A reinforced concrete support construction quality detection method based on data processing

Abstract

The present application belongs to the technical field of data processing, and particularly relates to a reinforced concrete support construction quality detection method based on data processing. The method comprises the following steps: acquiring raw material proportioning, mixing, pouring and vibrating, and curing environment multi-source real-time data of the whole construction process; performing improved DBSCAN clustering preprocessing (missing value filling, outlier elimination, and standardization); constructing a dynamic correlation model by using a neural network, and optimizing hyperparameters by using an improved starling optimization algorithm; calculating influence coefficients and quality correlation degrees of each link; determining the quality by comparing the quality correlation degrees with a threshold value, and iteratively updating the model until the construction is completed. The present application realizes real-time detection throughout the whole cycle, reduces manual deviation, avoids quality defect solidification, reduces repair cost, and improves the reliability of support structure construction.

Description

Technical Field

[0001] This invention belongs to the field of data processing technology, and in particular relates to a method for detecting the construction quality of reinforced concrete supports based on data processing. Background Technology

[0002] Reinforced concrete supports, as core load-bearing components in building construction, municipal engineering, and underground engineering, directly determine the stability and safety of the entire structural system. Concrete, as a key component of such supports, is affected by a combination of factors, including raw material proportions, mixing uniformity, pouring process, vibration effect, and curing conditions. Therefore, accurate testing of the construction quality of reinforced concrete supports is crucial for ensuring structural safety and preventing economic losses. Existing testing methods largely rely on non-destructive testing (NDT) techniques or laboratory sampling tests. The former focuses on finished product testing, while the latter can only reflect the performance of local samples. Furthermore, the data types and formats generated by the two methods differ significantly, and there is a lack of unified processing standards for multi-source data across the entire process. In addition, existing testing is mostly concentrated after construction is completed, which is a post-construction verification model and cannot capture dynamic risks such as insufficient vibration during pouring or sudden changes in the curing environment in real time. Even some process testing methods can only achieve real-time acquisition of single parameters and cannot combine data from the entire process to judge the impact of risks on the final quality. As a result, quality problems are discovered only after they have become established defects, which are costly and difficult to repair. Moreover, they rely heavily on human intervention, and the test results are easily affected by operational standardization and subjective judgment biases. Summary of the Invention

[0003] In view of the technical problems existing in the background art, the present invention proposes a method for detecting the construction quality of reinforced concrete supports based on data processing.

[0004] To achieve the above objectives, the technical solution adopted by the present invention includes the following steps:

[0005] Acquire multi-source real-time data of the entire construction process of reinforced concrete support, including raw material ratio data, mixing process parameter data, pouring and vibration data, and curing environment data;

[0006] The multi-source real-time data is preprocessed, and an improved DBSCAN clustering method is used to remove outliers and standardize the data to obtain standardized processed data.

[0007] Based on the standardized data, a dynamic correlation model of construction process quality indicators is constructed. By determining the correlation weight between data and quality indicators of each construction stage, the influence coefficient of each stage on the final quality indicator is determined.

[0008] Based on the dynamic correlation model of construction process quality indicators and the standardized processing data updated in real time, the quality correlation degree of the current construction stage is calculated. The quality correlation degree is the sum of the product of the influence coefficient corresponding to the data of each construction link and the compliance rate of the data of that link.

[0009] The quality correlation degree is compared with the set quality threshold. If the quality correlation degree is greater than or equal to the quality threshold, the current construction quality is determined to be qualified. If the quality correlation degree is less than the quality threshold, risk warning information is generated. The dynamic correlation model of the construction process quality indicators and the dynamic quality threshold are continuously iterated and updated until the entire process of reinforced concrete support construction is completed, and the full cycle quality inspection is completed.

[0010] The construction process quality index dynamic correlation model is constructed by using the root mean square error between the predicted quality index and the actual quality index as the objective function, using a neural network algorithm to construct the construction process quality index dynamic correlation model, and using an improved star-sparrow optimization algorithm to optimize the model, resulting in the optimized construction process quality index dynamic correlation model.

