Self-adaptive debugging grading method and system
Through adaptive debugging of grading methods and systems, the intelligence and automation of grading design are used to use machine learning and federated learning technology, solving the problem of relying on experience and lack of scientificity in traditional grading design, and achieving more efficient and reliable grading design and construction.
Patent Information
- Application Number
- CN202510359060.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-25
- Publication Date
- 2025-06-03
AI Technical Summary
Traditional grading design relies on experience and personal judgment, lacks scientificity and accuracy, resulting in inefficient design efficiency and insufficient credibility of results, and the inability to achieve intelligence and automation.
Adaptive debugging and grading methods and systems are adopted to obtain the mix grading data in historical engineering applications, define data nodes, perform data preprocessing and feature extraction, model training and global model aggregation based on machine learning and federated learning technology, and optimize and adjust it using real-time data and reinforcement learning technology to realize the debugging of real-time grading and grading solutions.
It improves the accuracy, efficiency and scientificity of grading design, realizes the intelligence and automation of grading design of mixtures, reduces uncertainty in human intervention and design, and improves construction quality and efficiency.
Smart Images
Figure CN120088589A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of intelligent construction technology, and particularly relates to an adaptive debugging gradation method and system. Background Art
[0002] Gradation refers to the distribution of particles with different particle sizes in a mixture of aggregates. A reasonable gradation design can make the mixture have good compactness, stability and mechanical properties, so as to meet various requirements in engineering construction. In practical applications, the optimization design of gradation needs to consider various factors, such as material characteristics, use environment and construction technology, etc., to achieve the best performance.
[0003] In traditional gradation design, it usually relies on experience and personal judgment. This method lacks scientificity and accuracy, requires a large number of repeated tests, and consumes a lot of time and resources. At the same time, the existing methods lack the analysis and mining of data, cannot achieve intelligence and automation, and are prone to problems such as human thinking deviation and interruption of technical experience inheritance, resulting in low design efficiency and insufficient credibility of results.
[0004] The information disclosed in this background art section is only intended to deepen the understanding of the overall background art of the present disclosure, and should not be regarded as an admission or any form of suggestion that this information constitutes the prior art known to those skilled in the art. Summary of the Invention
[0005] The present invention provides an adaptive debugging gradation method and system, thereby effectively solving the problems pointed out in the background art.
[0006] In order to achieve the above object, the technical solution adopted by the present invention is:
[0007] An adaptive debugging gradation method, comprising:
[0008] Obtaining the mixture gradation data in historical engineering applications, and defining data nodes, where the data nodes include enterprise nodes, laboratory nodes, project station nodes and data center nodes;
[0009] Preprocessing the mixture gradation data and performing feature extraction to obtain a historical data feature set;
[0010] Based on machine learning, at each of the data nodes, performing preliminary model training on the historical data in the historical data feature set;
[0011] Based on federated learning technology, uploading the model parameters of each of the data nodes to a central server for global model aggregation and optimization;
[0012] Automatically calculate the optimized mixture synthetic gradation based on real-time data and existing screening results, and adjust the optimized global model through real-time data feedback and reinforcement learning technology to obtain an adaptive debugging gradation model;
[0013] Based on the adaptive debugging gradation model, conduct real-time debugging on the gradation scheme to obtain a real-time gradation scheme.
[0014] Furthermore, obtain the mixture gradation data in historical engineering applications and define data nodes, including:
[0015] Identify and determine the data sources from enterprises, laboratories, project sites, and data centers;
[0016] Enterprises, laboratories, project sites, and data centers collect relevant historical engineering data and real-time data and store them in the corresponding data nodes.
[0017] Furthermore, preprocess the mixture gradation data and perform feature extraction to obtain a historical data feature set, including:
[0018] Clean the mixture gradation data, remove outliers, and perform standardization processing;
[0019] Through data analysis techniques, extract key features from the preprocessed data, and the key features include but are not limited to particle size distribution, material density, and environmental parameters.
[0020] Summarize the extracted features to generate a historical data feature set for subsequent model training and optimization.
