Large prefabricated T-beam strength prediction model, method and system and readable medium

Abstract

The invention relates to the technical field of production management and machine learning, in particular to a large prefabricated T-beam strength prediction model, method and system and a readable medium. According to the construction method of the large prefabricated T-beam strength prediction model, the strength prediction model is constructed based on the deep neural network, the weight of the strength prediction model is optimized in combination with the improved butterfly algorithm, and the complex relation between the concrete strength and the production line construction parameters is accurately captured; and a more reliable prediction result is provided for evaluation of the concrete structure. By applying the improved butterfly algorithm to optimize the deep neural network, a prefabrication enterprise can optimize a production plan, improve the production efficiency, reduce the cost and ensure the quality and safety of the prefabricated T-beam. According to the method, by optimizing the local search strategy of the butterfly algorithm, the convergence speed and precision of population iteration are remarkably improved, and meanwhile, the risk of falling into a local optimal solution is reduced.

Description

Technical Field

[0001] The present invention relates to the field of production management and machine learning technology, and in particular to a large prefabricated T-beam strength prediction model, method, system and readable medium. Background Art

[0002] Quality control is a core issue in the production and application of large precast T-beams, which is directly related to the structural safety and overall performance of bridges. As the core indicator of component quality, the stability and uniformity of concrete strength are crucial to the structural safety of precast T-beams. In concrete quality control, strength prediction is a core and challenging link. Traditional strength prediction methods rely on laboratory tests and statistical analysis, which are not only time-consuming and labor-intensive, but also easily limited by experimental conditions and sample quantity. Summary of the invention

[0003] In order to overcome the defects of the above-mentioned prior art in the production process of large prefabricated T-beams, that is, the concrete strength prediction is time-consuming and labor-intensive, and is easily limited by experimental conditions and sample quantity, the present invention proposes a method for constructing a large prefabricated T-beam strength prediction model, which greatly improves the model convergence speed and T-beam strength prediction accuracy.

[0004] The present invention proposes a method for constructing a large prefabricated T-beam strength prediction model, comprising:

[0005] S1. Sampling construction parameter data in combination with the T-beam process, constructing a learning sample {D', y}, where D' is a preprocessing sequence of construction parameters within a unit time, and y is the strength of the T-beam produced within the time, and dividing the learning sample into a training sample and a verification sample; constructing a strength prediction model based on a neural network model, and the strength prediction model predicts the strength of the T-beam produced at the corresponding time based on the data sequence D';

[0006] S2, initializing the strength prediction model and the butterfly algorithm population, the solution vector of each butterfly is the weight matrix and bias vector set of the strength prediction model;

[0007] S3, let the strength prediction model loaded with the solution vector of each butterfly perform machine learning on the training sample, then select the verification sample to substitute into the strength prediction model, obtain the strength prediction value output by the strength prediction model, and calculate the mean square error loss of the strength prediction model as the mean square error loss corresponding to the butterfly;

[0008] S4, select a random number r, r is between 0 and 1; determine whether r≤P;

[0009] If r≤P, use the global search algorithm to iterate the butterfly population; if r>P, use the following formula to iterate the butterfly population;

[0010]

[0011] in, represents the solution vector of the zth butterfly in population c after the t+1th iteration, represents the solution vector of the zth butterfly in population c after the tth iteration. The initial value of t is 0, and δ t+1 represents the step size factor of the t+1th search, Indicates the direction of exploration. represents the right search area of ​​the zth butterfly in population c after the tth iteration, represents the left search area of ​​the zth butterfly in population c after the tth iteration; Fit(.) represents the fitness function of minimizing the objective, and sign represents the sign function;

[0012] S5. Determine whether the number of iterations of the butterfly population reaches the set value; if not, return to step S3; if yes, output the optimal individual of the optimal population according to MSE as the final solution vector, assign the final solution vector to the intensity prediction model and fix the model.

