (DNN) architecture. The DNN includes an input layer, a hidden layer, and an output layer. Each hidden layer consists of a linear layer and a rectified linear unit (ReLU) activation function, and the last hidden layer contains a dropout layer. The number of neurons in the input layer is the same as the number of features in the dataset, and the number of neurons in the output layer is the same as the number of labels in the dataset. The pre-training module uses a source domain task similar to FRP composite confined concrete (such as unconstrained concrete) for model training, which is used to extract the influence of common features (such as unconstrained concrete strength and ultimate strain) of the source domain task (unconstrained concrete) and the target domain task (FRP composite confined concrete) on the constitutive law.

[0025] like Figure 5 As shown, transfer learning module (target domain task): The purpose of the transfer learning module is to transfer the knowledge obtained from the pre-training module to the target task with a limited dataset size. This knowledge transfer method is as follows: the transfer learning module copies the parameters of the first few hidden layers of the pre-training module.

[0026] The material performance prediction method of FRP-confined concrete based on summary statistics and transfer learning described in the preferred embodiment of the present invention is as follows: Figure 6 As shown, the FRP-confined concrete material performance prediction method based on summary statistics and transfer learning includes the following steps: Step S10: collecting a source domain data set and a target domain data set, wherein the source domain data set and the target domain data set both include multiple features and labels.

[0027] Specifically, a source domain data set and a target domain data set are collected. The source domain data set and the target domain data set contain features and labels. The source domain data set should cover as many features and labels as possible, and the features of the target domain data set should be the same or similar to those of the source domain data set.

[0028] That is, in the publicly published literature, the data of FRP-confined concrete compression test and the data of traditional FRP-confined concrete cylinder compression test are collected as the source domain data set, and the data of traditional FRP-confined concrete non-cylindrical and LRS-FRP concrete column compression test are collected as the target domain data set.

[0029] The characteristics of the specimen (i.e. the specimen to be tested) are divided into geometric characteristics and material characteristics. The geometric characteristics include: the width (b), height (h), length (l), corner radius (r) of the (b×h) section of the concrete specimen, and the FRP thickness t f , material characteristics include: the ultimate strength of concrete (f co ), the ultimate strain of concrete e co 、FRP fracture strain e f , elastic modulus of FRP (E1), elastic modulus of FRP (E2). Finally, the geometric features (5-dimensional vector) and material features (5-dimensional vector) of the specimen are combined into input features (10-dimensional vector).

[0030] The compressive performance prediction of concrete cylinders confined by (carbon fiber and glass fiber) FRP is used as the source task, and the compressive performance prediction of concrete square cylinders confined by (carbon fiber and glass fiber) FRP and large strain FRP (LRS-FRP) concrete cylinders are used as two target tasks. The source domain data set and the target domain data set have 10 features and 21 labels including (20 for stress and 1 for ultimate strain).

[0031] Step S20: select a source domain deep neural network model according to the source domain, select a target domain deep neural network model according to the target domain, and determine parameters of the source domain deep neural network model and the target domain deep neural network model.

[0032] Specifically, a source domain deep neural network model is selected according to the source domain, and a target domain deep neural network model is selected according to the target domain. The number of neuron layers, the number of neurons in each layer, the activation function of the nonlinear layer, and the dropout rate of the dropout layer of the source domain deep neural network model and the target domain deep neural network model are determined. The dropout rate represents the random inactivation of neurons in the dropout layer in the deep neural network model (temporarily not participating in the training process and not being updated), which can be used to calculate the probability distribution of output data in the subsequent Monte Carlo simulation. The first layer of the source domain deep neural network model and the target domain deep neural network model is the input layer, and the number of neurons in the input layer is the same as the number of features (10). The last layer is the output layer, and the number of neurons in the output layer is the same as the number of labels (21). The middle layer is the hidden layer, and the number of neurons in the hidden layer is a hyperparameter, which is determined when the hyperparameter is selected later.

