Method for constructing a trusted industrial data space

Abstract

The application discloses a construction method of a trusted industrial data space, and particularly relates to the technical field of industrial data trusted management and sharing; identity characteristic data and behavior data of an industrial subject are collected to establish a behavior data pool; a behavior view atlas is generated based on a behavior ontology model; a behavior consensus index model is constructed to evaluate the historical behavior credibility; the behavior credibility index is semantically aligned with the current access intention, a credibility mapping function is constructed in combination with multiple source decision factors to generate a credibility rating value of the access behavior; a one-time pass token is generated according to the rating result or access is rejected, and behavior feedback is used for model optimization and atlas evolution, and finally an extensible trusted industrial data space is constructed; the method realizes dynamic trust judgment, behavior semantic understanding and cross-domain collaborative control of industrial data access, and improves the credibility, adaptive ability and structural governance ability of the data space.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of trusted management and sharing of industrial data, and in particular to a method for constructing a trusted industrial data space. BACKGROUND

[0002] With the continuous advancement of industrial internet and digital transformation, a large amount of industrial data is circulated and interacted within and across enterprises, and data sharing, circulation and collaborative computing have become an important basis for promoting intelligent manufacturing. However, the current industrial data has serious problems such as data island, lack of trusted sharing mechanism, data tampering risk, etc., which makes it difficult for data resources to be effectively circulated and applied.

[0003] Especially when it comes to core industrial data such as equipment operation data, process control data and production process data, there is a lack of trust mechanism among enterprises. Data providers are worried about data leakage and misuse, and data users are worried about the authenticity and unverifiability of data, which seriously affects the collaboration efficiency and intelligent decision-making ability of cross-organizational industrial data. SUMMARY

[0004] The purpose of the present application is to provide a method for constructing a trusted industrial data space to solve the problems in the background art.

[0005] In order to achieve the above-mentioned purpose, the present application provides the following technical solution: a method for constructing a trusted industrial data space, comprising:

[0006] S100, collecting identity feature data I and data behavior data B of a plurality of industrial subjects, including data call records, policy change records and behavior time sequence features, and establishing a behavior data pool H;

[0007] S200, based on a preset behavior ontology model, feature encoding is performed on the behavior data pool H to generate a behavior view graph G, wherein G reflects the data behavior path and behavior coupling relationship of each industrial subject in different time periods;

[0008] S300, a trusted index model W based on behavior evolution is constructed, and the historical behavior consensus index CI of the subject is calculated according to the time sequence consistency, policy stability and behavior interaction collaboration of each behavior node in G;

[0009] S400, the historical behavior consensus index CI is semantically aligned with the current data access intention, and a multi-source decision factor set is constructed, n is the total number of decision factors, a trust mapping function M is constructed, which is used to dynamically generate a trusted rating value T of access behavior;

[0010] S500, judging whether to allow access to the requested data resource D according to a comparison result of the trust rating value T and the access threshold λ; if T ≥ λ, generating a one-time data authorization token TK and recording the access behavior to the behavior data pool H; otherwise, rejecting the access and updating the view graph G;

[0011] S600, optimizing W and M based on the real-time accumulated H and G, and constructing an extensible trusted industrial data space S, wherein S is represented as a set of data trust state graphs composed of {G, W, M, T}.

[0012] Preferably, the S100 further comprises:

[0013] S101, multi-dimensional feature coding is performed on the identity feature data I of each industrial subject, including subject registration source, historical participation behavior type, authentication frequency and role switching trajectory, to generate an identity vector set I_v for describing the evolution identity trajectory features of the subject;

[0014] S102, semantic extraction is performed on the policy change records in the data behavior data B, a policy behavior tensor model P_t is constructed based on the policy target, influence range and duration, and adaptive clustering is performed on the behavior intensive area to generate a policy change semantic graph P_g;

[0015] S103, the identity vector set I_v and the policy change semantic graph P_g are fused, the behavior confidence B_c of the subject in a set behavior period is calculated, and the behavior data pool H is established with the behavior confidence B_c as an index, so that H can represent the trusted behavior state set of the industrial subject at different time periods.

[0016] Preferably, the S200 comprises:

[0017] S201, based on a preset behavior ontology model, behavior unit nodes in the behavior data pool H are extracted, and semantic labeling is performed according to behavior type, data target, triggering mode and context intention to form an initial behavior ontology mapping set O, wherein each node in O corresponds to a specific industrial data interaction behavior;

[0018] S202, each behavior node in O is sliced according to time sequence to construct a time period sliding window structure W_t, and the state transition mode of the behavior in each window is identified, including behavior continuity, policy inheritance and abnormal jump features, so as to establish a behavior path chain group P under time dependence;

[0019] S203, according to the semantic overlap degree and the behavior time sequence interaction frequency between each node in the behavior path chain group P, a behavior coupling strength matrix C is calculated, and O, P and C are fused to generate a multi-level behavior view graph G, wherein each edge of G represents a coupling relationship, and each level represents a different behavior semantic abstraction layer.

[0020] Preferably, the S300 comprises:

[0021] S301, based on the behavior view graph G, extracting the time sequence execution mode of each behavior node in different time periods, constructing a behavior time consistency vector sequence T_v, representing the characteristic index of the behavior node in terms of periodic triggering, consistent path reproduction and time drift stability;

[0022] S302, performing stability analysis on the policy chain associated with the behavior node, generating a policy stability scalar set P_s according to the policy version iteration frequency, permission floating range and policy reference density, for reflecting the stable and reliable degree of the behavior decision layer;

[0023] S303, embedding the interaction relationship between behavior nodes into a high-dimensional behavior collaboration matrix C_m, calculating the behavior coupling density and behavior influence path length of each subject in the graph G, constructing an interaction collaboration vector I_v, and finally weighting and fusing T_v, P_s and I_v to generate a historical behavior consensus index CI.

