Model optimization data management system based on interactive feedback domain experience and its application

Abstract

The present invention relates to the field of industrial model management technology, and specifically to a model optimization data management system based on experience in the field of interactive feedback and its application. The present invention combs the debugging data generated by engineers in the actual use of industrial application models, captures the debugging data association between corresponding model parameter items and corresponding application output items according to the data change law of the industrial application model output results caused by parameter value changes on the corresponding model parameter items, constructs a debugging structure, and realizes the evaluation of the application normalization capability of the industrial application model. The application can carry out real-time monitoring of the application status of the industrial application model, and timely warn the industrial application model of poor performance when encountering new application scenarios, assist management personnel in optimizing decision-making and judgment on the industrial application model, so that the industrial application model can maintain high performance and a wide range of applications for a long time.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial model management, and in particular to a model optimization data management system based on interactive feedback field experience and its application. Background Art

[0002] An industrial application model refers to a mathematical model used to describe and analyze an industrial system. It can represent the relationships between the various parts of an industrial system through mathematical formulas or computer simulations. The establishment of an industrial application model can help us better understand and predict the operation of an industrial system and provide a scientific basis for decision-making.

[0003] Typically, industrial application models include production optimization models, which are used to optimize industrial production scheduling and reduce production costs; and equipment fault diagnosis models, which are used to provide timely warnings of abnormal conditions that exist during equipment operation, and provide companies with a basis for fault warnings and maintenance decisions. The generalization capability of an industrial application model refers to the ability of an industrial application model to correctly understand and predict new, unseen data. In the field of machine learning and artificial intelligence, the generalization capability of a model is one of the important indicators for evaluating model performance.

[0004] The commissioning phase of industrial application models mainly involves the evaluation of model accuracy and performance, as well as the adaptability and optimization of models in practical applications. The key to this phase is to ensure that the model can perform tasks accurately and efficiently in complex and changing industrial environments, while taking into account the interpretability of the model and user trust. Usually, before the formal evaluation begins, it is necessary to collect necessary information and data, including the model's training data set, documents, etc.; during the evaluation process, it is necessary to avoid over-reliance on a single indicator, such as accuracy, while ignoring other important performance indicators, such as precision and recall; the interpretability of the model is very important for building user trust and meeting regulatory requirements. It is necessary to limit the model with a standard prompt word framework to make the model's answers more in line with the requirements. According to the evaluation results, the model is optimized, including parameter adjustment, algorithm improvement, etc., to improve the performance of the model in specific industrial scenarios. Considering the complexity and diversity of the industrial environment, the model needs to have a certain generalization ability to adapt to different operating conditions and task requirements. Summary of the invention

[0005] The purpose of the present invention is to provide a model optimization data management system based on interactive feedback field experience and its application to solve the problems raised in the above background technology.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: a model optimization data management system based on interactive feedback field experience, the system includes a model commissioning data management module, a model commissioning structure identification module, a model commissioning anomaly judgment module, and a model evaluation feedback management module;

[0007] The model commissioning data management module is used to record each historical application of the industrial application model and sort out the commissioning data generated by engineers in the actual use of the industrial application model;

[0008] In the operation of rail transit network, the main industrial application models are BIM (Building Information Model) and digital twin technology. The application of these technologies provides intelligent and efficient tools for the full life cycle management of rail transit, playing an important role in all aspects from design, construction to operation and maintenance;

[0009] In the rail transit design stage, BIM technology can be used for collision inspection, pipeline integration, transfer scheme simulation, design scheme comparison, limit optimization design, and reserved and embedded inspection, so as to improve the design quality and reduce later design changes and construction rework; in the construction stage, BIM technology can be applied to site planning, construction simulation, etc., effectively guiding site layout and on-site construction, and reducing project construction risks and costs; in the operation stage, BIM models can be combined with real-time collected data to create digital twins, realize real-time monitoring and prediction of urban energy consumption, traffic flow, environmental quality, etc., optimize train scheduling, improve operational service quality, and perform preventive maintenance of facilities;