[0013] Preferred: δ t+1 =c1δ t +δ 0

[0014] Among them, δ t Indicates the step size factor of the t-th search; δ 0 represents the set initial step size, and c1 is the set constant.

[0015] Preferably, let d t represents the search distance of the butterfly in the tth iteration, d t =2R, R is the transition value, we have:

[0016] d t =δ t / c2

[0017]

[0018] Where c2 is a set constant, and [1] indicates a matrix whose elements are all 1.

[0019] Preferred:

[0020] Among them, rand represents a random value, rand(.) is a random function, rand(k', 1) represents generating a random k'×1-dimensional vector; ||.|| represents the bi-norm.

[0021] Preferably, the method for obtaining the construction parameter preprocessing sequence D' is: collecting the construction parameter data sequence in the T-beam production process, first performing missing value processing and outlier processing on the construction parameter data sequence, and then performing data normalization processing to obtain the construction parameter preprocessing sequence D'; the construction parameter data includes multiple construction parameters.

[0022] Preferably, the missing values ​​of the data are processed as follows: when a construction parameter at a certain time point is missing, the mean value of the construction parameter at the remaining time points within the unit time is taken to supplement the time point.

[0023] Preferably, the outlier processing method is: performing outlier judgment on each construction parameter in the construction parameter data sequence, and if a construction parameter is judged to be abnormal, deleting the construction parameter data containing the construction parameter from the construction parameter data sequence.

[0024] The present invention proposes a large prefabricated T-beam strength prediction method using the large prefabricated T-beam strength prediction model construction method, which is characterized in that the strength prediction model is first constructed using the large prefabricated T-beam strength prediction model construction method; then the construction parameter data sequence in the T-beam production process is collected and preprocessed, the preprocessed data sequence is input into the strength prediction model, and the strength prediction value output by the strength prediction model is obtained.

[0025] A large prefabricated T-beam strength prediction system proposed in the present invention comprises a memory and a processor, wherein a computer program and a strength prediction model constructed by adopting the method for constructing the large prefabricated T-beam strength prediction model are stored in the memory, and the processor is connected to the memory, and the processor is used to execute the computer program to implement the method for predicting the strength of large prefabricated T-beams.

[0026] The present invention provides a readable medium storing a computer program, which is used to implement the strength prediction method of a large prefabricated T-beam when executed.

[0027] The advantages of the present invention are:

[0028] (1) The method for constructing a strength prediction model for large prefabricated T-beams proposed in the present invention constructs a strength prediction model based on a deep neural network, and optimizes the weights of the strength prediction model in combination with an improved butterfly algorithm, accurately capturing the complex relationship between concrete strength and production line construction parameters, and providing more reliable prediction results for the evaluation of concrete structures. By applying the improved butterfly algorithm to optimize the deep neural network, prefabrication companies can optimize production plans, improve production efficiency, reduce costs, and ensure the quality and safety of prefabricated T-beams. In the present invention, by optimizing the local search strategy of the butterfly algorithm, the convergence speed and accuracy of the population iteration are significantly improved, while the risk of falling into a local optimal solution is reduced.

[0029] (2) The present invention proposes a method for predicting concrete strength in the production process of large prefabricated T-beams. The method extracts the key control points of each process on the production line through preprocessing of construction parameter data sequences, and uses prefabrication process data to construct an immediate concrete strength prediction model after steaming based on an improved deep neural network. The model weights are optimized in combination with an improved butterfly algorithm, which accurately captures the complex relationship between concrete strength and related factors, and provides more reliable prediction results for the evaluation of concrete structures.

[0030] (3) In the present invention, the population iteration termination condition can be specifically set according to the needs, which is conducive to taking into account the comprehensive evaluation of different process combinations and improving the calculation efficiency.