[0033] Step S30: copy the neuron signals of the lower neural perception layer in the source domain deep neural network model to the target domain deep neural network model, and keep them unchanged during the training process of the target domain deep neural network model.

[0034] Specifically, the neuron signals (weights, error functions) of the lower (close to the input layer) neural perception layer in the source domain (unconstrained concrete) deep neural network model are copied to the target domain (FRP composite material confined concrete) deep neural network model and remain unchanged during the training process of the deep neural network model.

[0035] Step S40: determine the performance criteria for evaluating the source domain deep neural network model and the target domain deep neural network model, and select mean square error as the performance evaluation indicator of the source domain deep neural network model and the target domain deep neural network model.

[0036] Specifically, the standard for evaluating the performance of the model is determined. The prediction of the constitutive law of FRP composite material confined concrete involved in the present invention is a regression task, and the mean square error is selected as the performance evaluation index of the model. The mean square error is: ; ; ; in, represents the mean square error of the total prediction, represents the mean square error of the predicted ultimate strain, represents the weight of stress prediction, represents the mean square error of the predicted ultimate stress, represents the ultimate strain of the experiment, represents the predicted ultimate strain, represents the number of selected stress points, Indicates The experimental stress, Indicates Predicted stress.

[0037] Step S50: determine the optimization algorithm and hyperparameters of the source domain deep neural network model, use the source domain data set to train the source domain deep neural network model, and migrate the parameters of the trained source domain deep neural network model to the target domain deep neural network model.

[0038] Specifically, the optimization algorithm of the source domain deep neural network model uses a gradient descent algorithm (for example, the present invention uses the Adam algorithm); the hyperparameters of the source domain deep neural network model include the number of layers of the neural network hidden layer, the number of neurons in each hidden layer, and the learning rate. The number of layers of the hidden layer of the source domain deep neural network model, the number of neurons in each hidden layer, the learning rate and other hyperparameters are determined by cross-validation. The cross-validation method first searches the applicable range of the hyperparameters through a grid. Five-fold CV (dividing the data into five parts, four parts of the data are used for training, and one part is folded (leaving one part for verification model)) divides the training set into five subsets of equal size. In each iteration of the cross-validation process, one of the folds is used for verification, and the remaining four folds are used as training sets. The model is trained on the training set, and then the model is evaluated on the verification set to calculate the error. This process is repeated five times, and each fold is used as a verification set only once. The final performance indicator is the average of the indicators obtained in each iteration.

[0039] The source domain data set is divided into a training set and a validation set according to a preset ratio; in each iteration of the cross-validation process, the training set is used to train the source domain deep neural network model, and the validation set is used to validate the source domain deep neural network model; the loss value is calculated according to the loss function, and when the loss value is minimized, the trained source domain deep neural network model is obtained. The specific model training process of the source domain deep neural network model is: 1. Initialize the model; 2. Prepare an iterable data set; 3. Determine the hyperparameters; 4. Put the data into the model operation, and obtain the loss value according to the loss function (evaluation indicator MSE function); 5. Update the model parameters according to the loss value and the model optimization algorithm; 6. Repeat 4 and 5 until the loss value is minimized or the loss value reaches the expected requirements; 7. Save the model and the training is completed.

[0040] After that, the parameters of the trained source domain deep neural network model are directly transferred to the target domain deep neural network model. Transfer learning is based on the hierarchical structure and parameter sharing characteristics of deep neural networks, and realizes knowledge transfer by extracting common features and specific features of data layer by layer. The underlying features are universal and weakly related to specific tasks, so they are suitable for direct transfer, while high-level features are closely related to tasks and need to be adjusted according to the target task. The transfer learning implementation method usually includes three steps: first, pre-train the neural network in the source task and learn a set of common parameters (such as weights and biases); then transfer the underlying features to the target task to reuse existing knowledge; finally, fine-tune the high-level features or retrain the top-level model according to the requirements of the target task. This method can effectively reduce the data requirements and training costs of the target task, while improving the performance and generalization ability of the model.

