Durability evaluation method of maritime work cementing material and training method and device of model

Abstract

The invention provides a maritime work cementing material durability evaluation method and a model training method and device.The evaluation method comprises the steps that application data, sent by a user through an interaction end, of a to-be-evaluated maritime work cementing material are received, and the application data comprise production process data, batch data and application environment data; extracting features from the application data to obtain application features; and inputting the application characteristics into a durability evaluation model to obtain a durability evaluation result of the maritime work cementing material. According to the invention, high-accuracy and high-efficiency evaluation of the durability of the maritime work cementing material can be realized.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of data processing technology, and in particular to a durability assessment method for marine engineering cementitious materials, a model training method and a device. Background Art

[0002] The embodiments of the present invention relate to the field of data processing technology, and in particular to a durability assessment method for marine engineering cementitious materials, a model training method and a device. Summary of the invention

[0003] The embodiments of the present invention provide a durability assessment method for marine engineering cementitious materials, a model training method and a device to solve the problems that although the existing methods can reflect the durability of marine engineering materials to a certain extent, the durability assessment of each type of marine engineering cementitious materials requires manual physical tests, resulting in low assessment efficiency; and the mathematical model is limited by the limitations of experience and has low assessment accuracy.

[0004] In order to solve the above-mentioned technical problems, the present invention is achieved as follows: In a first aspect, an embodiment of the present invention provides a method for evaluating the durability of a marine cementitious material, comprising: Receiving application data of the marine cementitious material to be evaluated sent by the user through the interactive terminal, the application data including: production process data, batch data and application environment data; Extracting features from the application data to obtain application features; The application characteristics are input into a durability evaluation model to obtain a durability evaluation result of the marine engineering cementitious material.

[0005] Optionally, extracting features from the application data to obtain application features includes: Using a feature extraction model to extract features from the application data to obtain primary features; A feature dimension reduction model is used to perform feature dimension reduction processing on the primary features, so that the primary features after the feature dimension reduction processing are used as the application features.

[0006] In a second aspect, an embodiment of the present invention provides a method for training a durability assessment model for marine cementitious materials, comprising: Acquire historical application data of each marine cementitious material in the marine cementitious material set, wherein the historical application data includes: production process data, batch data, and application environment data; extract features from the historical application data to obtain historical application features; Obtaining durability results corresponding to the marine cementitious materials one by one; and annotating the historical application features of the corresponding marine cementitious materials with the durability results, and combining all the annotated historical application features to obtain a training data set; The classifier model is trained using the training data set to obtain a durability evaluation model.

[0007] Optionally, extracting features from the historical application data to obtain historical application features includes: Using a feature extraction model to extract features from the historical application data to obtain historical primary features; A feature dimensionality reduction model is used to perform feature dimensionality reduction processing on the historical primary features, so as to use the historical primary features after the feature dimensionality reduction processing as the historical application features.

[0008] Optionally, the classifier model is a classifier model of an extreme learning machine algorithm based on orthogonal tensor decomposition.

[0009] In a third aspect, an embodiment of the present invention provides a durability assessment device for marine cementitious materials, comprising: A receiving module, used to receive application data of the marine cementitious material to be evaluated sent by the user through the interactive terminal, wherein the application data includes: production process data, batch data and application environment data; A feature extraction module, used to extract features from the application data to obtain application features; The evaluation module is used to input the application characteristics into the durability evaluation model to obtain the durability evaluation result of the marine engineering cementitious material.

[0010] In a fourth aspect, an embodiment of the present invention provides a training device for a durability assessment model of marine cementitious materials, comprising: An acquisition module is used to acquire historical application data of each marine cementitious material in the marine cementitious material set, wherein the historical application data includes: production process data, batch data, and application environment data; extract features from the historical application data to obtain historical application features; An execution module is used to obtain durability results corresponding to the marine cementitious materials one by one; and annotate the historical application features of the corresponding marine cementitious materials with the durability results, and combine all the annotated historical application features to obtain a training data set; The training module is used to train the classifier model using the training data set to obtain a durability evaluation model.

[0011] In a fifth aspect, an embodiment of the present invention provides an electronic device, comprising a processor, a memory, and a program or instruction stored in the memory and executable on the processor, wherein the program or instruction, when executed by the processor, implements the steps in the durability assessment method for marine cementitious materials as described in any one of the first aspects; or, implements the steps in the training method for a durability assessment model for marine cementitious materials as described in any one of the second aspects.

[0012] In a sixth aspect, an embodiment of the present invention provides a readable storage medium storing a program or instruction on the readable storage medium, which, when executed by a processor, implements the steps in the durability assessment method for marine cementitious materials as described in any one of the first aspects; or, implements the steps in the training method for a durability assessment model for marine cementitious materials as described in any one of the second aspects.

[0013] In the seventh aspect, an embodiment of the present invention provides a computer program product, comprising computer instructions, which, when executed by a processor, implement the steps in the durability assessment method for marine cementitious materials as described in any one of the first aspect; or, implement the steps in the training method for the durability assessment model for marine cementitious materials as described in any one of the second aspect.

[0014] In an embodiment of the present invention, by receiving application data of the marine cementitious materials to be evaluated sent by the user through the interactive terminal, the application data includes: production process data, batch data and application environment data; extracting features from the application data to obtain application features; inputting the application features into the durability evaluation model to obtain the durability evaluation results of the marine cementitious materials, the embodiment of the present invention can achieve high-accuracy and high-efficiency evaluation of the durability of the marine cementitious materials. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Moreover, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings: Figure 1 It is a schematic flow chart of a durability evaluation method for marine engineering cementitious materials according to an embodiment of the present invention; Figure 2 Schematic diagram of the training process of the feature extraction model; Figure 3 It is a flow chart of a method for training a durability assessment model for marine engineering cementitious materials according to an embodiment of the present invention; Figure 4 is the original data distribution diagram; Figure 5 Expand the data distribution map for the traditional one; Figure 6 To improve the data distribution map; Figure 7 A schematic diagram showing the comparison between the original data, the traditional augmented data and the improved augmented data; Figure 8 This is one of the schematic diagrams of data point distribution; Fig. 9 This is the second diagram of data point distribution; Fig.10 Schematic diagram of data points generated for adjustment based on local variance of the data; Fig.11 Schematic diagram of data distribution generated for adjusting noise amplitude in combination with multi-feature sensitivity; Fig.12 A schematic diagram of the convergence process of the loss function of the dynamic adaptive oscillation during the training of the dynamic adaptive oscillation neural network; Fig.13 is a dynamic adaptive amplitude curve; Fig.14 Update curve for dynamic weights; Fig.15 Schematic diagram of the simulated loss function surface; Fig.16 It is a schematic diagram of the dynamic adjustment effect of singular value disturbance; Fig.17 This is a schematic diagram of the change of singular values ​​during the training process; Fig.18 Schematic diagram of the change of reconstruction quality; Fig.19 Schematic diagram of the value distribution of the original features and the reconstructed features; Fig. 20 Schematic diagram of the change in classification accuracy; Fig.21 Schematic diagram of the change of sparsity level; Fig. 22 Schematic diagram of the change of model complexity; Fig.23 This is a schematic diagram of the classification results; Fig.24 This is a principle block diagram of a durability assessment device for marine engineering cementitious materials according to an embodiment of the present invention; Fig.25 This is a principle block diagram of a training device for a durability assessment model of marine engineering cementitious materials according to an embodiment of the present invention; Fig.26 The figure is a principle block diagram of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0016] 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 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.