[0031] (4) In the present invention, the data is preprocessed, mainly including missing value processing, outlier detection and correction, and dimensionless data, ensuring that the data set provides a solid foundation for subsequent analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 A flow chart of the construction method of the strength prediction model for large prefabricated T-beams;

[0033] Figure 2(a) shows the model prediction strength of BAS-BOA-DNN.

[0034] Figure 2(b) shows the residual of BAS-BOA-DNN;

[0035] Figure 2(c) shows the model-predicted strength of RF;

[0036] Figure 2(d) shows the residual of RF;

[0037] Figure 2(e) shows the model prediction strength of SVR;

[0038] Figure 2(f) shows the residual of SVR;

[0039] Figure 3 is the relative error of the three groups of models;

[0040] Figure 4(a) shows the optimization process of the BOA algorithm;

[0041] Figure 4(b) shows the optimization process of the BAS-BOA algorithm. DETAILED DESCRIPTION

[0042] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0043] Reference Figure 1 The large prefabricated T-beam strength prediction method proposed in this embodiment predicts the strength of the T-beam based on the construction parameter data sequence in the T-beam production process through the trained deep neural network model. The training samples used in the deep neural network model training process are denoted as {D', y}, where D' is the construction parameter preprocessing sequence within a unit time, and y is the strength of the T-beam produced within the time.

[0044] During specific implementation, the construction parameter data include: raw material ratio, construction technology and curing conditions data, etc.; the raw materials include: characteristic fields of water, cement, coarse aggregate, fine aggregate and fly ash, etc., and the construction technology includes steaming time, steaming temperature and steaming humidity, etc.

[0045] The method for obtaining the construction parameter preprocessing sequence D' is as follows: the construction parameter data sequence in the T-beam production process is collected, and the data is preprocessed to obtain D'; the data preprocessing method is to firstly process the missing values ​​and outliers of the construction parameter data, and then perform dimensionless processing on the data, that is, normalization processing.

[0046] The construction parameter data sequence collected within a unit time t is recorded as D. After data preprocessing of D, the data sequence D' is obtained.

[0047] D={d(1),d(2),…,d(n),…,d(N)}

[0048] d(n)={d(n,1),d(n,2),…,d(n,h),…,d(n,H)}

[0049] N is the number of samples in the construction parameter data sequence, d(1), d(2) and d(N) represent the construction parameter data collected at the 1st, 2nd and Nth time points within a unit time t, respectively, and d(n) represents the construction parameter data collected at the nth time point within a unit time t;

[0050] H represents the number of construction parameters collected at a single time point, d(n,h) represents the hth construction parameter in d(n); d(n,1), d(n,2) and d(n,H) represent the 1st, 2nd and Hth construction parameters in d(n), respectively.

[0051] The missing value processing method of the data is: when a construction parameter at a certain time point is missing, the mean value of the construction parameter at the remaining time points in the unit time is taken to supplement the time point. Assuming that the kth construction parameter at the uth time point in the unit time is missing, d(u,k) is assigned according to the following formula:

[0052]

[0053] d(i,k) represents the kth construction parameter at the i-th time point in unit time.

[0054] The outlier processing method is: to judge the outlier value of each construction parameter in the construction parameter data sequence, if a construction parameter is judged to be abnormal, then delete the construction parameter data of the construction parameter from the construction parameter data sequence. That is, assuming that the qth construction parameter d(p,q) at the pth time point is judged to be abnormal, then delete the construction parameter data d(p) from the collected construction parameter data sequence.

[0055] The method for judging whether the construction parameters are abnormal is: calculate the mean of each construction parameter in the construction parameter data sequence D. When the absolute value of the difference between a construction parameter and the corresponding mean is greater than the set threshold, the construction parameter is judged to be abnormal. The formula is expressed as:

[0056] Let the mean of the qth construction parameter in D be when It means that the construction parameter d(p,q) is abnormal, and d(p) is deleted from D; σ q represents the set threshold value corresponding to the qth construction parameter;

[0057]

[0058] Among them, d(i,q) represents the qth construction parameter at the i-th time point.