[0041] Step S60: cluster the verification data into several groups according to the geometric characteristics and material characteristics, and compare the constitutive law of FRP composite material confined concrete predicted by the model with the test results.

[0042] Specifically, the source model determined by the training hyperparameters is verified, the verification data is clustered into several groups according to the geometric features and material features, the constitutive law (stress-strain curve) of FRP composite material confined concrete predicted by the model is compared with the test results, the constitutive law of FRP composite material is summarized and statistically processed, and sufficient information is extracted. The present invention uses a summary statistical method to extract key information about the constitutive law, including characteristic points (strength, ultimate strain, etc.) and xy relationship (x represents stress, y represents strain), and the test data is clustered (summary statistics) according to some characteristics and divided into two groups for verification, including direct comparison of calculated numerical evaluation indicators and drawings. The prediction results include five curves: one curve corresponds to the mean, two curves correspond to the mean ±1 times the standard deviation, and two curves correspond to the mean ±2 times the standard deviation, which basically cover the distribution range of the test results.

[0043] Step S70: obtaining the FRP composite material to be predicted, performing performance prediction on the predicted FRP composite material through the target domain deep neural network model, and outputting the FRP material performance prediction result.

[0044] The present invention proposes a method of transfer learning and summary statistics to solve the current problem of data scarcity. Transfer learning can enable the model to obtain good training results under limited data conditions. The technical principle of transfer learning is mainly based on the concepts of source domain and target domain, where the source domain refers to a known domain with rich data, and the target domain is a domain to be learned with relatively less data. Transfer learning helps the model learn faster and improve performance in the target domain by transferring the knowledge learned in the source domain (such as features and model parameters) to the target domain. By aggregating data from different sources or the results of multiple experiments, the accidental errors of a single data point are reduced, and more stable statistical indicators (such as mean value, variance, etc.) are obtained. The principle of summary statistics uses a fixed number of output layer neurons to express the widely varying laws of material properties, and combines the Dropout method to express random characteristics based on the average value of material properties. The summary statistics-transfer learning algorithm can effectively utilize limited data resources, so that the model does not rely on individual data points and learns the complete statistical characteristics of material properties through more robust statistical characteristics, thereby improving the generalization ability and prediction accuracy of the model under data scarcity conditions.

[0045] The present invention transfers the common features of source domain tasks and target tasks from the source domain task model to the numerical mapping relationship of the constitutive law of FRP composite material constrained concrete, greatly reducing the data required for training the target task model, so that the target task model can converge quickly on a small data set and obtain a good generalization effect.

[0046] The present invention proposes a summary statistical method, which uses a limited number of output layer neurons to fully express the full information of the constitutive law with a wide range of variations, that is, to numerically express the probability distribution of the corresponding relationship between the ultimate stress, the ultimate strain and the stress-strain before the failure state in the form of mean value and standard deviation.

[0047] Furthermore, the present invention currently only aims at the prediction of the constitutive curve of FRP composite material confined concrete, but its technical background and technical problems existing in related technologies are widely present in other composite materials. In view of the strong generalization ability of the technical solution proposed in the present invention, the data of other composite materials can be used to predict the corresponding material properties without changing the model architecture.

[0048] Furthermore, if Figure 7 As shown, based on the above-mentioned FRP-confined concrete material performance prediction method based on summary statistics and transfer learning, the present invention also provides a FRP-confined concrete material performance prediction system based on summary statistics and transfer learning, wherein the FRP-confined concrete material performance prediction system based on summary statistics and transfer learning includes: A data collection module 51 is used to collect a source domain data set and a target domain data set, wherein the source domain data set and the target domain data set both include multiple features and labels; A model selection module 52, configured to select a source domain deep neural network model according to a source domain, select a target domain deep neural network model according to a target domain, and determine parameters of the source domain deep neural network model and the target domain deep neural network model; A transfer learning module 53, used to copy the neuron signals of the lower neural perception layer in the source domain deep neural network model to the target domain deep neural network model, and keep them unchanged during the training process of the target domain deep neural network model; A model evaluation module 54 is used to determine a performance standard for evaluating the source domain deep neural network model and the target domain deep neural network model, and select mean square error as a performance evaluation indicator for the source domain deep neural network model and the target domain deep neural network model; A model training module 55 is used to determine the optimization algorithm and hyperparameters of the source domain deep neural network model, and use the source domain data set to train the source domain deep neural network model, and migrate the parameters of the trained source domain deep neural network model to the target domain deep neural network model; The effect comparison module 56 is used to cluster the verification data into several groups according to the geometric characteristics and material characteristics, and compare the constitutive law of FRP composite material confined concrete predicted by the model with the test results; The performance prediction module 57 is used to obtain the FRP composite material to be predicted, perform performance prediction on the predicted FRP composite material through the target domain deep neural network model, and output the FRP material performance prediction result.

