Intelligent evaluation method for quality of highway fabricated beam bridge based on grey correlation degree analysis

Abstract

This invention discloses an intelligent quality evaluation method for prefabricated beam bridges on highways based on grey relational analysis, belonging to the field of bridge engineering quality evaluation. The method includes: constructing a three-level quality evaluation index system for the beam bridge and obtaining the percentage scores for each secondary index; based on the historical percentage scores of all secondary indices, identifying the key quality parameters of each primary index through grey relational analysis, and calculating the relative correlation strength between each secondary index and the key quality parameters; constructing and training a grey relational-guided weighted neural network; and based on the optimized network weights and the percentage scores of each secondary index of the bridge to be evaluated, obtaining the comprehensive quality score of the bridge through weighted summation, and determining the quality level of the bridge. This invention significantly improves the accuracy, objectivity, and reliability of prefabricated beam bridge quality evaluation, providing a scientific and efficient technical solution for intelligent quality evaluation of prefabricated beam bridges on highways.

Description

Technical Field

[0001] This invention belongs to the field of bridge engineering quality evaluation, and in particular relates to an intelligent evaluation method for the quality of prefabricated beam bridges for highways based on grey relational analysis. Background Technology

[0002] Prefabricated beam bridges are widely used in highway construction due to their advantages such as short construction period, easy quality control, and minimal environmental impact. The construction quality of prefabricated beam bridges directly affects the safety, durability, and service performance of the bridge structure; therefore, establishing scientific and reasonable quality evaluation methods is crucial.

[0003] Traditional quality evaluation of prefabricated beam bridges mainly relies on expert experience and manual inspection, which has the following prominent problems: the evaluation standards are highly subjective, and different experts have significantly different evaluation results for the same bridge; the evaluation index system is incomplete and cannot fully cover the entire construction process; qualitative and quantitative indicators are mixed, and there is a lack of a unified quantitative processing method; there are correlations and redundancies among the indicators, and simple weighted summation will lead to some factors being calculated repeatedly; the weight determination method is unreasonable, and subjective weighting is easily limited by personal experience, while objective weighting methods cannot fully integrate expert knowledge.

[0004] In existing technologies, the analytic hierarchy process (AHP) relies excessively on expert subjective judgment. While the consistency test is rigorous, the determination of weights is still greatly influenced by subjective factors. Although the fuzzy comprehensive evaluation method can handle qualitative indicators, the determination of membership functions is highly arbitrary. Traditional neural network methods do not consider the correlation and redundancy between indicators, which can easily lead to distorted evaluation results.

[0005] Therefore, there is an urgent need to establish an intelligent quality evaluation method that can automatically identify key quality control parameters, eliminate indicator redundancy, and objectively determine weight coefficients. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention provides an intelligent quality evaluation method for prefabricated highway beam bridges based on grey relational analysis, comprising the following steps: A three-tiered quality evaluation index system for beam bridges, comprising target layer, criterion layer, and indicator layer, is constructed, and multiple primary indicators and their corresponding multiple secondary indicators are determined. Based on the measured data of the secondary indicators, the percentage scores of each secondary indicator are obtained through fuzzification and unification processing. Based on the historical percentage scores of all secondary indicators under each primary indicator, the key quality parameters of each primary indicator are identified through grey relational analysis, and the relative correlation strength between each secondary indicator and the key quality parameters is calculated. Based on the key quality parameters and the relative correlation strength, a weighted neural network guided by gray correlation is constructed and trained to obtain the optimized network weights. Based on the optimized network weights and the percentage scores of each secondary indicator of the bridge to be evaluated, the scores of each primary indicator and the comprehensive quality score of the finished bridge are obtained by weighted summation. The quality level of the bridge is determined based on the comprehensive quality score of the finished bridge.

[0007] Optionally, in the step of determining multiple primary indicators and their respective multiple secondary indicators: The primary indicators included in the criteria layer are: raw material quality control, precast component quality, construction equipment and technology, construction personnel management, transportation and storage, on-site installation quality, testing and acceptance, and environmental and safety management.