[0024] Preferably, the S400 comprises:

[0025] S401, extracting access intention features V from the current access request, including target data type, operation purpose, time requirement and controlled range, and constructing an access intention semantic vector V_s for describing the context semantic space position of the access behavior;

[0026] S402, performing semantic alignment between the historical behavior consensus index CI and the access intention semantic vector V_s, calculating the historical behavior semantic adaptation degree R, and judging whether the current behavior has semantic overlap with the historical high-trust behavior mode through semantic path reasoning in the graph space;

[0027] S403, constructing a trustworthiness mapping function M based on a multi-source decision factor set F, mapping CI, V_s and external decision factors into a trust rating value T, wherein T represents the comprehensive trust level of the current access behavior in multiple contexts.

[0028] Preferably, the S500 comprises:

[0029] S501, comparing the trust rating value T of the access behavior with the access threshold λ under the current policy configuration; if T is greater than or equal to λ, it is considered as a trusted access request, and the token generation stage is entered; otherwise, the behavior abnormal feedback mechanism is triggered and the access is rejected;

[0030] S502, in the case where the access request is determined to be trusted, a one-time data authorization pass token TK is dynamically generated according to the unique resource identifier of the access target data D, the subject identity certificate, and the semantic context of the behavior, and the token is bound and stored for the subsequent behavior audit and token tracing;

[0031] S503, according to the access determination result, a write operation is performed on the behavior data pool H, if the access is approved, a successful behavior path is written; if the access is rejected, the abnormal behavior characteristics and its context information are fed back to the behavior view graph G, and the graph structure is adjusted and updated.

[0032] Preferably, the S600 comprises:

[0033] S601, continuously monitor the change state of the behavior data pool H and the behavior view graph G, and dynamically adjust the parameter weight of the behavior consensus index model W according to the behavior frequency fluctuation, the policy evolution trajectory and the access mode migration trend;

[0034] S602, based on the access result feedback and the abnormal behavior clustering result, the multi-source decision factor weight in the trustworthiness mapping function M is re-estimated;

[0035] S603, the graph structure G, the behavior model W, the mapping function M and the trust rating value set T are uniformly organized into a trust state graph set, and a trusted industrial data space is constructed therefrom.

[0036] In the above technical solution, the technical effects and advantages provided by the present application are:

[0037] 1. By introducing the behavior view graph, the behavior consensus index model and the multi-source decision mapping function, the present application realizes the dynamic trusted modeling and access control of the industrial subject data behavior. Compared with the existing industrial data governance mode relying on static permissions and rule-driven, the present application has the behavior semantic modeling capability, the context self-adaptation capability and the decision structure explainability, and can effectively support the trusted data circulation and cooperation modeling requirements in complex industrial scenarios.

[0038] 2. By constructing the trust state graph set and evolving it into a data space with expandable structure and evolvable behavior, the present application realizes the structured management of the trusted mechanism and the cross-subject sharing capability. The system has the functions of self-learning, self-updating and cross-domain collaboration, and improves the practicality and engineering adaptability of the industrial data space in multi-party cooperation, safe sharing and dynamic trust management. BRIEF DESCRIPTION OF DRAWINGS

[0039] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed in the embodiments. Obviously, the drawings described below are only some embodiments of the present application, and other drawings can also be obtained by those skilled in the art based on these drawings.

[0040] Figure 1 The method flowchart of the present application. DETAILED DESCRIPTION

[0041] In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the present application.

[0042] Embodiment 1, please refer to Figure 1 The construction method of the trusted industrial data space described in this embodiment includes:

[0043] S100, identity feature data I and data behavior data B of a plurality of industrial subjects are collected, including data call records, policy change records and behavior time sequence features, and a behavior data pool H is established;

[0044] S200, based on a preset behavior ontology model, the behavior data pool H is feature-coded to generate a behavior view graph G, wherein G reflects the data behavior path and the coupling relationship between behaviors of each industrial subject in different time periods;

[0045] S300, a trusted index model W based on behavior evolution is constructed, and a historical behavior consensus index CI of the subject is calculated according to the time sequence consistency, policy stability and behavior interaction cooperativeness of each behavior node in G;

[0046] S400, the historical behavior consensus index CI is semantically aligned with the current data access intention, a multi-source decision factor set is constructed, n is the total number of decision factors, a trust mapping function M is constructed, which is used to dynamically generate a trusted rating value T of the access behavior;

[0047] S500, according to the comparison result of the trusted rating value T and the access threshold value λ, it is judged whether to allow access to the requested data resource D; if T ≥ λ, a one-time data authorization pass token TK is generated, and the access behavior is recorded to the behavior data pool H; otherwise, the access is rejected and the view graph G is updated;

[0048] S600, based on real-time accumulated H and G, optimizing W and M, and constructing an extensible trusted industrial data space S, wherein S is represented as a set of data trusted state atlas sets composed of {G, W, M, T}.

[0049] In the present application, the construction of the trusted industrial data space is based on the premise of identity trust, behavior traceability and behavior trust of the industrial subject. On this basis, the first step of the method is to collect the identity characteristic data and data behavior data of multiple industrial subjects, and to establish a dynamic evolution behavior data pool to support the behavior pattern recognition and trust calculation in the subsequent trusted modeling process.