[0059] In this embodiment, the construction parameter data sequence D may be firstly processed for missing values ​​to supplement the sampling data at each time point, and then processed for outliers to delete sampling points with abnormal data, thereby obtaining a data sequence D1, and then the data sequence D1 may be normalized to obtain a data sequence D';

[0060] Alternatively, the construction parameter data sequence D is first processed for outliers to delete sampling points with abnormal data; then the remaining sampling points are processed for missing values ​​to obtain the data sequence D1, and then the data sequence D1 is normalized to obtain the data sequence D'.

[0061] Obviously, due to the outlier processing, the sampling points contained in the data sequence D1 must not exceed the construction parameter data sequence D;

[0062] D1={d(1), d(2),…, d(m),…, d(M)}; M≤N

[0063] D'={d'(1), d'(2),..., d'(m),..., d'(M)}

[0064] d'(m)={d'(m,1), d'(m,2),..., d'(m,h),..., d'(m,H)}

[0065]

[0066] Among them, min(D1,h) represents the minimum value of the h-th construction parameter in each data in the data sequence D1, mad(D1,h) represents the maximum value of the h-th construction parameter in each data in the data sequence D1; d'(m,h) represents the h-th construction parameter at the m-th sampling point in D', d'(m,1), d'(m,2) and d'(m,H) ​​represent the 1st, 2nd and Hth construction parameters at the m-th sampling point in D' respectively.

[0067] The method for constructing a large prefabricated T-beam strength prediction model proposed in this embodiment includes the following steps:

[0068] S1. Sampling construction parameter data in combination with the T-beam process, constructing a learning sample {D', y}, and dividing the learning sample into a training sample and a validation sample; constructing a strength prediction model based on a neural network model, the strength prediction model predicts the strength of the T-beam produced at the corresponding time based on the data sequence D'; the strength prediction model can specifically adopt a DNN model.

[0069] S2, initializing the strength prediction model and the butterfly algorithm population, the solution vector of each butterfly is the weight matrix and bias vector set of the strength prediction model;

[0070] S3, let the strength prediction model loaded with the solution vector of each butterfly perform machine learning on the training sample, then select the verification sample to substitute into the strength prediction model, obtain the strength prediction value output by the strength prediction model, and calculate the mean square error loss of the strength prediction model as the mean square error loss corresponding to the butterfly;

[0071] S4, select a random number r, r is between 0 and 1; determine whether r≤P;

[0072] If r≤P, the following formula is used to iterate the butterfly population;

[0073]

[0074] in, represents the solution vector of the zth butterfly in population c after the t+1th iteration, represents the solution vector of the zth butterfly in population c after the tth iteration, and the initial value of t is 0. represents the solution vector of the zth butterfly in the initialized butterfly population c; represents the optimal solution in population c after the tth iteration, that is, the butterfly solution vector corresponding to the minimum mean square error loss in population c after the tth iteration; f c,z represents the amount of fragrance released by the zth butterfly in population c;

[0075] If r>P, the following formula is used to iterate the butterfly population;

[0076]

[0077] δ t+1 =c1δ t +δ 0

[0078] d t =δ t / c2

[0079]

[0080] Among them, δ t+1 represents the step size factor of the t+1th search, δ t Indicates the step size factor of the t-th search; δ 0 represents the initial step size, δ 0 It is set according to the length of the independent variable. In specific implementation, it can be set to the dimension of the independent variable, i.e. the butterfly solution vector; c1 and c2 are adjustment coefficients, which are constants and are related to the complexity of the function; represents the exploration direction, k' represents the dimension of the solution vector, rand(.) is a random function, rand(k', 1) represents the generation of a random k'×1-dimensional vector; ||.|| represents the bi-norm.