[0011] Preferably, the multi-source real-time data is preprocessed, and an improved DBSCAN clustering method is used for outlier removal and data standardization to obtain standardized data. This process includes:

[0012] Real-time data from multiple sources First, missing values ​​are handled by imputing with the mean to obtain the data after missing values ​​are filled. Choose the number of nearest neighbors k, and define the local density using the reciprocal of the average distance between k-nearest neighbors. For each sample... Its k-nearest neighbor set is ,in Indicates sample The set of k nearest neighbors, For the j-th sample in the dataset, it is The nearest neighbor candidate samples; N is the total number of samples in the dataset. Represented as Euclidean distance, inequality middle , represented as arrive The Euclidean distance is less than or equal to The Euclidean distance to all other non-self samples; then the sample Local density The larger the value, the more likely it is to represent The denser the surrounding samples, the more accurate the calculation of each sample. The distance to the farthest sample in the k-nearest neighbor set. ;

[0013] Sample-based Adaptive adjustment yields dynamic neighborhood radius ,in To adjust the coefficients; and to compare MinPts with the local density Related ;in, The minimum core point threshold, The density correlation coefficient, It is the average local density of all sample points. This is the floor function;

[0014] Based on the obtained dynamic parameters, core points, edge points, and noise points are defined. Edge points are then subjected to secondary filtering to detect outliers, which are then removed from the preprocessed data. Deleted, resulting in no abnormal data. Standardized data is obtained by applying standard deviation to data without anomalies.

[0015] Preferably, the implementation of defining core points, edge points, and noise points based on the obtained dynamic parameters, and performing secondary screening on edge points to detect outliers includes:

[0016] First, detect the core points to determine the sample. of The number of samples within the neighborhood is denoted as the core point C; for the remaining non-core points, if a core point exists... Make , denoted as the initial edge point Samples that are neither core points nor initial edge points are considered outliers and removed.

[0017] A secondary selection process is performed on the initial edge points. For each initial edge point... First, find the set of core points it describes. ;calculate arrive The average distance, and set a filtering threshold. ,in For threshold coefficient, The average neighborhood radius of the core point; Determine arrive If the average distance is greater than the filtering threshold, it is considered an outlier; otherwise, it is considered a valid value.

[0018] As a preferred approach, the implementation of model optimization using the improved sparrow optimization algorithm includes:

[0019] First, perform hyperparameter initialization: initialize the population size. Maximum number of iterations And the upper and lower bounds of hyperparameters;

[0020] Initial positions of individual star larks: A vector of hyperparameters for the neural network model corresponding to each individual star lark is generated using a random method. , where b is the index item;

[0021] Calculate the fitness value of individual starbirds: Substitute the parameter vector of each starbird into the neural network model to obtain the objective function value of the model, and use the reciprocal of the obtained objective function as the fitness value of each starbird.

[0022] Perform adaptive exploration: Calculate the variance of the current population fitness. ,like If the value exceeds the set threshold, proceed to the exploration phase; otherwise, proceed to the development phase.

[0023] Exploration phase: Hyperparameter vectors for each individual starbird Update: ,in , The initial step size, For Lévy distribution random numbers, For random disturbance coefficients, Let be the hyperparameter vector of a randomly selected individual starbird; after updating, Perform boundary clipping;

[0024] Development phase: Updating the hyperparameter vector of each individual starbird based on a dynamic hierarchical memory pool: ,in For inertial weights, The hyperparameter vector of the globally optimal starbird individual. This is a hyperparameter vector randomly selected from either the historical suboptimal layers or the current high-quality layers in the dynamic hierarchical memory pool.

[0025] Update the dynamic hierarchical memory pool and determine the global optimum: For the global optimum layer, if there is a star sparrow individual whose fitness function is greater than that of the optimum layer, it is replaced; otherwise, the current optimum layer individuals are retained. For the historical suboptimal layer, the original historical suboptimal layer and the top m individuals in fitness of the new population are used as the new historical suboptimal layer. For the current high-quality layer: select star sparrow individuals in the new population whose fitness is better than the average fitness as the current high-quality layer. Determine the hyperparameter vector of the global optimum star sparrow individual.