[0010] The model debugging structure identification module is used to collect the debugging data of all historical application records, capture the debugging data association between the corresponding model parameter items and the corresponding application output items according to the data change law of the industrial application model output results caused by the parameter value changes on the corresponding model parameter items, build the debugging structure, and identify and extract the characteristic debugging structure;

[0011] The model commissioning anomaly judgment module is used to collect all historical application records, classify and divide all historical application records based on the commissioning data, and obtain several historical application record sets; in each historical application record set, according to the distribution of the extracted characteristic commissioning structure, the historical application records with industrial application model commissioning anomalies are judged and identified;

[0012] The model evaluation feedback management module is used to set unit cycles, evaluate the application generalization capabilities of industrial application models based on the distribution of historical application records of abnormal industrial application model adjustments in each unit cycle, and provide feedback to management personnel to assist management personnel in making decisions on whether to optimize the industrial application model.

[0013] Preferably, the model commissioning data management module includes:

[0014] Whenever an engineer chooses to adjust the parameter value of at least one model parameter item in the industrial application model and performs a model application test, the parameter values ​​of each model parameter item are extracted in turn to obtain the model parameter value sequence P={Y 1 ,Y 2 ,...,Y n}, where Y 1 ,Y 2 ,...,Y n Respectively represent the parameter values ​​of the 1st, 2nd, ..., nth model parameter items;

[0015] Capture the initial parameter values ​​of each model parameter item corresponding to the industrial application model, and obtain the initial model parameter value sequence P of the industrial application model 0 , extract the corresponding model parameter value sequence for each model application test executed in each historical application record in chronological order, and collect the test sequence L={P 0 ,P 1 ,P 2 ,...,P n}, where P 1 ,P 2 ,...,P n Respectively represent the model parameter value sequences corresponding to the 1st, 2nd, ..., nth model application tests, and P n As a feature sequence of the corresponding historical application record.

[0016] Preferably, the model debugging structure identification module includes a debugging node information management unit and a debugging structure construction unit;

[0017] The test node information management unit is used to extract the test sequence L={P 0 ,P 1 ,P 2 ,...,P n}, set two adjacent model parameter value sequences in each test sequence as a test node; if in a test node, two adjacent model parameter value sequences A and B are in the i-th model parameter item F i The corresponding parameter values ​​are different, extract the i-th model parameter item F i As target parameter items, the output results obtained after the engineers complete the model application test according to A and B are obtained respectively, and the application output items with numerical deviations are extracted as target output items;

[0018] The debugging structure building unit is used to establish debugging data associations between each target parameter item extracted from each debugging node and each target output item, and build and generate a plurality of debugging structures.

[0019] Preferably, the model tuning structure identification module includes a characteristic tuning structure screening unit;

[0020] The characteristic tuning structure screening unit is used to calculate the tuning characteristic index α=g / M for each tuning structure, where g represents the total number of times each tuning structure is extracted cumulatively, and M represents the total number of tuning structures extracted from all tuning nodes; the tuning structure whose tuning characteristic index α is greater than the index threshold is judged as a characteristic tuning structure;

[0021] Because in the comparison results of the model parameter value sequences of two consecutive model application tests, there are often multiple model parameter items whose parameter values ​​are adjusted. Therefore, it can only be explained that the adjustment of the parameter values ​​of these multiple model parameter items is the reason for the different output results obtained after the two consecutive model application tests, but it is impossible to establish an absolute correlation between a certain application output item whose specific output value has changed and a certain model parameter item whose specific parameter value has been adjusted.

[0022] Preferably, the model debugging anomaly judgment module includes a historical application record classification management unit, which is used to calculate the similarity of feature sequences between every two historical application records, and classify the historical application records whose feature sequence similarities are less than a similarity threshold into one category for aggregation, to obtain a plurality of historical application record sets;

[0023] Used to extract the total application debugging time spent by engineers from each historical application record; obtain the average application time in each historical application record set;

[0024] Used to set, in each historical application record set, historical application records whose total application debugging duration is greater than the corresponding average application duration as first characteristic application records, and historical application records whose total application debugging duration is less than or equal to the corresponding average application duration as second characteristic application records;

[0025] The total application debugging duration is used as a reference for classifying historical application records. It is assumed that the time required to adjust the industrial application model to the best state that is most suitable for the current application should be similar. If the deviation is too large, that is, it exceeds the average application duration too much, it means that the current feature sequence, that is, the model parameter setting, can only make the industrial application model reach the best state that is most suitable for the current application. When it does not meet the best state of the current application, that is, the industrial application model is not suitable for the current application, or the current application exceeds the universal scope of the industrial application model.

