Intelligent risk monitoring and management and control system and method

Abstract

The invention relates to an intelligent risk monitoring and management and control method, and belongs to the technical field of intelligent risk monitoring and management and control. At present, problems such as operation jitter and safety risk are easy to occur. The method comprises the steps of performing cleaning, alignment and quality verification on operation data, equipment data and environment information under a unified time reference to form three types of read-only inputs of an equipment portrait, an environment portrait and a primary topology snapshot; performing abnormal preliminary screening based on internal and external bistable baselines, performing open set identification and rejection on samples crossing a robust band, performing modal consistency check on multi-modal evidence, outputting interval confidence, and generating a structured candidate event record; the method comprises the following steps: performing hierarchical disposal judgment by combining node topology criticality and disposal cooperative sensitivity, performing radial threshold and angle gating mapping on the node topology criticality and the disposal cooperative sensitivity in a polar coordinate domain to obtain a disposal level, and inhibiting jitter of a critical interval through asymmetric hysteresis and minimum residence time. And the accuracy of exception identification, the robustness of decision processing and the security of system operation are effectively improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent risk monitoring and control technology, and more specifically, to an intelligent risk monitoring and control system and method. Background Technology

[0002] In wide-area distributed scenarios such as power and industrial processes, operational monitoring data comes from diverse sources, has varying time granularity, and inconsistent quality. Furthermore, equipment operating status is affected by multiple factors coupled together, including primary topology, protection settings, and environmental stresses. Existing risk monitoring and control solutions mostly rely on fixed thresholds or local models, lacking a unified time benchmark and consistent processing. This leads to difficulties in aligning data across devices and sites, and the inability to reuse out-of-limit judgment criteria across different periods, resulting in high auditing and traceability costs.

[0003] In anomaly identification, common methods rely on historical statistics of single devices or a small rule base, making it difficult to simultaneously consider the steady-state baseline of local devices and the group reference baseline of other similar devices. They also lack open-set identification and rejection mechanisms for "new patterns outside of known classes." Conflicts and weak consistency among multimodal evidence are difficult to quantify as uncertainty, leading to both false alarms and false negatives. Regarding operational handling, existing systems typically trigger simple actions based on device-level thresholds, failing to incorporate protection coordination selectivity, short-circuit current pressure, and systemic sensitivity to power flow redistribution into the same decision-making process. Such actions may introduce secondary risks (such as protection settings being under pressure or local line load rates being increased).

[0004] In terms of handling pace control, the lack of jitter suppression mechanisms such as asymmetric hysteresis and minimum dwell time makes it prone to frequent escalation and de-escalation within the threshold neighborhood. When clock anomalies, unverified topology, or substandard input quality occur, there is a lack of auditable conservative degradation channels and fallback paths. Looking towards long-term evolution, existing solutions often lack evidence-based records and knowledge accumulation, making it difficult to transform uncertain or boundary samples into verifiable improvements. They also lack shadow channels and canary release mechanisms to ensure online security. These issues collectively lead to unstable risk identification, unexplainable handling, and difficulty in tracing the entire chain, failing to meet the comprehensive requirements for security, usability, and compliance. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art, embodiments of the present invention provide a risk intelligent monitoring and control system and method to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A risk intelligent monitoring and control method includes the following steps: The system cleans, aligns, and verifies the quality of operational data, equipment information, and environmental information under a unified clock, forming three types of read-only inputs: equipment profile, environmental profile, and a topological snapshot. It performs initial screening of anomalies based on internal and external bistable baselines, performs open set identification and rejection for samples that cross robust bands, verifies the modality consistency of multimodal evidence, outputs interval confidence, and generates structured candidate event records. A graded treatment determination is made by combining the node topological criticality and the treatment coordination sensitivity, where the treatment coordination sensitivity is obtained by fusing the protection coordination disturbance index and the power flow redistribution sensitivity. The node topological criticality and treatment coordination sensitivity are mapped radially by threshold and angle by gating in the polar coordinate domain to obtain the treatment levels of L0-L3, and the jitter in the critical interval is suppressed by asymmetric hysteresis and minimum dwell time.