[0021] Furthermore, based on machine learning, conduct preliminary model training on the historical data in each data node for the historical data feature set, including:
[0022] Allocate and check the local historical data feature set at the enterprise node, laboratory node, project site node, and data center node, and select and initialize the GBDT model parameters;
[0023] Use the local historical data for GBDT model training and optimize the model parameters through step-by-step iteration;
[0024] Adjust the hyperparameters of the GBDT model according to the training results and retrain the GBDT model using the validation data set that did not participate in the training;
[0025] Save the optimized GBDT model parameters and perform version control to obtain the final model.
[0026] Further, based on the federated learning technology, the model parameters of each of the data nodes are uploaded to the central server for global model aggregation and optimization;
[0027] After each of the data nodes completes the preliminary model training, the trained GBDT model parameters are uploaded to the central server through a secure communication protocol;
[0028] The central server receives the model parameters of each of the data nodes and uses the federated learning technology to aggregate the model parameters and fuse them into a unified global model;
[0029] The performance of the global model is tested using an independent validation dataset, and the finally optimized model is assigned to each of the data nodes.
[0030] Further, the method by which the central server receives the model parameters of each of the data nodes and uses the federated learning technology to aggregate the model parameters and fuse them into a unified global model includes:
[0031] Assign corresponding weights according to the data volume and data quality of each of the data nodes;
[0032] Using the calculated weights, perform weighted averaging on the model parameters of each node to obtain the aggregated global model parameters.
[0033] Further, the weight calculation adopts a hybrid method based on data volume and data quality, and the formula is:
[0034]
[0035] where ω i is the weight of the i-th data node, α is the balance coefficient, with a value range between 0 and 1, n i is the data volume of the i-th data node, N is the total data volume of all data nodes, q i is the data quality score of the i-th data node, is the sum of the data quality scores of all data nodes.
[0036] Further, according to the real-time data and the existing screening results, automatically calculate the optimized mixture composite gradation, and through real-time data feedback and reinforcement learning technology, adjust the optimized global model to obtain an adaptive debugging gradation model, including:
[0037] Obtain the existing screening result data, where the screening result data includes the target particle size distribution and physical properties of the mixture, and integrate it with the real-time data to obtain real-time comprehensive data;
[0038] Calculate the optimized mixture composite gradation based on the optimization algorithm and the real-time comprehensive data;
[0039] Establish a real-time feedback mechanism to collect the actual effects of the optimized grading scheme, and dynamically adjust and optimize the global model through reinforcement learning technology, update and adjust the global model to obtain an adaptive debugging grading model.
[0040] An adaptive debugging grading system, the system includes:
[0041] A data node definition module, which obtains the mixture grading data in historical engineering applications and defines data nodes, and the data nodes include enterprise nodes, laboratory nodes, project station nodes and data center nodes;
[0042] A feature set acquisition module, which preprocesses the mixture grading data and extracts features to obtain a historical data feature set;
[0043] A preliminary model training module, based on machine learning, trains the historical data in the historical data feature set at each of the data nodes;
[0044] A global model aggregation module, based on federated learning technology, uploads the model parameters of each of the data nodes to a central server for global model aggregation and optimization;
[0045] An adaptive model acquisition module, according to real-time data and existing screening results, automatically calculates the optimized mixture composite grading, and adjusts the optimized global model through real-time data feedback and reinforcement learning technology to obtain an adaptive debugging grading model;
[0046] A grading scheme deployment module, based on the adaptive debugging grading model, performs real-time debugging on the grading scheme to obtain a real-time grading scheme.
[0047] Further, the adaptive model acquisition module includes:
[0048] A comprehensive data acquisition unit, which acquires existing screening result data, and the screening result data includes the target particle size distribution and physical properties of the mixture, and integrates it with real-time data to obtain real-time comprehensive data;
[0049] A grading optimization calculation unit, which calculates the optimized mixture composite grading based on an optimization algorithm and the real-time comprehensive data;
[0050] Establish a real-time feedback mechanism to collect the actual effects of the optimized grading scheme, and dynamically adjust and optimize the global model through reinforcement learning technology, update and adjust the global model to obtain an adaptive debugging grading model.
[0051] Through the technical solution of the present invention, the following technical effects can be achieved:
[0052] Through the present invention, the accuracy, efficiency and scientificity of gradation design are improved, the problems of relying on experience, low efficiency and inaccurate results in traditional methods are overcome, and the intelligentization and automation of the gradation design of the mixture are realized. BRIEF DESCRIPTION OF THE DRAWINGS
[0053] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.
