CN115860118A - Safety construction method and system of intelligent pipe network knowledge model - Google Patents

Abstract

The invention discloses a safe construction method and a safe construction system of an intelligent pipe network knowledge model, which are characterized in that data, knowledge, an algorithm, an applicable scene, model requirements and multi-dimensional information of a service system are collected; carrying out identification processing based on an acquisition object on the multi-dimensional information to construct an information knowledge graph; performing security design according to the information knowledge graph, determining a security protection strategy, and constructing a knowledge model based on the security protection strategy; constructing model training data and model testing data, and performing model training and testing on the knowledge model; and uniformly storing all the knowledge models in a warehouse, establishing a model safety monitoring rule, and managing and optimizing all the knowledge models. The method solves the technical problem of high safety risk degree in the whole life cycle process of knowledge model development and operation in the prior art. The safety credibility design and implementation of the whole process of the knowledge model construction are achieved, the safety contract design is carried out by combining the knowledge map construction, and the safety credibility controllable effect of the whole modeling process is ensured.

Description

Safety construction method and system of intelligent pipe network knowledge model

Technical Field

The invention relates to the technical field of knowledge security credible management in the field of oil and gas storage and transportation, in particular to a security construction method and system of an intelligent pipe network knowledge model.

Background

With the development of the internet of things and intelligent sensing technology in recent years, oil and gas pipeline operators accumulate a large amount of professional data in the process of establishing and operating intelligent pipe networks. A large amount of manpower, material resources and financial cost are spent behind the data, and only after the knowledge calculation of statistical analysis and artificial intelligence modeling, a knowledge model can be formed for value output. However, these data assets have various deficiencies and security risks in the whole life cycle process of knowledge model development, calculation, construction, application, etc., including the aspects of security supervision, whole life cycle operation, knowledge enabling, contribution confirmation, etc.

Disclosure of Invention

In order to solve the technical problems, the application provides a safety construction method and a safety construction system for an intelligent pipe network knowledge model, and the method and the system are used for solving the technical problem that in the prior art, the safety risk degree is high in the whole life cycle process of knowledge model development and operation.

In view of the above problems, the present application provides a method and a system for safely constructing a knowledge model of an intelligent pipe network.

In a first aspect, the application provides a secure construction method of a knowledge model of an intelligent pipe network, the method including: collecting multidimensional information including data, knowledge, algorithm, applicable scene, model requirement and service system; carrying out identification processing based on an acquisition object on the multi-dimensional information to construct an information knowledge graph; performing security design according to the information knowledge graph, determining a security protection strategy, and constructing a knowledge model based on the security protection strategy; constructing model training data and model testing data, and performing model training and testing on the knowledge model; and uniformly storing all the knowledge models which are constructed, trained and tested in a warehouse, and establishing a model safety monitoring rule to manage and optimize all the knowledge models.

In a second aspect, the present application provides a system for safely building a knowledge model of a smart pipe network, the system including: the data access module accesses multi-source data in the intelligent pipe network data lake; the model development module is connected with the data access module and performs model development by using data in the data access module; the model operation and maintenance module is connected with the model development module and is used for managing, maintaining, operating and monitoring the model; the safety execution module is connected with the data access module, the model development module and the model operation and maintenance module and is used for carrying out safety audit and safety monitoring on the whole process of model construction, development and operation and maintenance; the algorithm module is connected with the model development module and recommends an algorithm in the modeling process; and the visualization module is respectively connected with the data access module, the model development module, the model operation and maintenance module and the safety execution module, and provides online model visualization service.

In a third aspect, the present application provides an electronic device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method of the first aspect.

One or more technical solutions provided in the present application have at least the following technical effects or advantages:

the method comprises the steps of collecting multidimensional information including data, knowledge, algorithms, applicable scenes, model requirements and a service system; carrying out identification processing based on an acquisition object on the multi-dimensional information to construct an information knowledge graph; performing security design according to the information knowledge graph, determining a security protection strategy, and constructing a knowledge model based on the security protection strategy; constructing model training data and model testing data, and performing model training and testing on the knowledge model; and uniformly storing all the knowledge models which are constructed, trained and tested in a warehouse, and establishing a model safety monitoring rule to manage and optimize all the knowledge models. The technical effects that the safety credibility design and implementation are carried out on the whole process of the knowledge model construction, the knowledge modeling safety contract design is carried out by combining the knowledge map construction, the safety credibility controllable metering of all the participating parties in the whole modeling process is ensured, and the knowledge model is effectively managed are achieved. Therefore, the technical problem of high safety risk degree in the whole life cycle process of knowledge model development and operation in the prior art is solved.

The foregoing description is only an overview of the technical solutions of the present application, and the present application can be implemented according to the content of the description in order to make the technical means of the present application more clearly understood, and the following detailed description of the present application is given in order to make the above and other objects, features, and advantages of the present application more clearly understandable.

Drawings

In order to more clearly illustrate the technical solutions in the present application or the prior art, the drawings needed to be used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only exemplary, and for those skilled in the art, other drawings can be obtained according to the provided drawings without inventive effort.

Fig. 1 is a schematic flowchart of a security construction method of a knowledge model of an intelligent pipe network according to an embodiment of the present disclosure;

FIG. 2 is a schematic flow chart illustrating development and training of a knowledge model in a security construction method for a knowledge model of an intelligent pipe network according to an embodiment of the present disclosure;

FIG. 3 is a schematic flow chart illustrating operation and maintenance management of a knowledge model in a security construction method for a smart pipe network knowledge model according to an embodiment of the present disclosure;

fig. 4 is a schematic structural diagram of a security construction system of a smart pipe network knowledge model according to an embodiment of the present disclosure;

fig. 5 is a schematic structural diagram of an electronic device according to an embodiment of the present application.

Description of the reference numerals: the system comprises a data access module 1, an off-line training module 2, an algorithm framework knowledge layer 3, an on-line service module 4, a model operation and maintenance module 5, a safety execution module 6, a data safety module 7 and a model safety module 8.

Detailed Description

The application provides a safe construction method and a safe construction system for an intelligent pipe network knowledge model, and solves the technical problem that in the prior art, the safety risk degree is high in the whole life cycle process of knowledge model development and operation.