[0081] sign represents the sign function; represents the right search area of ​​the zth butterfly in population c after the tth iteration, represents the left search area of ​​the zth butterfly in population c after the tth iteration; d t represents the search distance of the butterfly in the tth iteration;

[0082] Fit(.) represents the fitness function of minimizing the objective, that is, express The corresponding mean square error loss is the opposite number, express The corresponding mean square error loss is the opposite.

[0083] Let d t=2R, then:

[0084]

[0085] Where [1] represents a matrix whose elements are all 1, and R is a transition value.

[0086] S5. Determine whether the number of iterations of the butterfly population reaches the set value; if not, return to step S3; if yes, output the optimal individual of the optimal population according to MSE as the final solution vector, assign the final solution vector to the intensity prediction model and fix the model.

[0087] The method for constructing the above-mentioned large prefabricated T-beam strength prediction model (hereinafter referred to as the method of the present invention) is described and verified in conjunction with specific embodiments.

[0088] In this embodiment, the raw materials and proportions, construction technology, and curing conditions in the actual production of a prefabricated beam field are used as construction parameter data to verify the method of the present invention. In this embodiment, the parameters for predicting concrete strength are determined in combination with expert experience, and 5 material parameters and 3 construction process parameters are finally selected, including water, cement, coarse aggregate, fine aggregate, fly ash, steam curing time, steam curing temperature, and steam curing humidity. For the original data with fewer missing values, each attribute is filled with the average value of the attribute.

[0089] The statistical analysis of the data set parameters is shown in Table 1.

[0090] In this embodiment, in the data preprocessing process, missing values ​​are supplemented first, and then outlier processing is performed. Normalization processing uses min-max standardization, and a total of 723 complete concrete strength prediction data are obtained, and some of the data are shown in Table 2.

[0091] Table 1 Statistical analysis of data set parameters

[0092]

[0093] Note: Steaming time is the total time of the constant temperature stage; steaming temperature is the temperature of the constant temperature stage; the immediate strength after steaming is the average value measured by the rebound instrument.

[0094] Table 2 Prediction data of concrete strength after pretreatment

[0095]

[0096]

[0097] In this embodiment, a deep neural network (DNN) model is selected to construct a strength prediction model, with 9 input nodes and 1 output node. In this embodiment, 723 complete concrete strength prediction data are divided into a training set, a validation set, and a test set in a ratio of 8:1:1. During the model training process, machine learning is performed on the training set, and the mean square error loss is calculated on the validation set; after the model training is completed, the model performance is tested on the test set.

[0098] In this embodiment, Random Forest (RF) and Support Vector Regression (SVR) are selected as control algorithms; the prediction results of the model are evaluated by relative error.

[0099] In this embodiment, the prediction results and residual performance of the model BAS-BOA-DNN trained by the method of the present invention and the trained control algorithm on the test set are shown in Figures 2(a)-(f), and the relative errors are shown in Table 3.

[0100] As shown in Figure 2(a)-(f), the BAS-BOA-DNN model shows significant superiority over the RF model and the SVR model. The predicted values ​​of the BAS-BOA-DNN model are basically consistent with the trend of the true values. The residual analysis shows that the vast majority of residual values ​​are in the range of [-1,1]MPa, and only in very few cases does the residual exceed 2MPa. This result shows that the BAS-BOA-DNN model has excellent fitting ability and prediction accuracy. In contrast, the RF and SVR models show certain limitations in the prediction process. Although these two models can provide relatively accurate predictions in some cases, the residual analysis shows their shortcomings in prediction accuracy. It can be seen that the BAS-BOA-DNN model can better capture the full complexity of concrete strength changes than the RF and SVR models, thereby improving the prediction accuracy.

[0101] Table 3 Relative error of prediction model

[0102]

[0103] As shown in Table 3, considering the average error and maximum error, the BAS-BOA-DNN model performs best in predicting concrete strength, followed by the random forest model and the support vector regression model. Especially in extreme cases, the error of the BAS-BOA-DNN model remains within a controllable range. It can be seen that the performance of the BAS-BOA-DNN model is more stable and more suitable for application scenarios that require high-precision prediction.