[0054] Figure 1 It is a schematic flow chart of an adaptive debugging gradation method;
[0055] Figure 2 It is a schematic flow chart of obtaining the gradation data of the mixture in historical engineering applications and defining data nodes;
[0056] Figure 3 It is a schematic flow chart of obtaining a historical data feature set;
[0057] Figure 4 It is a schematic flow chart of performing preliminary model training on the historical data in the historical data feature set;
[0058] Figure 5 It is a schematic flow chart of globally aggregating and optimizing the model parameters of each data node;
[0059] Figure 6 It is a schematic flow chart of aggregating model parameters using federated learning technology;
[0060] Figure 7 It is a schematic flow chart of adjusting the optimized global model and obtaining an adaptive debugging gradation model. DETAILED DESCRIPTION OF THE EMBODIMENTS
[0061] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments.
[0062] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the technical field to which the present invention belongs. The terms used in the specification of the present invention are only for the purpose of describing specific embodiments, and are not intended to limit the present invention. The term "and / or" used herein includes any and all combinations of one or more of the related listed items.
[0063] Example 1
[0064] As Figure 1 shown, the present invention provides an adaptive debugging gradation method, and the method includes:
[0065] S1: Obtain the mixture gradation data in historical engineering applications and define data nodes, where the data nodes include enterprise nodes, laboratory nodes, project station nodes, and data center nodes;
[0066] Specifically, by comprehensively and systematically collecting and managing the mixture gradation data in historical engineering applications, a high-quality data foundation is provided for subsequent model training and optimization. By defining data nodes (including enterprise nodes, laboratory nodes, project station nodes, and data center nodes), the data format and standards are unified to achieve multiple goals. First, by collecting mixture gradation data from different sources (enterprises, laboratories, project sites), a widely covered historical dataset is constructed; second, the aggregation and management of data center nodes establish a unified data storage and management system, facilitating data integration, analysis, and utilization.
[0067] S2: Preprocess the mixture gradation data and perform feature extraction to obtain a historical data feature set;
[0068] Specifically, by cleaning and filling missing values, the integrity and accuracy of the data are ensured, and the impact of data noise and errors on model training is reduced; by standardizing and normalizing, the magnitude differences between different features are eliminated, improving the comparability of data and the convergence speed of the model; key features most contributing to model prediction are identified and extracted to improve the interpretability and prediction performance of the model; by feature selection and dimensionality reduction processing, redundant and unimportant features are removed, reducing the number of features and improving the model training efficiency.
[0069] S3: Based on machine learning, perform preliminary model training on the historical data in the historical data feature set at each data node;
[0070] Specifically, by independently performing model training on each data node, leveraging the characteristics of local data, the accuracy and prediction ability of the model in a specific environment and conditions are improved; using the computing resources of each data node, model training is carried out distributively, reducing the computing pressure on a single central server and improving the overall training efficiency. At the same time, by performing local training on each data node, the data does not need to be uploaded to the central server, effectively protecting data privacy and security.
[0071] S4: Based on federated learning technology, upload the model parameters of each data node to the central server for global model aggregation and optimization;
[0072] Specifically, Federated Learning is a distributed machine learning method that allows multiple decentralized data nodes (such as user devices, enterprise servers, etc.) to jointly train a global model without sharing the original data. The core goal of federated learning technology is to protect data privacy while leveraging distributed data resources to improve the performance and generalization ability of the model.
[0073] S5: According to the real-time data and the existing screening results, automatically calculate the optimized synthetic gradation of the mixture, and through real-time data feedback and reinforcement learning technology, adjust the optimized global model to obtain an adaptive debugging gradation model.
[0074] Specifically, by combining real-time data with the existing screening results, the optimal synthetic gradation of the mixture is automatically calculated to ensure the best performance of the mixture under different environments and construction conditions. Using the real-time data feedback mechanism and reinforcement learning technology, the global model is continuously and dynamically adjusted so that the model can continuously adapt to the changes in actual applications, improving the model's adaptability and accuracy. By real-time optimizing and adjusting the gradation scheme, the density, stability, and mechanical properties of the mixture during construction are ensured to reach the best, thereby improving the construction quality and efficiency, reducing material waste and costs. This step realizes the intelligentization and automation of the mixture gradation design and debugging, reduces human intervention, reduces the uncertainty and errors in design and construction, improves the overall work efficiency, and through the application of reinforcement learning technology, continuously optimizes the decision-making process of the model and improves the robustness and stability of the model in various complex environments.
