State prediction-oriented all-solid waste pervious concrete detection data processing method

Abstract

The invention relates to the technical field of data processing, in particular to a state prediction-oriented full-solid waste concrete detection data processing method, which adopts a generative adversarial network algorithm based on adversarial enhancement to carry out data expansion, generates a new sample with trueness, and solves the problem of poor model generalization ability caused by insufficient training data. A neural network algorithm of dynamic group evolution optimization is adopted for feature extraction, the problems of gradient disappearance, explosion and the like of a traditional neural network are effectively solved, accurate feature capture of high-dimensional data is achieved, and the model generalization ability and the training efficiency are improved; feature dimension reduction is carried out through the self-encoding neural network based on feature refinement, important data information is effectively reserved, the data dimension is reduced, and the subsequent classifier processing efficiency is improved; and a support vector machine algorithm of quantum phase space mapping is adopted to optimize kernel function parameters, so that the model is more flexible when processing nonlinear features, and the classification precision and the adaptability of decision boundaries are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a method for processing detection data of all-solid waste concrete oriented to state prediction. Background Art

[0002] With the acceleration of urbanization, rainwater pipes and flood control issues have become increasingly prominent, and permeable concrete has received widespread attention due to its excellent permeability. All-solid waste permeable concrete not only helps to improve urban drainage systems, but also can minimize the use of solid waste resources and reduce environmental pollution. Therefore, the detection of concrete is very important. Traditional detection and evaluation methods usually rely on small-scale test data, which makes it difficult to accurately predict the properties of concrete, such as compressive strength and permeability, in practical applications. This problem of insufficient data limits the generalization ability of the model and reduces the reliability of the prediction results. In addition, existing feature extraction and classification techniques are inefficient when processing high-dimensional data, which can easily lead to information loss.

[0003] The existing patent CN202311722685.6 proposes a method for predicting the working performance of fresh concrete based on big data, including the following steps: statistically analyzing the concrete mix ratio data of historical engineering projects and establishing a concrete working performance information database; establishing a solution equation for the excess amount of concrete mortar and the excess amount of net paste, and calculating the excess amount of net paste and the excess amount of mortar in the concrete working performance information database; conducting machine learning training on the variables affecting the working performance of concrete, establishing a mapping model of the excess amount of mortar, the excess amount of net paste, the amount of water used in concrete, the water-cement ratio, and the sand ratio to the slump of concrete and the slump solution expression SV2; solving the slump prediction value according to the concrete mix ratio information to be adopted and the expression SV2. The present invention can provide concrete with the required working performance for project construction under the premise of ensuring the quality of concrete, while reducing the workload of on-site test and inspection personnel and improving the quality and efficiency of engineering construction.

[0004] The existing patent CN202110718615.8 proposes a method for predicting the mechanical properties of ultra-high performance concrete. First, a micro-scale UHPC hydration microstructure model and a cement-fly ash-silica fume three-phase composite cement-based material hydration microstructure evolution model are established, and a corresponding UHPC finite element numerical model at the microscale is established to determine the mechanical properties of the UHPC hardened slurry at the microscale. Secondly, a UHPC aggregate stacking model and a UHPC finite element numerical model at the mesoscale are established. According to the mechanical properties of the UHPC hardened net slurry and the three-dimensional spatial distribution of aggregates, the mechanical properties of the UHPC mortar are analyzed and calculated; finally, a random dynamic stacking model of UHPC fibers is established, and a corresponding UHPC finite element numerical model at the mesoscale is established. According to the constitutive parameters of the UHPC mortar material and the three-dimensional spatial distribution of fibers, the macroscopic mechanical constitutive relationship of UHPC is calculated. The present invention has accurate calculation and high stability.

[0005] The existing patent CN202410423447.3 proposes a portable cold region concrete detection device based on machine learning, the device includes an instrument body, a measurement module, a processing module, a power supply, an electrode, an intelligent display screen, a USB interface and a connection slot, wherein: the measurement module is located on the right side of the instrument body and is connected to the instrument body through the connection slot; the upper part of the instrument body is provided with an intelligent display screen, and the lower part is provided with a power supply; the processing module is located on the rear side of the intelligent display screen; the electrode is located on the front side of the power supply; the USB interface is provided below the electrode. The present invention takes the surface resistivity affected by the moisture content of concrete as a characteristic value, establishes a concrete compressive strength prediction model through regression fitting, thereby predicting the compressive strength of engineering concrete, and can further improve the accuracy while performing non-destructive testing.

[0006] Although the above-mentioned prior art can accurately predict the performance of concrete, it still has the following shortcomings:

[0007] 1. In the process of data collection and labeling, the number of samples is often insufficient, which limits the generalization ability of the training model and increases the risk of overfitting;

[0008] 2. When processing high-dimensional data, traditional feature extraction methods are prone to gradient vanishing or exploding, which affects the stability and performance of the model and leads to low training efficiency;

[0009] 3. Feature dimensionality reduction technology fails to effectively retain important information of the data, resulting in reduced accuracy of the model in the classification stage and failure to fully utilize the potential value of the input data;

[0010] 4. The existing support vector machine methods lack flexibility in dealing with nonlinear features and are difficult to adapt to the complexity of high-dimensional data, which limits the recognition ability of the classifier and the adaptability of the decision boundary. Summary of the invention

[0011] In order to solve the above-mentioned technical problems, the present invention provides a data processing method for detecting all-solid waste permeable concrete for state prediction, so as to solve the problems of insufficient sample quantity, unstable feature processing and insufficient classification accuracy in the prior art through improved data expansion, feature extraction, feature dimension reduction and classification technology, so as to improve the accuracy and reliability of concrete state prediction and realize comprehensive evaluation and quality control of all-solid waste permeable concrete.

[0012] To achieve this technical purpose, the present invention adopts the following scheme:

[0013] A method for processing detection data of all-solid waste concrete for state prediction is carried out in the following steps:

[0014] S1. Data collection and annotation

[0015] The collection and labeling of all-solid waste permeable concrete test data to obtain the original data set;

[0016] S2. Data expansion

[0017] The adversarial enhancement-based generative adversarial network algorithm generates samples to achieve data expansion. The adversarial enhancement-based generative adversarial network consists of a generator and a discriminator. The generator is used to generate new samples, and the discriminator is responsible for evaluating the authenticity of the generated samples. New virtual samples are generated through adversarial sample enhancement technology.

[0018] S3. Feature extraction model training

[0019] A neural network algorithm based on dynamic population evolution optimization is used as a feature extraction model, and the expanded data is input into the feature extraction model for training the feature extraction model;

[0020] S4. Feature dimensionality reduction model training

[0021] The autoencoder neural network based on feature refinement is used as a dimensionality reduction model, and the data after feature extraction is input into the feature dimensionality reduction model for training the feature dimensionality reduction model;

[0022] S5. Classifier model training

[0023] The support vector machine algorithm based on quantum phase space mapping is used as the classification algorithm. The data after feature extraction is input into the classifier to train the classifier model. This method uses feature coupling technology to optimize the parameters of the kernel function and enhance the adaptability of the model to high-dimensional data.