[0007] By cleaning, aligning, and quality-verifying multi-source heterogeneous data under a unified clock, standardized device profiles, environmental profiles, and topology snapshots are generated. This provides high-quality, traceable, read-only input for subsequent analysis, fundamentally reducing misjudgments caused by data inconsistency or poor quality. Initial screening using internal and external bistable baselines, combined with open-set identification and rejection mechanisms, not only detects known abnormal patterns but also effectively identifies unknown novel faults, reducing the risk of missed and false alarms. Consistency verification of multimodal evidence and outputting interval-based confidence levels quantifies the uncertainty of judgments, making conclusions more reliable and interpretable. Combining node topological criticality (structural risk) with response coordination sensitivity (systemic secondary risk) and performing comprehensive mapping in polar coordinates makes the determination of response levels more refined and comprehensive. Employing asymmetric hysteresis and minimum dwell time mechanisms effectively suppresses frequent level switching near threshold critical points, avoiding oscillating operation of actuators and improving system stability.

[0008] In a preferred embodiment, the handling coordination sensitivity is obtained by fusing the protection coordination disturbance index and the power flow redistribution sensitivity. After obtaining the protection coordination disturbance index and the power flow redistribution sensitivity, a monotonically continuous and saturated smooth compression mapping is applied to each candidate handling action to obtain the action-level sensitivity. Robust quantile aggregation of the action-level sensitivity of all candidate actions at the same time is performed, combined with saturation pruning and interval reduction, to obtain the handling coordination sensitivity. The slope parameter of the mapping is positive, and the equilibrium point parameter is used to determine the neutral equilibrium position.

[0009] In a preferred embodiment, under the assumption that the handling action is effective, the selective time interval margin of the key protection pair is calculated along the protection coordination link; the selective time interval benchmark value is used to constrain the minimum action time difference between the upper and lower level protections, and is solidified and versioned according to the protection setting principle and setting table; under the assumed operating condition, based on the protection setting and coordination rules and combined with the current operating point, the relative over-limit degree of the fault current and setting current of each protection element is determined to measure the pressure intensity on the setting; the time interval margin and the relative over-limit degree are dimensionlessly normalized and weighted and fused, and the fusion result is pruned according to the supremacy of the most unfavorable case of the protection pair and the [0,1] interval is pruned to form the protection coordination disturbance index.

[0010] In a preferred embodiment, obtaining the power flow redistribution sensitivity includes, under the assumption that the action is effective, calling the solidified PTDF / LODF sensitivity library or the verified linearized power flow approximation to quickly estimate the change in the current carrying rate of each line; comparing the line power before and after the action using thermal stability or long-term current carrying capacity as the caliber to obtain the over-limit intensity of each line and reflect the extent of the increase in load factor accordingly; performing upper quantile aggregation on the over-limit intensity of the entire network, and combining saturation clipping and interval reduction to normalize the results to [0,1] to obtain the power flow redistribution sensitivity.

[0011] In a preferred embodiment, the minimum dwell time is determined by a weighted sum of two parameters and constrained by upper and lower bounds; one of them is the correlation time in the neighborhood of the threshold, which is obtained by autocorrelation analysis or equivalent first-order autoregression approximation of the radial composite index within a fixed-length proximal sliding window, and is used to characterize the time correlation scale of local dynamics; the other is a site conservative constant, which is numerically weighted by the environmental level according to a predetermined mapping.

[0012] In a preferred embodiment, in the polar coordinate domain, the node topological criticality and the sensitivity to collaborative handling are mapped to a radius, which is the radial synthesis index.

[0013] In a preferred embodiment, after determining the handling level, a handling list and expected trajectory are generated based on the handling level; strategies involving strong actions are packaged and issued as a single transaction after digital twin verification, with version signature and idempotent key attached; when clock anomalies, unverified topology, or substandard input quality are detected, a conservative degradation channel is entered, allowing only L1 and below conservative actions and shortening the review cycle; for uncertain samples, boundary samples, and conservative degradation samples, shadow channel micro-adjustment and gray-scale release are implemented; during the linkage execution process, the expected and actual reconciliation and correction or rollback are completed based on the execution receipt, the handling effectiveness and drift monitoring indicators are accumulated, and key versions and evidence chains are written into the searchable record chain.