The invention conception is as follows: the following safety risk problems exist in the whole life cycle process of the existing knowledge model development and operation: and (4) safety supervision aspect: knowledge model construction needs to be established on the basis of analysis and reasoning on the data sets in the enterprise. Due to the requirement of data safety compliance, enterprises do not want to share high-value acquired data and professional knowledge which expend a lot of energy, but want to acquire an external more professional or superior universal knowledge model to improve the accuracy and efficiency of the model, and the security risk of internal data leakage and external knowledge model introduction exists. How to build a knowledge model to exert data value in a safety compliance manner is a key problem to be solved.

Full life cycle operation aspect: the enterprise knowledge model has the advantages of multiple links, long duration, various related personnel and mechanisms, and complex environment and resource allocation in the processes of training, building, verifying and the like. Often, due to the reasons of personnel organization change, environment configuration upgrade, model version disorder, safe and credible attack and the like, a knowledge model cannot form safe and credible development, test, deployment and production processes in the development and construction process, is difficult to trace and audit, and becomes a common pain point in enterprise intelligent transformation. A set of knowledge model full life cycle safe and trusted construction platform needs to be established.

Knowledge enabling aspect: the original data and knowledge are overloaded, the structuralization is poor, the redundancy is high, conflicts exist, and the conversion and utilization of the knowledge to assets are difficult to carry out. With the complexity of the modeling applicable scene and the improvement of fineness, the requirements on the quality and the interpretability of the knowledge model input data set are synchronously increased. The modeling in the actual industrial application scene has the following difficulties: 1) The data fineness and completeness are not enough, a large number of sparse samples exist, and the robustness and the generalization capability of the model are influenced; 2) Constructing a lightweight knowledge model in a cost minimization manner in a multi-way, fast, good and provincial way; 3) Dynamically updating the model; 4) Providing a secure, trusted model interpretation. Knowledge is required to participate in the construction of a knowledge model, and the modeling process is improved and perfected.

Contribution confirmation aspect: there is a lack of a mature evaluation and measurement method for contribution to knowledge calculation. In the process of constructing the enterprise knowledge model, a data provider, a model builder, an algorithm provider, a platform provider, an industry demander, a safety supervisor and other roles are involved, and each role provides more or less value for the knowledge model. How to fairly evaluate the contribution degree of the distribution participants and provide credible contribution basis is the premise of the knowledge model for showing the foreign value.

In order to solve the problems and risks, the embodiment of the invention provides a safety construction method and a platform system for an intelligent pipe network knowledge model. The method is characterized by comprising the steps of combing data, professional knowledge and intelligent algorithms of oil and gas pipeline enterprises, classifying safely, fusing efficiently on the premise of safety compliance, and constructing a knowledge model for the intelligent pipe network to provide automatic and intelligent support decision-making service for each business scene of the intelligent pipe network.

According to the technical scheme, the data acquisition, storage, use, processing and the like meet relevant regulations of national laws and regulations.

In the following, the technical solutions in the present application will be clearly and completely described with reference to the accompanying drawings, and it is to be understood that the described embodiments are only a part of the embodiments of the present application, and not all of the embodiments of the present application, and it is to be understood that the present application is not limited by the example embodiments described herein. All other embodiments obtained by a person of ordinary skill in the art based on the embodiments of the present application without making any creative effort belong to the protection scope of the present application. It should be further noted that, for the convenience of description, only some but not all of the elements relevant to the present application are shown in the drawings.

Example one

Fig. 1 is a schematic flow chart of a security construction method of a knowledge model of a smart pipe network according to an embodiment of the present application, and as shown in fig. 1, the security construction method of the knowledge model of the smart pipe network includes:

s10, collecting multidimensional information including data, knowledge, algorithm, applicable scenes, model requirements and a service system;

specifically, the service application scene of the intelligent pipe network is complex, the data sources and the types are various, the current knowledge model construction situation in each link of the intelligent pipe network is collected and combed, and the safety construction basis of the knowledge model is established. The aim of the acquisition and carding process is to master the knowledge modeling resource foundation and the supporting capability owned by oil and gas pipeline enterprises. The collected information comprises data, knowledge, an algorithm, an applicable scene, model requirements, a service system and other multi-dimensions, wherein the collected objects comprise three types of existing resource types, target requirement types and platform capability types, specific details are shown in the following table 1, the intelligent pipe network knowledge model constructs a purchase information detail table, a survey and carding list is set, and the collection target and the collection content of each type of the collected objects are definite.

TABLE 1

/>

S20, carrying out identification processing based on an acquisition object on the multi-dimensional information to construct an information knowledge graph;

further, step S20, the identification processing based on the collected object is carried out on the multi-dimensional information, and an information knowledge graph is constructed, and the method comprises the following steps: step S201, carrying out preset data type information identification on the multi-dimensional information, and constructing a knowledge graph conceptual model; and S202, performing object relation carding on the multi-dimensional information based on the acquired objects by using the knowledge graph conceptual model, and outputting a resource distribution and circulation graph.

Specifically, after the acquisition is finished, the acquired multi-dimensional information should be further combed based on the acquisition object. The purpose of the combing is to correlate the acquisition objects. The knowledge map technology can structure disordered knowledge in an ordered and dominant manner, pull through the chimney type monomers of each business system, and get through the data barriers. And integrating various scattered fragmented resources to form an enterprise knowledge asset context. The resource carding can be carried out by adopting knowledge graph technology.

The method comprises the steps of carrying out knowledge processing on collected intelligent pipe network service, data, knowledge and platform data, identifying contents such as keywords, resource metadata, resource summaries, participants, platforms and service demand topics, constructing a knowledge graph conceptual model, and carrying out extraction, alignment, fusion, labeling and other processing. And then, establishing knowledge association, combing the relationship venation of the entity object, and outputting a resource distribution and circulation diagram. The dependency relationship and the upstream and downstream processes among the objects are cleared, and particularly, the hidden relationship among the objects is identified and found, so that a logical foundation is laid for the subsequent smooth construction of a knowledge model. Under specific conditions, graph computation needs to be carried out by combining a graph database, knowledge reasoning is carried out by using relevance reasoning operation among graph data, and hidden relations and knowledge are discovered.

S30, performing safety design according to the information knowledge graph, determining a safety protection strategy, and constructing a knowledge model based on the safety protection strategy;

further, step S30, carrying out security design according to the information knowledge graph, determining a security protection strategy, and constructing a knowledge model based on the security protection strategy, wherein the steps comprise: step S301, determining data, knowledge, algorithm, applicable scene, model requirement and safety level of a service system according to the information knowledge map to obtain resource safety hierarchical design; step S302, determining the safety protection strategy according to the resource safety hierarchical design, wherein the safety protection strategy comprises a participant engagement, a calculation engagement, a contribution engagement and a safety engagement; and S303, constructing a knowledge model based on the engagement of the participants, the calculation engagement, the contribution engagement and the safety engagement.