[0024] The training process of the support vector machine algorithm based on quantum phase space mapping is as follows:

[0025] S501, initialize the weight parameters and high-order kernel function parameters of the vector machine, expressed as:

[0026] ;

[0027] ;

[0028] In the formula, is the weight of the support vector machine, is the initial weight vector of the classifier, is the initial parameter of the high-order kernel function;

[0029] S502: The features of the training samples after dimensionality reduction The model is passed in and the similarity matrix is ​​calculated based on the high-order kernel function, expressed as:

[0030] ;

[0031] In the formula, For the Samples and The similarity between samples, is the feature mapping function, is the total number of samples, For the samples, For the samples;

[0032] The calculation method of the feature mapping function in the feature input is expressed as:

[0033] ;

[0034] In the formula, is the width of the Gaussian kernel, is a power, is the frequency of the sine function;

[0035] S503, mapping the similarity matrix to the quantum state space, generating the corresponding quantum feature map using the quantum superposition principle, and constructing the quantum state, which is expressed as:

[0036] ;

[0037] In the formula, is a quantum state, For samples The quantum amplitude of the correlation;

[0038] S504. In the quantum feature space, a decision boundary of a support vector machine is constructed, and the selection of support vectors is adjusted according to the training samples. The method of calculating the decision boundary is expressed as:

[0039] ;

[0040] In the formula, is the decision function, is the bias term of the support vector machine;

[0041] The calculation method of the bias term of the support vector machine is expressed as:

[0042] ;

[0043] In the formula, For the The true labels of samples, is the quantum state corresponding to the i-th sample;

[0044] S505, using the improved loss function, minimizing the classification error through the gradient descent method, and updating the weight parameters. The calculation method of the loss function is expressed as:

[0045] ;

[0046] In the formula, is the loss function of the support vector machine, is the quantum state corresponding to the i-th sample, For the The true labels of samples, is the regularization parameter of the support vector machine, represents the L2 norm;

[0047] The calculation method of the L2 norm of the support vector machine weight is expressed as:

[0048] ;

[0049] In the formula, is the feature dimension input to the support vector machine, is the weight vector Quantity;

[0050] S506. According to the feedback of the loss function, the parameters of the high-order kernel function are dynamically adjusted to optimize the performance of the kernel function. The updating method of the high-order kernel function is expressed as:

[0051] ;

[0052] In the formula, is the learning rate of the support vector machine, Indicates The parameters of the kernel function for the iteration, Indicates The parameters of the kernel function for the iteration, is the gradient of the loss function of the support vector machine with respect to the parameters of the high-order kernel function;

[0053] Furthermore, the gradient calculation method of the loss function with respect to the parameters of the high-order kernel function is expressed as:

[0054] ;

[0055] In the formula, is the derivative of the similarity matrix with respect to the high-order kernel parameters, indicating how the similarity is affected by parameter changes; is the maximum value function;

[0056] Furthermore, during the iterative training process, the method of updating the support vector machine weights is expressed as:

[0057] ;

[0058] In the formula, For the The weights of the support vector machine for the iteration, For the The weight of the support vector machine for the iteration;

[0059] Furthermore, the gradient calculation method of the loss function with respect to the weight is expressed as:

[0060] ;

[0061] In the formula, is the indicator function, indicating that when 1 if the prediction of a sample is incorrect, 0 otherwise;

[0062] S507, repeating the above steps until the preset stop iteration condition is met, indicating that the model training is completed;

[0063] S6. Concrete sample inspection

[0064] The trained model is used to test concrete samples.

[0065] Compared with the prior art, the present invention has the following beneficial effects:

[0066] This method uses a generative adversarial network algorithm based on adversarial enhancement for data expansion. Through the interaction between the generator and the discriminator, new samples with authenticity are generated, which solves the problem of poor model generalization ability caused by insufficient training data; a neural network algorithm based on dynamic population evolution optimization is used for feature extraction, which effectively addresses the problems of gradient vanishing and explosion in traditional neural networks, realizes accurate feature capture of high-dimensional data, and improves the generalization ability and training efficiency of the model; an autoencoder neural network based on feature refinement is used for feature dimensionality reduction, which effectively retains important data information, reduces data dimensions, and improves the processing efficiency of subsequent classifiers; a support vector machine algorithm based on quantum phase space mapping is used to optimize the kernel function parameters, making the model more flexible in processing nonlinear features, and improving classification accuracy and the adaptability of decision boundaries.

[0067] Further, the preferred embodiment of the present invention is:

[0068] The sources of data collected in step S1 mainly include laboratory tests, on-site construction monitoring and historical inspection records. The collected data is stored in a standardized structured format, specifically in CSV or database format; the collected data attributes include

[0069] Compressive strength reflects the ability of concrete to resist pressure, and the unit is MPa;

[0070] Permeability, which indicates the penetration rate of water in concrete, is measured in m / s;

[0071] Porosity refers to the ratio of void volume in concrete to total volume, with the unit of %;

[0072] Moisture content refers to the moisture content in concrete, with the unit being %;

[0073] Temperature: the ambient temperature during collection, in °C;

[0074] Humidity: the ambient humidity during collection, in %;

[0075] Curing time, the time from concrete pouring to testing, in days;

[0076] Solid waste ratio refers to the proportion of solid waste in concrete, in units of %;

[0077] Specimen size, indicating the length, width and height of the specimen, in mm;

[0078] Construction method, the construction technology used is identified by characters;

[0079] In step S1, the collected data is labeled, and the labeling categories include:

[0080] Qualified: It means that the concrete meets the relevant standards;

[0081] Unqualified: means the concrete does not meet the relevant standards;

[0082] To be tested: indicates samples that require further testing.