[0050] Firstly, the identity characteristic data of each industrial participant is collected, which includes but is not limited to the following contents: registration source (such as industrial internet platform, local system or alliance platform), authentication frequency (such as regular certificate update), historical role type (such as equipment party, data consumer, data processing party, etc.) and role switching trajectory (the change of the subject's responsibility in different projects, etc.).

[0051] In order to more effectively represent the above multi-dimensional identity attributes, they are converted into a unified format of identity vector expression structure, which is denoted as "identity vector set". In the construction process, a weighted feature embedding algorithm is introduced, so that each industrial subject is represented as a vector, which can dynamically reflect its historical behavior trend and trusted identity evolution characteristics, thereby providing accurate identity benchmark for subsequent behavior analysis and trust rating.

[0052] For example, for a subject who frequently participates in data exchange, frequently changes roles and has complete authentication history, its identity vector has a higher trust characteristic stability dimension; while for a subject who participates for the first time and has incomplete registration information, its vector shows the characteristics of sparsity and mutation with higher trust risk.

[0053] In the process of collecting data behavior data, the present application pays special attention to the semantic features of strategy change behavior. The so-called strategy change behavior refers to the strategic adjustment made by the subject to its data access control rules, data processing flow or cooperation agreement terms, etc. in different business stages. In the traditional approach, only the time and field changes of these behaviors are recorded, and there is a lack of understanding and context analysis of their semantic intent.

[0054] In order to overcome the above shortcomings, the present application introduces a strategy behavior semantic atlas construction method, which maps each strategy change behavior to a strategy intent unit, and associates it to multiple semantic dimensions such as target data type, control boundary, behavior duration, etc. On this basis, an adaptive clustering algorithm based on semantic similarity is used to aggregate the strategy behaviors that repeatedly appear in similar contexts into "strategy intent clusters".

[0055] For example, an industrial subject frequently modifies its permission settings and shortens the data access time window before and after accessing core production data, and this behavior can be aggregated into a "high-sensitive data temporary access mode" policy cluster. This clustering behavior not only represents its operation mode, but also reflects the purpose behind its data behavior.

[0056] Finally, the semantic graph represents the policy units as nodes and the semantic linkage relationship between the policies as edges, forming a clear structure of the policy semantic graph, which lays a semantic analysis foundation for behavior trust modeling.

[0057] Based on the above identity vector and policy semantic graph, the application further constructs a behavior confidence model of the subject. The model is used to comprehensively evaluate the trust degree of the behavior of a subject in a set period and a set semantic scene.

[0058] The behavior confidence is a dynamic value, and the following factors are referred to in the calculation: whether the behavior has appeared in history, whether there is semantic anomaly, whether there is identity state switching, etc. In implementation, a sliding time window mechanism is adopted to analyze the behavior sequence in a unit time and extract behavior confidence factors. Each behavior confidence factor is weighted and calculated according to parameters such as identity consistency, policy stability, and context continuity, and finally the behavior confidence value is obtained.

[0059] For example, in the last 24 hours, a subject continuously initiated multiple access requests for high-sensitive data, but its identity vector shows that it is a low-privilege subject, and there is a policy mutation behavior, so the behavior confidence of the subject will decrease significantly, and it will be recorded as "untrusted behavior".

[0060] All the behavior data after the confidence modeling will be uniformly put into the behavior data pool. The data pool is organized in a time series index manner, and a dedicated behavior view is constructed for each subject, so that the system can call and analyze the behavior state change trajectory of any subject in history at any time.

[0061] The behavior data pool is not only a storage container for static behavior data, but also a knowledge carrier in the dynamic evolution process of trustworthiness, with core capabilities such as iterative updating, cross-subject comparison, and behavior mode feedback. It provides a semantic layer support and behavior layer perception base structure for the entire trusted industrial data space.

[0062] In the trusted industrial data space construction method provided by the application, the structured expression and semantic graph modeling of behavior data are the key foundations for realizing behavior trust analysis. Therefore, after the establishment of the behavior data pool H, the system enters the construction phase of the behavior view graph G, i.e., step S200.

[0063] Firstly, the system pre-sets a set of industrial data behavior ontology model, which defines a category system of typical industrial behaviors, a correlation mode of behaviors and data resources, a trigger condition, a strategy direction and a context intention and the like structures.

[0064] On this basis, all original behavior records in the behavior data pool H are extracted and abstracted into a group of "behavior unit nodes". Each behavior unit node represents a specific data interaction behavior of an industrial subject, including data request, strategy setting, access path adjustment, data processing action and the like.

[0065] For each node, the system performs a semantic labeling operation based on the behavior ontology model. The semantic labeling includes:

[0066] behavior type labeling (such as data reading, data publishing, strategy negotiation and the like);

[0067] data target type labeling (such as device state data, environment monitoring data, production process parameters and the like);

[0068] behavior trigger mode labeling (such as periodic trigger, event trigger, external driving and the like);

[0069] context intention labeling (such as data verification, collaborative modeling, analysis and inference and the like).

[0070] Through the above labeling process, an initial behavior ontology mapping set is formed, denoted as O. Each node in the set O is a behavior unit with complete semantic labels, which constitutes a basic behavior structure that can be semantically explained and compared.

[0071] The credible analysis of industrial behaviors cannot be separated from the time dimension. In order to mine the behavior migration law of the subject in different time periods, the invention introduces a sliding time window mechanism.

[0072] The specific operation is as follows: the behavior nodes in the set O are sorted according to the time stamp, a time window length T_w and a step δ are set, the set O is sliced to generate a plurality of time period behavior subsets, and each subset represents the behavior set of the industrial subject in the time period [t, t+T_w], t being the time.