Specifically, based on the constructed information knowledge graph, the safety design of the resources is carried out according to the incidence relation. The purpose of safety design is to locate high-value and high-sensitivity resources, carry out strategy design and system planning in advance and avoid data safety leakage or asset loss caused by unsafe knowledge modeling.

Safety grading design: grading resources such as combed data, knowledge and algorithms of the intelligent pipe network, evaluating the resources according to the angles of acquisition cost, disclosure hazard and application value in a grading manner, distinguishing and positioning high-cost, high-sensitivity and high-value resources of the intelligent pipe network, identifying the resources in advance in a targeted manner, and designing a knowledge model to construct a corresponding grading safety protection strategy. It should be noted that the security protection classification strategy should not be performed only according to the security requirement dimension, but should be designed in combination with the aspects of comprehensive and dynamic properties, such as scene applicability, protection performance cost, damage acceptability, and technical maturity.

Algorithm security policy: the algorithm has the security and credibility risks of defective backdoor, malicious attack, reverse extraction, tampering hijacking, data virus throwing and the like. And carrying out safety protection and reinforcement design on the algorithm for constructing the knowledge model. The algorithm reinforcement work can be done from the following aspects:

enhancing the safety of the algorithm in vivo: the algorithm training and structure are improved, the input and output modes are adjusted, and the robustness, interpretability and fairness of the algorithm are enhanced. Such as countertraining, regularization, addition of random noise, rotational translation, cleaning of input data distributions, and the like.

Monitoring algorithm malicious codes: the safety monitoring is performed outside the algorithm. For example, input and output of the algorithm and application behaviors are monitored, and malicious behaviors such as countermeasure samples of the algorithm, algorithm backdoors, falsification and the like are identified. And auditing and safety testing are carried out on the algorithm codes, so that coding loopholes and back door implantation are prevented.

Technical protection response strategy: and carrying out safe and reliable technical protection on the whole process of knowledge modeling, and relating to a corresponding response strategy plan. According to the actual scene, technologies such as data desensitization, multi-factor verification, safety encryption, authority control, retroactive audit, privacy calculation, output degradation and the like can be adopted. For example, the safety certification and the credible traceability audit in the whole process of building the model are enhanced by combining the block chain technology, and the data safety of internal and external combined modeling is improved by using the methods of federal learning, multi-party safety calculation, credible environment design and the like.

The security design and the response strategy are precipitated into a knowledge model security construction platform through security architecture design, and the platform security capability is embodied.

And (3) designing a model trust, wherein fairness, interpretability, robustness, a model blood relationship and white-box property of the model are mainly considered. Fairness refers to that there should be no prejudice in building input and output by a model, and unfairness to some groups is avoided. Interpretability means that the interpretation of the output prediction result constructed by the model can be approved by the demander. The robustness means that the model is not easy to be tampered and the training data cannot be stolen reversely. The model blood relationship means that the context of the model is clear to the dependency resources and other models in the construction process. The white-box finger model construction process is transparent and traceable, and malicious implantation is avoided.

According to the safety hierarchical design, the safety protection strategy is determined, specifically, the contract design is carried out, and as the intelligent pipe network knowledge modeling usually relates to enterprises and participants on a plurality of chains in the whole industry. Before knowledge modeling is carried out, related contract appointment design is carried out in advance, and subsequent security disclosure and unauthorized utilization risks caused by unclear authority and responsibility, fuzzy boundary and the like are effectively avoided.

The safety protection strategy comprises a participant engagement, a calculation engagement, a contribution degree engagement and a safety engagement, wherein the participant engagement is as follows: for the construction of the knowledge model of the intelligent pipe network, the participating modeling party can be divided into 6 roles, namely a data provider, a knowledge consultant, an algorithm modeling party, a calculation force platform party, a demand receiver and a safety supervisor. Each role has specific role function positioning, and the data provider: data acquisition, data cleaning, data identification, data updating, knowledge consultant: business knowledge, domain knowledge, experience knowledge, expert consultation; an algorithm modeling party: time-space statistical analysis, image recognition, natural language processing and machine learning; a force calculation platform side: functional positioning in three aspects of computing power, openness and operation, wherein the computing power comprises computing resources, an algorithm framework and task scheduling, and the openness comprises model training, updating iteration and model evaluation; the operation comprises deployment test, version release and operation and maintenance monitoring; the demand receiver: model training, updating iteration and model evaluation; the safety monitoring party: auditing and storing certificates, credibility tracing, authority authentication and security contracts. According to the difference of actual situations, the roles and functions in the data provider roles may be overlapped or offset, for example, the demand receiver also takes on the functions of a part of the data provider roles, but does not influence the division of the actual role positioning. The method is characterized in that the role, range, boundary and the like of each participant are definitely agreed in the contract design, and subsequently, the security audit of the role of the participants is carried out in the whole process of establishing a knowledge model in the contract execution process.

And (3) calculating the convention: under different business scenarios and requirements, the algorithms and frameworks selected in the knowledge model construction process and the expected effects of the algorithms are different. The convention of the calculation mode should be carried out according to the safety classification, the safety design and the like. And safety constraint and regulation are carried out on the flow structure, the input and output mode, the convergence effect, the bearing platform and the like of the modeling algorithm, so that safety risks in the calculation process are prevented.

The contribution degree convention is as follows: in the process of building a knowledge model participated by two or more parties, the method for determining the contribution of each party needs to be agreed in the contract. For example, in the case of jointly constructing a knowledge model inside and outside a smart pipe network enterprise, from the contribution rate of modeling resources invested by participants, the agreement confirmation of the contribution degree and proportion can be performed by referring to the cost and market value of the contributed resources and the evaluation of the improvement effect on the accuracy, safety and efficiency of the model.

Safety contract: the contract is agreed on the security risk in the whole process of knowledge model construction, including the security protection requirements on the aspects of data, knowledge, algorithm, technology, platform and the like in the whole process of knowledge model construction. For example, the safety level and audience range of the knowledge model are limited, and the safety protection technology to be adopted is agreed; the rules of disclosure, confidentiality and use of the proprietary technology of the participants are made explicit, and audit, traceability and punishment mechanisms for dishonest and infringement activities are constructed based on blockchain techniques.