[0026] If the number of iterations reaches the maximum or the fitness function reaches the preset precision threshold, the optimization terminates and the globally optimal hyperparameter vector is output.

[0027] As a preferred approach, the impact coefficient of each stage on the final quality indicator is determined by the correlation weight between data and quality indicators at each construction stage.

[0028] The feature weights of each input parameter for the quality index are extracted by constructing a dynamic correlation model of construction process quality indicators.

[0029] Calculate the information entropy of each feature parameter in each construction stage, and obtain the weight ratio of each parameter from the information entropy;

[0030] For each construction stage, the initial influence coefficient of that stage is obtained by multiplying the feature weights of all its feature parameters by their corresponding weight proportions and summing the results.

[0031] Calculate the deviation between the current batch's process impact coefficient and the historical average coefficient. If the deviation is greater than the threshold, recalculate the initial impact coefficient; if it is less than or equal to the threshold, retain it.

[0032] The influence coefficients of the retained construction steps are weighted and normalized to the [0,1] interval to obtain the standardized influence coefficients.

[0033] Preferably, the quality correlation degree is the sum of the products of the influence coefficients corresponding to the data of each construction stage and the compliance rate of the data of that stage. Specifically, the deviation rate between the feature of each stage and the set benchmark is calculated. The average deviation is calculated for each feature in each construction stage. Then, the compliance rate of the data of that stage is obtained by subtracting the average deviation from 1. Then, the quality correlation degree is obtained by multiplying the influence coefficients corresponding to the data of each construction stage and the compliance rate of the data of that stage.

[0034] Compared with existing technologies, the advantages and positive effects of this invention are as follows: it collects multi-source real-time data from the entire construction process, overcoming the problem of information fragmentation; it improves DBSCAN clustering, dynamically adjusts parameters to achieve accurate outlier removal and data standardization, and improves data reliability; it constructs a neural network dynamic correlation model, combines an improved star-sparrow optimization algorithm to optimize hyperparameters, and reduces prediction errors; it dynamically calculates the impact coefficient of each stage and the correlation degree of quality, realizing real-time quality assessment and risk warning, and continuously iterates the model and thresholds to achieve full-cycle intelligent detection, reducing human bias, avoiding quality defects from becoming established problems, and reducing repair costs. Attached Figure Description

[0035] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0036] Figure 1 This is a schematic diagram of the structural process for a data processing-based method for inspecting the construction quality of reinforced concrete supports. Detailed Implementation

[0037] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described below in conjunction with the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0038] Numerous specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways than those described herein, and therefore the invention is not limited to the specific embodiments disclosed in the following specification.

[0039] In this embodiment, to achieve high-precision quality inspection of the entire reinforced concrete support construction process and overcome the problems of poor real-time performance, information fragmentation, and reliance on manual judgment in traditional inspection methods, this invention proposes a data processing-based method for inspecting the construction quality of reinforced concrete supports. This invention integrates multi-source real-time data, constructs a dynamic quality index correlation model, and introduces intelligent optimization algorithms to realize a fully intelligent quality inspection system covering the entire process from data acquisition, anomaly identification, feature modeling to dynamic early warning. This solution overcomes the limitations of traditional post-construction inspection modes, enabling dynamic monitoring, real-time diagnosis, and adaptive optimization control of construction quality, thereby significantly improving the reliability and safety of the support structure. The technical process is as follows: Figure 1 As shown.

[0040] Firstly, in order to achieve comprehensive monitoring of the quality of the entire construction process, multi-source real-time data of the entire reinforced concrete support construction process is obtained. The multi-source real-time data includes raw material proportion data (such as the proportion and moisture content of cement, water, sand, stone, and admixtures); mixing process parameters (such as rotation speed, mixing time, temperature change curve, etc.); pouring and vibration parameters (such as vibration frequency, duration, placement temperature, measured slump value, etc.); and curing environment parameters (such as ambient temperature and humidity, curing age, surface temperature rise rate, etc.).