[0073] In each time window, the state transition mode between the behavior nodes is analyzed. The so-called state transition refers to the execution order relationship, semantic inheritance relationship and strategy continuity between the behavior nodes. For example:

[0074] if behavior A follows behavior B and acts on the same data resource, it constitutes "continuous access";

[0075] if the strategy setting behavior P is set before the behavior D and is referenced in the subsequent access, it constitutes "strategy inheritance";

[0076] If a behavior is suddenly inserted into a high-risk access in the regular sequence, it is marked as "behavior mutation".

[0077] Through the above operation, each time window will generate a set of directed behavior paths, called behavior path chain set, denoted as P. Each path chain is represented by nodes (behavior units) and edges (state transitions) in the graph structure.

[0078] This mechanism can significantly improve the system's ability to identify behavior sequence patterns and behavior evolution trends, so that the behavior graph not only contains isolated nodes, but also has time evolution semantics.

[0079] After completing the behavior semantic modeling and time path identification, the invention further introduces a "behavior coupling relationship modeling" mechanism, aiming to mine the coupling relationship between different behaviors in the semantic space and execution frequency, and then build a behavior coupling strength matrix C.

[0080] The calculation method of this matrix includes the following dimensions:

[0081] Semantic overlap: by calculating the similarity of two behavior nodes in the label space, such as consistent behavior type, consistent target data or highly related context intent;

[0082] Timing interaction frequency: statistics the frequency of two behaviors being triggered jointly or executed sequentially within or across time windows;

[0083] Strategy path overlap: determine whether two behaviors share the same strategy link, or have common or inheritance relationship in strategy.

[0084] Each pair of behavior nodes obtains a coupling strength value according to the above indicators, forming a coupling strength matrix C. This matrix can be interpreted as a measure of implicit collaboration between behaviors, providing a weight basis for graph generation.

[0085] Finally, the system fuses the behavior ontology mapping set O, the behavior path chain set P and the coupling strength matrix C to generate a multi-level behavior view graph G. The structure of graph G has the following characteristics:

[0086] Node layer: semantic annotated behavior unit node;

[0087] Edge layer: time path edge (sequential relationship) and coupling edge (semantic and frequency coupling);

[0088] Hierarchical structure: behaviors can be divided into strategy behavior layer, operation behavior layer and control behavior layer according to the degree of semantic abstraction, realizing semantic abstraction and behavior clustering;

[0089] As the behavior data pool expands, G can dynamically grow, adjust edge weights and merge nodes, realizing the evolution and scalability of the graph.

[0090] The present application aims to construct a trusted index model for the industrial data trusted circulation scenario, so as to comprehensively evaluate the behavior trust level of industrial participants in different stages and different contexts. After the construction of the behavior view graph G is completed, the system enters the trusted index modeling stage, that is, step S300.

[0091] Trusted behavior usually shows certain time regularity and path repeatability. Therefore, the present application first extracts the time evolution mode of each behavior node from the graph G.

[0092] The specific method is as follows: assuming that each behavior node has a time stamp attribute, the system counts the periodic recurrence frequency, path reuse rate and execution time drift range of the node within a given time window, and combines the three types of indexes into a behavior time consistency vector, denoted as T_v.

[0093] The periodic recurrence frequency refers to the frequency of the occurrence of a behavior in the same time period (such as the same period every week);

[0094] The path reuse rate indicates whether the path where the behavior is located is shared by multiple behavior chains in different time periods;

[0095] The execution time drift range indicates the standard deviation between the actual execution time of the behavior and the historical average.

[0096] Through the construction of such a structure vector, the system can identify which behaviors are “stable and repetitive” and which behaviors are “sudden and random”. This behavior time characteristic can be used as an important dimension for behavior trust degree modeling, avoiding the volatility interference of single behavior on trust evaluation.

[0097] Strategic behavior is the core background condition affecting behavior trust degree. Frequent fluctuations in strategy setting and drastic adjustments in permission range usually mean potential trust risks. Therefore, in the present application, the system performs stability evaluation on the strategy chain relied on or referenced by the behavior node, and generates a set of strategy stability scalars, denoted as P_s.

[0098] The construction of P_s is based on the following three key indicators:

[0099] Strategy version iteration frequency: records the number of updates of the same strategy object (such as an access control list) of a subject in a specific time period;

[0100] Permission float range: measures the degree of access permission change after each strategy modification compared with the last time, which can be represented by the size of the set difference of resource access range;

[0101] Strategy reference density: measures the frequency of a certain strategy being reused by multiple behavior nodes. If the density is high, it indicates that the strategy is stable and the behavior is general. If the density is low, it indicates that the strategy is specific and has potential abnormal risks.

[0102] Finally, the system synthesizes the above indicators to form the strategy stability score of each subject, which is used as the second vector input in the behavior credibility calculation. Compared with the traditional method of using "whether the strategy exists" as the judgment standard, this method describes the credible performance of the strategy in actual operation from the perspective of dynamic evolution, enhancing the understanding ability of the model to the strategy behavior.

[0103] In the multi-agent collaborative scenario, the behavior credibility is often closely related to its relationship with other subjects or behavior nodes in the graph. Therefore, the invention further proposes a behavior interaction and collaboration analysis model to describe the semantic coupling and path linkage characteristics between behaviors.