S40, constructing model training data and model testing data, and performing model training and testing on the knowledge model;

further, step S40, the constructing model training data and model testing data, and performing model training and testing on the knowledge model, as shown in fig. 2, includes: step S401, determining a training data source, and acquiring data based on the training data source to obtain original acquisition data; s402, cleaning the original collected data to obtain preprocessed data; step S403, based on the security protection strategy, performing data identification according to the preprocessed data to obtain data identification; step S404, generating a unique version identification code for data according to the original collected data, the preprocessed data and the data identification, and constructing the model training data and the model testing data; and S405, determining a model safety target based on the safety protection strategy, and performing model training and testing on the knowledge model by using model training data and model testing data according to the model safety target until the requirements of the model safety target are met.

Specifically, after the knowledge model is constructed and designed, the implementation stage of the model is started, and the implementation stage comprises links such as development, test, deployment and maintenance. The model construction and development are carried out on the basis of scene safety design and can be divided into three links of data preparation, model construction and model training.

A data preparation link: the method comprises the processes of data acquisition, data cleaning, data identification, data updating and the like, and is mainly performed by a data provider organization. Data acquisition: the training data sources required for knowledge model construction by different business departments of an oil and gas pipeline enterprise are different, such as internal business data middleboxes, enterprise data lakes, real-time access of internet of things, manual inspection collection, external public data, open source data, third-party detection data and the like. The data types also include images, text, forms, voice, etc. In the acquisition process, a data provider is responsible for integrating scattered data of various sources and various types together by adopting a standardized, automatic or visual mode and rules according to modeling requirements, and merging and storing the scattered data. Data cleaning: in the knowledge model construction process, the quality of input training data directly determines the upper limit of the model prediction effect. The collected data must be cleaned and processed, so that the quality and safety of the input data are ensured. In the data cleaning process, a data provider searches, analyzes and corrects data, comprehensively and quickly mines internal information of the data, improves the data quality to the maximum extent through methods of cleaning, converting, fusing, enhancing and the like, forms data metadata, ensures that the data can be balanced, positioned and reused, and uses the cleaned data as preprocessing data. And in the aspect of safety and credibility, safety processing such as sensitive information desensitization, data consistency signature addition, identification removal and the like is mainly performed. Data identification: labeling data identified and distinguished by a machine for inputting an algorithm is the key for learning and accurate prediction of a model. Data identification often requires a great deal of time and effort, and data providers typically do the identification in conjunction with metadata for the cleansing phase. In the actual training process, due to the fact that the data characteristics and the decision-making correlation degree are different, the effect is good and good, the effect is bad, and the like, in order to improve the efficiency and the accuracy of identification, knowledge consultation is needed, input of knowledge rules is obtained from a knowledge provider, characteristic selection and processing are carried out, and data safety classification and classification identification is also carried out in the step.

Because data needs to be adapted to different model algorithms, and attributes of the data in the processes of cleaning, identification and the like frequently change, in order to facilitate data identification management, the generation of the version of the data in the embodiment of the application is called data version processing, that is, a unique version identification code is generated for the data, and the generation of the unique version identification code is important for being hung on different model versions. Particularly, the version updating management is needed to be performed on the data input for the model automation pipeline in the platform. The version is provided with a serial and standardized updating management and control mechanism aiming at data storage positions, hooking relations, identification numbers and the like, so that the traceability and reusability of data are ensured, and the execution speed and the calling efficiency are improved.

The knowledge model construction is mainly carried out by an algorithm modeling party, and comprises the selection and determination of a model algorithm and a frame. The method comprises the steps of determining a proper algorithm and a proper framework aiming at model prediction and recognition requirements, wherein the common algorithm comprises linear regression, decision trees, random forests, logistic regression, gradient lifting, SVM and the like, training the framework such as XGboost, MLflow, tensorFlow, transformer, pyTorch, scikit-lean, spark-MLlib and the like, and generally selecting the algorithm and the framework with the best effect through preliminary testing and cross validation. When a model is safely constructed, the safety of the algorithms and the frames is evaluated and graded, and the algorithms and the frames suitable for a specific scene are selected according to the safety design requirements in the design stage. If the seed model is agreed in conjunction with a design phase or a design contract, the subsequent training can be started directly as a construction starting point.

The data preparation function is used for determining data, training data and test data are established according to specific safety construction requirements and adaptive scenes of the model, the training data are used for training and testing the knowledge model, and the model training process is a process for repeatedly adjusting model parameters to achieve optimal effect output. The training process is completed by the algorithm modeling party on the resources provided by the training platform or the algorithm platform party. The whole model training process is auditable and traceable, the safety supervisor carries out integrity supervision on the algorithm modeler, the modeler is prevented from tampering and forging modeling parameters and other dishonest modeling behaviors, a black box model which cannot be detected and interpreted is avoided being built, and meanwhile, the participant is prevented from implanting a back door and other potential safety hazards to the model in the training process.

The model test stage mainly performs model test and evaluation. And testing the trained model by using a specially prepared functional test set to determine whether the requirements of deployment and online are met. The evaluation includes an effect evaluation, a security evaluation, a robustness evaluation, and the like. The effect evaluation is mainly used for judging the quality of the model by comparing the recognition effect of real data and predicted data after training, indexes, robustness, speed and the like. And the safety evaluation is carried out by a safety monitoring party or a third-party safety manufacturer, and the safety of the model is judged through penetration test and safety attack evaluation from a safety angle. The robustness evaluation evaluates the robustness and robustness of the model by resisting attacks, resisting unknown attacks and the like. The model evaluation is crucial to finding the defects, security holes and the like of the model, the model with high security level requirements is subjected to multi-round evaluation from different security requirement angles, and finally the model is ensured to meet the security level requirements.

Further, after the model training and testing are performed on the knowledge model, the method comprises the following steps: s601, obtaining the information of the force calculation platform side; step S602, acquiring a deployment environment based on the computing force platform side information, wherein the deployment environment comprises a data center, a cloud, an edge side and an end side; step S603, when the deployment environment is consistent with the training environment, the model is cut, compressed, packaged, integrated, safely layered, encrypted and cooperatively optimized by cloud edges, and the knowledge model is constructed and deployed; and S604, when the deployment environment is inconsistent with the training environment, adapting and converting the knowledge model, and cutting, compressing, encapsulating, integrating, layering safely, encrypting and performing cloud edge collaborative optimization on the adapted and converted knowledge model.