[0075] S6: Based on the adaptive debugging gradation model, conduct real-time debugging on the gradation scheme to obtain a real-time gradation scheme.
[0076] Specifically, first, collect the mixture characteristic data in real time from the construction site and relevant sensors, including particle size distribution, material density, and environmental parameters, etc., and transmit the real-time data to the central server or data processing center through a stable and fast data transmission channel. Load the trained and optimized adaptive debugging gradation model on the central server, use the real-time collected data to input into the model for calculation, generate a real-time mixture gradation scheme. Based on the input real-time data, the adaptive debugging gradation model automatically calculates the optimized gradation scheme to ensure that the scheme meets the requirements of the current construction environment and material characteristics, and outputs the calculated real-time gradation scheme to the relevant control system or operator for immediate application to the construction process.
[0077] Through the present invention, the accuracy, efficiency, and scientificity of the gradation design are improved, the problems of relying on experience, low efficiency, and inaccurate results in traditional methods are overcome, and the intelligentization and automation of the mixture gradation design are realized.
[0078] As a preference of the above embodiments, as Figure 2 shown, obtain the mixture gradation data in historical engineering applications and define data nodes, including:
[0079] A10: Identify and determine the data sources from enterprises, laboratories, project sites, and data centers;
[0080] A20: Enterprises, laboratories, project sites, and data centers collect relevant historical engineering data and real-time data and store them in the corresponding data nodes.
[0081] Specifically, obtain the mixture gradation data in historical engineering applications and define data nodes, including identifying and determining the data sources from enterprises, laboratories, project sites, and data centers, respectively collecting, storing, and managing relevant historical engineering data and real-time data through the enterprise, laboratory, project site, and data center nodes, uniformly defining the data format and performing data cleaning and standardization processing to ensure data consistency and quality, establishing an efficient data storage system and a secure data sharing mechanism, and finally using this high-quality data for model training and optimization to support the intelligence and automation of mixture gradation design.
[0082] As a preference of the above embodiments, as Figure 3 shown, in step S2, preprocess the mixture gradation data and perform feature extraction to obtain a set of historical data features, including:
[0083] S21: Clean the mixture gradation data, remove outliers, and perform standardization processing;
[0084] S22: Extract key features from the preprocessed data through data analysis techniques. The key features include, but are not limited to, particle size distribution, material density, and environmental parameters.
[0085] S23: Summarize the extracted features to generate a set of historical data features for subsequent model training and optimization.
[0086] Specifically, first, clean the mixture gradation data to ensure the accuracy and integrity of the data by detecting and removing outliers and filling in missing values; then, perform standardization to normalize the data to a unified scale to avoid the impact of magnitude differences between different features on the model training effect. Common methods include Min-Max standardization and Z-score standardization; next, through data analysis techniques, extract key features from the preprocessed data. These features include but are not limited to particle size distribution, material density, and environmental parameters. At the same time, apply feature engineering techniques to process the original features to generate new and more representative features; finally, summarize all the extracted and selected key features to form a complete set of historical data features and store them in a unified data management system to provide high-quality data support for subsequent model training and optimization.
[0087] As a preference of the above embodiment, as Figure 4 shown, in step S3, based on machine learning, at each data node, perform preliminary model training on the historical data in the set of historical data features, including:
[0088] S31: Allocate and check the local set of historical data features at the enterprise node, laboratory node, project site node, and data center node, and select and initialize the GBDT model parameters;
[0089] S32: Use the local historical data to train the GBDT model and optimize the model parameters through step-by-step iteration;
[0090] S33: Adjust the hyperparameters of the GBDT model according to the training results and retrain the GBDT model using the validation data set that did not participate in the training;
[0091] S34: Save the optimized GBDT model parameters and perform version control to obtain the final model.