[0041] To reduce the impact of outlier data caused by sensor errors and construction disturbances, and considering that the traditional DBSCAN algorithm, with its fixed neighborhood radius and minimum core point threshold, cannot adapt to the uneven density of construction data, this invention employs an improved DBSCAN clustering algorithm for outlier removal and data standardization. Its implementation includes processing multi-source real-time data... First, missing values ​​are handled by imputing with the mean to obtain the data after missing values ​​are filled. Choose the number of nearest neighbors k, and define the local density using the reciprocal of the average distance between k-nearest neighbors. For each sample... Its k-nearest neighbor set is ,in Indicates sample The set of k nearest neighbors, For the j-th sample in the dataset, it is The nearest neighbor candidate samples; N is the total number of samples in the dataset. Represented as Euclidean distance, inequality middle , represented as arrive The Euclidean distance is less than or equal to The Euclidean distance to all other non-self samples; then the sample Local density The larger the value, the more likely it is to represent The denser the surrounding samples, the more accurate the calculation of each sample. The distance to the farthest sample in the k-nearest neighbor set. Sample-based Adaptive adjustment yields dynamic neighborhood radius ,in To adjust the coefficients; and to compare MinPts with the local density Related ;in, The minimum core point threshold, The density correlation coefficient, It is the average local density of all sample points. This is a floor function; based on the obtained dynamic parameters, core points, edge points, and noise points are defined, and edge points are further filtered to detect outliers, which are then removed from the preprocessed data. Deleted, resulting in no abnormal data. Standardized data is obtained by applying standard deviation to data without anomalies. The improved DBSCAN algorithm ensures that the standardized data are all valid and reliable construction parameters, eliminating dimensional differences and avoiding interference from outliers. This provides a clean and unified data source for the subsequent construction of a dynamic correlation model of construction process quality indicators, improving the model's prediction accuracy and stability.

[0042] The implementation of defining core points, edge points, and noise points based on the obtained dynamic parameters, and performing secondary screening on edge points to detect outliers, includes: firstly, detecting core points and judging the sample... of The number of samples within the neighborhood is denoted as the core point C; for the remaining non-core points, if a core point exists... Make , denoted as the initial edge point Samples that are neither core points nor initial edge points are considered outliers and removed; a secondary screening is performed on the initial edge points, for each initial edge point... First, find the set of core points it describes. ;calculate arrive The average distance, and set a filtering threshold. ,in For threshold coefficient, The average neighborhood radius of the core point; Determine arrive If the average distance is greater than the filtering threshold, it is considered an outlier; otherwise, it is considered a valid value.

[0043] Then, based on the standardized data, a dynamic correlation model of construction process quality indicators is constructed. The influence coefficient of each stage on the final quality indicator is determined by the correlation weights between data and quality indicators at each construction stage. The construction of the dynamic correlation model uses the root mean square error between the predicted and actual quality indicators as the objective function. A neural network algorithm is used to construct the dynamic correlation model, and an improved sparrow optimization algorithm is employed to optimize the model, resulting in the optimized dynamic correlation model of construction process quality indicators.