[0049] Furthermore, if Figure 8 As shown, based on the above-mentioned method and system for predicting the performance of FRP-confined concrete materials based on summary statistics and transfer learning, the present invention also provides a terminal accordingly, and the terminal includes a processor 10, a memory 20 and a display 30. Figure 8 Only some components of the terminal are shown, but it should be understood that it is not required to implement all of the components shown, and more or fewer components may be implemented instead.

[0050] In some embodiments, the memory 20 may be an internal storage unit of the terminal, such as a hard disk or memory of the terminal. In other embodiments, the memory 20 may also be an external storage device of the terminal, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the terminal. Further, the memory 20 may also include both an internal storage unit of the terminal and an external storage device. The memory 20 is used to store application software and various types of data installed on the terminal, such as the program code of the installation terminal. The memory 20 may also be used to temporarily store data that has been output or is to be output. In one embodiment, the memory 20 stores a FRP-confined concrete material performance prediction program 40 based on summary statistics and transfer learning, which can be executed by the processor 10, thereby realizing the FRP-confined concrete material performance prediction method based on summary statistics and transfer learning in this application.

[0051] In some embodiments, the processor 10 may be a central processing unit (CPU), a microprocessor or other data processing chip, used to run the program code or process data stored in the memory 20, such as executing the FRP confined concrete material performance prediction method based on summary statistics and transfer learning.

[0052] In some embodiments, the display 30 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, an OLED (Organic Light-Emitting Diode) touch device, etc. The display 30 is used to display information on the terminal and to display a visual user interface. The processor 10, the memory 20, and the display 30 of the terminal communicate with each other via a system bus.

[0053] In one embodiment, when the processor 10 executes the FRP confined concrete material performance prediction program 40 based on summary statistics and transfer learning in the memory 20, the steps of the FRP confined concrete material performance prediction method based on summary statistics and transfer learning are implemented.

[0054] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a program for predicting the performance of FRP-constrained concrete materials based on summary statistics and transfer learning, and when the program for predicting the performance of FRP-constrained concrete materials based on summary statistics and transfer learning is executed by a processor, the steps of the method for predicting the performance of FRP-constrained concrete materials based on summary statistics and transfer learning as described above are implemented.