[0104] In this example, the root mean square error (RMSE), mean absolute error (MAE) and determination coefficient (R2 ), and the results are shown in Table 4.

[0105] Table 4 Prediction model evaluation indicators

[0106]

[0107] It can be seen that the BAS-BOA-DNN model has good performance in terms of root mean square error (RMSE), mean absolute error (MAE) and determination coefficient (R 2 ) performed best in all three evaluation indicators. Its MAE was 27.11% and 40.84% ​​lower than RF and SVR, respectively. The smaller mean absolute error means that the difference between the model's predicted value and the actual value is smaller; RMSE was 30.00% and 37.35% lower than RF and SVR, respectively. The low value of the mean square error further illustrates the reliability of the model's prediction and the uniform distribution of errors; and R 2 It is 8.00% and 19.57% higher than RF and SVR respectively. The improvement of correlation coefficient shows that the model can capture more patterns and trends behind the data.

[0108] In this embodiment, the BAS-BOA-DNN model is further compared with the non-optimized DNN prediction model (DNN) and the standard butterfly algorithm optimized DNN model (BOA-DNN), and the relative errors of the outputs of the three groups of models are statistically analyzed. Figure 3 And as shown in Table 5.

[0109] Table 5 Relative errors of the improved model and the control group

[0110]

[0111] Depend on Figure 3 As shown in Table 5, the average relative error of the BAS-BOA-DNN model is reduced by 18.11% compared with the BOA-DNN model and by 32.46% compared with the DNN model. The relative error of the BAS-BOA-DNN model is concentrated in the interval [0.01, 0.03] and approaches zero at multiple data points, indicating that the prediction results of the model are highly consistent with the actual observations, and the number of significant error points is greatly reduced, indicating that the BAS-BOA-DNN model is more robust in dealing with data noise and outliers.

[0112] Figure 4 (a) and Figure 4 (b) show the optimization process of BOA and BAS-BOA algorithms. In the early stage of iteration, the performance of the two algorithms is not much different. However, with the increase of the number of iterations, especially after the 50th round, the optimization speed of BAS-BOA algorithm is significantly accelerated, and the optimization effect gradually surpasses the traditional BOA algorithm. It can be seen that the method of the present invention not only accelerates the convergence speed, but also may improve the quality of the final solution, thereby bringing better performance in practical applications.

[0113] Of course, it is obvious to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, but also includes the same or similar structures that can be implemented in other specific forms without departing from the spirit or essential features of the present invention. Therefore, no matter from which point of view, the embodiments should be regarded as exemplary and non-limiting, and the scope of the present invention is defined by the appended claims rather than the above description, and it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims are included in the present invention. Any reference numerals in the claims should not be regarded as limiting the claims involved.

[0114] In addition, it should be understood that although the present specification is described according to implementation modes, not every implementation mode contains only one independent technical solution. This description of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment may also be appropriately combined to form other implementation modes that can be understood by those skilled in the art.

[0115] The techniques, shapes, and structural parts not described in detail in the present invention are all well-known techniques.