[0008] Optionally, the process of obtaining the percentage score for each secondary indicator based on the measured data of the secondary indicators through fuzzification and standardization includes: The measured data of the secondary indicators are divided into qualitative parameters and quantitative parameters; For qualitative parameters, a five-level evaluation set is established and represented using triangular fuzzy numbers. Then, the centroid method is used for defuzzification to convert them into a percentage score. For quantitative parameters, the deviation between the measured value and the standard value is linearly normalized and converted into a percentage score.

[0009] Optionally, based on the historical percentage scores of all secondary indicators under each primary indicator, the process of identifying key quality parameters for each primary indicator through grey relational analysis and calculating the relative correlation strength between each secondary indicator and the key quality parameters includes: For each primary indicator, a data sequence matrix including historical scores of all secondary indicators is constructed; the data sequence matrix is ​​mean- and dimensionless; each secondary indicator is used as a reference sequence, and the remaining secondary indicators are used as comparison sequences, and the correlation coefficient at each sample point is calculated; the average correlation coefficient of all sample points is calculated to obtain the grey correlation degree between each sequence; the average grey correlation degree between each secondary indicator and all other secondary indicators within the same primary indicator is calculated as the average correlation degree of the corresponding secondary indicator; the secondary indicator with the largest average correlation degree is selected as the key quality parameter of the corresponding primary indicator; the relative correlation strength between each secondary indicator and the key quality parameter is calculated.

[0010] Optionally, the process of constructing and training a weighted neural network guided by grey relational degree based on the key quality parameters and the relative correlation strength to obtain optimized network weights includes: A three-layer feedforward neural network structure consisting of an input layer, a hidden layer, and an output layer is constructed. The connection weights from the input layer to the hidden layer are initialized using the key quality parameters and the relative correlation strength. A loss function is constructed, which includes a mean squared error loss term, a gray relational structure preservation regularization term, and a weight decay regularization term. The network structure is trained using a backpropagation algorithm and an optimization algorithm to optimize and obtain the final network weights.

[0011] Optionally, the training process includes independent offline training and online evaluation phases; During the offline training phase, historical data collection and preprocessing, identification of key quality parameters, training of the neural network and weight optimization are performed, and the optimized weight parameters are saved. During the online evaluation phase, measured data of each secondary indicator of the bridge to be evaluated are obtained and converted into a percentage score. Optimized weight parameters are loaded, the comprehensive quality score of the finished bridge is calculated, and the final evaluation result is output.

[0012] This invention also proposes an intelligent quality evaluation system for prefabricated beam bridges on highways based on grey relational analysis, used to implement the method, including: The system construction module is used to construct a three-level quality evaluation index system for beam bridges, including the target layer, criterion layer, and index layer, and to determine multiple primary indicators and their corresponding multiple secondary indicators. The data processing module is used to obtain a percentage score for each secondary indicator based on the measured data of the secondary indicators through fuzzification and unification processing. The parameter identification module is used to identify the key quality parameters of each primary indicator based on the historical percentage scores of all secondary indicators under each primary indicator, and to calculate the relative correlation strength between each secondary indicator and the key quality parameters through grey relational analysis. The network training module is used to construct and train a weighted neural network guided by gray relational degree based on the key quality parameters and the relative correlation strength to obtain optimized network weights. The comprehensive evaluation module is used to calculate the scores of each primary indicator and the comprehensive quality score of the finished bridge by weighted summation based on the optimized network weights and the percentage scores of each secondary indicator of the bridge to be evaluated. The quality level of the bridge is determined based on the comprehensive quality score of the finished bridge.

[0013] The present invention also proposes a computer device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method.

[0014] The present invention also proposes a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method.

[0015] The present invention also proposes a computer program product, including a computer program that, when executed by a processor, implements the steps of the method.

[0016] Compared with the prior art, the present invention has the following advantages and technical effects: This invention constructs a three-tiered quality evaluation index system, employs fuzzy mathematics to unify qualitative and quantitative indicators, identifies key quality parameters for each process based on grey relational analysis, and builds a grey relational-guided weighted neural network. The network weights are initialized using the key parameter identification results and relative correlation strength, and a loss function incorporating grey relational structure preservation constraints is designed. Intelligent adaptive optimization of the weights is achieved through neural network training. This technical solution effectively overcomes the shortcomings of traditional evaluation methods, such as strong subjectivity, redundant indicators, and difficulty in determining weights. It achieves deep synergy between expert experience and data-driven methods, significantly improving the accuracy, objectivity, and reliability of prefabricated beam bridge quality evaluation, and providing a scientific and efficient technical solution for intelligent quality evaluation of highway prefabricated beam bridges. Attached Figure Description