[0014] In a preferred embodiment, an angle lower limit gate is set when classifying in the polar coordinate domain; when the processing coordination sensitivity is at a high level and the angle gate condition is not met, it is prohibited to directly upgrade from the current level to a higher level, and at most the conservative processing of the L1 level is executed.

[0015] In a preferred embodiment, the following modules are included: The data access and profiling modeling module is used to solidify steady-state baselines, threshold calibers, and related sensitivity libraries under a unified clock. The anomaly screening and out-of-distribution identification module is used to generate candidate events and output uncertainty and consistency conclusions. The graded treatment and conservative degradation module is used to determine the treatment level and perform hysteresis and dwell vibration suppression in the polar coordinate domain; The rapid adaptation and knowledge enhancement module is used for shadow channel fine-tuning, gray-scale release, and knowledge graph maintenance; The linkage execution and closed-loop evolution module is used for packaging and issuing a single transaction, reconciling execution receipts, correcting deviations or rolling back, and writing key versions and evidence chains into a searchable record chain.

[0016] The technical effects and advantages of this invention are as follows: This invention achieves robust anomaly detection and significantly reduces false alarms and missed alarms by completing data cleaning, alignment, and quality verification under a unified time reference, superimposed with internal and external bistable baselines, external distribution identification, and multimodal consistency verification. On the handling side, secondary risks are characterized by protection coordination disturbance indicators and power flow redistribution sensitivity, and are classified in the polar coordinate domain with node topology criticality. Combined with asymmetric hysteresis and adaptive minimum dwell time, it effectively suppresses frequent threshold shifts and the resulting operational jitter. For strong actions, digital twin verification and conservative degradation gating are introduced to reduce the probability of selective protection under pressure and line over-limit, and control the spread of risks to the system level.

[0017] Meanwhile, the solution uses shadow channels for incremental adjustments and canary releases to rapidly absorb new operating conditions and boundary samples. It also uses a knowledge graph of linked equipment, components, symptoms, operating conditions, and treatments to accumulate a searchable chain of evidence, improving interpretability and traceability. The linked execution process adopts a one-time transaction package distribution and execution receipt reconciliation to form a closed loop from strategy generation and distribution to correction / rollback. Key models and threshold versions, scenario parameters and quality tags can all be audited, facilitating cross-site reuse and long-term operation and maintenance governance, ensuring both security and availability and scalability. Attached Figure Description

[0018] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings; Figure 1 This is a flowchart illustrating a risk intelligent monitoring and control method according to the present invention; Figure 2 This is a schematic diagram of the structure of a risk intelligent monitoring and control system according to the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Example 1 This invention provides a method for intelligent risk monitoring and control, such as... Figure 1 As shown, it includes the following steps: Step S1: Data Access, Time Base Unification, and Profile Modeling This step cleans, aligns, and verifies the quality of operational data, equipment information, and environmental information under a unified time benchmark, forming three types of pre-inputs: equipment profile, environmental profile, and a topology snapshot. Based on this, the steady-state baseline and threshold caliber are solidified, providing a reusable and auditable foundation for candidate event identification and uncertainty handling in step S2, as well as for the graded disposal calculation in step S3.

[0021] First, the system establishes millisecond-level time synchronization and a three-layer identifier mapping for sites, intervals, and devices, aligning data from monitoring and protection devices, metering and online monitoring, operation and maintenance ledgers and repair records, meteorological and electromagnetic disturbance sources on the same timeline. For critical quantities, it performs missing measurement filling, outlier removal, and truncation robustness processing, while recording data quality tags and processing reason codes to ensure that subsequent calculations can be traced back to the original evidence. Externally, only the processed values ​​and corresponding quality tags are output, without exposing intermediate operator details in the event stream to avoid noise propagation.

[0022] Subsequently, the system generates equipment profiles and establishes a one-to-one mapping with the primary topology. Based on rated parameters, connection relationships, protection settings, historical defects, and maintenance windows, the equipment profiles derive several indicators that support subsequent decision-making, including: a summary of local redundancy and transfer capacity, typical operating condition profiles (divided by peak, off-peak, and seasonal categories), and node topological criticality. . This is used to express the structural-operational dual-domain importance of the node in the network: the structural side focuses on the impact of removing this node on connectivity and redundancy, while the operational side focuses on the power flow redistribution cost, protection coordination margin, and operational complexity required to restore equivalent power supply under typical operating conditions; when the site is in a high-humidity, high-temperature, salt spray, or strong electromagnetic disturbance environment, the operational side results are appropriately amplified. All profile quantities are fixed in a dimensionless manner of 0-1 and written into the record chain along with "topology version, profile version, and calibration description" as read-only inputs directly referenced by subsequent steps.