[0017] The terms "first", "second", etc. of the present application are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the terms used in this way are interchangeable where appropriate, so that the embodiments of the present application can be implemented in an order other than those illustrated or described herein, and the objects distinguished by "first" and "second" are generally of one type, and the number of objects is not limited, for example, the first object can be one or more. In addition, "or" in the present application represents at least one of the connected objects. For example, "A or B" covers three schemes, namely, Scheme 1: including A but not including B; Scheme 2: including B but not including A; Scheme 3: including both A and B. The character " / " generally indicates that the objects associated with each other are in an "or" relationship.

[0018] In addition, the technical features involved in the different embodiments of the present application described below can be combined with each other as long as they do not conflict with each other.

[0019] Marine cementitious materials are widely used in the field of marine engineering, and their durability is directly related to the safety and service life of engineering structures. However, the marine environment is complex and changeable. Marine cementitious materials will be affected by multiple factors such as salt spray corrosion, tidal scouring, temperature and humidity changes during long-term use, resulting in performance degradation. Therefore, accurately evaluating the durability of marine cementitious materials is the key to ensuring engineering quality and optimizing material design.

[0020] Existing durability assessment methods for marine cementitious materials are mostly manual physical tests, or evaluation using mathematical models based on empirical induction.

[0021] Although the existing methods can reflect the durability of marine engineering materials to a certain extent, the durability assessment of each type of marine engineering cementitious materials requires manual physical testing, which has low assessment efficiency; and the mathematical model is limited by the limitations of experience, and the assessment accuracy is low.

[0022] Regarding the statement that mathematical models are limited by experience, specifically, in existing assessment schemes that use mathematical models, only when the characteristics of a certain type of marine cementitious materials have been summarized into a mathematical model can a relatively accurate assessment result be obtained, and it is difficult to be widely applied to the durability assessment of various types of marine cementitious materials. In particular, for marine cementitious materials that are rarely used in the industry or are new, the assessment accuracy is low because the material characteristics have not been summarized into the mathematical model. In actual assessments, it is difficult to know in advance whether the characteristics of the marine cementitious materials to be assessed have been summarized into the mathematical model, that is, it is difficult to know in advance whether the mathematical model is applicable, resulting in the low overall assessment accuracy of the existing mathematical model method when the durability assessment is directly carried out on the marine cementitious materials to be assessed.

[0023] The embodiment of the present invention provides a method for evaluating the durability of marine cementitious materials. Figure 1 As shown, Figure 1 The figure is a flow chart of a durability evaluation method for marine cementitious materials according to an embodiment of the present invention. The durability evaluation method for marine cementitious materials comprises: Step 11: receiving application data of the marine cementitious material to be evaluated sent by the user through the interactive terminal, the application data including: production process data, batch data and application environment data; Step 12: Extract features from the application data to obtain application features; Step 13: Input the application characteristics into the durability assessment model to obtain the durability assessment results of the marine cementitious materials.

[0024] It should be noted that the production process data, batch data and application environment data are all parameters that can affect the durability of marine cementitious materials. It is understandable that the durability of marine cementitious materials prepared by different production processes (for example, process, raw materials) varies; considering that in reality, manufacturers of marine cementitious materials often purchase raw materials from multiple sources to avoid the high risk of production suspension caused by slow supply from a single raw material manufacturer, even if the raw materials are the same, the detailed characteristics of the raw materials are different from different raw material manufacturers, which will also affect the finished products of marine cementitious materials. Therefore, different production batches will also lead to differences in the durability of marine cementitious materials; in addition, when marine cementitious materials are used in different application environments, they are affected by the environment, and the durability of marine cementitious materials also varies.

[0025] The embodiment of the present invention can achieve accurate definition of marine cementitious materials by applying data including: production process data, batch data and application environment data, ensure that application characteristics can accurately characterize the characteristics of the actual marine cementitious materials to be evaluated, and ensure that high-accuracy durability evaluation results are obtained.

[0026] In some embodiments of the present invention, optionally, the training method of the durability assessment model includes: obtaining historical application data of each marine cementitious material in a marine cementitious material set, the historical application data including: production process data, batch data and application environment data; extracting features from the historical application data to obtain historical application features; obtaining durability results corresponding one to one to the marine cementitious materials; and marking the corresponding historical application features of the marine cementitious materials with the durability results, and combining all the marked historical application features to obtain a training data set; using the training data set to train the classifier model to obtain a durability assessment model, whereby the durability assessment model can obtain accurate durability assessment results based on the input application characteristics of the marine cementitious materials to be evaluated, and has higher assessment efficiency than the existing manual assessment.

[0027] In an embodiment of the present invention, by receiving application data of the marine cementitious material to be evaluated sent by the user through the interactive terminal, the application data includes: production process data, batch data and application environment data; extracting features from the application data to obtain application features; inputting the application features into the durability evaluation model to obtain the durability evaluation results of the marine cementitious material, it is possible to achieve a high-accuracy and high-efficiency evaluation of the durability of the marine cementitious material.

[0028] In some embodiments of the present invention, optionally, features are extracted from application data to obtain application features, that is, step 12 includes: Step a: Use the feature extraction model to extract features from the application data to obtain primary features; Step b: Use the feature dimensionality reduction model to perform feature dimensionality reduction processing on the primary features, and use the primary features that have completed the feature dimensionality reduction processing as application features.

[0029] Through the above step a: using the feature extraction model to extract features from the application data to obtain primary features; step b: using the feature dimension reduction model to perform feature dimension reduction processing on the primary features, and using the primary features that have completed the feature dimension reduction processing as application features, it can be ensured that the obtained application features accurately characterize the properties of the marine cementitious materials to be evaluated, and ensure that a highly accurate durability evaluation result is obtained.

[0030] The following describes a method for training a feature extraction model in conjunction with a specific embodiment.

[0031] Data expansion is performed on the marine engineering cementation data. The expanded marine engineering cementation data is input into the feature extraction model to train the feature extraction model. The present invention uses a 6-layer fully connected neural network for feature extraction. The number of nodes in each layer of the fully connected layer is 128, that is, the layer contains 128 neurons. In the prior art, some schemes use neural networks for feature extraction. In certain neural network structures, problems such as gradient vanishing, gradient explosion, or falling into local optimal solutions may be encountered, affecting the stability of training and the performance of the model. The present invention uses a neural network algorithm based on dynamic adaptive oscillation as a feature extraction model. By simulating the nonlinear oscillation behavior in physical phenomena, the algorithm can effectively explore and utilize local extreme values ​​in high-dimensional parameter space, thereby achieving the purpose of optimizing the neural network.