[0017] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a schematic diagram of the method flow according to an embodiment of the present invention. Detailed Implementation

[0018] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0019] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0020] Example 1 like Figure 1 As shown in the figure, this embodiment provides an intelligent quality evaluation method for prefabricated beam bridges for highways based on grey relational analysis, including the following steps: A three-tiered quality evaluation index system for beam bridges, comprising target layer, criterion layer, and indicator layer, is constructed, and multiple primary indicators and their corresponding multiple secondary indicators are determined. Based on the measured data of the secondary indicators, the percentage scores of each secondary indicator are obtained through fuzzification and unification processing. Based on the historical percentage scores of all secondary indicators under each primary indicator, the key quality parameters of each primary indicator are identified through grey relational analysis, and the relative correlation strength between each secondary indicator and the key quality parameters is calculated. Based on the key quality parameters and the relative correlation strength, a weighted neural network guided by gray correlation is constructed and trained to obtain the optimized network weights. Based on the optimized network weights and the percentage scores of each secondary indicator of the bridge to be evaluated, the scores of each primary indicator and the comprehensive quality score of the finished bridge are obtained by weighted summation. The quality level of the bridge is determined based on the comprehensive quality score of the finished bridge.

[0021] As a feasible implementation method, the specific steps include: Step 1: Construct a three-tiered quality evaluation index system, establishing a hierarchical evaluation framework consisting of a target layer, a criterion layer, and an indicator layer. The criterion layer includes eight primary indicators: raw material quality control, precast component quality, construction equipment and technology, construction personnel management, transportation and storage, on-site installation quality, testing and acceptance, and environmental and safety management. The indicator layer comprises specific quality control parameters for each primary indicator, covering elements such as personnel, machinery, materials, methods, environment, and measurement. Step 2: Quantify the evaluation data. Use fuzzy quantification to convert qualitative parameters into percentage scores and normalization to convert quantitative parameters into percentage scores, so as to achieve a unified quantitative expression of different types of parameters. Step 3: Identify key quality parameters based on grey relational analysis. By constructing a data sequence matrix, calculating the grey relational coefficient and grey relational degree, and calculating the average relational degree, extract the secondary indicators with the largest average relational degree under each primary indicator as key quality parameters, and calculate the relative correlation strength between each indicator and the key parameters. Step 4: Construct a weighted neural network guided by grey relational analysis. Design a three-layer feedforward network structure, initialize the network weights using grey relational analysis results, design a loss function that includes grey relational structure preservation constraints, and optimize the network weights under grey relational constraints through neural network training and intelligent weight optimization to achieve intelligent adaptive adjustment of the weights. Step 5: Comprehensive quality evaluation. The weighted summation method is used to calculate the scores of the primary indicators and the comprehensive score. The quality level is determined based on the comprehensive score.

[0022] In the feasible three-tier quality evaluation index system, the target layer is the comprehensive score of the finished bridge quality, and the criterion layer includes eight primary indicators: raw material quality control, precast component quality, construction equipment and technology, construction personnel management, transportation and storage, on-site installation quality, testing and acceptance, and environmental and safety management; the secondary indicators are the quality control parameters under each primary indicator.

[0023] Furthermore, the primary indicators include raw material quality control, precast component quality, construction equipment and technology, construction personnel management, transportation and storage, on-site installation quality, testing and acceptance, and environmental and safety management.

[0024] The secondary indicators for raw material quality control in the primary indicators include cement quality pass rate, steel mechanical properties, coarse and fine aggregate gradation, admixture performance, prestressed material quality, and completeness of raw material inspection records.

[0025] The secondary indicators of precast component quality in the primary indicators include beam geometric dimension deviation, concrete compressive strength, surface flatness, appearance quality, embedded part position accuracy, crack control, and component identification integrity.

[0026] The secondary indicators of construction equipment and technology in the primary indicators include the accuracy of the formwork system, the performance of the vibrating equipment, the calibration of the tensioning equipment, the control of the steam curing equipment, the standardization of the demolding process, and the compliance of the construction process.

