A Loss Boundary Constraint Access Control Method and System for Downlink Access in Multi-Source Heterogeneous Industrial Internet of Things

By introducing a loss boundary constraint access control method into a multi-source heterogeneous industrial Internet of Things, the problem of the disconnect between the control plane and the data plane in high-risk downlink access is solved. This achieves clear worst-case net loss upper bound control and continuous reduction of abuse risk, ensuring business continuity and security.

CN123122952APending Publication Date: 2026-11-17HARBIN INST OF TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611392352.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-09-09
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

In multi-source heterogeneous industrial IoT, existing technologies lack clear upper bound constraints on the worst-case net loss for high-risk downlink access control, leading to a disconnect between control plane decision-making and data plane execution, and making it difficult to control the risk of replay or continued abuse after access is granted.

Method used

The loss boundary constraint access control method is adopted. By defining request quadruples, matching access control loss boundary policies, calculating the worst-case net loss acceptable boundary and candidate online stop mechanism bundles, determining whether the upper bound of the request's net loss exceeds the boundary, selecting the mechanism bundle with the minimum overall cost, and binding the control plane authorization semantics and data plane execution paths through short-term authorization evidence.

Benefits of technology

It achieves clear consequence boundary control in high-risk downlink access, ensuring the availability of necessary services. At the same time, it continuously reduces the risk of subsequent abuse of granted capabilities through mechanisms such as rate limiting, budgeting, degradation and write protection, avoiding the lack of intermediate control states in traditional binary models.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123122952A_ABST
    Figure CN123122952A_ABST
Patent Text Reader

Abstract

This invention discloses a loss boundary constraint access control method and system for downlink access in multi-source heterogeneous industrial IoT, belonging to the field of industrial IoT security technology. It addresses the problems of insufficient binary authorization granularity, disconnect between control plane decision-making and data plane execution, and difficulty in controlling the risk of replay or continuous abuse after admission in multi-source heterogeneous industrial IoT downlink access. The method includes: bounded feasibility admission determination based on the upper bound of the worst-case net loss; when a mechanism constraint within the policy's allowed range ensures that the upper bound of the worst-case net loss of the request does not exceed the policy boundary, the system issues and binds short-term authorization evidence, and continuously executes online loss mitigation mechanisms such as rate limiting, budget control, output degradation, write protection, and reversal refresh on the agent side. This invention is applicable to various types of downlink requests, such as industrial telemetry reading, historical window export, and remote parameter updates, and can implement differentiated authorization for multiple source entities, including cloud, enterprise, edge, and third parties.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of industrial IoT security technology, and in particular to cloud-edge-device collaboration, access control, and risk constraint enforcement in industrial IoT security. Background Technology

[0002] As IoT systems evolve from isolated deployments to cross-domain collaborative deployments, various upstream entities, including cloud platforms, enterprise business systems, edge gateways, third-party services, and human-machine interfaces, are constantly initiating downlink access requests to the terminal side, device side, or industrial control side. These requests may involve operations such as telemetry data reading, historical process data export, device configuration writing, sampling strategy updates, and remote operation and maintenance. Unlike access to traditional information systems, once industrial IoT downlink access is granted, it not only affects the visibility of logical layer data but may also affect device status, control logic, business continuity, and the scope of sensitive information dissemination.

[0003] Existing access control methods primarily revolve around "who can access which resources under what conditions." For example, role-based access control, task-based access control, attribute-based access control, and usage control models can express roles, task states, attributes, contexts, and usage conditions. However, in high-risk downlink access, simply providing a binary decision of allow or deny is insufficient to guarantee security. Attackers may not bypass access control points but instead reuse legitimate credentials, cached tokens, or authorized sessions to accumulate damage by increasing request frequency, extending access time, repeatedly exporting window data, expanding the scope of write impact, or continuing to use acquired capabilities after context changes.

[0004] Therefore, the real problems to be solved in multi-source heterogeneous industrial IoT are not just whether the request hits the policy rules, but whether the execution after admission can be kept within the boundaries of clear, executable, and verifiable worst-case consequences; whether the constraints given by the control plane can be strongly bound to the actual execution path of the data plane; and whether the abuse window after admission can be continuously narrowed through online mechanisms. Existing technologies have made improvements in risk perception, zero-trust continuous verification, capability tokens, and blockchain auditing, but they usually still treat risk as a decision input or threshold, lacking admission semantics with a hard constraint of the upper bound of the worst-case net loss, and also lacking a complete closed loop that binds this boundary, mechanism bundle, and parameters as short-term evidence to the execution path.

[0005] Based on the above problems, there is an urgent need for an access control scheme for high-risk downlink access in the Industrial Internet of Things. Summary of the Invention

[0006] The purpose of this invention is to address the problems existing in downlink access of multi-source heterogeneous industrial IoT, such as insufficient binarized authorization granularity, disconnect between control plane decision-making and data plane execution, and difficulty in controlling the risk of replay or continuous abuse after admission. This invention provides a loss boundary constraint access control method and system for downlink access of multi-source heterogeneous industrial IoT.

[0007] The technical solution adopted by this invention to solve the above problems is: a loss boundary constraint access control method for downlink access in multi-source heterogeneous industrial IoT, the method comprising:

[0008] Step 1: Define downlink access requests from multi-source heterogeneous industrial IoT as request quadruplets containing subject, object, action and context; match effective policy rules from the pre-configured access control loss boundary policy set to obtain the worst-case net loss acceptable boundary B(r) corresponding to the request, and the pre-configured finite candidate online stop loss mechanism bundle set M(r) in the effective policy rules.

[0009] The candidate online stop mechanism bundle is a finite combination of online stop mechanisms. Each candidate online stop mechanism bundle includes a mechanism type, mechanism parameters or parameter range, scope of application, and an identifier indicating whether it causes request semantic contraction.

[0010] The worst-case net loss acceptable boundary B(r) refers to the maximum value that the system allows the upper bound of the worst-case net loss to reach after being constrained by the candidate online stop-loss mechanism, under the loss dimension, loss dimension weight, and evaluation period corresponding to request r. The worst-case net loss acceptable boundary B(r) is pre-written into the corresponding policy rules by the policy manager based on one or more of the business criticality, data sensitivity, impact of access actions, security consequences, compliance requirements, and organizational risk tolerance of the protected object, and after calibration based on the risk matrix, historical security events, or expert evaluation.