[0104] The specific process is as follows:

[0105] First, extract the in-edge and out-edge sets of all behavior nodes in the behavior graph G, and establish the coupling path between them and other nodes;

[0106] Then build a behavior collaboration matrix, denoted as C_m, where each matrix element represents the interaction intensity of two behavior nodes in multiple dimensions, including:

[0107] Path co-occurrence frequency;

[0108] Strategy sharing ratio;

[0109] Temporal closeness.

[0110] On this basis, a behavior collaboration vector is constructed for each subject, denoted as I_v, which describes the behavior centrality, influence path length, and collaboration frequency of the subject in the graph.

[0111] For example, if a subject's behavior is always located in the center of the collaborative path and its strategy is frequently referenced by other subjects, its collaboration credibility is high. Conversely, if a subject's behavior is isolated and frequently inserted into broken link behaviors, its credibility will be reduced.

[0112] After obtaining the above three core structures:

[0113] Behavior time consistency vector T_v,

[0114] Strategy stability scalar set P_s,

[0115] Behavior collaboration vector I_v,

[0116] Then, the system performs weighted fusion on the three, generating the final historical behavior consensus index CI.

[0117] The calculation of CI can be realized by a weighted function model. For example, if three input features have weights α, β and γ respectively (where α+β+γ=1), then: consensus index CI =α×time consistency score+β×strategy stability score+γ×synergy credibility score.

[0118] In practical applications, the system can flexibly adjust the weights according to different industry scenarios and safety requirements to meet diversified trust modeling requirements such as "behavior stability first" or "strategy credibility first".

[0119] Finally, the system includes the CI values of all subjects in a unified trust index model W. The model W has the following characteristics:

[0120] Dynamic evolution: with the update of the behavior data pool, the CI value can be periodically refreshed;

[0121] Graph dependence: all calculations rely on the behavior graph G to realize structure-aware trust judgment;

[0122] Multi-dimensional input: three dimensions of time, strategy and synergy are introduced to avoid single indicator judgment;

[0123] Comparability and interpretability: the CI value has a unified dimension, can be compared horizontally, and its calculation process can be traced.

[0124] In the construction process of the trusted industrial data space, only the historical behavior performance (i.e. historical behavior consensus index CI) is not enough to completely judge the trust degree of an access behavior in the current context. Because the trustworthiness of data access behavior not only depends on the trust record of the subject in the past, but also is influenced by many factors such as access intention, behavior context, data sensitivity, time requirement, etc.

[0125] Therefore, in step S400, the application proposes a dynamic trust rating mechanism based on semantic alignment and multi-source factor driving. The core goal is to align the historical behavior consensus index CI with the semantic vector of the current access request, and combine external decision factors to construct a generalizable and evolving trust degree mapping function M, which is used to generate a dynamic access behavior trust rating value T.

[0126] Firstly, for the data access request initiated by an industrial subject, the system extracts the access intention feature of the request to construct the semantic representation of the access behavior.

[0127] The access intention feature V includes but is not limited to the following contents:

[0128] Target data type: the data category involved in the access object, such as device state data, process parameters, real-time sensor data, etc.

[0129] Access operation type: e.g. read, modify, bulk download, real-time subscription, etc. operation behavior;

[0130] Operation purpose / application context: e.g. for device fault analysis, cross-enterprise modeling, quality traceability, etc.

[0131] Access timeliness requirement: e.g. real-time requirement, historical data window size;

[0132] Controlled range: whether the behavior is limited to a certain cooperation agreement, whether it crosses the sovereign domain boundary, etc.

[0133] Based on the above characteristics, the system uses a semantic embedding mechanism to convert structured fields into a unified vector representation, generating an access intent semantic vector, denoted as V_s.

[0134] The main role of V_s is to project the access request into the semantic space where the industrial behavior graph G is located, so that the similarity and deviation degree of the subject's historical behavior and current behavior can be compared in the unified semantic graph structure.

[0135] The system constructs a semantic alignment mechanism based on the graph G and the historical behavior consensus index CI, which is used to judge the semantic adaptability between the current access intent V_s and the historical trusted behavior.

[0136] This adaptation process is not based on static rules or template matching, but is completed with the help of graph path reasoning:

[0137] In the graph G, a group of historical behavior nodes most similar to the current subject behavior is identified, and the screening conditions include behavior type, operation target, strategy background, etc.

[0138] Take each matching behavior as the starting point, and trace back its context semantic path (e.g. strategy chain, coordination node, behavior evolution path);

[0139] Calculate the similarity between the V_s vector and the average embedding vector of the nodes in these semantic paths to get the semantic adaptation degree R.

[0140] The semantic adaptation degree R reflects the degree of semantic consistency between the current access behavior and the historical trusted behavior. If there is a high degree of semantic overlap between the current behavior and the historical behavior (e.g. same type of access, same type of data, same type of context), the R value is higher, indicating that it has good historical reference.

[0141] If the current behavior deviates significantly from the historical behavior in terms of access content, behavior method or context (e.g. first access to sensitive data, first appearance of data cross-domain access request), the R value will be reduced, prompting the system that the behavior lacks a trusted historical precedent.

[0142] After obtaining the historical behavior consensus index CI and the semantic adaptation degree R, the system further considers external context factors to realize a multi-source fusion trusted judgment.

[0143] To this end, the present application proposes a multi-factor driven trustworthiness mapping function M for mapping various indicators to a trust rating value T of the current access behavior.

[0144] The input of the mapping function M includes:

[0145] CI (historical behavior consensus index): measures the overall trust level of the subject's behavior history;

[0146] V_s (access intention semantic vector): represents the semantic properties of the current behavior;

[0147] R (semantic adaptation degree): represents the semantic correlation strength between the current behavior and the historical behavior;

[0148] F (external decision factor set): includes a series of dynamic context variables that affect access trust judgment, such as environmental security level, cooperation agreement terms, time sensitivity level, resource load state, data usage compliance, etc.