[0027] The secondary indicators for construction personnel management within the primary indicators include the rate of key personnel holding relevant certificates, the completeness of technical briefings, the implementation of operating procedures, the implementation of the quality responsibility system, and the completeness of training records.

[0028] The secondary indicators for transportation and storage in the primary indicators include component transportation support measures, damage control during transportation, storage site conditions, compliance of stacking methods, component protection measures, and clarity of hoisting markings.

[0029] The secondary indicators of on-site installation quality in the primary indicators include the accuracy of bearing installation position, the quality of bearing pad stones, the alignment control of beam installation, the quality of transverse connection system, the quality of wet joint concrete, the quality of bridge deck pavement, the installation accuracy of expansion joints, and the construction quality of waterproof layer.

[0030] The secondary indicators for testing and acceptance in the primary indicators include the pass rate of geometric dimension testing, the pass rate of strength testing, prestressing tension testing, acceptance of concealed works, appearance quality inspection, load testing, and the completeness of test reports.

[0031] The secondary indicators of environmental and safety management in the primary indicators include construction site environmental protection, noise and dust control, waste disposal, safety protection measures, completeness of emergency plans, and operation of environmental protection facilities.

[0032] The feasible, constructed three-tier quality evaluation system is as follows: Target layer (O layer): Overall quality score Q of the finished bridge; Criterion Layer (C Layer): First-level indicators ( i =1, 2, ..., 8); Indicator layer ( I Level 2: Secondary indicators ( j =1, 2, ..., ),in, i Indicates the first i One primary indicator, j Indicates the first j Each secondary indicator n This indicates the number of secondary indicators under the current primary indicator. Indicates the first i The number of secondary indicators under each primary indicator. See Table 1 for details of the secondary indicators under each primary indicator.

[0033] Let the first i Each primary indicator includes If there are 2 secondary indicators, then the total number of secondary indicators is: ; The feasible process of obtaining a percentage score for each secondary indicator based on the measured data of the secondary indicators, through fuzzification and standardization processing, includes: The measured data of the secondary indicators are divided into qualitative parameters and quantitative parameters. For the qualitative parameters, a five-level evaluation set is established and represented by triangular fuzzy numbers. Then, the centroid method is used to defuzzify the data and convert it into a percentage score. For the quantitative parameters, the deviation between the measured values ​​and the standard values ​​is linearly normalized and converted into a percentage score.

[0034] As a specific implementation method, the fuzzification and unification processing of the evaluation data in step S2 is as follows: A linear normalization method is used to convert the deviation between the measured value and the standard value into a percentage score. The smaller the deviation, the higher the score; when the deviation reaches the maximum allowable value in the standard, the score is zero. The specific steps include: Step S21, fuzzy quantization of qualitative parameters: For qualitative parameters Establish a five-level evaluation system: ; The corresponding membership vector is It satisfies the normalization condition: , ; in, Indicates the first i The first indicator j The indicator belongs to the first k The degree of the rating, k is the index of the rating, indicating the level of the rating. k Each rating level T For the transpose symbol, Transpose to a column vector.

[0035] The comment set is represented by triangular fuzzy numbers, and the fuzzy numbers are defined. Its membership function is: ; in, a This is the lower limit value. b The median value. c This is the upper limit.

[0036] The triangular fuzzy number corresponding to each comment is set as follows: excellent: =(80, 90, 100); good: =(70, 80, 90); qualified: =(60, 70, 80); Poor: =(50, 60, 70); Difference: =(0, 50, 60); The centroid method is used for defuzzification to convert the fuzzy evaluation results into precise numerical scores. ; Step S22, normalization of quantitative parameters: For quantitative parameters The scores are converted to a percentage scale through linear normalization. ; in, For each secondary indicator, To standardize the standard values, To specify the maximum allowable deviation.