Specifically, the knowledge model determined after the training test is deployed and released, the stage is usually performed by a computing platform side, and as the models output by the training stage of the knowledge model are continuously tested and optimized in the off-line training stage, the demander does not want to be able to directly predict and interact through the control terminal of the modeling side, so that the deployment and the release are required. The configuration and dependence requirements of the model deployment environment are consistent with those of the training environment, and the safety requirement is higher.

According to different scenes, the deployment environment may be located in a data center, a cloud end, an edge side and an end side. For the inconsistency between the deployment environment and the training environment, the model needs to be adapted and converted according to the difference and actual needs of the deployment environment. Such as cutting, compressing, packaging, integrating, safety layering, encrypting, cloud edge cooperative optimization and the like of the model. The converted model can enter a deployment and release link only after passing through the test evaluation in the aspects of performance, safety and the like again. In addition, in the deployment phase, besides the security of the model itself, the security protection capabilities of the deployment environment and the release environment need to be ensured.

And S50, uniformly storing all the knowledge models which are constructed, trained and tested in a warehouse, and establishing a model safety monitoring rule to manage and optimize all the knowledge models.

Further, as shown in fig. 3, step S50, the step of uniformly warehousing and storing all the knowledge models which are constructed, trained and tested, and establishing a model safety monitoring rule to manage and optimize all the knowledge models includes: step S501, all knowledge models are stored in a unified warehousing model warehouse, and the warehousing knowledge models are detected according to a preset warehousing standard; step S502, classifying the warehousing knowledge model in a grading way to obtain a model version, model attributes, training process tracing and model deployment on-line configuration; step S503, hanging the model deployment on-line configuration, and when the knowledge model is exported, evaluating and verifying the ex-warehouse knowledge model based on the model deployment; step S504, carrying out global monitoring on the operation of the knowledge model, comprising the following steps: running the number of models, model consumption resources, model state details and model calling statistics; step S505, when the number of the operation models, the model consumption resources, the model state details and the model calling statistical global monitoring data are abnormal, carrying out data isolation; and S506, establishing a model updating requirement, and updating and optimizing the knowledge model based on the model updating requirement.

Specifically, the knowledge model also needs to be monitored and maintained for a long time, and a large amount of tedious and repeated manual operation and maintenance work often causes risks such as operation errors, model failures, abnormal input and output and the like. The safe and efficient operation monitoring and maintenance updating of the knowledge model are guarantee that the knowledge model can better exert value and meet the service requirements of business application. The embodiment of the application carries out operation and maintenance management on the knowledge model, each model has service scenes and directions along with the increasing number of the constructed models, and the model has multiple states such as a warehoused model, a developing model, an online model and the like, so that the model needs to be uniformly and automatically supervised. From the perspective of safety construction, the method specifically comprises a model warehousing and storing link, a model operation monitoring link and a model recycling and destroying link.

Unified warehousing and storage: the model warehouse is a core functional component for security update maintenance. Detecting the warehousing model according to the warehousing standard requirement of the model formulated in the safety design stage; providing classified and graded management for the warehousing model; and carrying out partition storage and management and control according to the safety classification condition. The model version is controlled, and model attributes, training process tracing and the like are supported. The model is connected with on-line configuration for deployment, secondary evaluation is carried out on the release and ex-warehouse models, and safety evaluation and verification of the deployment model are realized;

operation monitoring management: the model operation monitoring provides overall monitoring of the business online model, and comprises the operation model quantity, the model consumption resources, the model state detail, the model calling statistics and the like. In the aspect of safety, abnormal input and output, malicious attack codes, service safety risk identification and the like in the running process of the model are mainly concerned. According to the requirement of the safe operation level of the model, the isolation of the data of the operation state of the model is provided, the input and output data are ensured not to be leaked in the operation process of the knowledge model, and the data can not be exchanged and transmitted without authorization of all parties involved; encrypting and transmitting the model parameters; and the model runs and retains the whole process audit log.

Recovery and destruction: when the model has an operation fault and needs to be rolled back, or safety abnormity or service defect is found and needs to be off-line, the model is timely recovered and destroyed. After the model is recovered or destroyed, the abnormal operation log and the data segment are submitted in a specific area for the safety supervisor to check, repair and trace.

After the model is constructed, stable output is not constant, business requirements change, safety requirements adjustment, algorithm iteration upgrading, platform dependent environment updating and other factors cause the model to need continuous and regular adjustment and updating. And establishing a model updating requirement according to requirements of safety hierarchical design, adaptive scenes, training data and the like of each knowledge model, and periodically adjusting, updating and optimizing the knowledge model based on the model updating requirement.

Further, the establishing of the model update requirement and the update optimization of the knowledge model based on the model update requirement include: step S5061, acquiring knowledge model version information, and updating the knowledge model according to the model updating requirement according to knowledge model training data and knowledge model version information; step S5062, acquiring a knowledge model before updating and a knowledge model after updating, fusing and packaging the knowledge model before updating and the knowledge model after updating, and auditing the fusion and packaging process; and S5063, establishing an incidence relation blood-related atlas between versions of the knowledge model based on the pre-update knowledge model and the post-update knowledge model, and carrying out safety tracking on the knowledge model based on the incidence relation blood-related atlas.

Specifically, when updating, optimizing and maintaining the knowledge model, the method can be divided into three links of version iteration, encapsulation integration and migration and model blood margin.

And (4) version iteration: the model is closely related to the constructed data, knowledge, algorithm and platform. The data updating improves the accuracy of the model, the knowledge updating improves the accuracy and the scientificity of the model, the algorithm updating improves the robustness and the interpretability of the model, and the platform updating improves the operational efficiency of the model. Model updates require each participant to continue to make new investments and contributions within their respective responsibilities pursuant to the contract. Model version iterative updates require attention from a security perspective: and safety log auditing in the whole process provides basis for updating safety responsibility and confirming contribution degree. The model updating should be evaluated again according to the new model, including the evaluation in the aspect of safety level, and the emphasis is on the incremental updating part.