[0055] In summary, the present invention provides a method, system, terminal and computer-readable storage medium for predicting the performance of FRP-confined concrete materials based on summary statistics and transfer learning. The method comprises: collecting a source domain data set and a target domain data set, wherein the source domain data set and the target domain data set both include multiple features and labels; selecting a source domain deep neural network model according to the source domain, selecting a target domain deep neural network model according to the target domain, and determining parameters of the source domain deep neural network model and the target domain deep neural network model; copying the neuron signal of the lower neural perception layer in the source domain deep neural network model to the target domain deep neural network model, and keeping it unchanged during the training process of the target domain deep neural network model; determining a method for evaluating the source domain deep neural network model. The performance criteria of the source domain deep neural network model and the target domain deep neural network model are selected, and the mean square error is selected as the performance evaluation index of the source domain deep neural network model and the target domain deep neural network model; the optimization algorithm and hyperparameters of the source domain deep neural network model are determined, and the source domain deep neural network model is trained using the source domain data set, and the parameters of the trained source domain deep neural network model are transferred to the target domain deep neural network model; the verification data are clustered into several groups according to geometric features and material features, and the constitutive law of FRP composite material confined concrete predicted by the model is compared with the test results; the FRP composite material to be predicted is obtained, the performance of the predicted FRP composite material is predicted by the target domain deep neural network model, and the FRP material performance prediction result is output. The present invention proposes a method of transfer learning and summary statistics to solve the current problem of data scarcity. Transfer learning can enable the model to obtain good training results under limited data conditions. Transfer learning helps the model learn faster and improve performance in the target domain by transferring knowledge learned in the source domain to the target domain. It reduces the accidental errors of a single data point by aggregating data from different sources or the results of multiple experiments, and obtains more stable statistical indicators. The summary statistics principle uses a fixed number of output layer neurons to express widely varying material performance laws, and combines the Dropout method to express random characteristics based on the average value of material performance. The summary statistics-transfer learning algorithm can effectively utilize limited data resources, so that the model does not rely on individual data points and learns the complete statistical characteristics of material performance through more robust statistical characteristics, thereby improving the generalization ability and prediction accuracy of the model under data scarcity conditions.

[0056] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or terminal including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or terminal. In the absence of further restrictions, an element defined by the sentence "includes a ..." does not exclude the existence of other identical elements in the process, method, article or terminal including the element.

[0057] Of course, those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing related hardware (such as a processor, a controller, etc.) through a computer program, and the program can be stored in a computer-readable storage medium that can be read by a computer, and the program can include the processes of the above-mentioned method embodiments when executed. The computer-readable storage medium can be a memory, a disk, an optical disk, etc.

[0058] It should be understood that the application of the present invention is not limited to the above examples. For ordinary technicians in this field, improvements or changes can be made based on the above description. All these improvements and changes should fall within the scope of protection of the claims attached to the present invention.

Claims

1. A method for predicting the performance of FRP-confined concrete materials based on summary statistics and transfer learning, characterized in that: The FRP-confined concrete material performance prediction method based on summary statistics and transfer learning includes: Collecting a source domain data set and a target domain data set, wherein the source domain data set and the target domain data set both include multiple features and labels; Selecting a source domain deep neural network model according to the source domain, selecting a target domain deep neural network model according to the target domain, and determining parameters of the source domain deep neural network model and the target domain deep neural network model; Copying neuron signals of a lower neural perception layer in the source domain deep neural network model to the target domain deep neural network model, and keeping them unchanged during the training process of the target domain deep neural network model; Determine a performance standard for evaluating the source domain deep neural network model and the target domain deep neural network model, and select mean square error as a performance evaluation indicator for the source domain deep neural network model and the target domain deep neural network model; Determine an optimization algorithm and hyperparameters of the source domain deep neural network model, train the source domain deep neural network model using the source domain data set, and migrate the parameters of the trained source domain deep neural network model to the target domain deep neural network model; The validation data were clustered into several groups according to geometric and material characteristics, and the constitutive laws of FRP composite confined concrete predicted by the model were compared with the experimental results. The FRP composite material to be predicted is obtained, the performance of the predicted FRP composite material is predicted by using the target domain deep neural network model, and the FRP material performance prediction result is output.

2. The FRP-confined concrete material performance prediction method based on summary statistics and transfer learning according to claim 1 is characterized in that: The first layer of the source domain deep neural network model and the target domain deep neural network model is the input layer, the number of neurons in the input layer is the same as the number of features, the last layer is the output layer, the number of neurons in the output layer is the same as the number of labels, the middle layer is the hidden layer, and the number of neurons in the hidden layer is a hyperparameter.

3. The FRP-confined concrete material performance prediction method based on summary statistics and transfer learning according to claim 1 is characterized in that: The mean square error is: ； ； ； in, represents the mean square error of the total prediction, represents the mean square error of the predicted ultimate strain, represents the weight of stress prediction, represents the mean square error of the predicted ultimate stress, represents the ultimate strain of the experiment, represents the predicted ultimate strain, represents the number of selected stress points, Indicates The experimental stress, Indicates Predicted stress.