[0026] Preferably, the model debugging anomaly judgment module includes a feature tag management unit, which is used to collect all feature debugging structures extracted from the corresponding test sequence for any historical application record, and respectively accumulate the total number of extractions corresponding to each feature debugging structure in any historical application record;

[0027] It is used to obtain the average total extraction times K corresponding to any feature tuning structure in all second feature application records in any historical application record set. When the total extraction times Q corresponding to a first feature application record in any historical application record set and the average total extraction times K satisfy QK>β, where β is the times threshold, the application output item in any feature tuning structure is marked as a feature corresponding to a first feature application record.

[0028] Preferably, the model debugging anomaly judgment module includes a debugging anomaly judgment management unit, which is used to calculate the anomaly index δ=h / W for each first feature application record in any historical application record set, wherein h represents the total number of application output items having feature tags corresponding to each first feature application record, and W represents the total number of application output items; if the anomaly index of a first feature application record is greater than the index threshold, it is judged that an industrial application model debugging anomaly is presented on the first feature application record.

[0029] Preferably, the model evaluation feedback management module includes:

[0030] The historical application records covered in each unit period are collected to obtain the total number of historical application records corresponding to each unit period U. The total number of historical application records with abnormal industrial application model commissioning in each unit period E is accumulated, and the abnormal distribution rate S=E / U in each unit period is calculated;

[0031] The abnormal distribution rates corresponding to each unit period are collected in chronological order to obtain an abnormal distribution rate sequence. When V consecutive times satisfying Sj+1-Sj>0 are captured in the abnormal distribution rate sequence, the management port is fed back to indicate that the application normalization capability of the industrial application model has declined, prompting the management personnel to optimize the industrial application model.

[0032] On the other hand, the present invention provides an application of a model optimization data management system based on interactive feedback field experience, which is applied in rail transit network operation. The application includes: using the system to manage application models involved in the rail transit design stage, rail transit construction stage and rail transit operation stage.

[0033] Compared with the prior art, the beneficial effects achieved by the present invention are as follows: the present invention sorts out the debugging data generated by engineers in the actual use of industrial application models, and captures the debugging data association between the corresponding model parameter items and the corresponding application output items according to the data change law of the industrial application model output results caused by the parameter value changes on the corresponding model parameter items, and constructs a debugging structure. At the same time, according to the distribution of historical application records of abnormal industrial application model debugging in each unit period, the application generalization capability of the industrial application model is evaluated. While conducting real-time monitoring of the application status of the industrial application model, the application can provide timely warnings when the industrial application model performs poorly when encountering new application scenarios, and assist managers in optimizing decision-making and judgment on the industrial application model, so that the industrial application model can maintain high performance and a wide range of applications for a long time. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0035] Figure 1 It is a structural diagram of a large model optimization data management system based on field experience of interactive feedback of the present invention. DETAILED DESCRIPTION

[0036] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0037] See also Figure 1 , the present invention provides a technical solution: a model optimization data management system based on interactive feedback field experience, the system includes a model commissioning data management module, a model commissioning structure identification module, a model commissioning anomaly judgment module, and a model evaluation feedback management module;

[0038] The model commissioning data management module is used to record each historical application of the industrial application model and sort out the commissioning data generated by engineers in the actual use of the industrial application model;

[0039] The model commissioning data management module includes:

[0040] Whenever an engineer chooses to adjust the parameter value of at least one model parameter item in the industrial application model and performs a model application test, the parameter values ​​of each model parameter item are extracted in turn to obtain the model parameter value sequence P={Y 1 ,Y 2 ,...,Y n}, where Y 1 ,Y 2 ,...,Y n Respectively represent the parameter values ​​of the 1st, 2nd, ..., nth model parameter items;