[0032] For details, see Figure 2 As shown, Figure 2 The figure is a schematic diagram of the training process of the feature extraction model. The training process of the feature extraction model based on the dynamic adaptive oscillation neural network algorithm includes: S301, the input of the neural network is the feature vector corresponding to the expanded marine cementing data, which includes attributes such as production process data, batch data, and use environment data, and the parameters of the neural network are initialized. At the same time, the hyperparameters of the dynamic adaptive oscillation are initialized, including the initial phase of the oscillation and the initial amplitude of the oscillation. In one embodiment, the initialization method is expressed as:

[0033]

[0034]

[0035]

[0036] In the formula, Indicates compliance with a specific distribution; It means the mean is 0 and the variance is Normal distribution of represents normal distribution; Represents the initialization variance of the neural network; is the initial weight of the neural network, is the initial bias of the neural network; is the initial phase of oscillation; is the initial amplitude of oscillation. Preferably, Set to 0.01.

[0037] S302, based on the current loss function, the oscillation frequency of each parameter is calculated. The adjustment of the oscillation frequency depends on the local curvature estimation of the loss surface. The adjustment is performed using the historical gradient information to make the parameter update smoother and enable more efficient search in the parameter space. The adjustment method is expressed as:

[0038] In the formula, is the historical gradient weight factor, which controls the influence of the historical gradient on the current frequency adjustment; is the oscillation frequency, For the The oscillation frequency of the iteration; For the The weights of the neural network at the iteration; For the The weights of the neural network at the iteration; is the loss function of the neural network; is the symbol of partial derivative; is the fundamental oscillation frequency; is a hyperparameter that adjusts the oscillation responsiveness. Preferably, Set to 5, Set to 0.3; the loss function of the neural network uses cross entropy loss, which is calculated by the preset Softmax on the output of the last layer of the neural network.

[0039] S303, the update of each parameter is affected by the phase difference at the same time. The phase difference is determined by the success rate of the previous update and the interaction strength between the parameters. The phase update depends on the effect of the previous parameter update, which is used to simulate the delayed effect of causality. The update method is expressed as:

[0040] In the formula, is the phase of the oscillation, For the The phase of the oscillation of the iteration; For the The phase of the oscillation of the iteration; is the phase adjustment factor; It is a hyperbolic tangent function, which can limit the adjustment range of the phase and avoid excessive adjustment that may cause instability in the algorithm. is a hyperparameter that adjusts phase sensitivity. Preferably, Set to 0.1, Set to 0.3.

[0041] S304, automatically adjust the amplitude according to the effect of parameter update in the past few iterations. If the update of a certain parameter continues to reduce the loss, increase its amplitude; otherwise, reduce the amplitude. The adaptive adjustment method of the amplitude is expressed as:

[0042] In the formula, is the amplitude of the oscillation, It is The amplitude of the oscillation of the iteration; It is The amplitude of the oscillation of the iteration; is the amplitude adjustment coefficient. Preferably, Set to 0.95.

[0043] S305. Calculate the updated value of the parameters of each neural network. Specifically, update the parameters in combination with the oscillation behavior, use the sine function and cosine function to simulate the oscillation behavior, and allow the parameters to be periodically explored in the gradient direction of the loss function. The update method is expressed as:

[0044]

[0045] In the formula, For the The weights of the neural network at the iteration; For the The bias of the neural network at this iteration; For the The weights of the neural network at the iteration; For the The bias of the neural network at this iteration; is the current iteration number of the neural network.

[0046] S306, repeat the above steps until the preset stop iteration condition is met, which means the model training is completed. In one embodiment, the preset stop iteration condition is reaching a preset maximum number of iterations, preferably, the preset maximum number of iterations is set to 1000 times.

[0047] A neural network algorithm based on dynamic adaptive oscillation is adopted, combined with adaptive adjustment of amplitude, frequency and phase, to solve the problem that neural networks are prone to falling into local extreme values ​​in high-dimensional parameter space, and to improve the stability and performance of training.

[0048] The following describes a method for training a feature dimensionality reduction model in conjunction with a specific embodiment.

[0049] The feature-extracted marine gel number (i.e., the training data is equivalent to the primary feature of the embodiment of the present invention) is input into the feature dimensionality reduction model to train the feature dimensionality reduction model. The present invention adopts an autoencoder algorithm based on singular value perturbation as the feature dimensionality reduction model, and uses the structure of the autoencoder to learn the compressed representation of the input data, so as to enhance the sensitivity of the model to small but important variations in the data, thereby improving the accuracy and robustness of the feature dimensionality reduction; The singular value perturbation means that in the autoencoder, the encoder not only automatically encodes the original features, but also adjusts its internal weights by introducing singular value perturbations, so that the network can capture subtle features that may be ignored due to oversimplification of the model while maintaining the main structure of the data, so that it can effectively reconstruct the data in a lower dimensional space while maintaining the integrity of key information.

[0050] Specifically, the training process of the autoencoder algorithm based on singular value perturbation is as follows: S401, initialize the initial weight and bias of the autoencoder, and the initialization method is expressed as:

[0051]

[0052] In the formula, is the weight of the encoding layer of the autoencoder; is the bias of the encoding layer of the autoencoder; is the dimension of the input layer of the autoencoder, is the dimension of the output layer of the autoencoder, A function for generating normally distributed random numbers.

[0053] S402, the feature-extracted marine cementing data is propagated through the encoding layer of the autoencoder, each layer is processed by the activation function, and finally a compressed feature representation is formed in the encoding layer. In the forward propagation process, the input is converted into the compressed representation of the code as follows:

[0054] In the formula, is the Sigmoid activation function; It is the feature representation after compression of the encoding layer; is the input feature of the autoencoder.

[0055] S403, applying singular value perturbation to the feature representation, by performing a small range of random perturbations on each feature output by the encoding layer, in order to detect and amplify small changes in the data, and enhance the model's ability to perceive subtle changes in the data, expressed as:

[0056]

[0057] In the formula, is the disturbance intensity parameter; It is the feature representation after singular value perturbation; is the left singular vector of the singular value decomposition, is the singular value diagonal matrix of the singular value decomposition, is the right singular vector determined by the singular value decomposition; is the transpose of the right singular vector defined by the singular value decomposition; is the singular value decomposition function; is the amount of fine-tuning of the singular values.

[0058] Furthermore, the amount of fine-tuning of the singular value is used to increase the influence of the disturbance, aiming to strengthen the secondary features of the data. The calculation method is expressed as:

[0059] In the formula, is a small perturbation parameter, is the identity matrix. Preferably, Set to 0.01.