[0023] Simultaneously, the system generates an environmental profile, converting information such as meteorological conditions, seasonal load characteristics, external electromagnetic disturbances, and geographical corrosion levels at the site into an environmental stress index and environmental level label. The environmental profile and equipment profile are strongly correlated and bound by site and time slice, which is used to explain the suspected causes of observed anomalies in step S2 and to constrain the window and pace of the response strategy in step S3.

[0024] Once the network profile is complete, the system creates a topology snapshot of the current network and aligns it with the device profile using key values, clarifying the superior, subordinate, and communication relationships of each device. Based on this, two types of steady-state baselines are established: one is the local operating band obtained by statistically analyzing devices / loops during historical stable periods; the other is the reference band obtained by comparing objects of the same type and level with institutional standards. The system configures upper and lower hysteresis and minimum dwell time based on a single threshold anchor point, serving as a unified rule for subsequent classification and handover jitter suppression. All thresholds, hysteresis widths, and dwell times are fixed with version numbers and marked with their effective range along with the topology snapshot, ensuring consistency in cross-cycle comparisons.

[0025] Step S2: Routine anomaly screening and out-of-distribution identification Based on step S1, candidate anomalies are formed and their distribution is identified using "event summary, equipment profile, and environmental profile" as inputs. The system matches the equipment profile with a baseline model and threshold levels, and uses climate, humidity, and electromagnetic interference features from the environmental profile as contextual parameters to conditionally adjust the thresholds and sensitivity. The system also solidifies the caliber and model version into metadata. Event summaries, provided that the unified clock and quality markers meet the requirements, enter the initial screening stage. The initial screening uses the historical stable period of this equipment and a group of similar equipment as internal and external baselines, rapidly comparing key quantities such as temperature rise, partial discharge texture, current and voltage fluctuations, three-phase imbalance, and harmonic distortion. Cases exceeding limits or with a rate of change exceeding the steady-state upper limit are marked as candidate anomalies. To avoid duplicate alarms and dense alarms caused by the same physical cause, the system merges adjacent anomalies within a short time window, retains a representative time window, and links it to the original event summary with a reference relationship to ensure a complete tracing chain.

[0026] After candidate events are generated, the system calculates the deviation from the internal and external baselines to distinguish between normal fluctuations and atypical characteristics. If the deviation mainly originates from the known influence range of environmental covariates, a conservative correction is applied to subsequent conclusions; if the deviation crosses the population robustness zone, it enters open set identification. Open set identification uses event features as input, searches the known fault sample library and cluster centers, and combines similarity and exclusion. When the similarity with any known class is low and it is difficult to form a consistent classification, the event is marked as suspected out-of-distribution, and evidence fragments such as "the closest classes, similarity measure, and rejection criteria" are retained. Samples that fall within known classes but have obvious boundaries are marked as boundary samples, and subsequent verification is combined with uncertainty and human-machine collaborative verification.

[0027] Uncertainty assessment is used to characterize the robustness of the judgment. Without altering the criteria, the system generates a set of independent judgment results for the same candidate event and calibrates them to interval-based confidence. Simultaneously, it cross-verifies multimodal evidence based on modal consistency. When directions are consistent, the overall confidence is increased; when conflicts or reversals in strength exist, the overall confidence is decreased, and the source of the conflict and its impact weight are noted. Apparent conflicts caused by time synchronization boundaries, configuration changes, or inconsistencies in unit calibers are disambiguated using the metadata from step S1. If necessary, the event is downgraded to "requires review" to avoid inadvertently entering a strong handling path. Finally, a structured candidate event record is generated, using equipment, component, and channel identifiers as primary keys, associated with model version, threshold file, and profile version. This record solidifies the initial screening conclusion, suspected out-of-distribution markers, interval-based confidence, main evidence citations, modal consistency conclusions, quality markers, and timestamps, along with a draft handling level for reference only. Key decision point traceability includes contextual parameters for threshold correction, deviation reference surfaces, open-set rejection criteria, uncertainty generation methods, and modal conflict adjudication criteria. Records suspected of being out of distribution, lacking sufficient overall confidence, or being boundary samples are transferred to step S3 along with their chain of evidence.