[0092] Specifically, first, allocate the local historical data feature sets at the enterprise nodes, laboratory nodes, project site nodes, and data center nodes to ensure the data integrity and consistency of each node, and check and clean the data to ensure there are no missing values and outliers, laying a solid foundation for model training; then, select the Gradient Boosting Decision Tree (GBDT) model as the training model and initialize the model parameters, including the number of trees, maximum depth, learning rate, etc., to prepare for preliminary training; next, use the local historical data to train the GBDT model. By gradually iterating and optimizing the model parameters, in each round of iteration, the model is trained according to the current parameters, and the training effect is evaluated through the loss function. The parameters are continuously adjusted to minimize the error. According to the preliminary training results, adjust the hyperparameters of the GBDT model, such as the number of trees, maximum depth, and learning rate, etc. Use the validation data set that did not participate in the initial training to test the model, evaluate its generalization ability and performance, and then further optimize the hyperparameters according to the validation results and retrain the model to ensure that the model performs equally well on new data; finally, save the optimized GBDT model parameters to ensure that the model's parameters, training data, and hyperparameter settings are completely recorded and version control is performed to record the detailed information of each model version to ensure the traceability and reproducibility of the model. The reason for choosing the GBDT (Gradient Boosting Decision Tree) model in this solution is its high prediction accuracy, robustness, and stability. It can effectively process various data types and provide feature importance evaluation, and at the same time has the ability to prevent overfitting. It is suitable for efficient training and prediction on large-scale data sets, and there are mature implementation tools and libraries to support it, which is convenient for quickly deploying and training models at each data node, thus ensuring the accuracy and reliability of the mixture gradation optimization.
[0093] As a preference of the above embodiment, as Figure 5 shown, in step S4, based on the federated learning technology, upload the model parameters of each data node to the central server for global model aggregation and optimization;
[0094] S41: After each data node completes the preliminary model training, upload the trained GBDT model parameters to the central server through the secure communication protocol;
[0095] S42: The central server receives the model parameters of each data node and aggregates the model parameters using the federated learning technology to fuse them into a unified global model;
[0096] S43: Use an independent validation data set to test the performance of the global model and distribute the finally optimized model to each data node.
[0097] Specifically, after each data node completes the preliminary model training, it uploads the trained GBDT model parameters, including key model information such as model weights and biases, to the central server through an encrypted secure communication protocol to ensure the confidentiality and integrity of the data during transmission. The central server receives the model parameters uploaded from all data nodes and performs data preprocessing to ensure that all parameter formats are consistent. Subsequently, using the aggregation algorithm in federated learning technology (such as Federated Averaging, FedAvg), the model parameters of each node are weighted averaged or other aggregation methods to fuse the individual model parameters into a unified global model. The aggregation process takes into account the data volume and importance of each node and assigns appropriate weights to ensure that the global model can fully reflect the data characteristics of all nodes. The performance of the aggregated global model is tested on an independent validation dataset on the central server to evaluate its performance on unseen data. According to the validation results, the global model parameters are further optimized, and the model hyperparameters (such as learning rate, regularization parameter, etc.) are adjusted to ensure that the generalization ability and prediction accuracy of the model meet the expected requirements. Finally, the optimized global model parameters are distributed to each data node to replace the initial local model parameters. Each data node receives the optimized global model parameters and performs local validation and application to ensure the effectiveness and performance of the model in the actual environment. Through these specific steps, the secure upload of the model parameters of each data node and the effective aggregation and optimization of the global model are achieved using federated learning technology, thereby generating an efficient and accurate global model and distributing it to each data node to improve the prediction ability and application effect of the overall system.
[0098] As a preference of the above embodiment, as Figure 6 shown, in step S42, the central server receives the model parameters of each data node and aggregates the model parameters using federated learning technology to fuse them into a unified global model, including:
[0099] S421: Assign corresponding weights according to the data volume and data quality of each data node;
[0100] S422: Use the calculated weights to perform weighted averaging on the model parameters of each node to obtain the aggregated global model parameters.
[0101] Specifically, as a preference of the above embodiment, the central server receives the model parameters of each data node and uses the aggregation method of federated learning technology to fuse them into a unified global model, including:
[0102] Assign corresponding weights according to the data volume and data quality of each data node;
[0103] Using the calculated weights, perform a weighted average on the model parameters of each node to obtain the aggregated global model parameters.