[0060] S404, calculate the loss function of the autoencoder, and then update the weights and biases of the network through the back propagation algorithm. The calculation method of the autoencoder loss function and the partial derivatives of its parameters is expressed as:

[0061]

[0062]

[0063]

[0064] In the formula, is the loss function of the autoencoder; is the L2 norm; is the reconstructed output through the decoding process, is the Sigmoid activation function; is the derivative of the Sigmoid activation function; is the weight of the autoencoder decoding layer; is the bias of the decoder layer of the autoencoder.

[0065] S405. In the weight update process, the quantum tunneling effect is simulated to allow the weight to jump to a possible area near the global optimal solution in some cases, which helps to prevent the algorithm from falling into the local minimum. The update method of the weight and bias of the encoding layer of the autoencoder is expressed as:

[0066]

[0067] In the formula, is the weight of the updated autoencoder encoding layer; is the learning rate of the autoencoder, It is the quantum tunneling probability adjusted according to the loss gradient, so that the tunneling probability is associated with the gradient size of the loss function, thereby reducing the tunneling effect in high-gradient regions and enhancing the tunneling effect in low-gradient regions; Adjusting parameters for the first quantum tunneling; is the second quantum tunneling tuning parameter, is the quantum tunneling effect function. Preferably, Set to 2, Set to 2.4.

[0068] Furthermore, the quantum tunneling effect function is used to randomly jump in the potential optimal weight space, simulating the tunneling phenomenon in quantum physics, allowing the weight to surpass the potential energy barrier during the gradient descent process to avoid the local minimum. The calculation method is expressed as:

[0069] In the formula, is the quantum tunneling coefficient, which is used to control the intensity of the tunneling effect; is the Hessian matrix of the autoencoder's loss function with respect to the encoding layer weights, representing the second-order derivative; is the first-order derivative of the autoencoder’s loss function with respect to the encoding layer weights.

[0070] S406, repeat the above steps until the preset stop iteration condition is met, which means the model training is completed. In one embodiment, the preset stop iteration condition is reaching a preset maximum number of iterations, preferably, the preset maximum number of iterations is set to 1000 times.

[0071] An autoencoder algorithm based on singular value perturbation is adopted to improve the sensitivity to small but important feature variations by applying singular value perturbations in the encoding layer.

[0072] The embodiment of the present invention provides a method for training a durability assessment model for marine cementitious materials. Figure 3 As shown, Figure 3 The figure is a flow chart of a method for training a durability assessment model of marine engineering cementitious materials according to an embodiment of the present invention. The training method includes: Step 21: Obtain historical application data of each marine cementitious material in the marine cementitious material set, the historical application data including: production process data, batch data and application environment data; extract features from the historical application data to obtain historical application features; Step 22: Obtain durability results corresponding to the marine cementitious materials one by one; and annotate the historical application features of the corresponding marine cementitious materials with the durability results, and combine all the annotated historical application features to obtain a training data set; Step 23: Use the training data set to train the classifier model to obtain a durability evaluation model.

[0073] In some embodiments of the present invention, optionally, the historical application data of each marine engineering cementitious material in the marine engineering cementitious material set is the historical application data after data expansion, so as to improve the representativeness and authenticity of each marine engineering cementitious material in the marine engineering cementitious material set.

[0074] The data expansion is described below in conjunction with a specific embodiment: It is understandable that the collection, labeling and preprocessing of marine cement training data are time-consuming and labor-intensive, and insufficient training samples can easily lead to poor model generalization ability and affect the accuracy of the model.

[0075] A bilinear interpolation algorithm based on self-calibration of fit is used to adaptively adjust the interpolation parameters to maximize the representativeness and authenticity of the marine cement data. The self-calibration strategy of fit can be dynamically adjusted during the execution of bilinear interpolation to evaluate the fit of each marine cement data point, which is calculated by comparing the difference between the original marine cement data point and the interpolated marine cement data point. According to the fit result, the interpolation parameters are automatically adjusted to improve the accuracy and representativeness of the marine cement data.

[0076] Specifically, the process of generating samples using the bilinear interpolation algorithm based on self-calibration of fit is as follows: S201, firstly, the basic parameters of bilinear interpolation are initialized, including setting the size and shape of the interpolation kernel, and the initial fitting parameters. The initialization method is expressed as:

[0077] In the formula, Represents a set of initialized parameters; is the initial coefficient of parameter adjustment, which is used to control the adaptive adjustment of the interpolation kernel size; is the initial size of the interpolation kernel; The baseline offset for shape adjustment ensures consistent interpolation across different marine cement data distributions.

[0078] S202, evaluating the fit of each marine engineering cementing data point, wherein the fit is calculated by comparing the difference between the original marine engineering cementing data point and the interpolated marine engineering cementing data point. Specifically, according to the fit result of the initial interpolation, the interpolation parameter is automatically adjusted to improve the accuracy and representativeness of the marine engineering cementing data. The expression is:

[0079]

[0080] In the formula, is the parameter adjustment amount; is the interpolation learning rate, which controls the speed of parameter update; is the number of marine cementation data points; For the The actual value of each marine cementitious data point, specifically, a data point with attributes such as production process data, batch data, and use environment data; For the The interpolation prediction value of each marine cementitious data point is, specifically, the data point obtained by fitting the attributes of production process data, batch data, use environment data, etc. is an updated parameter set, which includes attribute parameters such as production process data, batch data, and usage environment data. Set to 0.1.

[0081] Furthermore, the interpolation prediction value is calculated based on bilinear interpolation, and the expression is:

[0082]

[0083] In the formula, is the weight adjusted according to the distance between the adjacent marine engineering cementing data points of the i-th marine engineering cementing data point, and is used to calculate the interpolation value of the current marine engineering cementing data point; is a parameter that adjusts the sensitivity of this weight; For the Input value of marine cementation data points; For the Input value of marine cementation data points; For the Input value of marine cementation data point; For the Actual value of the marine cementing data point; For the Actual value of the marine cement data point.

[0084] S203. The importance of each feature is evaluated using the entropy weight method to determine which features are more critical to the prediction of material properties, thereby guiding the focus of the marine cementitious data expansion process. The expression is:

[0085]

[0086] In the formula, is the weight of the i-th feature; is the total number of features; Score the normalized importance of the jth feature; is the original value of the jth feature; entropy is used to quantify the uncertainty of the feature and further guide the selection and re-adjustment of the feature.

[0087] Furthermore, the calculation expression of importance score is:

[0088] In the formula, is a parameter that adjusts the sensitivity of normalized importance, is the original value of the kth feature.

[0089] S204, performing marine cement data expansion according to the self-calibrated interpolation parameters and the importance of the features, using directional interpolation based on feature importance adjustment to ensure that the enhanced marine cement data set can reflect the real environmental changes and material properties, expressed as:

[0090]

[0091] In the formula, This is the expanded new offshore cementitious data set; Expand functions for offshore cementitious data; is the original marine cement dataset; is the expansion coefficient function of offshore cement data adjusted according to parameters and weights; It is the maximum value in the offshore cementitious data set and is used to normalize the range of offshore cementitious data expansion; is the weight vector of sample features, whose elements are ; is the data gain of the i-th sample.