[0041] Capture the initial parameter values ​​of each model parameter item corresponding to the industrial application model, and obtain the initial model parameter value sequence P of the industrial application model 0 , extract the corresponding model parameter value sequence for each model application test executed in each historical application record in chronological order, and collect the test sequence L={P 0 ,P 1 ,P 2 ,...,P n}, where P 1 ,P 2 ,...,P n Respectively represent the model parameter value sequences corresponding to the 1st, 2nd, ..., nth model application tests, and P n as a sequence of features recorded in the corresponding historical applications;

[0042] The model debugging structure identification module is used to collect the debugging data of all historical application records, capture the debugging data association between the corresponding model parameter items and the corresponding application output items according to the data change law of the industrial application model output results caused by the parameter value changes on the corresponding model parameter items, build the debugging structure, and identify and extract the characteristic debugging structure;

[0043] Among them, the model debugging structure identification module includes a debugging node information management unit, a debugging structure construction unit, and a characteristic debugging structure screening unit;

[0044] The test node information management unit is used to extract the test sequence L={P 0 ,P 1 ,P 2 ,...,P n}, set two adjacent model parameter value sequences in each test sequence as a test node; if in a test node, two adjacent model parameter value sequences A and B are in the i-th model parameter item F i The corresponding parameter values ​​are different, extract the i-th model parameter item F iAs target parameter items, the output results obtained after the engineers complete the model application test according to A and B are obtained respectively, and the application output items with numerical deviations are extracted as target output items;

[0045] The debugging structure building unit is used to establish debugging data association between each target parameter item extracted from each debugging node and each target output item, and build and generate a plurality of debugging structures;

[0046] The characteristic tuning structure screening unit is used to calculate the tuning characteristic index α=g / M for each tuning structure, where g represents the total number of times each tuning structure is extracted cumulatively, and M represents the total number of tuning structures extracted from all tuning nodes; the tuning structure whose tuning characteristic index α is greater than the index threshold is judged as a characteristic tuning structure;

[0047] The model commissioning anomaly judgment module is used to collect all historical application records, classify and divide all historical application records based on the commissioning data, and obtain several historical application record sets; in each historical application record set, according to the distribution of the extracted characteristic commissioning structure, the historical application records with industrial application model commissioning anomalies are judged and identified;

[0048] The model debugging anomaly judgment module includes a historical application record classification management unit, which is used to calculate the similarity of feature sequences between every two historical application records, and classify the historical application records whose feature sequence similarities are less than the similarity threshold into one category to obtain several historical application record sets;

[0049] Used to extract the total application debugging time spent by engineers from each historical application record; obtain the average application time in each historical application record set;

[0050] Used to set, in each historical application record set, historical application records whose total application debugging duration is greater than the corresponding average application duration as first characteristic application records, and historical application records whose total application debugging duration is less than or equal to the corresponding average application duration as second characteristic application records;

[0051] Among them, the model debugging anomaly judgment module includes a debugging anomaly judgment management unit, which is used to calculate the anomaly index δ=h / W for each first feature application record in any historical application record set, wherein h represents the total number of application output items with feature tags corresponding to each first feature application record, and W represents the total number of application output items; if the anomaly index of a first feature application record is greater than the index threshold, it is judged that an industrial application model debugging anomaly is presented on a first feature application record;

[0052] The model evaluation feedback management module is used to set the unit cycle, evaluate the application normalization capability of the industrial application model according to the distribution of the historical application records of the industrial application model debugging anomalies in each unit cycle, and provide feedback to the management personnel to assist the management personnel in making decisions on whether to optimize the industrial application model;

[0053] Among them, the model evaluation feedback management module includes:

[0054] The historical application records covered in each unit period are collected to obtain the total number of historical application records corresponding to each unit period U. The total number of historical application records with abnormal industrial application model commissioning in each unit period E is accumulated, and the abnormal distribution rate S=E / U in each unit period is calculated;

[0055] The abnormal distribution rates corresponding to each unit period are collected in chronological order to obtain an abnormal distribution rate sequence; when V times of satisfying Sj+1-Sj>0 are captured in the abnormal distribution rate sequence in succession, the management port is fed back to indicate that the application normalization capability of the industrial application model has declined, prompting the management personnel to optimize the industrial application model;