[0011] Step 2: For each candidate online stop-loss mechanism bundle M, calculate or conservatively estimate the worst-case net loss upper bound U(r,M) of request r under adverse conditions, and determine whether U(r,M) exceeds the worst-case net loss acceptable boundary B(r). If U(r,M) does not exceed B(r), then the candidate online stop-loss mechanism bundle is a feasible mechanism bundle.

[0012] Step 3: When no feasible mechanism bundle exists, the requested control plane authorization semantics are determined to be DENY and no short-term authorization evidence is issued; when one or more feasible mechanism bundles exist, the mechanism bundle M* with the lowest synthesis cost is selected. If M* does not cause the request semantics to shrink, the requested control plane authorization semantics are determined to be ALLOW. If M* causes output degradation, write range reduction or other request semantics shrinkage, the requested control plane authorization semantics are determined to be DEGRADE, and a short-term authorization evidence binding the authorization semantics, B(r), M* and its mechanism parameters is issued for ALLOW or DEGRADE.

[0013] Step 4: The data plane agent receives the request carrying the short-term authorization evidence, verifies the signature, scope, validity period, and revocation status of the short-term authorization evidence; if any verification fails, DENY is output and the request is blocked from entering the protected service path; if the verification passes, the M* bound to the evidence and its mechanism parameters are loaded, and rate limiting, budget control, output degradation, write protection, reassessment, or revocation checks are implemented before and during the restricted forwarding of the request.

[0014] Step 5: If the verification and constraint execution in Step 4 both meet the requirements, and the control plane authorization semantics bound to the short-term authorization evidence are ALLOW, then the request is forwarded with restrictions according to the original request semantics and the ALLOW execution result is output; if the control plane authorization semantics bound to the short-term authorization evidence are DEGRADE, then the output is forwarded with restrictions according to the evidence-bound output degradation, write range reduction or other constraint execution and the DEGRADE execution result is output; if the online stop-loss mechanism triggers the blocking condition, or U(r,M*) exceeds B(r) after re-evaluation, then DENY is output and subsequent forwarding is stopped.

[0015] Furthermore, the request quadruple in step 1 is denoted as r=<s,o,a,x> Where s represents the access subject, o represents the protected object, a represents the access action, and x represents the dynamic context; the dynamic context includes at least one or more of the following: time window, alarm level, trust status, request frequency, remaining budget, anomaly flag, revocation flag, data sensitivity, or operating condition; the access control loss boundary policy includes at least a policy identifier, matching conditions, priority, B(r), M(r), and the parameter range and scope of each mechanism; the control plane first retrieves candidate policy rules based on action a and object o, then compares s, o, a, and x with the matching conditions of the candidate policy rules item by item, and determines the rule that satisfies all matching conditions as the hit rule; when multiple rules are hit at the same time, the effective policy rule is determined according to the rejection priority and the preset priority, and DENY is output when no rule is hit.

[0016] Furthermore, the method for obtaining the upper bound of the worst-case net loss in step 2 specifically includes:

[0017] The matching result of the effective strategy rule includes the worst-case net loss acceptable boundary B(r) and the set of finite candidate online stop loss mechanism bundles M(r) pre-configured in the rule; M(r) is a set of mechanism bundle templates that are explicitly associated with the corresponding object, action and context conditions when the strategy is issued. Each template specifies the type of mechanism, parameter range and scope of action that can be adopted, rather than generating a combination on an arbitrary subset of mechanisms at runtime.

[0018] For request r and adverse scenario w, the original loss L(r,w) is obtained based on the weighted loss of the request in the dimensions of confidentiality, integrity, availability and compliance. The loss absorption term G(r,M,w) is obtained based on the amount of loss reduction by each stop mechanism in the candidate online stop mechanism bundle M. U(r,M)=sup_w[L(r,w)-G(r,M,w)]+ is taken as the upper bound of the worst-case net loss of the request under the mechanism bundle M, where [z]+ represents max{z,0}.

[0019] Furthermore, in step 3, the method for calculating the comprehensive cost specifically includes:

[0020] The feasible mechanism bundle set is F(r) = {M belongs to M(r) and U(r,M) <= B(r)}, and the comprehensive cost is:

[0021] C(M;r)=alpha1*C_lat(M;r)+alpha2*C_util(M;r)+alpha3*C_ops(M;r)

[0022] Where C_lat represents latency cost, C_util represents business utility loss, C_ops represents operational complexity, and alpha1, alpha2, and alpha3 are non-negative weights.

[0023] Furthermore, the short-term authorization evidence in step 3 includes at least the request identifier, policy identifier, policy version, rule identifier, authorization decision, worst-case net loss acceptable boundary B(r), selected mechanism bundle M*, mechanism parameters of the selected mechanism bundle, scope of action, issuance time, expiration time, and signature field; the short-term authorization evidence also includes one or more of U(r,M*), remaining budget, obligation set, degradation mode, proxy identifier, and proxy epoch; the signature field is an integrity authentication value generated from the normalized evidence content based on a message authentication code or digital signature algorithm.

[0024] Furthermore, the online loss mitigation mechanism includes: a rate limiting mechanism based on the joint access identity of the subject, object, and action; a risk budgeting mechanism based on the cumulative consumption of authorized evidence; an output degradation mechanism that performs desensitization, aggregation, truncation, or precision reduction on the returned results; a write protection mechanism that limits writable parameters, value ranges, or state transition sets; and an undo and isolation mechanism based on the undo set, heartbeat, audit deviation, or agent health status.

[0025] Furthermore, the reassessment in step 4 specifically includes:

[0026] When a token is detected to be outdated, the risk score is increased, the anomaly flag is set, the revocation flag is set, the budget is insufficient, or the agent status is abnormal, the data plane agent triggers an online reassessment. After the reassessment, if U(r,M*) exceeds the worst-case net loss acceptable boundary B(r), then DENY is output, and subsequent requests are marked as budget overrun, rate overrun, token revocation, write protection blocking, or risk rejection.