[0092] Furthermore, adaptive data gain modulation is used to optimize the information richness of the marine engineering cement training data set by dynamically adjusting the data gain. Specifically, the data gain is adaptively adjusted based on the local variability of the marine engineering cement data points, and the calculation method is expressed as:

[0093] In the formula, is the data gain adjustment factor, which controls the sensitivity of the gain; is the variance function, Characterized in Variability of offshore cement data within the surrounding window, the window size is Preferably, Set to 0.2, Set to 2.

[0094] Furthermore, the marine cement data expansion function uses a linear combination of parameters and weights to adjust the expansion range of marine cement data, and the calculation method is expressed as:

[0095] In the formula, is the adjustment coefficient used to control the expansion intensity; is the threshold parameter used to activate the marine cementation data expansion process; For S The activation function is used to smooth the product of the parameter and the weight, making the marine cement data expansion more refined and reasonable. Preferably, Set to 0.2, Set to 0.5.

[0096] S205. Perform a quality assessment on the expanded offshore cementitious data set to ensure the consistency and reliability of the offshore cementitious data set. The assessment depends on the data quality score, which is calculated as follows:

[0097]

[0098] In the formula, Score data quality; A weight-based quality scoring function for a single marine cementitious data point is presented to characterize the combined impact of feature weights and prediction errors; is the hyperbolic tangent function; is a coefficient used to adjust the sensitivity of the quality score.

[0099] When the data quality score is greater than the preset threshold, it indicates that the generated marine cementitious data sample meets the quality requirements and is retained; otherwise, it is discarded and regenerated.

[0100] A bilinear interpolation algorithm based on self-calibration of fit is adopted to solve the influence of insufficient training samples on the generalization ability and accuracy of the model by dynamically adjusting the interpolation parameters to optimize the accuracy and representativeness of the data.

[0101] In the embodiment of the present invention, the durability results corresponding to the marine engineering cementitious materials may include: high durability (strong durability), medium durability (medium durability), and low durability (weak durability). It should be noted that the determination basis of the above high, medium, and low durability can be determined by experts in the marine engineering field based on the durability test results of the materials and their performance in actual applications.

[0102] Production process data, batch data and application environment data are all parameters that can affect the durability of marine cementitious materials. The embodiment of the present invention can accurately define marine cementitious materials through historical application data including: production process data, batch data and application environment data, ensure that the application characteristics can accurately characterize the characteristics of the actual marine cementitious materials to be evaluated, ensure that the classifier model can form accurate cognition of different marine cementitious materials during training, and ensure that the embodiment of the present invention trains a durability assessment model that can output high-accuracy durability assessment results.

[0103] In an embodiment of the present invention, historical application data of each marine cementitious material in a marine cementitious material set is obtained, the historical application data including: production process data, batch data and application environment data; features are extracted from the historical application data to obtain historical application features; durability results corresponding to the marine cementitious materials are obtained; and the corresponding historical application features of the marine cementitious materials are labeled with the durability results, and all the labeled historical application features are combined to obtain a training data set; the classifier model is trained using the training data set to obtain a durability assessment model. The embodiment of the present invention trains a durability assessment model that can output high-accuracy durability assessment results.

[0104] In some embodiments of the present invention, optionally, extracting features from historical application data to obtain historical application features includes: Step c: Use the feature extraction model to extract features from historical application data to obtain historical primary features; Step d: Use the feature dimensionality reduction model to perform feature dimensionality reduction processing on the historical primary features, so as to use the historical primary features after the feature dimensionality reduction processing as the historical application features.

[0105] Through the above-mentioned step c: using the feature extraction model to extract features from the historical application data to obtain historical primary features; step d: using the feature dimensionality reduction model to perform feature dimensionality reduction processing on the historical primary features, so as to use the historical primary features that have completed the feature dimensionality reduction processing as historical application features, it can be ensured that the obtained historical application features accurately characterize the properties of the marine engineering cementitious materials to be evaluated, ensure that the classifier model can form an accurate understanding of different marine engineering cementitious materials during the training process, and ensure that the embodiment of the present invention trains a durability assessment model that can output high-accuracy durability assessment results.

[0106] It should be noted that the training process of the feature extraction model in the embodiment of the present invention can be found in the above description (including: a specific embodiment of the training method of the feature extraction model, Figure 2 , steps S301-S306), which will not be repeated here.

[0107] It should be noted that, for the training process of the feature dimensionality reduction model in the embodiment of the present invention, please refer to the above description (including: a specific embodiment of the training method of the feature dimensionality reduction model, steps S401-S406), which will not be repeated here.

[0108] In some embodiments of the present invention, optionally, the classifier model is a classifier model of an extreme learning machine algorithm based on orthogonal tensor decomposition.

[0109] The following describes the mechanism of the classifier model of the extreme learning machine algorithm based on orthogonal tensor decomposition during the training process in combination with a specific embodiment.

[0110] The marine cementitious data after feature dimensionality reduction is input into the classifier model for model training. The present invention adopts an extreme learning machine algorithm based on orthogonal tensor decomposition as the classifier model. The orthogonal tensor decomposition can effectively reduce the risk of overfitting by processing the complex interactions in the marine cementitious data after dimensionality reduction. By dynamically adjusting the sparsity parameters in the tensor decomposition process, the complexity of the model can be adaptively adjusted to match the characteristics of different marine cementitious data.

[0111] Specifically, the training process of the extreme learning machine algorithm based on orthogonal tensor decomposition is as follows: S501, initialize the parameters of the extreme learning machine, set the initial parameters of the tensor decomposition, expressed as:

[0112]

[0113] In the formula, is the initial weight of the extreme learning machine, is the initial bias of the extreme learning machine, is the standard deviation initialization factor, is the number of input features of the extreme learning machine, is a function that generates initial random values, is the rank of the tensor.

[0114] S502. Based on the marine cementitious data after feature dimension reduction, the data is converted into a high-order tensor form so that it can express the complex relationship and interaction between the original features, which is expressed as:

[0115] In the formula, represents the offshore cementation data matrix obtained from the feature dimension reduction model; represents the outer product; is a function that folds a matrix into a tensor form; It is a third-order tensor converted from the data matrix obtained from the feature dimensionality reduction model, expressing the complex relationships and interactions between the original features.

[0116] S503, using orthogonal tensor decomposition technology to process high-order tensors, and dynamically adjusting the sparsity parameters to optimize the tensor decomposition process. Specifically, in each iteration, the sparsity parameters are dynamically adjusted according to the classification error to seek the optimal sparsity level and tensor rank. The constraints of the iterative process are:

[0117]

[0118] In the formula, is the loss function of the extreme learning machine, specifically the cross entropy loss; is the weight of the extreme learning machine; is the hidden layer output of the extreme learning machine; For the The hidden layer output of the extreme learning machine at the iteration; is the target output of the extreme learning machine, specifically the classification categories of marine engineering cementitious data, including: high durability (strong durability), medium durability (medium durability), low durability (weak durability), with 0, 1, and 2 corresponding to 3 categories respectively; Represents the first dimension tensor product; Represents the second dimension tensor product; Represents the third dimension tensor product; and Represent the Frobenius norm and L1 norm respectively; is the learning rate of the extreme learning machine; represents the gradient of the extreme learning machine weights; For the The weight of the extreme learning machine at the iteration; For the The weight of the extreme learning machine at the iteration; is the sparse parameter of the extreme learning machine. Preferably, Set to 2, Set to 0.01.