Claims

1. A method for constructing a strength prediction model for large prefabricated T-beams, characterized in that: include: S1. Sampling construction parameter data in combination with the T-beam process, constructing a learning sample {D', y}, where D' is a preprocessing sequence of construction parameters within a unit time, and y is the strength of the T-beam produced within the time, and dividing the learning sample into a training sample and a verification sample; constructing a strength prediction model based on a neural network model, and the strength prediction model predicts the strength of the T-beam produced at the corresponding time based on the data sequence D'; S2, initializing the strength prediction model and the butterfly algorithm population, the solution vector of each butterfly is the weight matrix and bias vector set of the strength prediction model; S3, let the strength prediction model loaded with the solution vector of each butterfly perform machine learning on the training sample, then select the verification sample to substitute into the strength prediction model, obtain the strength prediction value output by the strength prediction model, and calculate the mean square error loss of the strength prediction model as the mean square error loss corresponding to the butterfly; S4, select a random number r, r is between 0 and 1; determine whether r≤P; If r≤P, use the global search algorithm to iterate the butterfly population; if r>P, use the following formula to iterate the butterfly population; in, represents the solution vector of the zth butterfly in population c after the t+1th iteration, represents the solution vector of the zth butterfly in population c after the tth iteration. The initial value of t is 0, and δ t+1 represents the step size factor of the t+1th search, Indicates the direction of exploration. represents the right search area of ​​the zth butterfly in population c after the tth iteration, represents the left search area of ​​the zth butterfly in population c after the tth iteration; Fit(.) represents the fitness function of minimizing the objective, and sign represents the sign function; S5, determine whether the number of iterations of the butterfly population reaches the set value; if not, return to step S3; If yes, then the optimal individual of the optimal population is output according to the MSE as the final solution vector, the final solution vector is assigned to the intensity prediction model and the model is fixed.

2. The method for constructing a large prefabricated T-beam strength prediction model according to claim 1, characterized in that: d t+1 =c1δ t +d 0 Among them, δ t Indicates the step size factor of the t-th search; δ 0 represents the set initial step size, and c1 is the set constant.

3. The method for constructing a large prefabricated T-beam strength prediction model according to claim 2, characterized in that: Let d t represents the search distance of the butterfly in the tth iteration, d t =2R, R is the transition value, we have: d t =d t / c2 Where c2 is a set constant, and [1] indicates a matrix whose elements are all 1.

4. The method for constructing a large prefabricated T-beam strength prediction model according to claim 1, characterized in that: Among them, rand represents a random value, rand(.) is a random function, rand(k', 1) represents generating a random k'×1-dimensional vector; ||.|| represents the bi-norm.

5. The method for constructing a large prefabricated T-beam strength prediction model according to claim 1, characterized in that: The method for obtaining the construction parameter preprocessing sequence D' is as follows: the construction parameter data sequence in the T-beam production process is collected, the missing value processing and outlier processing of the construction parameter data sequence are first performed, and then the data is normalized to obtain the construction parameter preprocessing sequence D'; the construction parameter data includes multiple construction parameters.

6. The method for constructing a large prefabricated T-beam strength prediction model according to claim 5, characterized in that: The missing values ​​of the data are handled as follows: when a construction parameter at a certain time point is missing, the mean value of the construction parameter at the remaining time points within the unit time is taken to supplement the time point.

7. The method for constructing a large prefabricated T-beam strength prediction model according to claim 5, characterized in that: The outlier processing method is: to judge the outlier value of each construction parameter in the construction parameter data sequence, if a construction parameter is judged to be abnormal, the construction parameter data of the construction parameter is deleted from the construction parameter data sequence.

8. A method for predicting the strength of a large prefabricated T-beam using the method for constructing a large prefabricated T-beam strength prediction model as described in any one of claims 1 to 7, characterized in that: First, a strength prediction model is constructed by using the method for constructing a large prefabricated T-beam strength prediction model as described in any one of claims 1 to 7; then, a construction parameter data sequence in the T-beam production process is collected and preprocessed, and the preprocessed data sequence is input into the strength prediction model to obtain the strength prediction value output by the strength prediction model.

9. A large prefabricated T-beam strength prediction system, characterized in that: It includes a memory and a processor, the memory stores a computer program and a strength prediction model constructed by the method for constructing a large prefabricated T-beam strength prediction model as described in any one of claims 1 to 7, the processor is connected to the memory, and the processor is used to execute the computer program to implement the large prefabricated T-beam strength prediction method as described in claim 8.

10. A readable medium, characterized in that: A computer program is stored, and when the computer program is executed, it is used to implement the large prefabricated T-beam strength prediction method as claimed in claim 8.

Images