[0083] The training process of the adversarial enhancement-based generative adversarial network algorithm in step S2 is as follows:

[0084] S201, initialize and set the parameters of the generator and discriminator, including weights and biases. The bias is initialized to 0. The weight initialization method is expressed as:

[0085] ;

[0086] In the formula, represents the initial weight of the generator, represents the initial weight of the discriminator, The mean is 0 and the variance is The normal distribution of represents a normal distribution, Indicates compliance with a specific distribution;

[0087] S202, input the original data set into the generator to generate a certain number of virtual samples, expressed as:

[0088] ;

[0089] In the formula, For the generated samples, represents a generator function, is a noise vector randomly drawn from the latent space, Indicates The weight of the generator for the iteration;

[0090] S203: Input the generated sample into the discriminator for evaluation to obtain the authenticity score of the generated sample, which is expressed as:

[0091] ;

[0092] In the formula, is the score of the discriminator on the generated samples, represents the discriminator function, Indicates The weight of the discriminator at the iteration;

[0093] S204, adjusting the parameters of the discriminator through the optimization algorithm to improve its ability to distinguish the generated samples. For the optimization of the discriminator, the calculation method of its loss function is expressed as:

[0094] ;

[0095] In the formula, is the discriminator loss, is the real data distribution, To generate the sample distribution, For expectations, is real data;

[0096] S205: Feedback adjustment to the generator: According to the evaluation result of the discriminator, feedback is given to the generator to adjust the parameters of the generator so that the samples generated by it are closer to the real samples in terms of authenticity. The calculation method of the loss function is expressed as:

[0097] ;

[0098] In the formula, is the generator loss;

[0099] S206. Perform a small perturbation on the generated sample to obtain richer samples and achieve adversarial sample enhancement, which is expressed as:

[0100] ;

[0101] In the formula, To enhance the samples, is the disturbance amplitude, represents the gradient of the discriminator to the generated sample, is a symbolic function;

[0102] The calculation method of the gradient of the discriminator to the generated sample is expressed as:

[0103] ;

[0104] In the formula, The calculation method is:

[0105] ;

[0106] S207. Update the parameters of the generator and the discriminator according to the loss function. The updating method is expressed as:

[0107] ;

[0108] ;

[0109] In the formula, and are the learning rates of the discriminator and generator, respectively. For the The discriminator weights for the iterations, For the The generator weights for the iterations;

[0110] S208, repeating the above steps until the preset stop iteration condition is met, indicating that the data expansion model training is completed;

[0111] After the data expansion model training is completed, the trained data expansion model is used to increase the number of samples.

[0112] In step S3, the specific training process of the neural network algorithm based on dynamic population evolution optimization as the feature extraction model for the extended data is as follows:

[0113] S301. According to the bionic algorithm initialization method, in the initialization stage, an initial population is generated, and each individual represents a configuration of network weights. Specifically, the population size is , for The individual initialization weights and biases are expressed as:

[0114] ;

[0115] ;

[0116] In the formula, For the The weight of the neural network corresponding to each individual, For the The bias of the neural network corresponding to each individual, Indicates The weight matrix of each individual in the initial state; Indicates The bias of each individual in the initial state; represents the initialization variance; It means the mean is 0 and the variance is Normal distribution of is a normal distribution;

[0117] S302: For each individual in the population, use its corresponding neural network configuration to process the input training data, calculate the output of the model, and evaluate its performance according to a predetermined loss function. Specifically, for the first Individuals use their weights and biases to calculate the loss on the training data set, expressed as:

[0118] ;

[0119] In the formula, For the The loss of the neural network corresponding to each individual; Indicates the number of samples input in the current batch; represents the composite loss function, Represents a neural network model function; Indicates Characteristics of the samples; Indicates The labels of the samples;

[0120] The composite loss function contains a regularization term, which can increase the generalization ability of the model. The calculation method is expressed as: ;

[0121] In the formula, is the mean square error function, is the regularization term, is the regularization parameter;

[0122] Furthermore, the calculation method of the regularization term is expressed as:

[0123] ;

[0124] In the formula, For the first The neural network corresponding to each individual weights;

[0125] S303. According to the fitness of the individual, the best performing individual is selected from the current population and retained as the candidate solution for the next generation. Specifically, based on the fitness of the individual, the excellent individual has a higher probability of being selected. The calculation method of the probability of the individual being selected is expressed as:

[0126] ;

[0127] In the formula, Indicates The probability of an individual being selected; is the parameter that controls the selection pressure; For the The loss of the neural network corresponding to each individual;

[0128] S304. Generate new individuals through crossover and mutation operations. The crossover operation allows two excellent individuals to exchange some genes to generate new offspring; the mutation operation randomly changes some genes in an individual to increase the diversity of the population. Specifically, the crossover operation randomly selects two individuals for gene exchange, which is expressed as:

[0129] ;

[0130] ;

[0131] In the formula, is the crossover rate, is the weight of the neural network corresponding to the first individual selected, is the bias of the neural network corresponding to the first individual selected, is the weight of the neural network corresponding to the second individual selected, is the bias of the neural network corresponding to the second individual selected, is the weight of the neural network corresponding to the individual after the crossover operation, is the bias of the neural network corresponding to the individual after the crossover operation;

[0132] Moreover, the mutation operation performs a small random perturbation on the newly generated individual weights, which can be expressed as:

[0133] ;

[0134] ;

[0135] In the formula, represents the variance of the variation, is the weight of the neural network corresponding to the individual after the mutation operation, is the bias of the neural network corresponding to the individual after the mutation operation;

[0136] S305, repeat the above steps until the preset stop iteration condition is met, which means the model training is completed.

[0137] The autoencoder neural network based on feature refinement in step S4 is composed of three parts: an encoder, a decoder, and a feature adjustment module. The encoder is responsible for mapping high-dimensional input data to a low-dimensional feature space, and the decoder is used to reconstruct the reduced-dimensional features back to the original space to ensure the reversibility of the dimensionality reduction process; the adjustment module dynamically adjusts the reduced-dimensional feature space through recursive feature adaptive optimization, so that important features can be strengthened and minor features are gradually weakened, so that the reduced-dimensional feature representation has better simplicity while retaining data information;

[0138] The training process of the autoencoder neural network algorithm based on feature refinement is as follows:

[0139] S401. First, assume that the data input to the autoencoder neural network is , the encoder uses a multi-layer nonlinear mapping structure to map high-dimensional data to the initial low-dimensional feature space, which can be expressed as:

[0140] ;

[0141] In the formula, represents the initial low-dimensional features, is the weight matrix of the encoder, is the encoder bias vector, is the multi-layer Sigmoid activation function of the encoder;

[0142] S402, after the low-dimensional features are generated, the feature adjustment module automatically generates feature weights according to the feature importance in the current feature space. When the module is initialized, the same initial weights are assigned to all features so that they can be gradually adjusted according to feature contributions in subsequent steps. The calculation method of the initial weight matrix is ​​expressed as:

[0143] ;

[0144] In the formula, is the initial weight matrix; is the feature weight vector, Each element in Initialized to the same value, indicating that all features have the same importance in the initial stage; is a function for extracting the diagonal elements of a matrix;

[0145] Furthermore, the adjusted features can be expressed as:

[0146] ;

[0147] In the formula, is the feature representation after feature weight adjustment;

[0148] S403, the feature adjustment module recursively optimizes the initially generated low-dimensional features. In each round of iteration, the module adjusts the weights of each feature according to the performance of the previous round of features, gradually enhances the features with important influence, and gradually weakens the redundant or noise features. In the round iteration, the weight update rule is as follows:

[0149] ;

[0150] In the formula, Indicates The feature weights of the round iteration, Indicates The feature weights of the round iteration, is the learning rate of the autoencoder neural network, is the loss function of the autoencoder neural network, is the label data, Represents the gradient of the loss function with respect to the feature weight;