[0119] S504, repeat the above steps until the preset stop iteration condition is met, which means that the model training is completed. In one embodiment, the preset stop iteration condition is reaching a preset maximum number of iterations, preferably, the preset maximum number of iterations is set to 1000 times.

[0120] The extreme learning machine algorithm based on orthogonal tensor decomposition is adopted to capture the complex interactions between features in the form of high-order tensors, dynamically adjust the sparse parameters, optimize the tensor decomposition process, and reduce the risk of overfitting.

[0121] The technical effects of the embodiments of the present invention are described below in combination with actual measured data.

[0122] See Table 1, which shows the experimental comparison data of the improved algorithm (equivalent to the embodiment of the present invention) and the unimproved algorithm at various stages, including detailed experimental results of data expansion, feature extraction, dimensionality reduction, classification and overall evaluation.

[0123] Table 1 Experimental results 1

[0124]

[0125] See Table 2, which shows the comparative experimental data of the algorithms during the training process of the feature extraction model.

[0126] Table 2 Experimental results 2 Number of training rounds Initial loss value (improvement) Convergence loss value (improvement) Initial loss value (not improved) Convergence loss value (not improved) Training time (seconds) (improved) Training time (seconds) (unimproved) Model accuracy (improvement) Model accuracy (not improved) 500 0.560 0.120 0.570 0.150 180 170 88.50% 82.30% 1000 0.550 0.098 0.565 0.130 360 340 91.20% 86.50% See Table 3, which shows the comparative experimental data of the algorithms during the training process of the feature dimensionality reduction model.

[0127] Table 3 Experimental results 3 method Feature Dimension Reconstruction error (improvement) Reconstruction error (not improved) Important feature retention rate (improvement) Important feature retention rate (no improvement) Calculation time (seconds) (improved) Computation time (seconds) (unimproved) PCA 3 0.12 0.15 86.50% 80.30% 0.80 0.75 Singular Value Perturbation Autoencoder 3 0.08 0.12 93.70% 88.10% 1.20 1.10 See Table 4, which shows the comparative experimental data of the algorithms during the training process of the classifier model.

[0128] Table 4 Experimental results 4

[0129] See Table 5, which shows the comprehensive evaluation experimental results of the algorithm proposed in the present invention.

[0130] Table 5 Experimental results 5 Evaluation indicators Improved algorithm raw data Original data without improved algorithm Improve the algorithm and expand the data No algorithm improvement to expand data Improved algorithm for dimensionality reduction data No improved algorithm for dimensionality reduction Improved algorithm synthesis data Unimproved algorithm comprehensive data Data representativeness score 0.85 0.78 0.91 0.83 0.89 0.81 0.93 0.85 Data consistency scoring 0.87 0.80 0.93 0.84 0.90 0.83 0.95 0.87 Classifier overall accuracy (%) 78.90% 73.40% 84.90% 79.50% 86.70% 81.60% 91.40% 85.30% In addition, the differences and improvement effects among the original data, traditional augmented data, and improved augmented data can be intuitively observed through the charts of the experimental data.

[0131] See also Figures 4 to 6 As shown, the two-dimensional spatial distribution of feature 1 and feature 2 of the offshore cementing data and the changes in their target values ​​are shown, where: Figure 4 is the original data distribution diagram, Figure 5 This is the traditional extended data distribution diagram. Figure 6 To improve the expanded (that is, expanded by the embodiment of the present invention) data distribution map.

[0132] See Figure 4 As shown in the figure, the original data has a certain distribution pattern, but the data points are relatively sparse, especially in areas with higher and lower target values, where there is a lack of sufficient data point coverage. Sparsity may limit the ability to capture boundary conditions during model training, thereby affecting the generalization of the model.

[0133] Next, see Figure 5 As shown in the figure, the traditional augmented data distribution graph (through the traditional SMOTE algorithm) generates new data points by simply adding random noise. Although the number of data points is enriched, it can be seen from the figure that the distribution of these newly generated data points is relatively irregular, especially in the feature space, some areas appear densely packed, while other areas are still relatively sparse.

[0134] In contrast, see Figure 6 As shown, the improved expanded data distribution diagram proposed by the present invention generates new data points with more reasonable distribution through the bilinear interpolation method with self-calibration of fitting degree, and the distribution of data points in the feature space is more uniform, especially in areas where the original data is sparse, and the newly generated data points effectively fill the gaps.

[0135] See also Figure 7 As shown, Figure 7 Schematic diagram of the comparison between original data, traditional augmented data and improved augmented data. Figure 7 The characteristic differences of the three data distributions are quantified through the variance comparison chart. The variance of the original data reflects the inherent characteristics of its distribution. The traditional expansion method does not consider the correlation between features and the consistency of the target value, which leads to abnormal fluctuations in the variance of the expanded data on some features. The variance of the expanded data of the method proposed in the present invention is closer to the original data, indicating that the improved method effectively maintains the characteristic distribution characteristics of the original data during the expansion process, while increasing the diversity of samples.

[0136] See also Figure 8 and Fig. 9 As shown in , by dynamically adjusting the interpolation weights, a smooth transition of data points in the feature space is achieved. Figure 8 As shown in , experiments show that the data points generated by the method proposed in the present invention are more evenly distributed, especially in areas where the original data are sparse. Fig. 9 As shown in the figure, the noise is weighted by using the importance score of the feature to enhance the representativeness of the key features. The experimental results show that the data generated by entropy weight expansion is more concentrated in the areas with high-importance features, thereby further improving the data density in terms of key characteristics.

[0137] See also Fig.10 As shown in the figure, adjustments are made based on the local variance of the data to generate more data points that can reflect changes in local features. In the experiment, local variability expansion can well capture tiny regional changes in the original data. The generated data effectively retains the differences in local characteristics and enhances the model's ability to learn small-scale changes.

[0138] See also Fig.11 As shown, the noise amplitude is adjusted in combination with multi-feature sensitivity to generate samples that are more in line with the actual data distribution. Experimental results show that the data point distribution generated by the method proposed in the present invention not only covers the sparse area of ​​the original data, but also can adapt to the different effects of different features on the target value, which significantly improves the adaptability and quality of the expanded data.