Package integration and migration: the new model and the old model are fused and packaged, for example, the simplest mean fusion is carried out on the prediction results of a plurality of models, so that the robustness and the generalization capability of the service prediction model can be improved. On the basis of the existing model, the new scene model is constructed by using transfer learning, so that the construction training speed of the model can be improved. And (3) cutting and compressing the existing model, and quickly constructing a light model aiming at a specific service operation scene. In the link, an algorithm modeling party mainly carries out customization development according to the requirement of a required receiving party. The safety supervision party focuses on auditing in the process of packaging integration or transferring the cutting model, such as safety defects and implantation which may be introduced, and performs safety assessment before deployment and release.

Model blood relationship: the establishment of the blood relationship map of the incidence relation between the versions of the model updating and maintenance is the best means for carrying out safe, reliable and efficient updating and maintenance on the model. And the decision reasoning support of model construction can be effectively carried out according to data, knowledge, algorithms, platforms, participants, model metadata, evaluation attributes and the like which are depended by the construction processes of different versions of the model. In the aspect of safety, the model blood relationship graph is utilized, and safety risks can be positioned and tracked through measures such as safety monitoring rules, safety risk tracing, abnormal subgraph matching and the like. If the attribute mismatching monitoring rule is evaluated in a safety grading way, when the model is found to be hooked and called mismatching, particularly when the model is built or updated by high-safety-level resources, abnormity is automatically monitored and blocked by alarming.

Further, the method comprises: step S701, acquiring real-time knowledge updating information; step S702, determining a knowledge influence model stage according to the real-time knowledge updating information; and S703, based on the knowledge influence model stage, utilizing the real-time knowledge updating information to enable knowledge in the corresponding stage of the knowledge model and updating the knowledge model.

Specifically, in the embodiment of the application, in the whole cycle of construction, operation and maintenance and updating of the knowledge model, in order to ensure the safety and the content reliability of the model, knowledge energization is carried out, and an automatic construction scheme of the knowledge model is designed. First, the knowledge driving ability provided by the knowledge provider also goes through the whole process of the above construction and development. Knowledge investment in the early preparation stage can effectively reduce investment and cost of an algorithm modeling party in the process of constructing algorithm input and training. For example, the professional knowledge and the expert experience of the oil pipe service can be combined, and a high-quality training and testing data set can be pertinently and quickly filtered and constructed in the data preparation stage.

In the later training deployment stage, knowledge assistance can accelerate knowledge model construction, recommend an optimal algorithm and a framework, accelerate parameter optimization, and promote business effects and problem solution. For example, common model parameters include learning rate, the number of network layers, the number of network layer nodes and the like, an algorithm modeling party accumulates a large amount of parameter optimization knowledge in a large amount of modeling training processes, repetitive tasks such as super-parameter adjustment can be automated by using the contribution of the knowledge, adaptive parameters can be automatically adjusted according to model prediction results, even a knowledge model can be specially constructed to assist training in constructing the knowledge model, and parameter configuration with the highest positioning accuracy is achieved.

Second, knowledge modeling also requires a significant investment of expertise in industry enabling. For example, aiming at the requirements of business characteristics and deployment environment of the oil and gas pipe network industry, by combining industry professional knowledge, a development kit, a pre-training model, a high-quality data set and the like adapted to a customized construction industry knowledge model, an automatic, visualized and low-code knowledge model rapid construction testing tool is provided for a large number of business personnel. And on the basis, the accurate knowledge model asset safety monitoring and protection facing the industry business are provided.

The knowledge energizing long-term target realizes the whole engineering and automation of construction training, deployment, updating, maintenance and implementation, including the automation of safety monitoring of knowledge model construction. And (3) learning and analyzing the steps of establishing each stage of the knowledge model, establishing a standardized operation line, and fully utilizing knowledge resources contributed or acquired by each party. The contribution rate of the knowledge can be calculated by comparing the promotion rates of the effects before and after the introduction of the knowledge. And the safety construction implementation process also comprises safety processing and auditing of knowledge invested in modeling development.

In summary, the embodiment of the present application has the following beneficial effects:

by safely constructing, collecting, combing and classifying and grading, fragmented resources such as original data assets, knowledge assets, algorithm assets and platform assets can be centrally managed, grading management is carried out from the aspects of safety compliance and value utilization, and the foundation of safety compliance of a knowledge model is strengthened. The method is characterized by comprising the steps of combing data, professional knowledge and intelligent algorithms of oil and gas pipeline enterprises, classifying safely, fusing efficiently on the premise of safety compliance, and constructing a knowledge model for the intelligent pipe network to provide automatic and intelligent support decision-making service for each business scene of the intelligent pipe network. The method provided by the invention is used for carrying out safe and reliable design and implementation on the whole process of knowledge model construction, and realizing and bearing the safe construction of the knowledge model through platform safety protection technical measures. And the knowledge modeling safety contract design is carried out by combining the knowledge map construction and the block chain auditing technology, so that the safety, credibility, controllability and measurement of all the participating parties in the whole modeling process are ensured.

Safety design aspect: designing a safety strategy aiming at modeling input data, knowledge and algorithm pertinence, and utilizing a knowledge graph construction and block chain auditing technology. And (4) designing a knowledge modeling security contract to ensure the security, credibility and controllability of the whole modeling process.

And (3) construction and deployment aspects: technical measures such as data security processing, algorithm security evaluation, whole-process security audit, model security test, prevention of malicious implantation and the like which run through the whole process of development, test and deployment of the knowledge model are set, and the security construction implementation of the knowledge model is guaranteed.

And (3) management and control aspects of the model: the safe and efficient operation monitoring and maintenance updating of the knowledge model are important guarantees that the knowledge model meets the service requirements of business application. And establishing safety rules of model warehousing, version management, operation monitoring and updating maintenance, monitoring and tracing the dependency relationship of model circulation operation by using the model blood-related atlas, and managing the model safely and efficiently.

Knowledge enabling aspect: the internal and external professional knowledge and expert experience of an enterprise are combed and utilized, and the knowledge model is built in an early stage, so that the training cost is saved, and the knowledge support is provided for safe and efficient building and operation of the model. And the knowledge value is fully enabled in the model building and service process.

The platform bears the weight of the aspect: an automatic asset checking tool can be provided through the platform, and visual monitoring and operation of assets are achieved. Meanwhile, through a model development and operation integrated standard flow, tedious and complex offline manual operations such as model verification, approval, deployment, online and the like are reduced, an online automatic generation and execution knowledge model construction contract is generated, and the safety, the high efficiency, the accuracy and the credibility of the whole knowledge model construction process are guaranteed.