[0149] The number of external decision factor set F is scalable, set as where n represents the total number of factors. Each factor has an independent weight, which can be learned from historical access results or obtained by pre-configured strategy.

[0150] The goal of the mapping function M is to convert the input variable set (CI, V_s, R, F) into a trust score value T, which can be set to 0~1 or equivalent trust level grading, such as a five-level scoring system (untrusted, low trust, medium trust, high trust, fully trusted).

[0151] In one implementation, M can be trained using data-driven methods such as logistic regression function, support vector regression model, graph neural network mapping model, etc. It can also use rule-driven models (such as expert scoring function or rule atlas).

[0152] For example: where , , are weight parameters; this T value will be the core score basis for whether the current access behavior can be trusted, participating in subsequent access control processes, such as whether to generate an access token, whether to trigger manual review, whether to temporarily shrink permissions, etc.

[0153] ​In the construction process of the trusted industrial data space, only the trusted modeling and semantic scoring of the historical behavior of the industrial subject is still insufficient to support the complete data security circulation mechanism. The system also needs to have an executable, controllable and traceable behavior response mechanism to make a clear authorization or rejection judgment on the industrial data access behavior, and to ensure that the trusted score can play an effective role in actual operation.

[0154] To this end, in step S500, the present application proposes an access control decision mechanism based on the comparison of the trusted rating value T and the access threshold λ, combined with the behavior pass token generation, behavior binding record and graph adaptive update mechanism, to build a dynamic access management system with a decision-making closed loop.

[0155] Firstly, the system has calculated a comprehensive trusted rating value T for the current access request behavior in the previous stage, which is obtained by fusing multi-source information such as behavior consensus index, access intention semantics, external context decision factor, etc. In order to ensure that the access behavior meets the requirements of the security policy, the system sets an access threshold corresponding to each type of data resource, denoted as λ.

[0156] The access threshold λ can be dynamically set according to the following parameters:

[0157] The sensitivity level of the requested data D;

[0158] The domain or sovereignty boundary to which the data belongs;

[0159] The risk situation of the current time period (such as automatically increasing λ in high-risk time intervals);

[0160] The role level and authorized range of the visitor.

[0161] The system compares the trusted rating value T with the access threshold λ:

[0162] If T ≥ λ, it means that the current access behavior meets or exceeds the access threshold of the requested data in terms of comprehensive trustworthiness, and is judged as "trusted behavior", allowing to enter the token generation phase;

[0163] If T < λ, it means that the current behavior is not trustworthy enough, and the system will reject this access and trigger the subsequent behavior feedback and graph update mechanism.

[0164] This judgment logic is not only a static evaluation of the current behavior trustworthiness, but also a dynamic policy decision result combined with the data itself attributes and access context conditions.

[0165] When a certain access behavior is judged as trusted, the system will enter the token generation phase. Unlike traditional fixed credential authorization methods, the invention introduces the concept of "one-time data authorization access token (Token, hereinafter referred to as TK)", which is only valid under the current behavior semantics and time window, has strong binding and time limit, and prevents token transfer, reuse or abuse.

[0166] In the process of generating the access token TK, the system writes the following elements into the token structure:

[0167] The identity credential ID of the access request subject (such as encrypted identification or DID);

[0168] The unique resource identifier of the target data D (such as URI, data object hash, etc.);

[0169] The semantic summary of the current behavior (including access intent, operation type, policy context);

[0170] The trusted rating value T and its calculation path summary;

[0171] The valid time window (such as valid within 10 minutes);

[0172] The one-time use identifier (non-reusable marker).

[0173] The above information will be signed or encrypted and packaged into a structured token TK as the unique authorization access credential for this access operation. The token needs to be submitted with the request in the subsequent data access interface, and the system will verify its legitimacy.

[0174] In addition, the token structure will also be recorded in the behavior data pool H as an audit basis for behavior events, realizing the behavior-credential-data ternary binding, and providing structural support for subsequent behavior traceability and safe review.

[0175] If the current access behavior is rejected (i.e. T is less than λ), the system will not simply discard the request, but will feed the behavior as an "abnormal behavior event" to the behavior data pool H, and trigger the graph update mechanism.

[0176] The mechanism includes two sub-processes:

[0177] Abnormal behavior semantic record: write the access features (V_s), context environment (F), rating value (T), target data D, etc. of the current behavior node into the behavior data pool H, mark it as a "rejected" event, and associate the failure reason label (such as "low permission", "intent deviation", "policy conflict").

[0178] Behavior view graph G update: the system updates the graph structure according to the rejection result of the behavior, including but not limited to the following operations:

[0179] Abnormal behavior nodes are marked as "low trust behavior";

[0180] Coupling conflict detection is performed on the policy chain on its path, upstream and downstream behavior relationship;

[0181] If multiple subjects are continuously denied access to a certain type of data in a similar context, a "risk path identification area" is formed in the graph;

[0182] If the same subject fails multiple times but has high behavior consistency, the system can perform "trusted warning and release processing" on its behavior node, that is, add a "to be promoted" weight node in the graph for weight adjustment in the next behavior analysis.

[0183] Through the graph feedback mechanism, the system not only rejects the current access operation, but also "learns" the characteristics of a failed behavior once, providing a structural reference for the next assessment of similar behavior, thereby realizing the continuous evolution of the behavior graph and the enhancement of the strategy adaptability.