[0044] Specifically, the implementation of model optimization using the improved star-sparrow optimization algorithm includes: firstly, hyperparameter initialization: initializing the population size. Maximum number of iterations And the upper and lower bounds of the hyperparameters; generating the initial position of the starbird individual: using a random method to generate the vector of hyperparameters of the neural network model corresponding to each starbird individual. Where b is the index item; Calculate the fitness value of each starling individual: Substitute the parameter vector of each starling individual into the neural network model to obtain the objective function value of the model, and use the reciprocal of the obtained objective function as the fitness value of each starling individual; Perform adaptive exploration: Calculate the variance of the current population fitness. ,like If the value exceeds the set threshold, proceed to the exploration phase; otherwise, proceed to the development phase. Exploration phase: Hyperparameter vector for each individual Star Sparrow. Update: ,in , The initial step size, For Lévy distribution random numbers, For random disturbance coefficients, Let be the hyperparameter vector of a randomly selected individual starbird; after updating, Boundary pruning is performed; Development phase: The hyperparameter vector of each individual starbird is updated based on a dynamic hierarchical memory pool: ,in For inertial weights, The hyperparameter vector of the globally optimal starbird individual. The process involves randomly selecting hyperparameter vectors from either the historical suboptimal layers or the current high-quality layers in a dynamic hierarchical memory pool. The dynamic hierarchical memory pool is then updated, and the global optimum is determined: for the globally optimal layer, if a starling individual satisfies a fitness function greater than that of the optimal layer, it is replaced; otherwise, the current optimal layer individuals are retained. For historical suboptimal layers, the top m individuals in fitness from the original historical suboptimal layer and the new population are used as the new historical suboptimal layer. For the current high-quality layer, starling individuals with fitness better than the average fitness in the new population are selected as the current high-quality layer. The hyperparameter vector of the globally optimal starling individual is determined. Optimization terminates when the number of iterations reaches its maximum or the fitness function reaches a preset accuracy threshold, and the globally optimal hyperparameter vector is output. This process balances global search and local optimization capabilities through adaptive exploration and development, while the dynamic hierarchical memory pool enhances the utilization of high-quality parameter information. Compared to traditional optimization algorithms, its hyperparameter optimization efficiency is improved, enabling it to quickly find the parameter combination that optimizes the neural network's prediction accuracy, significantly reducing model prediction errors, and providing reliable algorithmic support for dynamic assessment of construction quality.

[0045] Next, by determining the correlation weights between data from each construction stage and quality indicators, the impact coefficients of each stage on the final quality indicators are determined. This includes: extracting the feature weights of each input parameter for the quality indicators through the constructed dynamic correlation model of construction process quality indicators; calculating the information entropy of each feature parameter within each construction stage and obtaining the weight ratio of each parameter from the information entropy; for each construction stage, multiplying the feature weights of all its feature parameters by their corresponding weight ratios to obtain the initial impact coefficient of that stage; calculating the deviation between the stage impact coefficient of the current batch and the historical average coefficient; if the deviation is greater than a threshold, recalculating the initial impact coefficient; if it is less than or equal to the threshold, retaining it; and weighting and normalizing the retained impact coefficients of the construction stages to the [0,1] interval to obtain the standardized impact coefficients. Specifically, in order to achieve a quantitative assessment of the impact of each construction stage on the final construction quality, a dynamic correlation model of construction process quality indicators is constructed to extract the correlation strength between parameters of each stage and the final quality indicators in multi-source construction data. First, multi-dimensional feature parameters are extracted from each construction stage, and the feature weights of each feature parameter for the target quality indicator are calculated based on the dynamic correlation model. The dynamic correlation model calculates the sensitivity gradient of input features to output quality indicators to obtain the feature weight of each input parameter, characterizing the degree of influence of feature changes on quality indicators. Secondly, to further eliminate the interference of different parameter dimensions and units on weight calculation, information entropy theory is used to calculate the information entropy of each feature parameter, measuring the degree of parameter uncertainty. Specifically, the probability distribution of each parameter within the construction stage is calculated. The larger the information entropy value, the more dispersed the parameter's influence on the quality indicator and the lower its weight; conversely, the smaller the information entropy, the more stable the parameter and the higher its contribution. The information entropy of each feature parameter is inversely mapped to its weight proportion to obtain the relative importance distribution among parameters. Then, within each construction stage, the feature weights of all feature parameters are multiplied by their corresponding information entropy weight proportions and summed to obtain the initial influence coefficient of that stage, reflecting the comprehensive contribution of that stage to the overall quality indicator. Furthermore, to ensure the dynamic stability of the impact coefficients, the deviation between the impact coefficients calculated for the current batch and the historical average coefficients is evaluated. When the deviation exceeds a set threshold, it indicates that there is fluctuation or anomaly in the current process, and the initial impact coefficients need to be recalculated. If the deviation is within the threshold range, the impact coefficient of that process is retained to ensure the temporal continuity and robustness of the model. Finally, the retained construction process impact coefficients are normalized, linearly mapping each process impact coefficient to the [0,1] interval to obtain standardized process impact coefficients. Based on the normalization results, the overall quality correlation of the construction is calculated and compared with a preset quality threshold. When the quality correlation is greater than or equal to the quality threshold, the current construction quality is deemed qualified; when the quality correlation is less than the threshold, a risk warning is generated, prompting the supervisor and the construction party to review or adjust the construction parameters.The system continuously iterates and updates the dynamic correlation model and quality threshold of the construction process quality indicators in the subsequent construction stages, realizing dynamic quality detection and self-learning optimization of the entire process of reinforced concrete support construction.