[0027] Secondly, the present invention provides a system based on the loss boundary constraint access control method for downlink access in multi-source heterogeneous industrial IoT as described above, the system comprising:

[0028] The control plane host is configured to perform effective policy rule matching, worst-case net loss acceptable boundary acquisition, worst-case net loss upper bound estimation, mechanism bundle selection, control plane authorization semantic determination, and short-term authorization evidence issuance.

[0029] The data plane agent is configured to perform evidence verification, implement online loss mitigation mechanisms, reassess, determine the final execution result, and restrict forwarding.

[0030] Terminals or protected industrial services are configured to accept only access requests constrained by a data plane proxy;

[0031] The audit module is configured to record authorization evidence, execution results, budget consumption, revocation status, and abnormal events.

[0032] Thirdly, the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor runs the computer program stored in the memory, it performs the steps of a loss boundary constraint access control method for downlink access of a multi-source heterogeneous industrial Internet of Things as described above.

[0033] Fourthly, the present invention provides a computer-readable storage medium for storing a computer program that executes a loss boundary constraint access control method for downlink access in a multi-source heterogeneous industrial Internet of Things as described above.

[0034] The beneficial effects of this invention are:

[0035] This invention transforms the authorization problem from traditional rule satisfaction determination into a bounded feasibility determination of "whether there exists a mechanism that ensures the upper bound of the net loss of a request in the worst-case scenario does not exceed the policy boundary," and binds the feasibility result to the actual execution path through short-term authorization evidence. This allows for ensuring the availability of necessary business operations while controlling the consequences of access requests within the policy-defined loss boundaries, and continuously reducing the risk of subsequent abuse of granted access capabilities through online mitigation mechanisms such as rate limiting, budgeting, degradation, write protection, and reversal.

[0036] 1. This invention extends access control from the traditional allow / deny decision to a bounded feasibility access decision based on the upper bound of the worst-case net loss, so that high-risk downlink access has a clear consequence boundary.

[0037] 2. This invention achieves strong consistency between the authorization semantics of the control plane and the actual execution path of the data plane by binding boundaries, mechanism bundles, parameters, scope and validity period with short-term authorization evidence.

[0038] 3. This invention introduces three types of authorization semantics: ALLOW, DEGRADE, and DENY. In boundary risk scenarios, necessary business continuity can be maintained by degrading services, avoiding the lack of intermediate control states between full allow and full deny in traditional binary models.

[0039] 4. This invention continuously shrinks the replay window, long-term abuse window, and cumulative loss window after admission by means of online loss prevention mechanisms such as rate limiting, budget control, output degradation, write protection, and undo refresh.

[0040] 5. This invention is applicable to various downlink requests such as industrial telemetry reading, historical window export, and remote parameter updates, and can implement differentiated authorization for multiple source entities such as cloud, enterprise, edge, and third parties.

[0041] 6. In cases of proxy anomalies, delayed revocation, or outdated tokens, this invention employs failure shutdown, isolation, and service contraction strategies to reduce the risk of protected industrial services continuing to be exposed under uncertain conditions. Attached Figure Description

[0042] To more clearly illustrate the technical solution of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0043] Figure 1 This is a schematic diagram of the overall DBAC architecture and the decision binding path from the control plane to the data plane, provided in an embodiment of the present invention.

[0044] Figure 2This is a schematic diagram of the control plane loss boundary feasibility assessment and mechanism bundle selection process provided in an embodiment of the present invention;

[0045] Figure 3 A schematic diagram illustrating the data plane evidence verification, online loss prevention execution, and restricted forwarding process provided in this embodiment of the invention;

[0046] Figure 4 This is a schematic diagram of the undo refresh, proxy isolation, and audit closed-loop process provided in an embodiment of the present invention. Detailed Implementation

[0047] Specific Implementation Method 1: A loss boundary constraint access control method for downlink access in a multi-source heterogeneous industrial IoT system, as described in this implementation method, includes:

[0048] The method is applicable to systems consisting of a request source, a control plane host, a data plane proxy, terminal devices, or protected industrial services, and the method includes the following steps:

[0049] S1, Request Modeling and Policy Matching: Abstract downlink access requests from cloud platforms, enterprise systems, edge gateways, human-machine interfaces, or third-party services into request quadruplets containing subject, object, action, and context; match effective policy rules from a pre-configured set of access control loss boundary policies to obtain the worst-case net loss acceptable boundary B(r) corresponding to the request, and a pre-configured set of finite candidate online stop-loss mechanism bundles M(r) from the effective policy rules; the online stop-loss mechanism bundle is a finite combination of online stop-loss mechanisms, including at least one or more of rate limiting, budget control, output degradation, write protection, and undo refresh.

[0050] Specifically, the requested quadruple is denoted as r=<s,o,a,x> Where s represents the access subject, o represents the protected object, a represents the access action, and x represents the dynamic context; the dynamic context includes at least one or more of the following: time window, alarm level, trust status, request frequency, remaining budget, anomaly flag, revocation flag, data sensitivity, or operating condition.

[0051] Specifically, the online loss mitigation mechanism includes: a rate limiting mechanism based on the joint access identity of the subject, object, and action; a risk budgeting mechanism based on the cumulative consumption of authorized evidence; an output degradation mechanism that performs desensitization, aggregation, truncation, or precision reduction on the returned results; a write protection mechanism that limits writable parameters, value ranges, or state transition sets; and a revocation and isolation mechanism based on the revocation set, heartbeat, audit deviation, or agent health status.

[0052] Specifically, the access control loss boundary policy is pre-configured in the form of policy records, including at least a policy identifier, subject, object, action and context matching conditions, priority, worst-case net loss acceptable boundary B(r), a finite set of candidate online stop-loss mechanism bundles M(r), and the parameter range and scope of each mechanism. The control plane first retrieves candidate policy rules based on the action and object, then compares the request quadruple with the matching conditions of each candidate policy rule item by item, and determines the rule that satisfies all matching conditions as the hit rule; when multiple rules are hit simultaneously, the effective policy rule is determined according to rejection priority and preset priority, and DENY is output when no rule is hit. B(r) is the maximum value that the upper bound of the worst-case net loss can reach under the corresponding loss dimension, weight, and evaluation period, after the system allows the mechanism bundle constraints. It is pre-written into the policy as a numerical threshold after being calibrated by the policy manager based on the business criticality, data sensitivity, impact of access actions, security consequences, compliance requirements, and organizational risk tolerance of the protected object, and based on the risk matrix, historical security events, or expert evaluation. M(r) is a finite set of mechanism bundle templates that are explicitly associated with effective policy rules when the policy is published, rather than a combination of arbitrary mechanism subsets at runtime.