[0184] In the process of industrial data trusted exchange and governance, it is difficult to adapt to the changing behavior scenarios and dynamic strategies in complex industrial environments by relying only on static rules or one-time evaluation mechanisms. To realize a data governance system that is long-term running, heterogeneous collaboration and multi-agent participation, the present application proposes a model optimization mechanism based on real-time behavior feedback and a trusted state graph set construction method in step S600, and then forms an extensible trusted industrial data space S.

[0185] Specifically, the trusted industrial data space S is a trusted space set with semantic interpretability, state evolution capability and cross-agent adaptability, which is constructed by continuous evolution, collaborative optimization and structural governance, with behavior graph G, consensus index model W, trusted mapping function M and dynamic trusted rating value T as the core components.

[0186] The present application realizes the continuous optimization of the historical behavior consensus index model W by real-time monitoring of the structural changes of the behavior data pool H and the behavior graph G.

[0187] Specifically, as the behavior data pool H continues to accumulate, the system can identify the following behavior evolution characteristics:

[0188] Behavior frequency fluctuation: for example, a certain type of access behavior significantly increases or decreases in a certain time period;

[0189] Strategy chain evolution: for example, the control strategy used for the same data request is frequently changed, and the strategy path jumps;

[0190] Access path migration trend: that is, the behavior path of the subject migrates from a certain part of the graph G to another area, and the behavior coupling relationship is restructured.

[0191] After recognizing the above evolution trend, the system adaptively adjusts the weight parameters in the behavior consensus index model W. For example, if the historical confidence of a certain access behavior is low but it appears frequently recently and is authorized successfully, the system can increase its time series weight to adapt to the new behavior pattern; if a certain strategy path is gradually abandoned, its influence factor will be weakened.

[0192] The above mechanism can be regarded as a graph behavior trend driven model learning path, which makes the model W continuously adapt to the dynamic behavior evolution scenario and has the ability of self-learning and self-correction.

[0193] In addition to behavior evolution, the accuracy of the trusted mapping function M is also affected by external context factors (such as policy adjustment, cooperation agreement change, data sensitivity change, etc.).

[0194] Therefore, the application introduces a context-aware factor re-estimation mechanism to dynamically adjust the weights of various decision factors in the mapping function M, and enhances the response ability of the function to environmental changes.

[0195] The original structure of the mapping function M contains multiple external decision factors (denoted as to ), such as the current data sensitivity level, the access time period security level, the subject role change, the cooperation agreement validity, etc. The system monitors the actual influence of these factors in real time during operation, for example:

[0196] If the "cooperation agreement validity" significantly increases the influence on access decision at a certain stage, its weight is automatically increased;

[0197] If the "access time period" has little influence on the decision, the influence coefficient of the factor in the function M is correspondingly reduced.

[0198] The system can also introduce a sliding window aggregation analysis to learn the historical contribution of each factor, thereby dynamically constructing a factor sensitivity curve to enhance the context adaptability of the function M.

[0199] Finally, the output trusted rating value T of the function M will more accurately reflect the trusted state of the behavior in the current actual environment and semantic context, not only relying on historical model reasoning, but also integrating real-time context evolution, and having a strategy response force.

[0200] After completing the optimization of the model W and the function M, the application proposes to organize the graph G, the model W, the function M and the dynamic rating value set T into a trusted state graph set, and to build a trusted industrial data space S with it as the logical core.

[0201] The specific steps are as follows:

[0202] Each industrial subject or business domain builds its own graph structure unit, including its behavior view graph , consensus model , mapping function and real-time rating value sequence . Multiple units can form a distributed sub-graph, constituting a global graph set.

[0203] To adapt to the multi-agent deployment environment, the trusted state graph set supports graph-level sharding deployment, and the sub-graphs are interconnected through identity indexing, policy link interface and behavior path interworking protocol.

[0204] When the subject migrates the operation behavior between multiple graph units, its behavior history and rating state are synchronized through the state migration interface to ensure that the behavior trusted information is not lost due to platform changes.

[0205] If multiple subjects have overlapping business paths or joint cooperation scenarios, the graph set will fuse or couple the behavior nodes and policy chains to form cross-domain trusted paths to support joint modeling, collaborative analysis and other industrial scenarios.

[0206] Finally, the trusted industrial data space S constructed is defined as: , where i represents the graph unit number, and h is the number of industrial subjects or trust domains participating in the current system.

[0207] This space has the following characteristics:

[0208] Structural scalability: new subjects can access the system and quickly build their own graph units;

[0209] Behavior evolution compatibility: support behavior pattern migration and evolution, and the system automatically adjusts the model parameters;

[0210] State semantic uniformity: through the standardized G-W-M-T architecture, each graph unit can be interoperable;

[0211] Policy autonomous controllability: each graph unit can adjust its access policy, trusted threshold λ and other configurations according to business rules.

[0212] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto, and any skilled person in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be covered within the protection scope of the present application.