[0028] Step S3: Risk and cost-based tiered treatment and conservative downgrade This step, based on the candidate events and uncertainty conclusions output in step S2, and combining the device profile, environment profile, and a topology snapshot solidified in step S1, completes the calculation and control issuance of the handling level under clearly defined boundary conditions. To ensure stable criteria and simplicity of implementation, the handling criteria rely on only two dimensionless quantities: node topological criticality. The sensitivity to handling coordination is Ψ(t). The node topology criticality is the node importance index calculated offline in step S1 and fixed in version; the sensitivity to handling coordination represents the intensity of secondary risk introduced by the execution of standard handling actions at time t to relay protection coordination and power flow redistribution; the larger the value, the more sensitive it is and the less suitable it is to take aggressive measures.

[0029] First, the system reads the equipment profile and a topology snapshot (including protection settings, time limit settings, selective criterion CTIref, line thermal stability limit, and PTDF / LODF sensitivity library) solidified in step S1 under a unified time reference, as well as the current operating point output in step S2 (real-time current carrying rate of each line, bus injection, quality mark, and synchronization status). This is based on the system's pre-set set of standard handling actions. (Such as on-site flow restriction, segmented isolation, load transfer, etc.), evaluate each action one by one. Two types of components: Protect the coordination disturbance index, in action Assuming it takes effect, calculate the selective time interval margin of critical protection pairs along the protection coordination link. The calculation formula is as follows: ;in This is a selective time interval reference value used to constrain the minimum time difference between upstream and downstream protection systems. It is solidified and versioned from the protection setting principles and setting table in step S1. , To calculate the action time or equivalent time limit of the corresponding protection pair p obtained through protection coordination calculation under the new operating conditions after the assumed action is taken, based on the setting and coordination rules in step S1 and combined with the current operating point in step S2.

[0030] The degree of pressure exerted on the setting by the action is measured by the relative exceedance of the fault current and the setting current. ;in To protect the setting current of component p, The fault current is calculated using short-circuit conditions under the assumed topology and operating point where action a is effective. The relative over-limit ratio reflects the pressure exerted by the fault current on the setting current.

[0031] The selective time interval margin and the relative over-limit ratio are normalized and integrated into the protection coordination disturbance index for this action. ,in , ≥0, + =1, for example ; Determined and versioned by site type and protection setting system; To protect the coordination disturbance index, the dimensionless quantity after taking the supremum of the most unfavorable protection pair and pruning it to 0–1 represents the intensity of the influence of action a on protection coordination.

[0032] The power flow redistribution sensitivity is estimated by using the PTDF / LODF sensitivity fixed in step S1 or the calibrated linearized power flow approximation to estimate the change in current carrying capacity of each line under action a. in This is the thermal stability limit or long-term current carrying capacity limit for line e. , The current line power and the line power after the assumption of action a take effect (consistent with the stability limit, which can be active or apparent power). The former is given by the operating point in step S2, and the latter is quickly estimated based on the PTDF / LODF sensitivity library in step S1 or a calibrated linearized power flow model; The change in current carrying capacity reflects the extent to which the action increases the line load rate.

[0033] Systematic sensitivity was characterized by fractional polymerization exceeding the limit strength. in (•) represents the upper quantile (e.g., α=0.9 to weaken individual abnormal line segments). The saturation constant on the power flow side. >0. Reduce the quantile aggregation strength to [0,1], and calibrate and version it according to the network level in step S1. The dimensionless quantity after quantization and saturation clipping, representing the sensitivity to trend redistribution, indicates the tendency of action a to exceed the systematic limit.

[0034] After obtaining the two types of components, define the action-level sensitivity. in For smooth compression mapping (γ>0 is the slope, b is the equilibrium point).

[0035] To avoid optimistic / pessimistic biases towards a single "optimal / worst" action, the system uses robust quantile aggregation to obtain the collaborative sensitivity of the handling. .