[0104] Specifically, corresponding weights are assigned according to the data volume and data quality of each data node. Nodes with a large data volume will be assigned higher weights, and nodes with high data quality will also receive relatively high weights. Then, using the calculated weights, a weighted average is performed on the model parameters of each node. By multiplying the model parameters of each node by the corresponding weights and summing them up, the aggregated global model parameters are obtained. Through these steps, the central server can effectively utilize the data volume and quality information of each data node, reasonably allocate weights, and perform a weighted average to generate a global model parameter that synthesizes the advantages of all nodes. This method not only ensures the accuracy and generalization ability of the global model but also fully utilizes the characteristics and advantages of different data nodes, improving the prediction ability and application effect of the overall system.
[0105] As a preference of the above embodiment, the weight calculation adopts a hybrid method based on data volume and data quality, and the formula is:
[0106]
[0107] where ω i is the weight of the i-th data node, α is the balance coefficient, and its value range is between 0 and 1, n i is the data volume of the i-th data node, N is the total data volume of all data nodes, q i is the data quality score of the i-th data node, and is the sum of the data quality scores of all data nodes.
[0108] Specifically, determine the data volume and data quality scores of each data node. The data volume is usually determined by the number of data samples provided by each node, and the data quality score can be evaluated according to the performance indicators of the node model (such as accuracy, mean square error, etc.). Then, select an appropriate balance coefficient, which is used to balance the influence of data volume and data quality in weight calculation, and its value range is between 0 and 1. Then, use the above formula to calculate the weight of each data node. This weight combines the information of both data volume and data quality, ensuring that the contribution of each node to the overall model is comprehensively considered. After the calculation, use these weights to perform a weighted average on the model parameters of each node. The specific method is to multiply the model parameters of each node by the corresponding weights and then sum up all the weighted parameters to obtain the final global model parameters. Through this hybrid method of weight calculation, the two factors of data volume and data quality can be effectively considered comprehensively, enabling the global model to more accurately reflect the characteristics and advantages of each data node and improving the overall performance and generalization ability of the model.
[0109] As a preference of the above embodiments, as Figure 7 shown, in step S5, according to the real-time data and the existing screening results, the optimized synthetic gradation of the mixture is automatically calculated, and through real-time data feedback and reinforcement learning technology, the optimized global model is adjusted to obtain an adaptive debugging gradation model, including:
[0110] S51: Obtain the existing screening result data, where the screening result data includes the target particle size distribution and physical properties of the mixture, and integrate it with the real-time data to obtain real-time comprehensive data;
[0111] S52: Calculate the optimized synthetic gradation of the mixture based on the optimization algorithm and the real-time comprehensive data;
[0112] S53: Establish a real-time feedback mechanism, collect the actual effects of the optimized gradation scheme, and dynamically adjust and optimize the global model through reinforcement learning technology, update and adjust the global model to obtain an adaptive debugging gradation model.
[0113] Specifically, first, obtain the existing screening result data, which includes the target particle size distribution and physical properties of the mixture, and integrate these screening result data with the mixture property data (such as particle size distribution, material density, environmental parameters, etc.) collected in real time from the construction site and sensors to form real-time comprehensive data. This step ensures the comprehensiveness and timeliness of the data required for the optimized calculation; then, calculate the optimized synthetic gradation of the mixture based on the optimization algorithm and the real-time comprehensive data. By using appropriate optimization algorithms (such as linear programming, genetic algorithms, etc.), calculate the optimal synthetic gradation scheme of the mixture that can meet the current construction requirements and environmental conditions. This step ensures the high efficiency and adaptability of the mixture in practical applications; then, establish a real-time feedback mechanism, continuously collect the effect data of the optimized gradation scheme in actual construction, analyze these real-time feedback data to evaluate the actual performance of the optimized gradation scheme, and identify areas that need improvement; finally, apply reinforcement learning technology to dynamically adjust and optimize the global model according to the real-time feedback data. Reinforcement learning enables the model to gradually adapt to the changes and feedback in actual applications through continuous iteration and learning, update and adjust the global model, and finally obtain an adaptive debugging gradation model. Through these specific steps, it is possible to automatically calculate the optimized synthetic gradation of the mixture according to the real-time data and the existing screening results, and dynamically adjust and optimize the global model through real-time data feedback and reinforcement learning technology to ensure the accuracy and adaptability of the mixture gradation scheme, thereby achieving efficient and high-quality construction.