[0151] S404, in order to ensure the effectiveness of the dimensionality reduction process, the decoder remaps the low-dimensional features back to the high-dimensional space to ensure that the dimensionality reduction process does not lose important information. The reconstruction process of the decoder is expressed as:

[0152] ;

[0153] In the formula, is the reconstructed high-dimensional data, is the weight matrix of the decoder, is the bias vector of the decoder, is the multi-layer Sigmoid activation function of the decoder;

[0154] S405, repeat the above steps until the preset stop iteration condition is met, which means the model training is completed. BRIEF DESCRIPTION OF THE DRAWINGS

[0155] Figure 1 It is a schematic diagram of the process of the present invention;

[0156] Figure 2 It is a schematic diagram of the model training process and model practical process of the present invention;

[0157] Figure 3 A flowchart of the neural network algorithm training based on dynamic population optimization in step S3;

[0158] Figure 4 is a flowchart of the training of the autoencoder neural network algorithm based on feature refinement in step S4; DETAILED DESCRIPTION

[0159] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are described in detail below in conjunction with the accompanying drawings. In the following description, many specific details are set forth to facilitate a full understanding of the present application. However, the present application can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without violating the connotation of the present application, so the present application is not limited by the specific embodiments disclosed below.

[0160] A method for processing detection data of all-solid waste concrete for state prediction is carried out in the following steps:

[0161] S1. Data collection and annotation

[0162] The data collection and labeling of all-solid waste permeable concrete testing data includes laboratory testing, on-site construction monitoring and historical inspection records. The collected data is stored in a standardized structured format, specifically in CSV or database format;

[0163] During specific collection, the collected data attributes mainly include:

[0164] Compressive strength (Ra1), reflects the ability of concrete to resist pressure, unit is MPa;

[0165] Permeability (Ra2), which indicates the penetration rate of water in concrete, in m / s;

[0166] Porosity (Ra3) indicates the ratio of void volume in concrete to total volume, expressed in %;

[0167] Moisture content (Ra4) indicates the moisture content in concrete, in units of %;

[0168] Temperature (Ra5), ambient temperature during acquisition, in °C;

[0169] Humidity (Ra6), the ambient humidity during collection, unit is %

[0170] Curing time (Ra7), the time from concrete pouring to testing, in days;

[0171] Solid waste ratio (Ra8) indicates the proportion of solid waste in concrete, in units of %;

[0172] Specimen size (Ra9), which indicates the length, width and height of the specimen, in mm;

[0173] Construction method (Ra10), the construction technology used, such as pouring, vibrating, etc., is identified by characters;

[0174] The data attributes collected above are only one of the data formats and types. In actual applications, the data attributes usually exceed 10 attributes, and the number of data attributes may reach dozens or even hundreds;

[0175] The collected data is labeled. The labeling method of this embodiment is manual labeling. The labeling categories include:

[0176] Qualified: It means that the concrete meets the relevant standards;

[0177] Unqualified: means the concrete does not meet the relevant standards;

[0178] To be tested: indicates samples that require further testing.

[0179] S2. Data expansion

[0180] The collection, acquisition, labeling and preprocessing of training data are time-consuming and labor-intensive. Insufficient training samples can easily lead to poor generalization ability of the model and affect the accuracy of the model. Therefore, an adversarial enhanced generative adversarial network algorithm is used to generate samples and then achieve data expansion. The adversarial enhanced generative adversarial network consists of a generator and a discriminator. The generator is used to generate new samples, and the discriminator is responsible for evaluating the authenticity of the generated samples. Based on the generated samples, new virtual samples are generated through adversarial sample enhancement technology to enhance the diversity of the data set, thereby improving the quality of the generated samples.

[0181] The training process of the generative adversarial network algorithm based on adversarial enhancement is as follows:

[0182] S201, initializing and setting the parameters of the generator and the discriminator, including weights and biases. In one embodiment, the bias is initialized to 0, and the weight initialization method is expressed as:

[0183] ;

[0184] In the formula, represents the initial weight of the generator, represents the initial weight of the discriminator, The mean is 0 and the variance is The normal distribution of represents a normal distribution, Indicates that it obeys a specific distribution.

[0185] S202, input the original data set into the generator to generate a certain number of virtual samples, expressed as:

[0186] ;

[0187] In the formula, For the generated samples, represents a generator function, is a noise vector randomly drawn from the latent space, Indicates The weight of the generator for this iteration.

[0188] S203: Input the generated sample into the discriminator for evaluation to obtain the authenticity score of the generated sample, which is expressed as:

[0189] ;

[0190] In the formula, is the score of the discriminator on the generated samples, represents the discriminator function, Indicates The weights of the discriminator at this iteration.

[0191] S204, adjusting the parameters of the discriminator through the optimization algorithm to improve its ability to distinguish the generated samples. For the optimization of the discriminator, the calculation method of its loss function is expressed as:

[0192] ;

[0193] In the formula, is the discriminator loss, is the real data distribution, To generate the sample distribution, For expectations, For real data.

[0194] S205: Feedback adjustment to the generator: According to the evaluation result of the discriminator, feedback is given to the generator to adjust the parameters of the generator so that the samples generated by it are closer to the real samples in terms of authenticity. The calculation method of the loss function is expressed as:

[0195] ;

[0196] In the formula, is the generator loss.

[0197] S206. Perform a small perturbation on the generated sample to obtain richer samples and achieve adversarial sample enhancement, which is expressed as:

[0198] ;

[0199] In the formula, To enhance the samples, is the disturbance amplitude, represents the gradient of the discriminator to the generated sample, is a symbolic function;

[0200] The calculation method of the gradient of the discriminator to the generated sample is expressed as:

[0201] ;

[0202] In the formula, The calculation method is:

[0203] .

[0204] S207. Update the parameters of the generator and the discriminator according to the loss function. The updating method is expressed as:

[0205] ;

[0206] ;

[0207] In the formula, and are the learning rates of the discriminator and generator, respectively. For the The discriminator weights for the iterations, For the The generator weight of the iteration, preferably, and Both are set to 0.01.

[0208] S208, repeat the above steps until the preset stop iteration condition is met, which means that the data expansion model training is completed. 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.

[0209] After the data augmentation model training is completed, the trained data augmentation model is used to increase the number of samples. For example, assuming that the original collected samples are 800, the data augmentation model generates 200 samples by augmentation, and the augmented data set contains 1000 samples.

[0210] S3. Feature extraction model training

[0211] The expanded data is input into the feature extraction model for training the feature extraction model, and a 6-layer fully connected neural network is used for feature extraction. Some solutions in the prior art use neural networks for feature extraction. In some 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. In this embodiment, a neural network algorithm based on dynamic population evolution optimization is used as a feature extraction model to replace the method based on fixed rules for the evolution of all individuals in the traditional population evolution algorithm. The self-correction mechanism is used to automatically adjust the evolution rules according to the characteristics of the current training data, and the probability of crossover and mutation is adjusted according to the changing trend of the loss function, so that the algorithm can adapt to different data distributions more flexibly and improve the generalization ability and training efficiency of the model.