[0139] See also Fig.12As shown in the figure, during the training of the dynamic adaptive oscillating neural network, the convergence process of the loss function of the dynamic adaptive oscillation shows a smooth and fast convergence effect, while maintaining moderate oscillation and avoiding falling into the local optimum. Fig.12 The orange dotted line in the middle represents the loss convergence process of fixed oscillation. Although there is a certain convergence trend, the fixed oscillation amplitude makes it impossible to fully optimize in some stages.

[0140] See also Fig.13 As shown in the figure, the dynamic adaptive amplitude curve shows the changing trend of the amplitude during the training process. By adaptively adjusting the amplitude, the model's exploration ability in high-dimensional parameter space is enhanced. The fixed amplitude curve remains constant, resulting in the model being unable to flexibly adjust the exploration intensity at different stages.

[0141] See also Fig.14 As shown in the figure, the dynamic weight update curve demonstrates the flexibility of weight adjustment, and the oscillation behavior changes dynamically with the training stage, which is conducive to the efficient optimization of parameters. The fixed weight update curve is relatively stable and lacks the ability to adapt to changes in the loss surface.

[0142] See also Fig.15 As shown in the figure, the simulated loss function surface shows the local extreme value distribution of the high-dimensional parameter space. Dynamic adaptive oscillation can explore the surface more efficiently and avoid falling into the local extreme value area.

[0143] See also Fig.16 As shown in the figure, during the training process of the autoencoder based on singular value perturbation, the loss function gradually decreases with the number of training rounds, indicating that the model is constantly optimizing. The smooth downward trend of the loss value, accompanied by small fluctuations, reflects the dynamic adjustment effect of the singular value perturbation.

[0144] See also Fig.17 As shown in Figure 3, the changing pattern of singular values ​​during the training process shows that with the increase in the number of training rounds, the singular values ​​gradually tend to be stable. At the same time, the simulated small disturbances amplify the secondary information of the features and enhance the model's ability to capture subtle variations.

[0145] See also Fig.18 As shown in Figure 3, the improvement trend of reconstruction quality indicates that the model continuously optimizes the low-dimensional representation of features during the learning process, and finally achieves higher reconstruction accuracy. The model can effectively maintain the main information of the original features while paying attention to the detailed features.

[0146] See also Fig.19 As shown, Fig.19 By comparing the value distributions of the original features and the reconstructed features, the two curves are highly consistent, indicating that the autoencoder successfully captured the main structural information of the original features, and the reconstructed features have slight adjustments in details, reflecting the enhancing effect of singular value perturbation on feature capture.

[0147] The autoencoder based on singular value perturbation has high accuracy and robustness in feature dimensionality reduction tasks. It can not only maintain the global structure of the data, but also effectively enhance the sensitivity to subtle feature changes, providing a more representative low-dimensional data representation for downstream tasks.

[0148] See also Fig. 20 As shown in the figure, regarding the training process of the extreme learning machine classifier model based on orthogonal tensor decomposition, the classification accuracy changes with the number of training rounds, showing a gradual improvement trend of the classification accuracy. The model gradually tends to be stable with the increase of the number of training rounds. The classification accuracy fluctuates greatly in the early training stage, and then gradually converges after 500 rounds, and finally approaches the theoretical optimal level.

[0149] See also Fig.21 As shown in the figure, the change of sparsity level with the number of training rounds shows the dynamic adjustment process of sparsity parameters. In the early stage of training, the sparsity level is low. As the training progresses, it gradually increases and tends to be stable. The dynamic adjustment effectively controls the complexity of the model and reduces the risk of overfitting.

[0150] See also Fig. 22 As shown in the figure, the change of model complexity with the number of training rounds reflects the changing law of model complexity. The model shows a higher complexity in the early stage, and as the training progresses, the model complexity gradually decreases through the dynamic adjustment of the sparsity parameters, indicating that orthogonal tensor decomposition can adaptively optimize the model structure.

[0151] See also Fig.23 As shown in the figure, the confusion matrix of the classification results shows the prediction performance of the classifier on different categories. The classification results of high durability, medium durability and low durability categories are generally accurate, but there is some confusion on the boundary samples, indicating that the model still needs to be further optimized.

[0152] The extreme learning machine classifier based on orthogonal tensor decomposition has great advantages in processing marine engineering cementitious data after dimensionality reduction. This method shows good adaptability in dynamically adjusting model complexity and sparsity parameters, providing an effective means to improve classification performance and generalization ability.

[0153] In addition, as shown in Table 6, the experimental data of the single module algorithm are compared on the experimental data set. In the data expansion and interpolation algorithms: Table 6 Experimental results 6 Extension Method Total number of samples The number of samples after expansion Data consistency scoring Representative score Original acquisition data 100 - 0.85 ± 0.03 0.82 Bilinear interpolation expansion 100 300 0.90 ± 0.02 0.88 Entropy weight interpolation expansion 100 300 0.91 ± 0.01 0.89 Dynamic Data Gain Expansion 100 300 0.93 ± 0.02 0.91 In the feature extraction model training, the comparative experimental results are shown in Table 7.

[0154] Table 7 Experimental results 7 Number of training rounds Initial loss value Convergence loss value Number of gradient updates Number of gradient disappearances Training time (seconds) Model accuracy 500 0.560 0.120 450 5 180 88.50% 1000 0.550 0.098 920 3 360 91.20% In the feature dimension reduction model training, the comparative experimental results are shown in Table 8.

[0155] Table 8 Experimental results 8 method Feature Dimension Reconstruction error Important feature retention rate Calculation time (seconds) PCA 3 0.12 86.50% 0.80 Singular Value Perturbation Autoencoder 3 0.08 93.70% 1.20 In the classifier model training, the comparative experimental results are shown in Table 9.

[0156] Table 9 Experimental results 9 Classification Category Original data accuracy Accuracy of expanded data Accuracy of dimensionality reduction data Comprehensive model accuracy Classification time (seconds) High durability 85.20% 89.10% 90.30% 92.50% 0.10 Medium durability 78.50% 83.70% 85.90% 88.40% 0.12 Low durability 70.30% 75.20% 77.50% 81.30% 0.15 The above experimental data fully verified the effectiveness of each link, from data expansion, feature extraction to optimization of classifier performance. The combined model based on self-calibration interpolation of fit, singular value perturbation autoencoder and orthogonal tensor decomposition extreme learning machine performed excellently in improving the accuracy and robustness of durability assessment of marine engineering cementitious materials.

[0157] The embodiment of the present invention provides a durability evaluation device for marine cementitious materials, see Fig.24 As shown, Fig.24 : is a principle block diagram of a durability evaluation device for marine engineering cementitious materials according to an embodiment of the present invention. The durability evaluation device 30 includes: The receiving module 31 is used to receive the application data of the marine cementitious material to be evaluated sent by the user through the interactive terminal, wherein the application data includes: production process data, batch data and application environment data; A feature extraction module 32, used to extract features from the application data to obtain application features; The evaluation module 33 is used to input the application characteristics into the durability evaluation model to obtain the durability evaluation result of the marine engineering cementitious material.