4. The method for predicting FRP-confined concrete material properties based on summary statistics and transfer learning according to claim 1 is characterized in that: The optimization algorithm of the source domain deep neural network model uses a gradient descent algorithm; the hyperparameters of the source domain deep neural network model are determined by cross-validation, and the hyperparameters of the source domain deep neural network model include the number of hidden layers of the neural network, the number of neurons in each hidden layer, and the learning rate.

5. The method for predicting FRP-confined concrete material properties based on summary statistics and transfer learning according to claim 1 is characterized in that: The using the source domain data set to train the source domain deep neural network model specifically includes: Dividing the source domain data set into a training set and a validation set according to a preset ratio; In each iteration of the cross-validation process, the training set is used to train the source domain deep neural network model, and the validation set is used to validate the source domain deep neural network model; The loss value is calculated according to the loss function, and when the loss value is minimized, the trained source domain deep neural network model is obtained.

6. The method for predicting FRP-confined concrete material properties based on summary statistics and transfer learning according to claim 1 is characterized in that: The geometric features include the width, height, length, corner radius of the cross section and FRP thickness of the concrete specimen; The material characteristics include the ultimate strength of concrete, the ultimate strain of concrete, the fracture strain of FRP, the elastic modulus of FRP and the elastic modulus of FRP.

7. The method for predicting FRP-confined concrete material properties based on summary statistics and transfer learning according to claim 1 is characterized in that: The source domain deep neural network model and the target domain deep neural network model have the same number of neuron layers, the number of neurons in each layer, the activation function of the nonlinear layer, and the dropout rate of the dropout layer.

8. A material performance prediction system for FRP-confined concrete based on summary statistics and transfer learning, characterized in that: The FRP-confined concrete material performance prediction system based on summary statistics and transfer learning includes: A data collection module, used to collect a source domain data set and a target domain data set, wherein the source domain data set and the target domain data set both include multiple features and labels; A model selection module, used to select a source domain deep neural network model according to the source domain, select a target domain deep neural network model according to the target domain, and determine parameters of the source domain deep neural network model and the target domain deep neural network model; A transfer learning module, used to copy the neuron signals of the lower neural perception layer in the source domain deep neural network model to the target domain deep neural network model, and keep them unchanged during the training process of the target domain deep neural network model; A model evaluation module, used to determine a performance standard for evaluating the source domain deep neural network model and the target domain deep neural network model, and select mean square error as a performance evaluation indicator for the source domain deep neural network model and the target domain deep neural network model; A model training module, used to determine the optimization algorithm and hyperparameters of the source domain deep neural network model, and use the source domain data set to train the source domain deep neural network model, and migrate the parameters of the trained source domain deep neural network model to the target domain deep neural network model; Effect comparison module, which is used to cluster the validation data into several groups according to geometric and material characteristics, and compare the constitutive law of FRP composite confined concrete predicted by the model with the test results; The performance prediction module is used to obtain the FRP composite material to be predicted, perform performance prediction on the predicted FRP composite material through the target domain deep neural network model, and output the FRP material performance prediction result.

9. A terminal, characterized in that: The terminal includes: a memory, a processor, and a FRP-confined concrete material performance prediction program based on summary statistics and transfer learning stored in the memory and executable on the processor. When the FRP-confined concrete material performance prediction program based on summary statistics and transfer learning is executed by the processor, the steps of the FRP-confined concrete material performance prediction method based on summary statistics and transfer learning as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a program for predicting the performance of FRP-confined concrete materials based on summary statistics and transfer learning. When the program for predicting the performance of FRP-confined concrete materials based on summary statistics and transfer learning is executed by a processor, the steps of the method for predicting the performance of FRP-confined concrete materials based on summary statistics and transfer learning as described in any one of claims 1 to 7 are implemented.

Images