[0114] Embodiment 2
[0115] Based on the same inventive concept as an adaptive debugging gradation method in the foregoing embodiments, the present invention also provides an adaptive debugging gradation system, which includes:
[0116] A data node definition module, which obtains the mixture gradation data in historical engineering applications and defines data nodes, including enterprise nodes, laboratory nodes, project station nodes, and data center nodes;
[0117] A feature set acquisition module, which preprocesses the mixture gradation data and extracts features to obtain a historical data feature set;
[0118] A preliminary model training module, which based on machine learning, conducts preliminary model training on the historical data in the historical data feature set at each data node;
[0119] A global model aggregation module, which based on federated learning technology, uploads the model parameters of each data node to a central server for global model aggregation and optimization;
[0120] An adaptive model acquisition module, which automatically calculates the optimized mixture composite gradation according to real-time data and existing screening results, and adjusts the optimized global model through real-time data feedback and reinforcement learning technology to obtain an adaptive debugging gradation model;
[0121] A gradation scheme deployment module, which conducts real-time debugging on the gradation scheme based on the adaptive debugging gradation model to obtain a real-time gradation scheme.
[0122] The above-mentioned gradation system in the present invention can effectively implement the adaptive debugging gradation method, and the technical effects that can be achieved are as described in the foregoing embodiments, which will not be elaborated here.
[0123] As a preference of the above embodiment, the adaptive model acquisition module includes:
[0124] A comprehensive data acquisition unit, which obtains existing screening result data, including the target particle size distribution and physical properties of the mixture, and integrates it with real-time data to obtain real-time comprehensive data;
[0125] A gradation optimization calculation unit, which calculates the optimized mixture composite gradation based on an optimization algorithm and real-time comprehensive data;
[0126] Establish a real-time feedback mechanism, collect the actual effects of the optimized gradation scheme, dynamically adjust and optimize the global model through reinforcement learning technology, update and adjust the global model to obtain an adaptive debugging gradation model.
[0127] Similarly, for the above optimization schemes of the system, the corresponding optimization effects of the method in the first embodiment can also be respectively achieved, which will not be elaborated here either.
[0128] Although the present application has been described in connection with specific features and their embodiments, it will be apparent that various modifications and combinations can be made without departing from the spirit and scope of the present application. Accordingly, the present specification and drawings are merely exemplary illustrations of the present application as defined by the appended claims, and are considered to cover any and all modifications, variations, combinations, or equivalents within the scope of the present application. Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalent technologies, the present application is intended to include these changes and modifications.
Claims
1. An adaptive debugging grading method, characterized in that: include: Acquire mixture gradation data in historical engineering applications and define data nodes, which include enterprise nodes, laboratory nodes, project station nodes and data center nodes; Preprocessing the mixture gradation data and performing feature extraction to obtain a historical data feature set; Based on machine learning, at each of the data nodes, preliminary model training is performed on the historical data in the historical data feature set; Based on federated learning technology, the model parameters of each data node are uploaded to the central server for global model aggregation and optimization; According to the real-time data and the existing screening results, the optimized mixture synthesis gradation is automatically calculated, and the optimized global model is adjusted through real-time data feedback and reinforcement learning technology to obtain an adaptive debugging gradation model; The grading scheme is debugged in real time based on the self-adaptive debugging grading model to obtain a real-time grading scheme.
2. The adaptive debugging grading method according to claim 1, characterized in that: Obtain mixture gradation data from historical engineering applications and define data nodes, including: Identify and determine data sources from enterprises, laboratories, project sites, and data centers; Enterprises, laboratories, project sites and data centers collect relevant historical engineering data and real-time data and store them in the corresponding data nodes.
3. The adaptive debugging grading method according to claim 1, characterized in that: The mixture gradation data is preprocessed and feature extracted to obtain a historical data feature set, including: Cleaning the mixture gradation data, removing outliers, and performing standardization processing; Extracting key features from the preprocessed data using data analysis techniques, including but not limited to particle size distribution, material density, and environmental parameters; The extracted features are aggregated to generate a historical data feature set for subsequent model training and optimization.