[0037] Furthermore, in step S3, the key quality parameters based on grey relational analysis are identified as follows: Constructing a data sequence matrix: For each primary indicator, construct a data sequence matrix by representing the scores of all secondary indicators under this process for all historical samples; Dimensionless data processing: The data series is processed to eliminate the influence of differences in the dimensions and numerical ranges of different indicators by using the mean method. Calculate the grey relational coefficient: Select each secondary indicator as a reference sequence and the remaining indicators as a comparison sequence, and calculate the correlation coefficient at each sample point according to the grey relational coefficient calculation method. Calculate the grey relational degree: average the correlation coefficients of all sample points to obtain the grey relational degree between the reference sequence and the comparison sequence; Calculate the average correlation degree: Calculate the average of the grey correlation degree between each secondary indicator and all other secondary indicators in the same process, and use it as the average correlation degree of that indicator; Extract key quality parameters: Select the secondary indicator with the highest average correlation as the key quality parameter of the primary indicator, and calculate the relative correlation strength between each secondary indicator and the key parameter.

[0038] Specifically, the following steps are included: Step S31, construct the data sequence matrix: ; in, L The total number of samples, For the first i The number of secondary indicators contained in each primary indicator; For the l-th sample, the first... i The first-level indicator j The score value of each secondary indicator.

[0039] Step S32, Dimensionless data processing: ; Step S33, determine the reference sequence and comparison sequence: Select the first j The secondary indicators serve as a reference sequence: ; The remaining indicators are comparison sequences: ; in, The reference sequence is indicated by the superscript 0. For comparing sequences, superscript j Indicates the relationship with the first j The two reference sequences were compared.

[0040] Step S34, calculate the correlation coefficient: For the l-th sample, the correlation coefficient between the reference sequence and the comparison sequence at that point is: ; in, The resolution coefficient.

[0041] Step S35, calculate the grey relational degree. : ; Step S36, calculate the average correlation degree : ; Step S37, extract key quality parameters: The larger the value, the more significant the effect. j The stronger the correlation between a secondary indicator and other indicators within the same process, the better. Key quality parameters are defined as those with the highest average correlation. ; Let the key quality parameters be: For the first i One primary indicator, one secondary indicator With key quality parameters The relative correlation strength is defined as: ; in, .

[0042] Furthermore, in step S4, the weighted neural network guided by grey relational analysis is constructed as follows: By constructing a weighted neural network guided by grey relational analysis, a three-layer feedforward network structure is designed. The network weights are initialized using the results of grey relational analysis. A loss function containing grey relational structure preservation constraints is designed. The network weights are optimized under grey relational constraints through neural network training and intelligent weight optimization, thereby achieving intelligent adaptive adjustment of the weights.

[0043] Specifically, the following steps are included: Step S41, Network Architecture Design: The constructed neural network consists of an input layer (48 secondary indicators), hidden layers (8 primary indicators), and an output layer (comprehensive score), employing a grouped fully connected structure, i.e., the... i Each primary indicator is determined solely by its corresponding... The calculations for the secondary indicators show that the different primary indicator groups are independent of each other. The constructed block diagonal weight matrix is ​​as follows: ; in, No. i The weight submatrix corresponding to each primary indicator It is a zero matrix.

[0044] Step S42, weight initialization based on grey relational degree: when Time (key quality parameter): ; in, The adjusted weights for key quality parameters, As the initial weights for key quality parameters, As a key parameter enhancement factor, .

[0045] when Time (non-critical quality parameter): ; in, The adjusted weights for non-critical quality parameters. Initial weights for non-critical quality parameters, The attenuation coefficient is... .

[0046] Step S43, weight normalization processing: To ensure that the adjusted weights meet the normalization criteria, normalization is performed: ; in, For the first i The first-level indicator j Normalized weights of each secondary indicator.

[0047] At this time, the i The initial weight vector corresponding to each primary indicator It can be represented as: ; in, For the first i The first-level indicator The weights of each secondary indicator.

[0048] Step S44, Hidden Layer Design: The eight neurons in the hidden layer correspond to eight primary metrics. The hidden layer to the output layer uses a fully connected structure. The weight vector of the second layer can be represented as: , ; in, Indicates the first i The weights of each primary indicator are determined based on expert experience.

[0049] The bias term is initialized to zero: ; in, This is the hidden layer bias vector. This is the output layer bias scalar. Let H represent the zero vector of dimension H, where H=8 is the number of neurons in the hidden layer.

[0050] The activation function chosen here is the Sigmoid activation function. ; Step S45, Loss function design: The total loss function is: ; in, For mean square error loss, To maintain the regularization term for the grey relational structure, This is the L2 weight decay regularization term. and is the regularization coefficient.