[0056] The model debugging anomaly judgment module includes a feature tag management unit, which is used to collect all feature debugging structures extracted from the corresponding test sequence for any historical application record, and respectively accumulate the total number of extractions corresponding to each feature debugging structure in any historical application record;

[0057] Used to obtain, in any historical application record set, the average total number of extractions K corresponding to any feature tuning structure in all second feature application records in any historical application record set; when the total number of extractions Q corresponding to a first feature application record in any historical application record set by any feature tuning structure and the average total number of extractions K satisfy QK>β, where β is a number threshold, the application output item in any feature tuning structure is marked as a feature corresponding to a first feature application record;

[0058] For example, a historical application record set includes {historical application record 1, historical application record 2, historical application record 3, historical application record 4, historical application record 5};

[0059] Among them, historical application records 1, historical application records 3, and historical application records 5 are all second characteristic application records, and historical application records 2 and historical application records 4 are all first characteristic application records;

[0060] Among them, the feature debugging structure: {model parameter item a, application output item b} is extracted a total of 10 times in historical application record 1, a total of 12 times in historical application record 2, a total of 8 times in historical application record 3, a total of 16 times in historical application record 4, and a total of 9 times in historical application record 5.

[0061] In summary, the feature tuning structure: {model parameter item a, application output item b} corresponds to the average total number of extractions in all second feature application records K = (10 + 8 + 9) / 3 = 9;

[0062] When the number threshold β=3, it can be seen from the above that the total number of extractions corresponding to the feature tuning structure: {model parameter item a, application output item b} in historical application record 2 is Q=12, which satisfies 12-9=3=β; the total number of extractions corresponding to the feature tuning structure: {model parameter item a, application output item b} in historical application record 4 is Q=16, which satisfies 16-9=5>β. Therefore, the application output item b in the feature tuning structure 1: {model parameter item a, application output item b} needs to be marked as the feature corresponding to historical application record 4.

[0063] On the other hand, the present invention provides an application of a model optimization data management system based on interactive feedback field experience in rail transit network operation, which includes: using the system to manage BIM models involved in the rail transit design stage, rail transit construction stage and rail transit operation stage.