[0053] S2, Feasibility assessment of loss boundary: For each candidate online stop loss mechanism bundle M, calculate or conservatively estimate the worst-case net loss upper bound U(r,M) of request r under adverse conditions, and determine whether there is a feasible mechanism bundle that makes U(r,M) not exceed the worst-case net loss acceptable boundary B(r).

[0054] Specifically, the worst-case net loss upper bound is calculated as follows: the original loss L(r,w) is obtained based on the weighted loss of the request in the dimensions of confidentiality, integrity, availability and compliance; the loss absorption term G(r,M,w) is obtained based on the amount of loss reduction by each stop-loss mechanism in the candidate mechanism bundle M; and U(r,M)=sup_w[L(r,w)-G(r,M,w)]+ is taken as the worst-case net loss upper bound of the request under the mechanism bundle M, where [z]+ represents max{z,0}.

[0055] S3, Mechanism Bundle Selection, Control Plane Authorization Semantics Determination, and Evidence Issuance: When no feasible mechanism bundle exists, the requested control plane authorization semantics are determined to be DENY and no short-term authorization evidence is issued; when one or more feasible mechanism bundles exist, the mechanism bundle M* with the lowest overall cost is selected based on latency cost, business utility loss, and operational complexity. If M* does not cause request semantic contraction, the requested control plane authorization semantics are determined to be ALLOW. If M* causes output degradation, write range reduction, or other request semantic contraction, the requested control plane authorization semantics are determined to be DEGRADE, and short-term authorization evidence is issued for ALLOW or DEGRADE, binding request identifier, policy identifier, authorization semantics, B(r), U(r,M*), M*, mechanism parameters, scope of application, issuance time, expiration time, and signature.

[0056] Specifically, the short-term authorization evidence includes at least a request identifier, a policy identifier, a policy version, a rule identifier, an authorization decision, a worst-case net loss acceptable boundary B(r), a selected mechanism bundle M*, the mechanism parameters of the selected mechanism bundle, its scope of application, issuance time, expiration time, and a signature field; the short-term authorization evidence may also include one or more of U(r,M*), remaining budget, obligation set, degradation mode, proxy identifier, and proxy epoch; the signature field is an integrity authentication value generated from the normalized evidence content based on a message authentication code or a digital signature algorithm.

[0057] Specifically, the mechanism bundle selection step finds the mechanism bundle with the minimum comprehensive cost function C(M;r) = alpha1*C_lat(M;r) + alpha2*C_util(M;r) + alpha3*C_ops(M;r) from the feasible set F(r) = {M belongs to M(r) and U(r,M) <= B(r)}, where C_lat represents the delay cost, C_util represents the business utility loss, C_ops represents the operational complexity, and alpha1, alpha2, and alpha3 are non-negative weights.

[0058] S4, Data Plane Evidence Verification and Constraint Enforcement: The data plane agent receives a request carrying the short-term authorization evidence, verifies the signature, scope, validity period, and revocation status of the evidence; if any verification fails, DENY is output and the request is blocked from entering the protected service path; if the verification passes, the M* bound to the evidence and its mechanism parameters are loaded, and rate limiting, budget control, output degradation, write protection, reassessment, or revocation checks are implemented before the request is restricted and during the execution process.

[0059] Specifically, the data plane agent processes requests in the following order: parsing evidence, verifying signatures, checking expiration time, querying revocation status, checking scope, determining whether a reassessment is triggered, and executing local obligations. When any verification condition is not met, the data plane agent adopts a failure shutdown strategy and outputs DENY. When token staleness, increased risk score, anomaly flag set, revocation flag set, insufficient budget, or abnormal agent status is detected, the data plane agent triggers online reassessment. After reassessment, if U(r,M*) exceeds the worst-case net loss acceptable boundary B(r), DENY is output, and subsequent requests are marked as budget overrun, rate overrun, token revocation, write protection blocking, or risk rejection.

[0060] S5, Restricted Forwarding and Final Execution Result Output: If the verification and constraint execution in S4 both meet the requirements, and the control plane authorization semantics bound to the short-term authorization evidence are ALLOW, then restricted forwarding is performed according to the original request semantics and the ALLOW execution result is output; if the control plane authorization semantics bound to the short-term authorization evidence are DEGRADE, then restricted forwarding is performed according to the evidence-bound output degradation, write range reduction, or other constraint execution and the DEGRADE execution result is output; if the online stop-loss mechanism triggers the blocking condition, or U(r,M*) exceeds B(r) after re-evaluation, then DENY is output and subsequent forwarding is stopped.

[0061] Implementation Method 2: A loss boundary constraint access control method for downlink access in a multi-source heterogeneous industrial IoT according to this implementation method includes:

[0062] The method is applicable to industrial internet systems consisting of control plane hosts, data plane proxies, terminal devices or protected industrial services, and multi-source requesters, including cloud analytics services, enterprise applications, edge gateways, human-machine interfaces, and third-party business integration services.

[0063] Step S1, Request Modeling and Policy Matching. The system normalizes the downlink access request into r=<s,o,a,x> Here, s represents the subject, o represents the access object, a represents the action, and x represents the context. The control plane first retrieves candidate policy rules based on a and o, then compares s, o, a, and x with the matching conditions of the candidate rules item by item; the rule that satisfies all conditions is determined as the hit rule. When multiple rules hit simultaneously, the effective policy rule is determined according to rejection priority and preset priority. The effective policy rule p(r) outputs the worst-case net loss acceptable boundary B(r) corresponding to the request, as well as the pre-configured finite candidate mechanism bundle set M(r) in the rule; if no rule is hit, the request is determined as DENY.

[0064] Step S2: Worst-case net loss upper bound assessment. The control plane calculates or conservatively estimates U(r,M) for each candidate mechanism bundle M based on context snapshots, historical requests, remaining budget, alarm status, data sensitivity, and operating conditions. If there exists an M such that U(r,M) <= B(r), the request is considered loss boundary feasible; otherwise, the control plane authorization semantics of the request are determined to be DENY, and short-term authorization evidence is not issued.