[0212] Specifically, Figure 3 As shown in the figure, the training process of the neural network algorithm based on dynamic population evolution optimization is as follows:

[0213] S301. According to the bionic algorithm initialization method, in the initialization stage, an initial population is generated, and each individual represents a configuration of network weights. Specifically, the population size is , for The individual initialization weights and biases are expressed as:

[0214] ;

[0215] ;

[0216] In the formula, For the The weight of the neural network corresponding to each individual, For the The bias of the neural network corresponding to each individual, Indicates The weight matrix of each individual in the initial state; Indicates The bias of each individual in the initial state; represents the initialization variance; It means the mean is 0 and the variance is Normal distribution of For normal distribution, preferred, Set to 0.01.

[0217] S302: For each individual in the population, use its corresponding neural network configuration to process the input training data, calculate the output of the model, and evaluate its performance according to a predetermined loss function. Specifically, for the first Individuals use their weights and biases to calculate the loss on the training data set, expressed as:

[0218] ;

[0219] In the formula, For the The loss of the neural network corresponding to each individual; Indicates the number of samples input in the current batch; represents the composite loss function, Represents a neural network model function; Indicates Characteristics of the samples; Indicates The labels of samples.

[0220] The composite loss function contains a regularization term, which can increase the generalization ability of the model. The calculation method is expressed as:

[0221] In the formula, is the mean square error function, is the regularization term, is the regularization parameter.

[0222] Furthermore, the calculation method of the regularization term is expressed as:

[0223] ;

[0224] In the formula, For the first The neural network corresponding to each individual A weight.

[0225] S303. According to the fitness of the individual, the best performing individual is selected from the current population and retained as the candidate solution for the next generation. Specifically, based on the fitness of the individual, the excellent individual has a higher probability of being selected. The calculation method of the probability of the individual being selected is expressed as:

[0226] ;

[0227] In the formula, Indicates The probability of an individual being selected; is the parameter that controls the selection pressure; For the The loss of the neural network corresponding to each individual; preferably, Set to 2.

[0228] S304. Generate new individuals through crossover and mutation operations. The crossover operation allows two excellent individuals to exchange some genes to generate new offspring; the mutation operation randomly changes some genes in an individual to increase the diversity of the population. Specifically, the crossover operation randomly selects two individuals for gene exchange, which is expressed as:

[0229] ;

[0230] ;

[0231] In the formula, is the crossover rate, is the weight of the neural network corresponding to the first individual selected, is the bias of the neural network corresponding to the first individual selected, is the weight of the neural network corresponding to the second individual selected, is the bias of the neural network corresponding to the second individual selected, is the weight of the neural network corresponding to the individual after the crossover operation, is the bias of the neural network corresponding to the individual after the crossover operation; preferably, Set to 0.3.

[0232] Moreover, the mutation operation performs a small random perturbation on the newly generated individual weights, which can be expressed as:

[0233] ;

[0234] ;

[0235] In the formula, represents the variance of the variation, is the weight of the neural network corresponding to the individual after the mutation operation, is the bias of the neural network corresponding to the individual after the mutation operation; preferably, Set to 0.04.

[0236] S305, repeating the above steps until a preset stop iteration condition is met, indicating that the model training is completed, wherein 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.

[0237] S4. Feature dimensionality reduction model training

[0238] The data after feature extraction is input into the feature dimensionality reduction model to train the feature dimensionality reduction model. This embodiment adopts an autoencoding neural network based on feature refinement as the dimensionality reduction model, wherein the autoencoding neural network based on feature refinement consists of three parts: an encoder, a decoder, and a feature adjustment module. The encoder is responsible for mapping the high-dimensional input data to a low-dimensional feature space, and the decoder is used to reconstruct the reduced-dimensional features back to the original space to ensure the reversibility of the dimensionality reduction process; the feature adjustment module dynamically adjusts the reduced-dimensional feature space through recursive feature adaptive optimization, so that important features can be strengthened and secondary features are gradually weakened, so that the reduced-dimensional feature representation has better simplicity while retaining data information.

[0239] Specifically, Figure 4 As shown in the figure, the training process of the autoencoder neural network algorithm based on feature refinement is as follows:

[0240] S401. First, assume that the data input to the autoencoder neural network is , the encoder uses a multi-layer nonlinear mapping structure to map high-dimensional data to the initial low-dimensional feature space, which can be expressed as:

[0241] ;

[0242] In the formula, represents the initial low-dimensional features, is the weight matrix of the encoder, is the encoder bias vector, is the multi-layer Sigmoid activation function of the encoder.

[0243] S402, after the low-dimensional features are generated, the feature adjustment module automatically generates feature weights according to the feature importance in the current feature space. When the module is initialized, the same initial weights are assigned to all features so that they can be gradually adjusted according to feature contributions in subsequent steps. The calculation method of the initial weight matrix is ​​expressed as:

[0244] ;

[0245] In the formula, is the initial weight matrix; is the feature weight vector, Each element in Initialized to the same value, indicating that all features have the same importance in the initial stage; is a function for extracting the diagonal elements of a matrix;

[0246] Furthermore, the adjusted features can be expressed as:

[0247] ;

[0248] In the formula, is the feature representation after feature weight adjustment.

[0249] S403, the feature adjustment module recursively optimizes the initially generated low-dimensional features. In each round of iteration, the module adjusts the weights of each feature according to the performance of the previous round of features, gradually enhances the features with important influence, and gradually weakens the redundant or noise features. In the round iteration, the weight update rule is as follows:

[0250] ;

[0251] In the formula, Indicates The feature weights of the round iteration, Indicates The feature weights of the round iteration, is the learning rate of the autoencoder neural network, is the loss function of the autoencoder neural network, is the label data, represents the gradient of the loss function with respect to the feature weight; preferably, the loss function of the autoencoder neural network is the reconstructed error loss function, Set to 0.01.

[0252] S404, in order to ensure the effectiveness of the dimensionality reduction process, the decoder remaps the low-dimensional features back to the high-dimensional space to ensure that the dimensionality reduction process does not lose important information. The reconstruction process of the decoder is expressed as:

[0253] ;

[0254] In the formula, is the reconstructed high-dimensional data, is the weight matrix of the decoder, is the bias vector of the decoder, is the multi-layer Sigmoid activation function of the decoder.

[0255] S405, repeat the above steps until the preset stop iteration condition is met, which means that the model training is completed; the preset stop iteration condition is reaching a preset maximum number of iterations, preferably, the preset maximum number of iterations is 1000 times.