[0046] Based on the dynamic correlation model of construction process quality indicators and the real-time updated standardized processing data, the quality correlation degree of the current construction stage is calculated. This quality correlation degree is the sum of the products of the influence coefficients corresponding to the data of each construction stage and the data compliance rate of that stage. Specifically, this is achieved by calculating the deviation rate between each stage's characteristics and the set benchmark. The average deviation is calculated for each characteristic within each construction stage, and then 1 - the average deviation is used to obtain the data compliance rate of that stage. Finally, the quality correlation degree is obtained by multiplying the influence coefficients corresponding to the data of each construction stage by the data compliance rate of that stage and summing the results. Specifically, the deviation rate is defined as the ratio of the absolute value of the difference between the characteristic value and the corresponding benchmark value to the benchmark value, reflecting the degree to which the parameter deviates from the target standard. The average deviation of multiple characteristic parameters under the same construction stage is obtained by averaging the deviation rates, and then the data compliance rate of that stage is obtained through the formula "data compliance rate = 1 - average deviation". This compliance rate characterizes the overall quality stability and standard compliance of the current construction stage; the closer the value is to 1, the better the quality. Next, the influence coefficient corresponding to each construction stage is multiplied by the data compliance rate of that stage to represent the weighted contribution of that stage to the overall quality. Finally, the weighted contribution values ​​of all construction stages are summed to obtain the overall construction stage quality correlation index. This index comprehensively reflects the combined impact of each construction stage on the final structural quality.

[0047] Finally, the quality correlation degree is compared with the set quality threshold. If the quality correlation degree is greater than or equal to the quality threshold, the current construction quality is determined to be qualified. If the quality correlation degree is less than the quality threshold, risk warning information is generated, and the dynamic correlation model of the construction process quality indicators and the dynamic quality threshold are continuously iterated and updated until the entire process of reinforced concrete support construction is completed, and the full cycle quality inspection is completed.