Claims

1. A method for constructing a trusted industrial data space, characterized by: include: S100. Collect identity feature data I and data behavior data B from multiple industrial entities, including data call records, policy change records and behavioral time sequence characteristics, and establish a behavioral data pool H. S200. Based on the preset behavior ontology model, feature encoding is performed on the behavior data pool H to generate a behavior view map G, where G reflects the data behavior path and the coupling relationship between behaviors of each industrial entity in different time periods. S300. Construct a credibility index model W based on behavioral evolution, and calculate the historical behavioral consensus index CI of the subject based on the temporal consistency, strategy stability and behavioral interaction synergy of each behavioral node in G. S400. Semantically align the historical behavior consensus index (CI) with the current data access intent, based on a multi-source decision factor set. , where n is the total number of decision factors, construct a credibility mapping function M to dynamically generate credibility rating values ​​T for access behavior; S500. Based on the comparison between the trust rating value T and the access threshold λ, determine whether to allow access to the requested data resource D. If T ≥ λ, then generate a one-time data authorization token TK and record the access behavior to the behavior data pool H; otherwise, deny access and update the view graph G. S600 optimizes W and M based on real-time accumulated H and G to construct a scalable trusted industrial data space S, where S represents a set of trusted data state maps consisting of {G, W, M, T}.

2. The method for constructing a trusted industrial data space according to claim 1, characterized in that: S100 further includes: S101. Perform multi-dimensional feature encoding on the identity feature data I of each industrial entity, including entity registration source, historical participation behavior type, authentication frequency and role switching trajectory, to generate identity vector set I_v, which is used to describe the evolutionary identity trajectory features of the entity. S102. Semantically extract the strategy change records in the data behavior data B, construct a strategy behavior tensor model P_t based on the strategy goal, scope of influence and duration, and adaptively cluster the dense behavior areas to generate a strategy change semantic map P_g. S103. Integrate the identity vector set I_v and the policy change semantic graph P_g, calculate the behavioral confidence B_c of the subject within a set behavioral period, and use this as an index to establish a behavioral data pool H, so that H can represent the set of credible behavioral states of industrial subjects in different time periods.

3. The method for constructing a trusted industrial data space according to claim 2, characterized in that: S200 includes: S201. Based on the preset behavior ontology model, extract the behavior unit nodes in the behavior data pool H, and perform semantic annotation according to behavior type, data target, triggering method and contextual intent to form an initial behavior ontology mapping set O, where each node in O corresponds to a specific industrial data interaction behavior. S202. Slice each behavior node in O according to the time series, construct a time-sliding window structure W_t, and identify the state transition pattern of behavior in each window, including behavior continuity, strategy inheritance and abnormal jump characteristics, so as to establish a time-dependent behavior path chain group P. S203. Based on the semantic overlap and behavioral temporal interaction frequency between nodes in the behavioral path chain group P, calculate the behavioral coupling strength matrix C, and merge O, P and C to generate a multi-level behavioral view graph G, where each side of G represents a type of coupling relationship, and each level represents a different behavioral semantic abstraction layer.

4. The method for constructing a trusted industrial data space according to claim 3, characterized in that: The S300 includes: S301. Based on the behavior view graph G, extract the temporal execution mode of each behavior node in different time periods, construct the behavior time consistency vector sequence T_v, and characterize the behavior node in terms of periodic triggering, consistent path reproduction and time drift stability. S302. Perform stability analysis on the policy chain associated with the behavior node. Based on the policy version iteration frequency, permission fluctuation range and policy reference density, generate a policy stability scalar set P_s to reflect the stability and reliability of the behavior decision layer. S303. Embed the interaction relationship between behavioral nodes into the high-dimensional behavioral collaboration matrix C_m, calculate the behavioral coupling density and behavioral influence path length of each subject in the graph G, construct the interaction collaboration vector I_v, and finally weight and fuse T_v, P_s and I_v to generate the historical behavioral consensus index CI.

5. The method for constructing a trusted industrial data space according to claim 4, characterized in that: The S400 includes: S401. Extract the access intent feature V from the current access request, including the target data type, operation purpose, timeliness requirements and controlled scope, and construct the access intent semantic vector V_s to describe the contextual semantic space location of the access behavior. S402. Semantically align the historical behavior consensus index CI with the access intention semantic vector V_s, calculate its historical behavior semantic fit degree R, and determine whether the current behavior has semantic overlap with the historical high-confidence behavior pattern through semantic path reasoning in the graph space. S403. Construct a credibility mapping function M based on the multi-source decision factor set F, which maps CI, V_s and external decision factors together to a credibility rating value T, where T represents the comprehensive credibility level of the current access behavior in multiple contexts.

6. The method for constructing a trusted industrial data space according to claim 5, characterized in that: The S500 includes: S501. Compare the trust rating value T of the access behavior with the access threshold λ under the current policy configuration; if T is greater than or equal to λ, it is considered a trusted access request and enters the token generation stage; otherwise, the behavior anomaly feedback mechanism is triggered and access is denied. S502. When the access request is deemed trustworthy, a one-time data authorization token TK is dynamically generated based on the unique resource identifier of the access target data D, the subject identity certificate, and the semantic context of the behavior. The token is then bound to the current behavior event for storage and subsequent behavior auditing and token tracing. S503. Based on the access determination result, perform a write operation on the behavior data pool H. If the access is approved, write the successful behavior path; if the access is denied, feed back the abnormal behavior characteristics and their context information to the behavior view graph G, and adjust and update the graph structure.

7. The method for constructing a trusted industrial data space according to claim 6, characterized in that: The S600 includes: S601. Continuously monitor the changes in the behavior data pool H and the behavior view graph G, and dynamically adjust the parameter weights of the behavior consensus index model W based on the fluctuations in behavior frequency, the evolution trajectory of strategies, and the migration trend of access patterns. S602. Based on the feedback of access results and the clustering results of abnormal behavior, the weights of multi-source decision factors in the credibility mapping function M are re-evaluated. S603. The graph structure G, behavior model W, mapping function M and trust rating value set T are uniformly organized into a trustworthy state graph set, and a trustworthy industrial data space is constructed using it.

Images