Example two

Based on the same inventive concept as the safety construction method of the intelligent pipe network knowledge model in the foregoing embodiment, as shown in fig. 4, the present application further provides a safety construction system of the intelligent pipe network knowledge model, where the system includes:

the data access module 1 is used for accessing multi-source data in the data lake of the intelligent pipe network;

specifically, data access module 1 accesses multisource data in the wisdom pipe network data lake. Common business system relational databases and real-time transaction data access modes should be supported. Such as JDBC, hive, HBase, remote distributed cluster. Common file management system data access should also be supported, such as HDFS, NFS, etc. For the requirement of the association analysis of the externally imported knowledge graph, the access of data such as an external graph database and the like is supported.

The model development module is connected with the data access module 1 and performs model development by using data in the data access module 1;

specifically, the model development module comprises an offline training module 2, an algorithm framework knowledge layer 3 and an online service module 4, and a model development construction part of the platform is constructed. During the model construction, the three modules need to be frequently intercommunicated and interconnected. The knowledge model construction generally comprises the steps of firstly carrying out offline training, extracting offline data by a data access layer, selecting a scene adaptive algorithm and a scene adaptive framework from an algorithm framework knowledge layer, and carrying out model training and optimization under the assistance of knowledge. The offline trained models, after optimization and lightweight, are cured into online services to provide automated real-time services. In some cases, the model needs to be reflowed, and the online service returns to the offline state for improvement and upgrade.

The model operation and maintenance module 5 is connected with the model development module and is used for managing, maintaining, operating and monitoring the model;

specifically, the knowledge model is managed, maintained, operated and monitored, and is responsible for maintaining the model base. Each model generated in the development stage is required to have a unique identification code, and the model operation and maintenance module is used for carrying out unified management and operation and maintenance on the models, and establishing function supports such as version upgrading, integrated release, test evaluation, state monitoring, automatic deployment, continuous optimization and the like.

The safety execution module 6 is connected with the data access module, the model development module and the model operation and maintenance module, and is used for carrying out safety audit and safety monitoring on the whole process of model construction, development and operation and maintenance;

specifically, the safety execution module 6 is a core guarantee module of the system, and has a platform function required by the knowledge model to construct safety design. The system is composed of functions of main body safety, system safety, contract safety and the like. The main body safety function is mainly responsible for performing services such as identity authentication, authority distribution, safety audit and the like on each main body participating in construction. The system security function is mainly responsible for providing basic services such as security encryption, privacy calculation, security protection and the like. The contract security function mainly aims at services such as security audit, security flow, security monitoring and the like in the knowledge model construction process.

The safety execution module also comprises two branch modules of a data safety module 7 and a model safety module 8, which are safety function branch modules derived from the safety execution module 6. The data security module 7 performs security processing on the data constructed by the knowledge model, such as data security hierarchical marking, data map drawing, data update tracing and the like. The model security module 8 is responsible for security processing and reinforcement of the model output in the development stage, such as hierarchical processing of the model, establishment of a blood relationship of the model, control of release security detection of the model version, and the like.

The algorithm module is connected with the model development module and recommends an algorithm in the modeling process;

specifically, after the platform is built, the following functional components can be optimized and upgraded to reflect the automation and intelligence of the safety model building in the later stage. The algorithm module divides the modeling algorithm into minimum units which can be scheduled according to functions, and the training process of the machine learning model is quickly built through combination and association of various minimum units; and a recommendation type modeling knowledge base is constructed, algorithm unit recommendation in a modeling process is provided for a user, a knowledge model construction threshold is reduced, and modeling test efficiency is improved.

And the visualization module is respectively connected with the data access module, the model development module, the model operation and maintenance module and the safety execution module, and provides online model visualization service.

Specifically, the visualization module supports the adoption of a dragging mode, interactively data preprocessing is carried out on a knowledge model algorithm in a visualization mode, a machine learning framework is selected on line, and low coding of model training, evaluation and iteration is achieved. And the model training result is solidified, packaged into a mirror image and released as an online model prediction service.

The visualization module may be implemented to include model consanguinity display: and (3) associating the model with data, knowledge, algorithms and the like from the model, and constructing a blood relationship map behind the model. And a seed model and an iteration basis are provided for inheritance, reproduction and collaborative development of the model. If the model data source can be improved, the parameter reproduction is carried out on the model training process, and the model structure flow is perfected.

Visual presentation of workflow designs may also be made: and (4) carrying out contract automatic task scheduling by combing and optimizing the flow of constructing the knowledge model. A series of standardized main and branch flows such as data processing, model training, hyper-parameter optimization, model creation, model registration, model reasoning, model quality inspection, release flows and the like are established, and the construction efficiency of the knowledge model is improved.

The specific example of the security construction method for the intelligent pipe network knowledge model in the first embodiment is also applicable to the security construction system for the intelligent pipe network knowledge model in the present embodiment, and through the detailed description of the security construction method for the intelligent pipe network knowledge model, a person skilled in the art can clearly know the security construction system for the intelligent pipe network knowledge model in the present embodiment, so for the brevity of the description, detailed description is omitted here. The device disclosed by the embodiment corresponds to the method disclosed by the embodiment, so that the description is simple, and the relevant points can be referred to the method part for description.

EXAMPLE III

Based on the same inventive concept as the safety construction method of the intelligent pipe network knowledge model in the foregoing embodiment, as shown in fig. 5, the present application further provides an electronic device 300, where the electronic device 300 includes a memory 301 and a processor 302, a computer program is stored in the memory 301, and when the computer program is executed by the processor 302, the steps of the method in the embodiment are implemented.

The electronic device 300 includes: processor 302, communication interface 303, memory 301. Optionally, the electronic device 300 may also include a bus architecture 304. Wherein, the communication interface 303, the processor 302 and the memory 301 may be connected to each other through a bus architecture 304; the bus architecture 304 may be a Peripheral Component Interconnect (PCI) bus, an Extended Industry Standard Architecture (EISA) bus, or the like. The bus architecture 304 may be divided into an address bus, a data bus, a control bus, and the like. For ease of illustration, only one thick line is shown in FIG. 5, but this is not intended to represent only one bus or type of bus.

Processor 302 may be a CPU, microprocessor, ASIC, or one or more integrated circuits for controlling the execution of programs in accordance with the teachings of the present application.

The communication interface 303 may be any device, such as a transceiver, for communicating with other devices or communication networks, such as an ethernet, a Radio Access Network (RAN), a Wireless Local Area Network (WLAN), a wired access network, and the like.