[0065] Step S3: Mechanism Bundle Selection and Control Plane Authorization Semantics Determination. For all feasible mechanism bundles, the control plane selects the optimal mechanism bundle M* according to the comprehensive cost function. The comprehensive cost can consist of latency cost, business utility loss, and operational complexity. If M* does not cause externally visible semantic contraction, the requested control plane authorization semantic is determined as ALLOW; if M* requires output degradation, write range reduction, local latency, or other semantic contraction, the requested control plane authorization semantic is determined as DEGRADE. Thus, the subject of ALLOW or DEGRADE is the request, and M* is the execution constraint scheme that makes the request satisfy B(r) and determines the authorization semantics accordingly.

[0066] Step S4, short-term authorization evidence issuance. The control plane constructs authorization evidence pi only for requests with the control plane authorization semantics of ALLOW or DEGRADE. Fields such as request identifier, policy identifier, worst-case net loss acceptable boundary B(r), worst-case net loss upper bound U(r,M*), selected mechanism bundle M*, mechanism parameters, scope of action, issuance time, expiration time, authorization decision, and signature are written into the evidence. This evidence has a short validity period, which can reduce the time window for capability replay or long-term abuse.

[0067] Step S5: Evidence verification, online loss prevention execution, and final execution result output. The data plane agent parses the evidence and verifies the signature, scope, validity period, and revocation status; if any verification fails, DENY is output and the request is blocked. After successful verification, the agent, according to the M* bound to the evidence and its mechanism parameters, implements rate limiting, budgeting, degradation, write protection, revocation refresh, reassessment, and audit logging before and during the request enters the PLC, terminal device, or protected service path; if the blocking condition is not triggered, restricted forwarding is executed according to the control plane authorization semantics bound to the evidence, and ALLOW or DEGRADE execution result is output; if the blocking condition is triggered or U(r,M*) exceeds B(r) after reassessment, DENY is output.

[0068] Step S6: Revocation, Isolation, and Auditing. The control plane or security operations team can add specific evidence, agent identifiers, or agent epochs to the revocation set; the data plane periodically refreshes the revocation status and rejects requests after evidence is revoked. If the agent's health status is abnormal, the system stops issuing new evidence to that agent path, revokes related evidence, and switches high-risk export or write operations to rejection or manual mode.

[0069] The set of loss dimensions can include confidentiality, integrity, availability, and compliance. For request r and adverse case w, the original loss can be expressed as L(r,w)=sum(lambda_q*l_q(r,w)), where lambda_q is the weight of the loss dimension and l_q is the loss function of the corresponding dimension.

[0070] The mechanism bundle M is a finite combination of online stop-loss mechanisms, including at least one or more of rate limiting, budget control, output degradation, write protection, and undo refresh. Each mechanism m has a parameter theta_m and can absorb or reduce part of the loss g_m(r,theta_m,w). Therefore, the loss absorption term of the mechanism bundle can be expressed as G(r,M,w)=sum(g_m(r,theta_m,w)).

[0071] The worst-case net loss upper bound can be expressed as U(r,M)=sup_w [L(r,w)-G(r,M,w)]+. Here, B(r) refers to the acceptable boundary of the worst-case net loss output by the effective policy rule. When the feasible set F(r)={M ∈ M(r) and U(r,M)<=B(r)} is empty, the control plane will determine the request as DENY and will not issue evidence; when F(r) is not empty, the minimum cost mechanism bundle is selected as M*.

[0072] The authorized evidence pi is used to ensure consistency between control plane decisions and data plane execution. The data plane agent only allows a request to enter the actual execution path when Valid(pi,r,t)=1, where Valid(pi,r,t) requires that the signature verification is successful, the current time is between the evidence issuance time and the expiration time, the request matches the scope of the evidence, and the evidence has not been revoked.

[0073] The data plane execution trajectory must satisfy the evidence-induced acceptable execution set Gamma(pi), meaning the actual execution must not exceed the scope of the object, time, output semantics, write domain, rate limit, and budget window declared by the evidence. Therefore, the loss boundary obtained from the control plane is no longer confined to the policy semantic layer but is forcibly implemented in the online execution path.

[0074] Example 1: System Composition and Request Abstraction

[0075] like Figure 1As shown, in this embodiment, the system includes a Host control plane, a Proxy data plane, a Terminal, a PLC simulator, or a protected industrial service. The request source can be a cloud analytics service, enterprise application, edge gateway, human-machine interface, or third-party business system. The Host is responsible for policy matching, context snapshots, historical request and budget status collection, worst-case net loss upper bound estimation, mechanism selection, control plane authorization semantic determination, and authorization evidence issuance; the Proxy is responsible for evidence verification, revocation queries, online obligation execution, final execution result determination, restricted forwarding, and auditing.

[0076] Downlink access requests are uniformly abstracted as r=<s,o,a,x> Where s represents the request subject, o represents the access object, such as the telemetry interface, the history window export interface, or the device configuration interface; a represents the action, such as ReadTelemetry, ExportWindow, or WriteSamplingProfile; and x represents the context, such as risk score, sensitivity, export window length, remaining budget, anomaly flag, cancellation flag, and alarm status.

[0077] The above abstraction enables requests from different domains, protocols, and business interfaces to enter a unified authorization process. The effective policy rule p(r) outputs two core results: first, the worst-case net loss acceptable boundary B(r), and second, the pre-configured set of finite candidate mechanism bundles M(r) in the rule. For example, for telemetry reads, a configurable B(r) that is easily satisfied but still subject to rate and budget constraints is available; for historical window derivation, a configurable B(r) that requires budget and degradation constraints to satisfy is available; and for remote writes, a strict B(r) corresponding to write protection obligations is available.

[0078] Example 2: Feasibility Assessment of Loss Boundary

[0079] like Figure 2 As shown, the control plane evaluates each candidate mechanism bundle to determine whether it can compress the worst-case net loss upper bound U(r,M) of the request to within the worst-case net loss acceptable boundary B(r). The original loss can be obtained by weighting multiple dimensions such as confidentiality, integrity, availability, and compliance. For data export requests, the weights of confidentiality and compliance can be higher; for remote write requests, the weights of integrity and availability can be higher.