[0256] S5. Classifier model training

[0257] The classifier model training adopts the support vector calculation method based on quantum phase space mapping as the classification algorithm, and the data after feature extraction is input into the classifier to train the classifier model; this embodiment proposes an improved method of high-order kernel function to address the limitations of traditional support vector machines in processing traditional nonlinear features, and uses feature coupling technology to optimize the parameters of the kernel function to enhance the model's adaptability to high-dimensional data. In addition, the superposition and entanglement characteristics of quantum states are introduced to improve the flexibility and accuracy of the decision boundary.

[0258] Specifically, the training process of the support vector machine algorithm based on quantum phase space mapping is as follows:

[0259] S501, initialize the weight parameters and high-order kernel function parameters of the vector machine, expressed as:

[0260] ;

[0261] ;

[0262] In the formula, is the weight of the support vector machine, is the initial weight vector of the classifier, is the initial parameter of the high-order kernel function.

[0263] S502: The features of the training samples after dimensionality reduction The model is passed in and the similarity matrix is ​​calculated based on the high-order kernel function, expressed as:

[0264] ;

[0265] In the formula, For the Samples and The similarity between samples, is the feature mapping function, is the total number of samples, For the samples, For the samples;

[0266] The calculation method of the feature mapping function in the feature input is expressed as:

[0267] ;

[0268] In the formula, is the width of the Gaussian kernel, is a power, is the frequency of the sine function.

[0269] S503, mapping the similarity matrix to the quantum state space, generating the corresponding quantum feature map using the quantum superposition principle, and constructing the quantum state, which is expressed as:

[0270] ;

[0271] In the formula, is a quantum state, For samples The quantum amplitude of the correlation.

[0272] S504. In the quantum feature space, a decision boundary of a support vector machine is constructed, and the selection of support vectors is adjusted according to the training samples. The method of calculating the decision boundary is expressed as:

[0273] ;

[0274] In the formula, is the decision function, is the bias term of the support vector machine.

[0275] The calculation method of the bias term of the support vector machine is expressed as:

[0276] ;

[0277] In the formula, For the The true labels of samples, is the quantum state corresponding to the i-th sample;

[0278] S505, using the improved loss function, minimizing the classification error through the gradient descent method, and updating the weight parameters. The calculation method of the loss function is expressed as:

[0279] ;

[0280] In the formula, is the loss function of the support vector machine, is the quantum state corresponding to the i-th sample, For the The true labels of samples, is the regularization parameter of the support vector machine, represents the L2 norm; preferably, Set to 0.3.

[0281] The calculation method of the L2 norm of the support vector machine weight is expressed as:

[0282] ;

[0283] In the formula, is the feature dimension input to the support vector machine, is the weight vector A quantity.

[0284] S506. According to the feedback of the loss function, the parameters of the high-order kernel function are dynamically adjusted to optimize the performance of the kernel function. The updating method of the high-order kernel function is expressed as:

[0285] ;

[0286] In the formula, is the learning rate of the support vector machine, Indicates The parameters of the kernel function for the iteration, Indicates The parameters of the kernel function for the iteration, is the gradient of the loss function of the support vector machine with respect to the parameters of the high-order kernel function;

[0287] Furthermore, the gradient calculation method of the loss function with respect to the parameters of the high-order kernel function is expressed as:

[0288] ;

[0289] In the formula, is the derivative of the similarity matrix with respect to the high-order kernel parameters, indicating how the similarity is affected by parameter changes; is the maximum value function.

[0290] Furthermore, during the iterative training process, the method of updating the support vector machine weights is expressed as:

[0291] ;

[0292] In the formula, For the The weights of the support vector machine for the iteration, For the The weight of the support vector machine for the iteration;

[0293] Furthermore, the gradient calculation method of the loss function with respect to the weight is expressed as:

[0294] ;

[0295] In the formula, is the indicator function, indicating that when It is 1 if the prediction of a sample is incorrect, and 0 otherwise.

[0296] S507, repeat the above steps until the preset stop iteration condition is met, which means that the model training is completed; wherein 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.

[0297] S6. Concrete sample inspection

[0298] The trained model is used to inspect concrete samples. In one embodiment, the collected raw data is input into the trained feature extraction and feature dimension reduction model for feature processing. Further, the processed features are input into the classifier model for classifier training to obtain classification results. The categories of classification results include:

[0299] Qualified: It means that the concrete meets the relevant standards;

[0300] Unqualified: means the concrete does not meet the relevant standards;

[0301] To be tested: indicates samples that require further testing.

[0302] The present invention adopts a generative adversarial network algorithm based on adversarial enhancement for data expansion, generates new samples with authenticity through the interaction of the generator and the discriminator, and further increases the diversity of the data set by perturbing the adversarial sample enhancement on the basis of the samples generated by the generative adversarial network, so that the generated samples are more similar to the real samples in the feature space, thereby improving the quality of the generated sample data set and solving the problem of poor generalization ability of the model due to insufficient training data; adopts a neural network algorithm based on dynamic population evolution optimization for feature extraction, and automatically adjusts the evolution rules according to the characteristics of the training data through a self-correction mechanism, thereby avoiding the problems of gradient vanishing, gradient exploding or falling into the local optimal solution in the traditional neural network during the training process, especially the self-correction mechanism dynamically adjusts the probability of crossover and mutation according to the change of the loss function, realizes accurate feature capture of high-dimensional data, improves the generalization ability and training efficiency of the model, makes the algorithm more flexibly adapt to different data distributions, and improves training efficiency; on the basis of the autoencoder A feature adjustment module is adopted to dynamically adjust feature weights through recursive optimization, strengthen important features, weaken noise features, ensure that the features after dimensionality reduction have good information retention ability and simplicity, reduce data dimensions, and improve the processing efficiency of subsequent classifiers; samples are mapped to quantum state space, and the superposition and entanglement characteristics of quantum states are utilized, and the parameters of high-order kernel functions are optimized through feature coupling, so that the support vector machine can better process high-dimensional data, and the construction of quantum states improves the flexibility and accuracy of decision boundaries; in particular, for the design of the support vector machine model based on quantum phase space mapping proposed by the present invention, in order to solve the limitations of traditional support vector machines in processing high-dimensional nonlinear data, the present invention enhances the nonlinear feature processing capability of the support vector machine by mapping samples to quantum state space and utilizing the superposition and entanglement characteristics of quantum states. In addition, the parameters of high-order kernel functions are optimized through feature coupling, so that the support vector machine can better process high-dimensional data, and the construction of quantum states improves the flexibility and accuracy of decision boundaries.

[0303] In the training process of support vector machine based on quantum phase space mapping, the similarity matrix of training samples is calculated based on high-order kernel function, and the similarity matrix is ​​mapped to quantum state space by quantum feature mapping to generate quantum feature mapping. In addition, the classification error is minimized by gradient descent method to optimize support vector machine weights and kernel function. Based on this, quantum feature mapping is used to enhance the flexibility of decision boundary by utilizing quantum superposition and entanglement characteristics, so that support vector machine can better handle complex nonlinear data and improve classification accuracy. The improved high-order kernel function and quantum state mapping enable support vector machine to accurately identify the boundaries between different categories, which is especially suitable for the classification task of high-dimensional and nonlinear characteristic all-solid waste permeable concrete inspection data, and significantly improves the classification performance of the model.