[0051] The mean squared error loss function is: ; Where M is the sample size. For the first m The prediction quality score for each sample This is a genuine rating.

[0052] The gray relational structure loss function is: ; in, For the first i Group current weight vector, For the first i Group relative correlation strength vector, It is an L2 norm.

[0053] The expanded form can be represented as: ; in: , ; To prevent overfitting, L2 weight decay regularization is used: ; Step S46, Forward Propagation Design: The sample input vector is: ; in, Only includes the first i The first primary indicator Each secondary indicator score, Indicates the firsti The first-level indicator j Scoring of each secondary indicator.

[0054] No. i The weighted inputs (before activation) of each hidden layer neuron are obtained by weighted summation of the secondary index scores of the corresponding group: ; in, The vector dot product is equivalent to a weighted summation. For bias terms, Indicates the first i The raw scores of each primary indicator.

[0055] After the Sigmoid activation function, the... i The output of each hidden layer neuron is: ; To ensure that the scores for the primary metrics conform to the percentage-based system used in quality assessments, the hidden layer output will be linearly scaled: ; in, For the first i The final score of each primary indicator.

[0056] The outputs of all hidden layer neurons are combined into a vector: ; The weighted inputs to the output layer neurons are obtained by weighted summation of scores from eight primary metrics: ; Final overall quality score prediction: ; in, For the first m The predicted comprehensive quality score for each sample.

[0057] The MSE loss is calculated as follows: ; in, M This indicates the number of prefabricated beam bridge samples.

[0058] Step S47, Backpropagation and Gradient Calculation: Backpropagation i (on) is the core algorithm for training neural networks. Its basic idea is: through the chain rule, the error of the output layer is propagated backward to each layer, the gradient of the loss function with respect to the weights of each layer is calculated, and then the weights are updated along the negative gradient direction to reduce the loss.

[0059] The partial derivative of the MSE loss with respect to the predicted output, i.e., the output layer error term. : ; For the second layer weights Its gradient consists of two parts: the gradient of the MSE loss and the gradient of the L2 regularization. By the chain rule, the total gradient is: ; in, The regularization coefficient is an adjustable hyperparameter set before model training.

[0060] Similarly, the gradient of the second layer bias is: ; To calculate the gradient of the first layer weights, the error needs to be backpropagated from the output layer to the hidden layers. i The error term for each hidden layer neuron is: ; Based on the basic backpropagation algorithm, the MSE loss pair The gradient is: ; Normalized weights and correlation strength : ; The gray relational regularization gradient is: ; L2 regularization pairs The gradient is: ; Adding the three gradients above, we obtain the total gradient of the first layer weights: ; Step S48, Weight Update and Optimization: Adopting Adam (Adapt) i ve Moment Est i mat i The optimization algorithm updates the weights for the (on) th ... t In the next iteration, the update formula for the first-order moment estimate is: ; in, This is the first moment estimate of the gradient at the t-th iteration. For the first t The gradient value of the next iteration. It represents the first-order moment decay rate.

[0061] The update formula for the second-order moment estimate is: ; in, No. t The second moment estimate of the squared gradient at the next iteration. The element-wise square of the gradient. It represents the second-order moment decay rate.

[0062] because and Initialize to zero, in the first few iterations and It will bias towards zero. To eliminate this bias, it is corrected using the first and second moment biases: ; ; The final weight update strategy for the first and second layers is as follows: ; ; in, For the first i The first-level indicator j The weights of each secondary indicator in the first layer (hidden layer) For the first i The first-level indicator j The weights of the secondary indicators in the second layer (output layer) For learning rate, The gradient mean, Let be the mean of the squared gradient. To prevent the constant from having a denominator of 0, it is 10 here. -8 .

[0063] The final optimal weight vector is: ; in, For the first i The weight vector of each secondary indicator under each primary indicator. For the first i The first-level indicator The weights of each secondary indicator.

[0064] In practice, the training process in this embodiment includes an independent offline training phase and an online evaluation phase. During the offline training phase, historical data collection and preprocessing, identification of key quality parameters, training of the neural network and weight optimization are performed, and the optimized weight parameters are saved. During the online evaluation phase, measured data of each secondary indicator of the bridge to be evaluated are obtained and converted into a percentage score. Optimized weight parameters are loaded, the comprehensive quality score of the finished bridge is calculated, and the final evaluation result is output.