The memory 301 may be, but is not limited to, a ROM or other type of static storage device that can store static information and instructions, a RAM or other type of dynamic storage device that can store information and instructions, an Electrically Erasable Programmable Read Only Memory (EEPROM), a compact disc read only memory (CD ROM) or other optical disk storage, optical disk storage (including compact disc, laser disc, optical disc, digital versatile disc, blu-ray disc, etc.), a magnetic disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and that can be accessed by a computer. The memory may be self-contained and coupled to the processor through a bus architecture 304. The memory may also be integral to the processor.

The memory 301 is used for storing computer-executable instructions for implementing the present application, and is controlled by the processor 302 to execute. The processor 302 is used for executing the computer-executable instructions stored in the memory 301, thereby implementing the steps of the method in the first embodiment of the present application.

As will be appreciated by one skilled in the art, embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.

The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the application. Thus, the present application is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims (10)

1. A safety construction method of an intelligent pipe network knowledge model is characterized by comprising the following steps:

collecting multidimensional information including data, knowledge, algorithm, applicable scene, model requirement and service system;

carrying out identification processing based on an acquisition object on the multi-dimensional information to construct an information knowledge graph;

performing security design according to the information knowledge graph, determining a security protection strategy, and constructing a knowledge model based on the security protection strategy;

constructing model training data and model testing data, and performing model training and testing on the knowledge model;

and uniformly storing all the knowledge models which are constructed, trained and tested in a warehouse, and establishing a model safety monitoring rule to manage and optimize all the knowledge models.

2. The method of claim 1, wherein the performing an acquisition object-based recognition process on the multi-dimensional information to construct an information knowledgegraph comprises:

carrying out preset data type information identification on the multi-dimensional information, and constructing a knowledge graph conceptual model;

and performing object relation carding on the multi-dimensional information based on the acquired object by using the knowledge graph conceptual model, and outputting a resource distribution and circulation graph.

3. The method of claim 1, wherein performing security design from the information knowledgegraph, determining a security protection policy, and building a knowledge model based on the security protection policy comprises:

determining data, knowledge, an algorithm, an application scene, a model requirement and the safety level of a service system according to the information knowledge graph to obtain a resource safety hierarchical design;

determining the safety protection strategy according to the resource safety hierarchical design, wherein the safety protection strategy comprises participant engagement, calculation engagement, contribution degree engagement and safety engagement;

and constructing a knowledge model based on the participant engagement, the calculation engagement, the contribution engagement and the safety engagement.

4. The method of claim 1, wherein the constructing model training data, model testing data, and model training, testing the knowledge model comprises:

determining a training data source, and acquiring data based on the training data source to obtain original acquisition data;

cleaning the original collected data to obtain preprocessed data;

based on a safety protection strategy, performing data identification according to the preprocessed data to obtain data identification;

generating a unique version identification code for data according to the original collected data, the preprocessed data and the data identification, and constructing the model training data and the model test data;

and determining a model safety target based on the safety protection strategy, and performing model training and testing on the knowledge model by using model training data and model testing data according to the model safety target until the requirement of the model safety target is met.

5. The method of claim 4, wherein after model training and testing the knowledge model, comprising:

obtaining computing force platform side information;

obtaining a deployment environment based on the computing force platform side information, wherein the deployment environment comprises a data center, a cloud, an edge side and an end side;

when the deployment environment is consistent with the training environment, cutting, compressing, packaging, integrating, layering safely, encrypting and cloud edge collaborative optimization are carried out on the model, and the knowledge model is constructed and deployed;

and when the deployment environment is inconsistent with the training environment, adapting and converting the knowledge model, and cutting, compressing, packaging, integrating, layering safely, encrypting and cloud edge collaborative optimization are carried out on the adapted and converted knowledge model.

6. The method of claim 1, wherein the step of uniformly storing all the knowledge models after construction, training and testing in a warehouse and establishing model security monitoring rules to manage and optimize all the knowledge models comprises the steps of:

all knowledge models are uniformly stored in a warehouse of the model, and the knowledge models are detected according to preset warehousing standards;

classifying the warehousing knowledge model in a grading way to obtain a model version, a model attribute, a training process tracing and model deployment online configuration;

the method comprises the steps of carrying out on-line configuration on the deployment of a hanging model, and when the knowledge model is exported, evaluating and verifying the exported knowledge model based on model deployment;

carrying out global monitoring on the operation of the knowledge model, comprising the following steps: running the number of models, model consumption resources, model state details and model calling statistics;

when the number of the operation models, the model consumption resources, the model state details and the model calling statistical global monitoring data are abnormal, data isolation is carried out;

the requirements for the update of the model are established, and updating and optimizing the knowledge model based on the model updating requirement.

7. The method of claim 6, wherein the establishing model update requirements, and the update optimizing of the knowledge model based on the model update requirements, comprises:

acquiring knowledge model version information, and updating the knowledge model according to the model updating requirement according to knowledge model training data and knowledge model version information;

acquiring a knowledge model before updating and a knowledge model after updating, fusing and packaging the knowledge model before updating and the knowledge model after updating, and auditing the fusion and packaging process;

and establishing an incidence relation blood-related graph among all versions of the knowledge model based on the pre-update knowledge model and the post-update knowledge model, and carrying out safety tracking on the knowledge model based on the incidence relation blood-related graph.

8. The method of claim 1, wherein the method comprises:

acquiring real-time knowledge updating information;

determining a knowledge influence model stage according to the real-time knowledge updating information;

and enabling knowledge of the corresponding stage of the knowledge model by using the real-time knowledge updating information based on the knowledge influence model stage, and updating the knowledge model.

9. A safety construction system of a smart pipe network knowledge model is characterized by comprising the following components:

the data access module accesses multi-source data in the intelligent pipe network data lake;

the model development module is connected with the data access module and performs model development by using data in the data access module;

the model operation and maintenance module is connected with the model development module and is used for managing, maintaining, operating and monitoring the model;

the safety execution module is connected with the data access module, the model development module and the model operation and maintenance module and is used for carrying out safety audit and safety monitoring on the whole process of model construction, development and operation and maintenance;

the algorithm module is connected with the model development module and recommends an algorithm in the modeling process;

and the visualization module is respectively connected with the data access module, the model development module, the model operation and maintenance module and the safety execution module, and provides online model visualization service.

10. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the steps of the method according to any of claims 1-8 are implemented when the processor executes the program.

Images