[0158] In some embodiments of the present invention, optionally, The feature extraction module 32 is further used to extract features from the application data using a feature extraction model to obtain primary features; The feature extraction module 32 is further configured to use a feature dimension reduction model to perform feature dimension reduction processing on the primary features, so as to use the primary features after the feature dimension reduction processing as the application features.

[0159] The durability evaluation device for marine cementitious materials provided in the embodiment of the present application can achieve Figure 1 The various processes implemented by the method embodiment and achieving the same technical effect are not described here to avoid repetition.

[0160] The embodiment of the present invention provides a training device for a durability assessment model of marine cementitious materials, see Fig.25 As shown, Fig.25The principle block diagram of the training device of the durability evaluation model of marine cementitious materials according to the embodiment of the present invention is shown in FIG. 40. The training device 40 includes: The acquisition module 41 is used to acquire the historical application data of each marine cementitious material in the marine cementitious material set, wherein the historical application data includes: production process data, batch data and application environment data; extract features from the historical application data to obtain historical application features; An execution module 42 is used to obtain durability results corresponding to the marine cementitious materials one by one; and annotate the historical application features of the corresponding marine cementitious materials with the durability results, and combine all the annotated historical application features to obtain a training data set; The training module 43 is used to train the classifier model using the training data set to obtain a durability evaluation model.

[0161] In some embodiments of the present invention, optionally, The acquisition module 41 is further used to extract features from the historical application data using a feature extraction model to obtain historical primary features; The acquisition module 41 is further configured to use a feature dimension reduction model to perform feature dimension reduction processing on the historical primary features, so as to use the historical primary features after the feature dimension reduction processing as the historical application features.

[0162] In some embodiments of the present invention, optionally, the classifier model is a classifier model of an extreme learning machine algorithm based on orthogonal tensor decomposition.

[0163] The training device for the durability evaluation model of marine engineering cementitious materials provided in the embodiment of the present application can achieve Figure 3 The various processes implemented by the method embodiment and achieving the same technical effect are not described here to avoid repetition.

[0164] An embodiment of the present invention provides an electronic device 50, see Fig.26 As shown, Fig.26 The electronic device 50 of an embodiment of the present invention includes a processor 51, a memory 52, and a program or instruction stored in the memory 52 and executable on the processor 51. When the program or instruction is executed by the processor, the steps of any method for evaluating the durability of marine cementitious materials of the present invention are implemented, or the steps of any method for training a durability evaluation model for marine cementitious materials of the present invention are implemented.

[0165] An embodiment of the present invention provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, the various processes of an embodiment of a method for evaluating the durability of marine cementitious materials as described above are implemented, or the various processes of an embodiment of a method for training a durability evaluation model for marine cementitious materials as described above are implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0166] The readable storage medium is, for example, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc. In some examples, the readable storage medium may be a non-transitory readable storage medium.

[0167] An embodiment of the present invention also provides a computer program product, including computer instructions, which, when executed by a processor, implement the various processes of any of the above-mentioned embodiments of the method for evaluating the durability of marine cementitious materials, or implement the various processes of any of the above-mentioned embodiments of the method for training a durability evaluation model for marine cementitious materials, and can achieve the same technical effect. To avoid repetition, they are not described here.

[0168] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the enlightenment of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the present invention and the claims, all of which are within the protection of the present invention.

Claims

1. A method for evaluating the durability of marine cementitious materials, characterized in that: include: Receiving application data of the marine cementitious material to be evaluated sent by the user through the interactive terminal, the application data including: production process data, batch data and application environment data; Extracting features from the application data to obtain application features; The application characteristics are input into a durability evaluation model to obtain a durability evaluation result of the marine engineering cementitious material.

2. The method for evaluating the durability of marine cementitious materials according to claim 1, characterized in that: Extracting features from the application data to obtain application features includes: Using a feature extraction model to extract features from the application data to obtain primary features; A feature dimension reduction model is used to perform feature dimension reduction processing on the primary features, so that the primary features after the feature dimension reduction processing are used as the application features.

3. A method for training a durability assessment model for marine cementitious materials, characterized in that: include: Acquire historical application data of each marine cementitious material in the marine cementitious material set, wherein the historical application data includes: production process data, batch data, and application environment data; extract features from the historical application data to obtain historical application features; Obtaining durability results corresponding to the marine cementitious materials one by one; and annotating the historical application features of the corresponding marine cementitious materials with the durability results, and combining all the annotated historical application features to obtain a training data set; The classifier model is trained using the training data set to obtain a durability evaluation model.

4. The method for training a durability assessment model for marine cementitious materials according to claim 3, characterized in that: Extracting features from the historical application data to obtain historical application features includes: Using a feature extraction model to extract features from the historical application data to obtain historical primary features; The feature dimension reduction model is used to perform feature dimension reduction processing on the historical primary features, so as to use the historical primary features after the feature dimension reduction processing as the historical application features.

5. The method for training a durability assessment model for marine cementitious materials according to claim 3, characterized in that: The classifier model is a classifier model of an extreme learning machine algorithm based on orthogonal tensor decomposition.

6. A durability assessment device for marine cementitious materials, characterized in that: include: A receiving module, used to receive application data of the marine cementitious material to be evaluated sent by the user through the interactive terminal, wherein the application data includes: production process data, batch data and application environment data; A feature extraction module, used to extract features from the application data to obtain application features; The evaluation module is used to input the application characteristics into the durability evaluation model to obtain the durability evaluation result of the marine engineering cementitious material.

7. A training device for a durability assessment model of marine cementitious materials, characterized in that: include: An acquisition module is used to acquire historical application data of each marine cementitious material in the marine cementitious material set, wherein the historical application data includes: production process data, batch data, and application environment data; extract features from the historical application data to obtain historical application features; An execution module is used to obtain durability results corresponding to the marine cementitious materials one by one; and annotate the historical application features of the corresponding marine cementitious materials with the durability results, and combine all the annotated historical application features to obtain a training data set; The training module is used to train the classifier model using the training data set to obtain a durability evaluation model.

8. An electronic device, characterized in that: It comprises a processor, a memory, and a program or instruction stored in the memory and executable on the processor, wherein when the program or instruction is executed by the processor, the steps in the method for durability assessment of marine cementitious materials as described in any one of claims 1 to 2 are implemented; or, the steps in the method for training a durability assessment model for marine cementitious materials as described in any one of claims 3 to 5 are implemented.

9. A readable storage medium, characterized in that: The readable storage medium stores programs or instructions, and when the programs or instructions are executed by the processor, the steps in the durability assessment method of marine cementitious materials as described in any one of claims 1 to 2 are implemented; or, the steps in the training method of the durability assessment model of marine cementitious materials as described in any one of claims 3 to 5 are implemented.

10. A computer program product, characterized in that The method comprises computer instructions, which, when executed by a processor, implement the steps in the method for durability assessment of marine cementitious materials as described in any one of claims 1 to 2; or, implement the steps in the method for training a durability assessment model for marine cementitious materials as described in any one of claims 3 to 5.

Images