[0080] In a reproducible example, during the policy release phase, the policy manager, based on the business criticality, data sensitivity, impact of access actions, security consequences, compliance requirements, and organizational risk tolerance of the protected object, and using a risk matrix, historical security events, or expert assessments, writes B(r) as a numerical threshold consistent with the loss assessment scale into the policy. During the operational phase, the system can use an external risk assessor or a configurable risk matrix to generate a conservative U(r,M) estimate based on sensitivity, risk score, export window length, insufficient remaining budget flag, anomaly flag, and revocation flag. This risk assessor can be calibrated using historical event data, expert knowledge graphs, or an enterprise risk matrix; this invention does not limit its specific implementation, only requiring that its output be a conservative upper bound that can be selected and used by the mechanism.

[0081] If there exists a candidate mechanism bundle M for a request such that U(r,M)<=B(r), then the request is boundedly feasible; if not, it means that no candidate online stop-loss mechanism bundle pre-configured in the effective policy rules can compress the consequences of the request to within the worst-case net loss acceptable boundary B(r), and the system should determine the request as DENY and not issue evidence before execution.

[0082] Example 3: Mechanism Bundle Selection and Authorization Evidence Issuance

[0083] When multiple feasible mechanism bundles exist, the control plane does not arbitrarily choose one, but rather selects the bundle with the lowest overall cost from the feasible set. The overall cost can simultaneously consider execution latency, loss of business utility, and operational complexity. For example, for low-risk telemetry reads, rate limiting is sufficient to meet the boundary conditions; for historical data export with boundary risks, budget control and output degradation may be required simultaneously; for remote writes, write protection, rollback, or direct rejection may be necessary.

[0084] The control plane generates short-term authorization evidence based on the selected mechanism bundle. The evidence content includes at least the fields request_id, policy_id, policy_version, rule_id, decision, acceptable_bound, mechanism_bundle, mechanism_params, scope, issued_at, exp, and signature, and may include upper_bound, obligations, budget_remaining, degrade_mode, target proxy identifier, and proxy epoch. Acceptable_bound corresponds to B(r), upper_bound corresponds to U(r, M*), and mechanism_bundle and mechanism_params correspond to M* and its mechanism parameters, respectively. After normalization and serialization, the evidence is signed by the control plane or a message authentication code is generated.

[0085] By using short-term authorization evidence, the system avoids a full control plane re-adjudication for every request, while limiting the window of opportunity for abuse to within the evidence's validity period. The scope and obligations fields in the evidence allow the data plane to directly access executable constraints, preventing control plane decisions from degenerating into mere log entries for reference only.

[0086] Example 4: Data Plane Validation and Online Stop-Loss Execution

[0087] like Figure 3 As shown, after receiving a request carrying authorization evidence, the data plane agent first parses the evidence and verifies the signature or message authentication code; then it checks whether the evidence has expired, whether it is in the revocation set, and whether the request falls within the scope of the evidence. Any verification failure results in a failure shutdown and the output of DENY, meaning the request will not be forwarded to the terminal device, PLC, or protected service.

[0088] After successful verification, the agent executes the online loss mitigation mechanism according to the evidentiary obligation. The rate limiting mechanism counts the number of requests for the same subject-object-action combination within a given window; requests exceeding the upper limit are rejected. The budget mechanism accumulates the risk consumption of the same evidence or within the same policy window; requests exceeding the budget are rejected. The output degradation mechanism aggregates, truncates, de-identifies, or reduces the precision of responses. The write protection mechanism limits the allowed set of parameters, value ranges, and state transition sets. The revocation mechanism immediately blocks requests after evidence is revoked.

[0089] The agent can also trigger a reassessment when the token age exceeds a threshold, the risk score increases, the anomaly flag is set, the revocation flag is set, or the budget is exhausted. After the reassessment, if there is still a candidate mechanism bundle that makes U(r,M)<=B(r), the original ALLOW request can be converted to DEGRADE and an updated evidence can be issued or exchanged; if not, DENY is output, thereby continuously narrowing the risk exposure window after admission.

[0090] Example 5: Revocation, Isolation, and Auditing

[0091] like Figure 4 As shown, when the control plane detects credential leakage, missing agent heartbeats, audit deviations, abnormal marking by operations personnel, or escalation of business risks, it can add authorized evidence to the revocation set, or add the target agent identifier and agent epoch to the isolation scope. Afterward, the host stops issuing new evidence with that agent as the endpoint and revokes any related evidence that has not yet expired.

[0092] For data plane nodes suspected of being abnormal, the system isolates them from the normal service path; rejects high-risk operations such as export and write operations; and for extremely low-risk read-only telemetry, the visible policy requires limited service to be provided through alternative trusted paths or manual local mode. This mechanism cannot undo actions that have already occurred before detection, but it can transform the residual exposure window after detection into a controllable service contraction and isolation process.

[0093] The audit module records evidence issuance, proxy verification, mechanism triggering, budget consumption, degradation output, write protection interception, revocation hit, and final execution results, providing a data foundation for post-event tracking, strategy optimization, and risk assessor calibration.

[0094] Example 6: Typical Industrial Scenario

[0095] In telemetry burst read scenarios, edge-side requesters continuously read device telemetry data within a short period. Traditional binary authorization models may continue to allow a large number of requests after initial permission. This invention uses an evidence-based rate limiting mechanism to mark subsequent requests as RateLimited and reject them after exceeding the upper frequency limit, thereby shortening the burst abuse window.

[0096] In scenarios involving repeated exports from historical windows, enterprise applications or third-party services repeatedly request historical window data after gaining export capabilities. This invention records the cost of each export through budget control. When the cumulative cost exceeds the evidence or policy window budget, subsequent requests are marked as BudgetExceeded and rejected. For export requests with boundary risks but still serviceable, the system can perform degraded exports, such as shortening the window, reducing field precision, or outputting aggregated results.

[0097] In remote sampling configuration write scenarios, requests may affect the state of industrial processes. The system checks the target parameters, value ranges, and state transition sets based on write protection constraints; if the write would cause the system state transitions to exceed the evidence-allowed set, it is rejected or redirected to a rollback path.