[0048] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments for application in other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for inspecting the construction quality of reinforced concrete supports based on data processing, characterized in that, Includes the following steps: Acquire multi-source real-time data of the entire construction process of reinforced concrete support, including raw material ratio data, mixing process parameter data, pouring and vibration data, and curing environment data; The multi-source real-time data is preprocessed, and an improved DBSCAN clustering method is used to remove outliers and standardize the data to obtain standardized processed data. Based on the standardized data, a dynamic correlation model of construction process quality indicators is constructed. By determining the correlation weight between data and quality indicators of each construction stage, the influence coefficient of each stage on the final quality indicator is determined. Based on the dynamic correlation model of construction process quality indicators and the standardized processing data updated in real time, the quality correlation degree of the current construction stage is calculated. The quality correlation degree is the sum of the product of the influence coefficient corresponding to the data of each construction link and the compliance rate of the data of that link. The quality correlation degree is compared with the set quality threshold. If the quality correlation degree is greater than or equal to the quality threshold, the current construction quality is determined to be qualified. If the quality correlation degree is less than the quality threshold, risk warning information is generated. The dynamic correlation model of the construction process quality indicators and the dynamic quality threshold are continuously iterated and updated until the entire process of reinforced concrete support construction is completed, and the full cycle quality inspection is completed. The construction process quality index dynamic correlation model is constructed using the root mean square error between the predicted quality index and the actual quality index as the objective function. The model is constructed using a neural network algorithm and optimized using an improved star-sparrow optimization algorithm to obtain the optimized construction process quality index dynamic correlation model. The implementation of model optimization using the improved star-sparrow optimization algorithm includes: First, perform hyperparameter initialization: initialize the population size. Maximum number of iterations And the upper and lower bounds of hyperparameters; Initial positions of individual star larks: A vector of hyperparameters for the neural network model corresponding to each individual star lark is generated using a random method. , where b is the index item; Calculate the fitness value of individual starbirds: Substitute the parameter vector of each starbird into the neural network model to obtain the objective function value of the model, and use the reciprocal of the obtained objective function as the fitness value of each starbird. Perform adaptive exploration: Calculate the variance of the current population fitness. ,like If the value exceeds the set threshold, proceed to the exploration phase; otherwise, proceed to the development phase. Exploration phase: Hyperparameter vectors for each individual starbird Update: ,in , The initial step size, For Lévy distribution random numbers, For random disturbance coefficients, Let be the hyperparameter vector of a randomly selected individual starbird; after updating, Perform boundary clipping; Development phase: Updating the hyperparameter vector of each individual starbird based on a dynamic hierarchical memory pool: ,in For inertial weights, The hyperparameter vector of the globally optimal starbird individual. This is a hyperparameter vector randomly selected from either the historical suboptimal layers or the current high-quality layers in the dynamic hierarchical memory pool. Update the dynamic hierarchical memory pool and determine the global optimum: For the global optimum layer, if there is a star sparrow individual whose fitness function is greater than that of the optimum layer, it is replaced; otherwise, the current optimum layer individuals are retained. For the historical suboptimal layer, the original historical suboptimal layer and the top m individuals in fitness of the new population are used as the new historical suboptimal layer. For the current high-quality layer: select star sparrow individuals in the new population whose fitness is better than the average fitness as the current high-quality layer. Determine the hyperparameter vector of the global optimum star sparrow individual. If the number of iterations reaches the maximum or the fitness function reaches the preset precision threshold, the optimization terminates and the globally optimal hyperparameter vector is output. The process of determining the influence coefficient of each stage on the final quality indicator by associating the data of each construction stage with the quality indicators includes: By constructing a dynamic correlation model of construction process quality indicators, the correlation strength between parameters of each link in multi-source construction data and the final quality indicators is extracted. Multi-dimensional feature parameters are extracted in each construction link. Based on the dynamic correlation model, the feature weight of each feature parameter to the target quality indicator is calculated. The dynamic correlation model obtains the feature weight of each input parameter by calculating the sensitivity gradient of the input feature to the output quality indicator, which characterizes the degree of influence of the feature change on the quality indicator. Calculate the information entropy of each feature parameter in each construction stage, and inversely map the information entropy of each feature parameter to the weight ratio to obtain the relative importance distribution among the parameters; For each construction stage, the initial influence coefficient of that stage is obtained by multiplying the feature weights of all its feature parameters by their corresponding weight proportions and summing the results. Calculate the deviation between the current batch's process impact coefficient and the historical average coefficient. If the deviation is greater than the threshold, recalculate the initial impact coefficient; if it is less than or equal to the threshold, retain it. The influence coefficients of the retained construction steps are weighted and normalized to the [0,1] interval to obtain standardized influence coefficients. The quality correlation degree, which is the sum of the products of the influence coefficients corresponding to the data of each construction stage and the compliance rate of the data of that stage, is specifically implemented as follows: calculate the deviation rate of each characteristic parameter of each stage from the corresponding set benchmark, take the average value of the deviation rates of each characteristic parameter under the same construction stage to obtain the average deviation of that stage, and then use 1-average deviation to obtain the compliance rate of the data of that stage; then multiply the influence coefficients corresponding to the data of each construction stage by the compliance rate of the data of that stage and sum them to obtain the quality correlation degree.

Images