4. The adaptive debugging grading method according to claim 1, characterized in that: Based on machine learning, preliminary model training is performed on the historical data in the historical data feature set at each of the data nodes, including: Allocate and check local historical data feature sets at enterprise nodes, laboratory nodes, project site nodes, and data center nodes, and select and initialize GBDT model parameters; Use local historical data to train the GBDT model and optimize the model parameters through gradual iteration; Adjust the hyperparameters of the GBDT model according to the training results, and retrain the GBDT model using a validation dataset that was not involved in the training; The optimized GBDT model parameters are saved and version controlled to obtain the final model.
5. The adaptive debugging grading method according to claim 1, characterized in that: Based on federated learning technology, the model parameters of each data node are uploaded to the central server for global model aggregation and optimization; After completing the preliminary model training, each of the data nodes uploads the trained GBDT model parameters to the central server through a secure communication protocol; The central server receives the model parameters of each of the data nodes, and aggregates the model parameters using federated learning technology to merge them into a unified global model; The performance of the global model is tested using an independent validation dataset, and the final optimized model is assigned to each of the data nodes.
6. The adaptive debugging grading method according to claim 1, characterized in that: The central server receives the model parameters of each of the data nodes and aggregates the model parameters using federated learning technology to merge them into a unified global model, including: Allocate corresponding weights according to the data volume and data quality of each of the data nodes; The calculated weights are used to perform weighted averaging on the model parameters of each node to obtain the aggregated global model parameters.
7. The adaptive debugging grading method according to claim 6, characterized in that: The weight calculation adopts a hybrid method based on data quantity and data quality, and the formula is: Among them, ω i is the weight of the i-th data node, α is the balance coefficient, ranging from 0 to 1, n i is the amount of data of the ith data node, N is the total amount of data of all data nodes, q i Score the data quality of the i-th data node, The sum of the data quality scores of all data nodes.
8. The adaptive debugging grading method according to claim 1, characterized in that: According to the real-time data and the existing screening results, the optimized mixture synthesis gradation is automatically calculated, and the optimized global model is adjusted through real-time data feedback and reinforcement learning technology to obtain an adaptive debugging gradation model, including: Acquire existing screening result data, which includes target particle size distribution and physical properties of the mixture, and integrate it with real-time data to obtain real-time comprehensive data; Calculate the optimized mixture synthesis gradation based on the optimization algorithm and the real-time comprehensive data; A real-time feedback mechanism is established to collect the actual effects of the optimized grading scheme, dynamically adjust and optimize the global model through reinforcement learning technology, update and adjust the global model, and obtain an adaptive debugging grading model.
9. An adaptive debugging grading system, characterized in that: The system comprises: A data node definition module obtains mixture gradation data in historical engineering applications and defines data nodes, including enterprise nodes, laboratory nodes, project station nodes and data center nodes; A feature set acquisition module preprocesses the mixture gradation data and performs feature extraction to obtain a historical data feature set; A preliminary model training module, based on machine learning, performs preliminary model training on the historical data in the historical data feature set at each of the data nodes; A global model aggregation module, based on federated learning technology, uploads the model parameters of each data node to the central server to perform global model aggregation and optimization; The adaptive model acquisition module automatically calculates the optimized mixture synthesis gradation according to the real-time data and the existing screening results, and adjusts the optimized global model through real-time data feedback and reinforcement learning technology to obtain an adaptive debugging gradation model; The grading scheme adjustment module performs real-time debugging on the grading scheme based on the adaptive debugging grading model to obtain a real-time grading scheme.
10. The adaptive debugging grading system according to claim 9, characterized in that: The adaptive model acquisition module comprises: A comprehensive data acquisition unit acquires existing screening result data, the screening result data including target particle size distribution and physical properties of the mixture, and integrates the data with the real-time data to obtain real-time comprehensive data; A gradation optimization calculation unit calculates an optimized mixture synthesis gradation based on an optimization algorithm and the real-time comprehensive data; A real-time feedback mechanism is established to collect the actual effects of the optimized grading scheme, dynamically adjust and optimize the global model through reinforcement learning technology, update and adjust the global model, and obtain an adaptive debugging grading model.