[0065] Furthermore, in step S5, the comprehensive evaluation of the quality of the prefabricated beam bridge for highways is as follows: Step S51, calculate the overall score: The scores for each primary indicator are calculated using a weighted summation method: ; in, For the first i The score of each primary indicator.

[0066] Step S52, calculate the score of the primary indicator: Based on the scores of the primary indicators and the weights of each primary indicator given by experts. Calculate the overall score Q of the finished bridge: ; The indicators are categorized as shown in Tables 1 and 2: Table 1

[0067] Table 2

[0068] This embodiment constructs a three-tiered quality evaluation index system encompassing all elements of prefabricated beam bridge construction, from raw materials to finished products. It systematically covers the entire construction process of prefabricated beam bridges. A key quality parameter identification method based on grey relational analysis is proposed, enabling quantitative analysis of the correlation between secondary indicators within each process, providing the optimal path for weight optimization. A weight adaptive decay mechanism based on key parameters is established. Addressing the correlation and redundancy issues among indicators, the weights of key parameters remain unchanged, while the weights of non-key parameters adaptively decay according to the correlation strength, improving evaluation accuracy. A weighted neural network structure guided by grey relational analysis is designed, initializing network weights using key parameter identification results and relative correlation strength, avoiding the training instability caused by random initialization in traditional neural networks and accelerating convergence. Historical evaluation data is fully utilized, and weight coefficients are objectively determined through a data-driven approach, overcoming the limitations of traditional subjective and purely objective weighting, and reducing the influence of human factors. This embodiment improves the accuracy and robustness of prefabricated beam bridge finished product quality evaluation, providing an efficient and reliable technical solution for intelligent quality evaluation of prefabricated beam bridges.

[0069] Example 2 Based on the same general inventive concept, this invention also provides an intelligent quality evaluation system for prefabricated beam bridges for highways based on grey relational analysis. The system provided by this invention is described below, and the system described below can be referred to in conjunction with the method described above. The system includes: The system construction module is used to construct a three-level quality evaluation index system for beam bridges, including the target layer, criterion layer, and index layer, and to determine multiple primary indicators and their corresponding multiple secondary indicators. The data processing module is used to obtain a percentage score for each secondary indicator based on the measured data of the secondary indicators through fuzzification and unification processing. The parameter identification module is used to identify the key quality parameters of each primary indicator based on the historical percentage scores of all secondary indicators under each primary indicator, and to calculate the relative correlation strength between each secondary indicator and the key quality parameters through grey relational analysis. The network training module is used to construct and train a weighted neural network guided by gray relational degree based on the key quality parameters and the relative correlation strength to obtain optimized network weights. The comprehensive evaluation module is used to calculate the scores of each primary indicator and the comprehensive quality score of the finished bridge by weighted summation based on the optimized network weights and the percentage scores of each secondary indicator of the bridge to be evaluated. The quality level of the bridge is determined based on the comprehensive quality score of the finished bridge.

[0070] Example 3 This embodiment also discloses a computer device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method described in Embodiment 1.

[0071] Example 4 This embodiment also discloses a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the method described in Embodiment 1.

[0072] Example 5 This embodiment also discloses a computer program product, including a computer program that, when executed by a processor, implements the steps of the method described in Embodiment 1.

[0073] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for intelligent quality evaluation of prefabricated beam bridges for highways based on grey relational analysis, characterized in that, Includes the following steps: A three-tiered quality evaluation index system for beam bridges, comprising target layer, criterion layer, and indicator layer, is constructed, and multiple primary indicators and their corresponding multiple secondary indicators are determined. Based on the measured data of the secondary indicators, the percentage scores of each secondary indicator are obtained through fuzzification and unification processing. Based on the historical percentage scores of all secondary indicators under each primary indicator, the key quality parameters of each primary indicator are identified through grey relational analysis, and the relative correlation strength between each secondary indicator and the key quality parameters is calculated. Based on the key quality parameters and the relative correlation strength, a weighted neural network guided by gray correlation is constructed and trained to obtain the optimized network weights. Based on the optimized network weights and the percentage scores of each secondary indicator of the bridge to be evaluated, the scores of each primary indicator and the comprehensive quality score of the finished bridge are obtained by weighted summation. The quality level of the bridge is determined based on the comprehensive quality score of the finished bridge.