[0064] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0065] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A model optimization data management system based on interactive feedback domain experience, characterized by: The system includes a model debugging data management module, a model debugging structure identification module, a model debugging anomaly judgment module, and a model evaluation feedback management module; The model commissioning data management module is used to sort out the commissioning data generated by engineers in the process of actually using the industrial application model for each historical application record of the industrial application model; The model debugging structure identification module is used to collect the debugging data of all historical application records, capture the debugging data association between the corresponding model parameter items and the corresponding application output items according to the data change law of the industrial application model output results caused by the parameter value change on the corresponding model parameter items, construct the debugging structure, and identify and extract the characteristic debugging structure; The model commissioning anomaly judgment module is used to collect all historical application records, classify and divide all historical application records based on the commissioning data, and obtain several historical application record sets; in each historical application record set, according to the distribution of the extracted characteristic commissioning structure, judge and identify the historical application records with industrial application model commissioning anomalies; The model evaluation feedback management module is used to set the unit cycle, evaluate the application normalization capability of the industrial application model according to the distribution of the historical application records of the industrial application model debugging anomalies in each unit cycle, and provide feedback to the management personnel port to assist the management personnel in making a decision on whether to optimize the industrial application model; The model debugging data management module includes: Whenever an engineer chooses to adjust the parameter value of at least one model parameter item in the industrial application model and performs a model application test, the parameter values ​​of each model parameter item are extracted in turn to obtain a model parameter value sequence P={Y1,Y2,...,Y n }, where Y1,Y2,...,Y n Respectively represent the parameter values ​​of the 1st, 2nd, ..., nth model parameter items; Capture the initial parameter values ​​of each model parameter item corresponding to the industrial application model to obtain the initial model parameter value sequence P0 of the industrial application model. Extract the corresponding model parameter value sequence for each model application test executed in each historical application record in chronological order, and collect the test sequence L={P0,P1,P2,...,P n }, where P1, P2, ..., P n Respectively represent the model parameter value sequences corresponding to the 1st, 2nd, ..., nth model application tests, and P n as a sequence of features recorded in the corresponding historical applications; The model debugging anomaly judgment module includes a historical application record classification management unit, which is used to calculate the similarity of feature sequences between every two historical application records, and classify the historical application records whose feature sequence similarities are less than the similarity threshold into one category to obtain several historical application record sets; Used to extract the total application debugging time spent by engineers from each historical application record; obtain the average application time in each historical application record set; Used to set, in each historical application record set, historical application records whose total application debugging duration is greater than the corresponding average application duration as first characteristic application records, and historical application records whose total application debugging duration is less than or equal to the corresponding average application duration as second characteristic application records; The model debugging anomaly judgment module includes a feature tag management unit, which is used to collect all feature debugging structures extracted from the corresponding test sequence for any historical application record, and respectively accumulate the total number of extractions corresponding to each feature debugging structure in the any historical application record; Used to obtain, in any historical application record set, the average total number of extractions K corresponding to any feature tuning structure in all second feature application records in the any historical application record set, when the total number of extractions Q corresponding to a first feature application record in the any historical application record set by the any feature tuning structure satisfies QK>β with the average total number of extractions K, where β is a number threshold, the application output item in the any feature tuning structure is marked as a feature corresponding to the first feature application record; The model debugging anomaly judgment module includes a debugging anomaly judgment management unit, which is used to calculate the anomaly index δ=h / W for each first feature application record in any historical application record set, wherein h represents the total number of application output items having feature tags corresponding to the first feature application records, and W represents the total number of application output items; if the anomaly index of a first feature application record is greater than the index threshold, it is judged that an industrial application model debugging anomaly is presented on the first feature application record.

2. The model optimization data management system based on interactive feedback field experience according to claim 1, characterized in that: The model debugging structure identification module includes a debugging node information management unit and a debugging structure construction unit; The debugging node information management unit is used to extract the test sequence L={P0,P1,P2,...,P n }, set two adjacent model parameter value sequences in each test sequence as a test node; if in a test node, two adjacent model parameter value sequences A and B are in the i-th model parameter item F i The corresponding parameter values ​​are different, and the i-th model parameter item F is extracted. i As target parameter items, the output results obtained after the engineers complete the model application test according to A and B are obtained respectively, and the application output items with numerical deviations are extracted as target output items; The debugging structure construction unit is used to establish debugging data associations between each target parameter item extracted from each debugging node and each target output item, and construct and generate a plurality of debugging structures.

3. The model optimization data management system based on interactive feedback field experience according to claim 2 is characterized in that: The model tuning structure identification module includes a characteristic tuning structure screening unit; The characteristic tuning structure screening unit is used to calculate the tuning characteristic index α=g / M for each tuning structure respectively, wherein g represents the total number of times each tuning structure is extracted cumulatively, and M represents the total number of tuning structures extracted from all tuning nodes; and the tuning structure whose tuning characteristic index α is greater than the index threshold is judged to be a characteristic tuning structure.

4. The model optimization data management system based on interactive feedback field experience according to claim 1, characterized in that: The model evaluation feedback management module includes: The historical application records covered in each unit period are collected to obtain the total number of historical application records corresponding to each unit period U. The total number of historical application records with abnormal industrial application model commissioning in each unit period E is accumulated, and the abnormal distribution rate S=E / U in each unit period is calculated; The abnormal distribution rates corresponding to each unit period are collected in chronological order to obtain an abnormal distribution rate sequence; when V consecutive times satisfying Sj+1-Sj>0 are captured in the abnormal distribution rate sequence, a feedback management port is provided to display that the application generalization ability of the industrial application model has declined, prompting the management personnel to optimize the industrial application model.

5. The application of the model optimization data management system based on interactive feedback field experience according to any one of claims 1 to 4, characterized in that: Its application in rail transit network operation includes: using the system to manage application models involved in the rail transit design stage, rail transit construction stage and rail transit operation stage.

Images