[0098] In a maintenance token revocation scenario, after the host revokes a maintenance token, the proxy invalidates the evidence by refreshing the revocation set or querying the revocation status during the next verification. The shorter the token's validity period and the smaller the revocation refresh interval, the fewer the number of residual successful requests after revocation.

[0099] Experimental results and feasibility description:

[0100] In the four-node prototype, the Host can be deployed on a general-purpose server or virtual machine, the Proxy can be deployed as a containerized proxy, the Terminal can be deployed on a Raspberry Pi terminal, and the PLC simulator can be deployed on a standalone Raspberry Pi or a protected server. The system carries standardized requests via HTTP or industrial gateway protocols and records authorization and execution results with structured logs.

[0101] In workloads such as telemetry reads, historical window exports, and remote writes, this method can maintain differentiated authorization results across source domains and actions: low-risk telemetry reads are usually ALLOW, historical window exports are ALLOW, DEGRADE, or DENY depending on the source and risk context, and remote writes are strictly constrained by write protection and risk boundaries.

[0102] Compared to representative models such as RBAC, TBAC, ABAC, and UCON, the key difference in this method lies not in simply increasing the expressive power of rules, but in using a clearly defined worst-case consequence boundary as the admission condition, and continuously binding this boundary to the execution path through short-term evidence and an online stop-loss mechanism. Experimental results show that in scenarios involving sudden reads and repeated exports, this method can significantly reduce the number of requests that successfully execute after being triggered, and provides limited degradation services in boundary risk scenarios.

[0103] The above embodiments only illustrate several implementations of the present invention and do not limit the scope of protection of the present invention. Any technical solution that, under the concept of the present invention, transforms access permission into a worst-case net loss boundary feasibility determination, binds the policy boundary, mechanism bundle, parameters, scope, and validity period to the data plane execution path through short-term evidence, and simultaneously utilizes one or more mechanisms among rate limiting, budgeting, degradation, write protection, or revocation for online loss mitigation, should fall within the scope of protection of the present invention.

[0104] This invention discloses a downside access control method and system suitable for multi-source heterogeneous industrial IoT environments, referred to as DBAC (Downside-Bounded Access Control). The method includes: a request modeling and effective policy rule matching stage, a loss boundary feasibility assessment stage, mechanism bundle selection, control plane authorization semantic determination and short-term authorization evidence issuance stage, data plane evidence verification, online loss mitigation execution and final execution result output stage, and a revocation and auditing stage. This method extends the allow / deny decision in traditional access control to a bounded feasibility access decision based on the worst-case net loss upper bound. When the candidate mechanism bundle M* pre-configured in the effective policy rules ensures that U(r,M*) does not exceed the worst-case net loss acceptable boundary B(r), the control plane determines the authorization semantic of the request as ALLOW or DEGRADE and issues short-term authorization evidence binding B(r), U(r,M*), M*, mechanism parameters, scope of action, and validity period. When no feasible mechanism bundle exists, the request is determined as DENY and no evidence is issued. The data plane proxy verifies evidence and continuously executes online mitigation mechanisms such as rate limiting, budget control, output degradation, write protection, and undo / refresh, outputting a final execution result consistent with the evidence constraints. Through three authorization semantics—ALLOW, DEGRADE, and DENY—this invention can limit the risks of replay, long-term abuse, and scope expansion of permitted capabilities while ensuring business continuity. Experiments show that this invention maintains differentiated authorization results in workloads such as industrial telemetry reading, historical window export, and remote writing, and significantly reduces the number of successful requests after triggering. It is suitable for high-risk downlink access governance in industrial internet, cloud-edge-device collaboration, and cross-domain business systems.

Claims

1. A loss boundary constraint access control method for downlink access in multi-source heterogeneous industrial IoT, characterized in that, The method includes: Step 1: Define downlink access requests from multi-source heterogeneous industrial IoT as request quadruplets containing subject, object, action and context; match effective policy rules from the pre-configured access control loss boundary policy set to obtain the worst-case net loss acceptable boundary B(r) corresponding to the request, and the pre-configured finite candidate online stop loss mechanism bundle set M(r) in the effective policy rules. The candidate online stop mechanism bundle is a finite combination of online stop mechanisms. Each candidate online stop mechanism bundle includes a mechanism type, mechanism parameters or parameter range, scope of application, and an identifier indicating whether it causes request semantic contraction. The worst-case net loss acceptable boundary B(r) refers to the maximum value that the system allows the upper bound of the worst-case net loss to reach after being constrained by the candidate online stop-loss mechanism, under the loss dimension, loss dimension weight, and evaluation period corresponding to request r. The worst-case net loss acceptable boundary B(r) is pre-written into the corresponding policy rules by the policy manager based on one or more of the business criticality, data sensitivity, impact of access actions, security consequences, compliance requirements, and organizational risk tolerance of the protected object, and after calibration based on the risk matrix, historical security events, or expert evaluation. Step 2: For each candidate online stop-loss mechanism bundle M, calculate or conservatively estimate the worst-case net loss upper bound U(r,M) of request r under adverse conditions, and determine whether U(r,M) exceeds the worst-case net loss acceptable boundary B(r). If U(r,M) does not exceed B(r), then the candidate online stop-loss mechanism bundle is a feasible mechanism bundle. Step 3: When no feasible mechanism bundle exists, the requested control plane authorization semantics are determined to be DENY and no short-term authorization evidence is issued; when one or more feasible mechanism bundles exist, the mechanism bundle M* with the lowest synthesis cost is selected. If M* does not cause the request semantics to shrink, the requested control plane authorization semantics are determined to be ALLOW. If M* causes output degradation, write range reduction or other request semantics shrinkage, the requested control plane authorization semantics are determined to be DEGRADE, and a short-term authorization evidence binding the authorization semantics, B(r), M* and its mechanism parameters is issued for ALLOW or DEGRADE. Step 4: The data plane agent receives the request carrying the short-term authorization evidence, verifies the signature, scope, validity period, and revocation status of the short-term authorization evidence; if any verification fails, DENY is output and the request is blocked from entering the protected service path; if the verification passes, the M* bound to the evidence and its mechanism parameters are loaded, and rate limiting, budget control, output degradation, write protection, reassessment, or revocation checks are implemented before and during the restricted forwarding of the request. Step 5: If the verification and constraint execution in Step 4 both meet the requirements, and the control plane authorization semantics bound to the short-term authorization evidence are ALLOW, then the request is forwarded with restrictions according to the original request semantics and the ALLOW execution result is output; if the control plane authorization semantics bound to the short-term authorization evidence are DEGRADE, then the output is forwarded with restrictions according to the evidence-bound output degradation, write range reduction or other constraint execution and the DEGRADE execution result is output; if the online stop-loss mechanism triggers the blocking condition, or U(r,M*) exceeds B(r) after re-evaluation, then DENY is output and subsequent forwarding is stopped.