2. The method according to claim 1, characterized in that, In the step of determining multiple primary indicators and their corresponding multiple secondary indicators: The primary indicators included in the criteria layer are: raw material quality control, precast component quality, construction equipment and technology, construction personnel management, transportation and storage, on-site installation quality, testing and acceptance, and environmental and safety management.

3. The method according to claim 1, characterized in that, Based on the measured data of the secondary indicators, the process of obtaining the percentage score for each secondary indicator through fuzzification and standardization includes: The measured data of the secondary indicators are divided into qualitative parameters and quantitative parameters; For qualitative parameters, a five-level evaluation set is established and represented using triangular fuzzy numbers. Then, the centroid method is used for defuzzification to convert them into a percentage score. For quantitative parameters, the deviation between the measured value and the standard value is linearly normalized and converted into a percentage score.

4. The method according to claim 1, characterized in that, Based on the historical percentage scores of all secondary indicators under each primary indicator, the process of identifying key quality parameters for each primary indicator through grey relational analysis and calculating the relative correlation strength between each secondary indicator and the key quality parameters includes: For each primary indicator, a data sequence matrix including historical scores of all secondary indicators is constructed; the data sequence matrix is ​​mean- and dimensionless; each secondary indicator is used as a reference sequence, and the remaining secondary indicators are used as comparison sequences, and the correlation coefficient at each sample point is calculated; the average correlation coefficient of all sample points is calculated to obtain the grey correlation degree between each sequence; the average grey correlation degree between each secondary indicator and all other secondary indicators within the same primary indicator is calculated as the average correlation degree of the corresponding secondary indicator; the secondary indicator with the largest average correlation degree is selected as the key quality parameter of the corresponding primary indicator; the relative correlation strength between each secondary indicator and the key quality parameter is calculated.

5. The method according to claim 1, characterized in that, The process of constructing and training a weighted neural network guided by grey relational degree based on the key quality parameters and the relative correlation strength to obtain the optimized network weights includes: A three-layer feedforward neural network structure consisting of an input layer, a hidden layer, and an output layer is constructed. The connection weights from the input layer to the hidden layer are initialized using the key quality parameters and the relative correlation strength. A loss function is constructed, which includes a mean squared error loss term, a gray relational structure preservation regularization term, and a weight decay regularization term. The network structure is trained using a backpropagation algorithm and an optimization algorithm to optimize and obtain the final network weights.

6. The method according to claim 5, characterized in that: The training process includes two independent phases: offline training and online evaluation. During the offline training phase, historical data collection and preprocessing, identification of key quality parameters, training of the neural network and weight optimization are performed, and the optimized weight parameters are saved. During the online evaluation phase, measured data of each secondary indicator of the bridge to be evaluated are obtained and converted into a percentage score. Optimized weight parameters are loaded, the comprehensive quality score of the finished bridge is calculated, and the final evaluation result is output.

7. A smart quality evaluation system for prefabricated beam bridges for highways based on grey relational analysis, characterized in that, For implementing the method according to any one of claims 1-6, comprising: The system construction module is used to construct a three-level quality evaluation index system for beam bridges, including the target layer, criterion layer, and index layer, and to determine multiple primary indicators and their corresponding multiple secondary indicators. The data processing module is used to obtain a percentage score for each secondary indicator based on the measured data of the secondary indicators through fuzzification and unification processing. The parameter identification module is used to identify the key quality parameters of each primary indicator based on the historical percentage scores of all secondary indicators under each primary indicator, and to calculate the relative correlation strength between each secondary indicator and the key quality parameters through grey relational analysis. The network training module is used to construct and train a weighted neural network guided by gray relational degree based on the key quality parameters and the relative correlation strength to obtain optimized network weights. The comprehensive evaluation module is used to calculate the scores of each primary indicator and the comprehensive quality score of the finished bridge by weighted summation based on the optimized network weights and the percentage scores of each secondary indicator of the bridge to be evaluated. The quality level of the bridge is determined based on the comprehensive quality score of the finished bridge.

8. A computer device comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the method according to any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the steps of the method according to any one of claims 1-6.

10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the steps of the method according to any one of claims 1-6.

Images