[0304] Finally, it should be noted that the above are only preferred embodiments of the present invention. Of course, those skilled in the art can make changes and modifications to the present invention. If these modifications and variations fall within the scope of the claims of the present invention and their equivalent technologies, they should be considered to be within the scope of protection of the present invention.

Claims

1. A method for processing detection data of all-solid waste concrete for state prediction, characterized in that: Follow these steps: S1. Data collection and annotation The collection and labeling of all-solid waste permeable concrete test data to obtain the original data set; S2. Data expansion The adversarial enhancement-based generative adversarial network algorithm generates samples to achieve data expansion. The adversarial enhancement-based generative adversarial network consists of a generator and a discriminator. The generator is used to generate new samples, and the discriminator is responsible for evaluating the authenticity of the generated samples. New virtual samples are generated through adversarial sample enhancement technology. S3. Feature extraction model training A neural network algorithm based on dynamic population evolution optimization is used as a feature extraction model, and the expanded data is input into the feature extraction model for training the feature extraction model; S4. Feature dimensionality reduction model training The autoencoder neural network based on feature refinement is used as a dimensionality reduction model, and the data after feature extraction is input into the feature dimensionality reduction model for training the feature dimensionality reduction model; S5. Classifier model training The support vector machine algorithm based on quantum phase space mapping is used as the classification algorithm. The data after feature extraction is input into the classifier to train the classifier model. This method uses feature coupling technology to optimize the parameters of the kernel function and enhance the adaptability of the model to high-dimensional data. The training process of the support vector machine algorithm based on quantum phase space mapping is as follows: S501, initialize the weight parameters and high-order kernel function parameters of the vector machine, expressed as: ; ; In the formula, is the weight of the support vector machine, is the initial weight vector of the classifier, is the initial parameter of the high-order kernel function; S502: The features of the training samples after dimensionality reduction The model is passed in and the similarity matrix is ​​calculated based on the high-order kernel function, expressed as: ; In the formula, For the Samples and The similarity between samples, is the feature mapping function, is the total number of samples, For the samples, For the samples; The calculation method of the feature mapping function in the feature input is expressed as: ; In the formula, is the width of the Gaussian kernel, is a power, is the frequency of the sine function; S503, mapping the similarity matrix to the quantum state space, generating the corresponding quantum feature map using the quantum superposition principle, and constructing the quantum state, which is expressed as: ; In the formula, is a quantum state, For samples The quantum amplitude of the correlation; S504. In the quantum feature space, a decision boundary of a support vector machine is constructed, and the selection of support vectors is adjusted according to the training samples. The method of calculating the decision boundary is expressed as: ; In the formula, is the decision function, is the bias term of the support vector machine; The calculation method of the bias term of the support vector machine is expressed as: ; In the formula, For the The true labels of samples, is the quantum state corresponding to the i-th sample; S505, using the improved loss function, minimizing the classification error through the gradient descent method, and updating the weight parameters. The calculation method of the loss function is expressed as: ; In the formula, is the loss function of the support vector machine, is the quantum state corresponding to the i-th sample, For the The true labels of samples, is the regularization parameter of the support vector machine, represents the L2 norm; The calculation method of the L2 norm of the support vector machine weight is expressed as: ; In the formula, is the feature dimension input to the support vector machine, is the weight vector Quantity; S506. According to the feedback of the loss function, the parameters of the high-order kernel function are dynamically adjusted to optimize the performance of the kernel function. The updating method of the high-order kernel function is expressed as: ; In the formula, is the learning rate of the support vector machine, Indicates The parameters of the kernel function for the iteration, Indicates The parameters of the kernel function for the iteration, is the gradient of the loss function of the support vector machine with respect to the parameters of the high-order kernel function; Furthermore, the gradient calculation method of the loss function with respect to the parameters of the high-order kernel function is expressed as: ; In the formula, is the derivative of the similarity matrix with respect to the high-order kernel parameters, indicating how the similarity is affected by parameter changes; is the maximum value function; Furthermore, during the iterative training process, the method of updating the support vector machine weights is expressed as: ; In the formula, For the The weights of the support vector machine for the iteration, For the The weight of the support vector machine for the iteration; Furthermore, the gradient calculation method of the loss function with respect to the weight is expressed as: ; In the formula, is the indicator function, indicating that when 1 if the prediction of a sample is incorrect, 0 otherwise; S507, repeating the above steps until the preset stop iteration condition is met, indicating that the model training is completed; S6. Concrete sample inspection The trained model is used to test concrete samples.

2. The method for processing detection data of all-solid waste concrete for state prediction according to claim 1 is characterized by: The sources of data collected in step S1 mainly include laboratory tests, on-site construction monitoring and historical inspection records. The collected data is stored in a standardized structured format, specifically in CSV or database format; the collected data attributes include Compressive strength reflects the ability of concrete to resist pressure, and the unit is MPa; Permeability, which indicates the penetration rate of water in concrete, is measured in m / s; Porosity refers to the ratio of void volume in concrete to total volume, with the unit of %; Moisture content refers to the moisture content in concrete, with the unit being %; Temperature: the ambient temperature during collection, in °C; Humidity: the ambient humidity during collection, in %; Curing time, the time from concrete pouring to testing, in days; Solid waste ratio refers to the proportion of solid waste in concrete, in units of %; Specimen size, indicating the length, width and height of the specimen, in mm; Construction method, the construction technology used is identified by characters; In step S1, the collected data is labeled, and the labeling categories include: Qualified: It means that the concrete meets the relevant standards; Unqualified: means the concrete does not meet the relevant standards; To be tested: indicates samples that require further testing.