2. The loss boundary constraint access control method for downlink access in a multi-source heterogeneous industrial IoT according to claim 1, characterized in that: The request quadruple in step 1 is denoted as r=<s,o,a,x> Where s represents the access subject, o represents the protected object, a represents the access action, and x represents the dynamic context; the dynamic context includes at least one or more of the following: time window, alarm level, trust status, request frequency, remaining budget, anomaly flag, revocation flag, data sensitivity, or operating condition; the access control loss boundary policy includes at least a policy identifier, matching conditions, priority, B(r), M(r), and the parameter range and scope of each mechanism; the control plane first retrieves candidate policy rules based on action a and object o, then compares s, o, a, and x with the matching conditions of the candidate policy rules item by item, and determines the rule that satisfies all matching conditions as the hit rule; when multiple rules are hit at the same time, the effective policy rule is determined according to the rejection priority and the preset priority, and DENY is output when no rule is hit.

3. The loss boundary constraint access control method for downlink access in a multi-source heterogeneous industrial IoT according to claim 1, characterized in that: The method for obtaining the upper bound of the worst-case net loss in step 2 specifically includes: The matching result of the effective strategy rule includes the worst-case net loss acceptable boundary B(r) and the set of finite candidate online stop loss mechanism bundles M(r) pre-configured in the rule; M(r) is a set of mechanism bundle templates that are explicitly associated with the corresponding object, action and context conditions when the strategy is issued. Each template specifies the type of mechanism, parameter range and scope of action that can be adopted, rather than generating a combination on an arbitrary subset of mechanisms at runtime. For request r and adverse scenario w, the original loss L(r,w) is obtained based on the weighted loss of the request in the dimensions of confidentiality, integrity, availability and compliance. The loss absorption term G(r,M,w) is obtained based on the amount of loss reduction by each stop mechanism in the candidate online stop mechanism bundle M. U(r,M)=sup_w[L(r,w)-G(r,M,w)]+ is taken as the upper bound of the worst-case net loss of the request under the mechanism bundle M, where [z]+ represents max{z,0}.

4. The loss boundary constraint access control method for downlink access in a multi-source heterogeneous industrial IoT according to claim 1, characterized in that: Step 3, the method for calculating the overall cost, specifically includes: The feasible mechanism bundle set is F(r) = {M belongs to M(r) and U(r,M) <= B(r)}, and the comprehensive cost is: C(M;r)=alpha1*C_lat(M;r)+alpha2*C_util(M;r)+alpha3*C_ops(M;r) Where C_lat represents latency cost, C_util represents business utility loss, C_ops represents operational complexity, and alpha1, alpha2, and alpha3 are non-negative weights.

5. The loss boundary constraint access control method for downlink access in a multi-source heterogeneous industrial IoT according to claim 1, characterized in that: The short-term authorization evidence in step 3 includes at least the request identifier, policy identifier, policy version, rule identifier, authorization decision, worst-case net loss acceptable boundary B(r), selected mechanism bundle M*, mechanism parameters of the selected mechanism bundle, scope of action, issuance time, expiration time, and signature field; the short-term authorization evidence also includes one or more of U(r,M*), remaining budget, obligation set, degradation mode, proxy identifier, and proxy epoch; the signature field is an integrity authentication value generated from the normalized evidence content based on a message authentication code or digital signature algorithm.

6. The loss boundary constraint access control method for downlink access in a multi-source heterogeneous industrial IoT according to claim 1, characterized in that: The online loss mitigation mechanism includes: a rate limiting mechanism based on the joint access identity of the subject, object, and action; a risk budgeting mechanism based on the cumulative consumption of authorized evidence; an output degradation mechanism that performs desensitization, aggregation, truncation, or precision reduction on the returned results; a write protection mechanism that limits writable parameters, value ranges, or state transition sets; and a revocation and isolation mechanism that is executed based on the revocation set, heartbeat, audit deviation, or agent health status.

7. The loss boundary constraint access control method for downlink access in a multi-source heterogeneous industrial IoT according to claim 1, characterized in that: Step 4, the reassessment, specifically includes: When a token is detected to be outdated, the risk score is increased, the anomaly flag is set, the revocation flag is set, the budget is insufficient, or the agent status is abnormal, the data plane agent triggers an online reassessment. After the reassessment, if U(r,M*) exceeds the worst-case net loss acceptable boundary B(r), then DENY is output, and subsequent requests are marked as budget overrun, rate overrun, token revocation, write protection blocking, or risk rejection.

8. A system based on any one of claims 1-7 for loss boundary constraint access control method for downlink access in multi-source heterogeneous industrial IoT, characterized in that: The system includes: The control plane host is configured to perform effective policy rule matching, worst-case net loss acceptable boundary acquisition, worst-case net loss upper bound estimation, mechanism bundle selection, control plane authorization semantic determination, and short-term authorization evidence issuance. The data plane agent is configured to perform evidence verification, implement online loss mitigation mechanisms, reassess, determine the final execution result, and restrict forwarding. Terminals or protected industrial services are configured to accept only access requests constrained by a data plane proxy. The audit module is configured to record authorization evidence, execution results, budget consumption, revocation status, and abnormal events.

9. A computer device, characterized in that: The device includes a memory and a processor, wherein the memory stores a computer program, and when the processor runs the computer program stored in the memory, it performs the steps of the loss boundary constraint access control method for downlink access of a multi-source heterogeneous industrial Internet of Things as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that: The storage medium is used to store a computer program that executes a loss boundary constraint access control method for downlink access in a multi-source heterogeneous industrial Internet of Things according to any one of claims 1-7.