3. The method for processing detection data of all-solid waste concrete for state prediction according to claim 1 is characterized by: The training process of the adversarial enhancement-based generative adversarial network algorithm in step S2 is as follows: S201, initialize and set the parameters of the generator and discriminator, including weights and biases. The bias is initialized to 0. The weight initialization method is expressed as: ; In the formula, represents the initial weight of the generator, represents the initial weight of the discriminator, The mean is 0 and the variance is The normal distribution of represents a normal distribution, Indicates compliance with a specific distribution; S202, input the original data set into the generator to generate a certain number of virtual samples, expressed as: ; In the formula, For the generated samples, represents a generator function, is a noise vector randomly drawn from the latent space, Indicates The weight of the generator for the iteration; S203: Input the generated sample into the discriminator for evaluation to obtain the authenticity score of the generated sample, which is expressed as: ; In the formula, is the score of the discriminator on the generated samples, represents the discriminator function, Indicates The weight of the discriminator at the iteration; S204, adjusting the parameters of the discriminator through the optimization algorithm to improve its ability to distinguish the generated samples. For the optimization of the discriminator, the calculation method of its loss function is expressed as: ; In the formula, is the discriminator loss, is the real data distribution, To generate the sample distribution, For expectations, is real data; S205, feedback adjustment to the generator, based on the evaluation result of the discriminator, feedback to the generator, adjust the parameters of the generator, so that the samples generated by it are closer to the real samples in terms of authenticity. The calculation method of its loss function is expressed as: ; In the formula, is the generator loss; S206. Perform a small perturbation on the generated sample to obtain richer samples and achieve adversarial sample enhancement, which is expressed as: ; In the formula, To enhance the samples, is the disturbance amplitude, represents the gradient of the discriminator to the generated sample, is a symbolic function; The calculation method of the gradient of the discriminator to the generated sample is expressed as: ; In the formula, The calculation method is: ； S207. Update the parameters of the generator and the discriminator according to the loss function. The updating method is expressed as: ; ; In the formula, and are the learning rates of the discriminator and generator, respectively. For the The discriminator weights for the iterations, For the The generator weights for the iterations; S208, repeating the above steps until the preset stop iteration condition is met, indicating that the data expansion model training is completed; After the data expansion model training is completed, the trained data expansion model is used to increase the number of samples.

4. The method for processing detection data of all-solid waste concrete for state prediction according to claim 1 is characterized by: In step S3, the specific training process of the neural network algorithm based on dynamic population evolution optimization as the feature extraction model for the extended data is as follows: S301. According to the bionic algorithm initialization method, in the initialization stage, an initial population is generated, and each individual represents a configuration of network weights. Specifically, the population size is , for The individual initialization weights and biases are expressed as: ; ; In the formula, For the The weight of the neural network corresponding to each individual, For the The bias of the neural network corresponding to each individual, Indicates The weight matrix of each individual in the initial state; Indicates The bias of each individual in the initial state; represents the initialization variance; It means the mean is 0 and the variance is Normal distribution of is a normal distribution; S302: For each individual in the population, use its corresponding neural network configuration to process the input training data, calculate the output of the model, and evaluate its performance according to a predetermined loss function. Specifically, for the first Individuals use their weights and biases to calculate the loss on the training data set, expressed as: ; In the formula, For the The loss of the neural network corresponding to each individual; Indicates the number of samples input in the current batch; represents the composite loss function, Represents a neural network model function; Indicates Characteristics of the samples; Indicates The labels of the samples; The composite loss function contains a regularization term, which can increase the generalization ability of the model. The calculation method is expressed as: ; In the formula, is the mean square error function, is the regularization term, is the regularization parameter; further, the calculation method of the regularization term is expressed as: ; In the formula, For the first The neural network corresponding to each individual weights; S303. According to the fitness of the individual, the best performing individual is selected from the current population and retained as the candidate solution for the next generation. Specifically, based on the fitness of the individual, the excellent individual has a higher probability of being selected. The calculation method of the probability of the individual being selected is expressed as: ; In the formula, Indicates The probability of an individual being selected; is the parameter that controls the selection pressure; For the The loss of the neural network corresponding to each individual; S304. Generate new individuals through crossover and mutation operations. The crossover operation allows two excellent individuals to exchange some genes to generate new offspring; the mutation operation randomly changes some genes in an individual to increase the diversity of the population. Specifically, the crossover operation randomly selects two individuals for gene exchange, which is expressed as: ; ; In the formula, is the crossover rate, is the weight of the neural network corresponding to the first individual selected, is the bias of the neural network corresponding to the first individual selected, is the weight of the neural network corresponding to the second individual selected, is the bias of the neural network corresponding to the second individual selected, is the weight of the neural network corresponding to the individual after the crossover operation, is the bias of the neural network corresponding to the individual after the crossover operation; Moreover, the mutation operation performs a small random perturbation on the newly generated individual weights, which can be expressed as: ; ; In the formula, represents the variance of the variation, is the weight of the neural network corresponding to the individual after the mutation operation, is the bias of the neural network corresponding to the individual after the mutation operation; S305, repeat the above steps until the preset stop iteration condition is met, which means the model training is completed.

5. The method for processing detection data of all-solid waste concrete for state prediction according to claim 1 is characterized by: The autoencoder neural network based on feature refinement in step S4 is composed of three parts: an encoder, a decoder, and a feature adjustment module. The encoder is responsible for mapping high-dimensional input data to a low-dimensional feature space, and the decoder is used to reconstruct the reduced-dimensional features back to the original space to ensure the reversibility of the dimensionality reduction process; the adjustment module dynamically adjusts the reduced-dimensional feature space through recursive feature adaptive optimization, so that important features can be strengthened and minor features are gradually weakened, so that the reduced-dimensional feature representation has better simplicity while retaining data information; The training process of the autoencoder neural network algorithm based on feature refinement is as follows: S401. First, assume that the data input to the autoencoder neural network is , the encoder uses a multi-layer nonlinear mapping structure to map high-dimensional data to the initial low-dimensional feature space, which can be expressed as: ; In the formula, represents the initial low-dimensional features, is the weight matrix of the encoder, is the encoder bias vector, is the multi-layer Sigmoid activation function of the encoder; S402, after the low-dimensional features are generated, the feature adjustment module automatically generates feature weights according to the feature importance in the current feature space. When the module is initialized, the same initial weights are assigned to all features so that they can be gradually adjusted according to feature contributions in subsequent steps. The calculation method of the initial weight matrix is ​​expressed as: ; In the formula, is the initial weight matrix; is the feature weight vector, Each element in Initialized to the same value, indicating that all features have the same importance in the initial stage; is a function for extracting the diagonal elements of a matrix; Furthermore, the adjusted features can be expressed as: ; In the formula, is the feature representation after feature weight adjustment; S403, the feature adjustment module recursively optimizes the initially generated low-dimensional features. In each round of iteration, the module adjusts the weights of each feature according to the performance of the previous round of features, gradually enhances the features with important influence, and gradually weakens the redundant or noise features. In the round iteration, the weight update rule is as follows: ; In the formula, Indicates The feature weights of the round iteration, Indicates The feature weights of the round iteration, is the learning rate of the autoencoder neural network, is the loss function of the autoencoder neural network, is the label data, Represents the gradient of the loss function with respect to the feature weight; S404, in order to ensure the effectiveness of the dimensionality reduction process, the decoder remaps the low-dimensional features back to the high-dimensional space to ensure that the dimensionality reduction process does not lose important information. The reconstruction process of the decoder is expressed as: ; In the formula, is the reconstructed high-dimensional data, is the weight matrix of the decoder, is the bias vector of the decoder, is the multi-layer Sigmoid activation function of the decoder; S405, repeat the above steps until the preset stop iteration condition is met, which means the model training is completed.

Images