[0036] To convert the two parameters into a disposal level, first ( , Mapped to the polar coordinate plane. Definition , ,calculate .

[0037] in, It is a radial composite index that represents the overall magnitude of the "response intensity" (the more critical the node and the lower its sensitivity, the larger ρ is). Used for angle gating in highly sensitive scenarios. Given three radial thresholds. and the lower limit of the angle ,For example, =0.1, =0.3, =1, = The treatment level is determined according to the following rules: when ρ(t) < ρ1, it is judged as L0 (observation); when ρ1 ≤ ρ(t) < ρ2 and The condition is classified as L1 (light treatment); when ρ2≤ρ(t)<ρ3 and It is determined to be L2 (medium treatment); when ρ(t)≥ρ3 and The case is classified as L3 (strong handling). If < If the sensitivity is considered too high, the system will not be aggressively upgraded; at most, it will perform conservative L1 handling.

[0038] To suppress frequent switching of the threshold neighborhood, an asymmetric hysteresis band and a minimum dwell time are configured for each radial threshold. The upgrade condition is... The downgrade condition is: The downgrade must be maintained at a continuous dwell time of not less than the minimum dwell time τmin for the downgrade to take effect. It is effective only if the dwell time is maintained continuously for a period not less than the minimum dwell time τmin. Radial threshold, and Let represent the hysteresis band width, specifically the hysteresis band width on the "upshift side" (upward) and "downshift side" (downward) of the k-th gear; for example... =0.02, =0.01.

[0039] To achieve adaptive control with minimum residence time, this step sets it as a dynamic quantity constrained by upper and lower bounds, consisting of a weighted sum of two parameters. Let the minimum residence time of the k-th criterion be... The definition is as follows: ;in ,For example . The residency limit is not less than an integer multiple of the sampling period. This represents the maximum residency limit. The meanings of the two weighted parameters are clearly defined as follows: The correlation time of the k-th threshold neighborhood is obtained by approximating the autocorrelation function of ρ(t) or the equivalent AR(1) within a fixed-length proximal sliding window, representing the time correlation scale of local dynamics. The environmental level is a conservative constant for the site, determined by the environmental grade. Let the environmental grade of the site be denoted as . (A is the best, D is the worst), its numerical mapping is as follows: Based on this, a site conservatism constant is defined. .

[0040] Environmental rating refers to the classification of the strength of external environmental risks to equipment or scenarios, used to guide conservative settings such as protection settings, maintenance strategies, and operating methods. For example, an indicator system can be established based on multi-source environmental and operating condition data, and after standardization and weighted scoring, it can be divided into several levels (such as A–D) according to preset thresholds.

[0041] After the classification is completed, the system generates a list of actions and corresponding expected trajectories based on the current level L∈{L0, L1, L2, L3}: When L=L0, only observation and review plans are formulated, and no control is issued; when L=L1, conservative measures that do not change the topology (such as local current limiting, temporary load reduction, and enhanced temperature / dehumidification) are implemented, and the minimum observation period and review cycle are set simultaneously; when L=L2, under the premise of meeting protection coordination and power flow upper limit constraints, limited topology adjustments such as small-scale segmented isolation and load transfer are included, and all strong actions must be verified by digital twin before being included in the expected trajectory; when L=L3, strong actions are issued according to the expected trajectory, and the preconditions, mutual exclusion relationships, and fallback paths are clearly defined for each action; if the above conditions are not met, the system automatically downgrades to L1 and enters the conservative downgrade channel. Control commands are issued in a single transaction package, accompanied by a version signature, idempotent key, and time validity window; the edge controller executes the commands after completing signature verification and local consistency checks. When network jitter or short-term disconnection occurs, the edge side maintains low-risk autonomy without exceeding the conservative boundaries, and retransmits execution logs and quality tags after the link is restored.

[0042] If boundary conditions such as clock anomalies, incomplete topology change verification, substandard input quality, or failure of human-machine review are detected, the system will forcibly enter a conservative downgrade: freezing strong actions, retaining only L1 and lower conservative actions, and shortening the review cycle until the anomaly is resolved. The handling level, handling list, and expected trajectory formed in this step, along with the complete evidence chain, are simultaneously transferred to step S4 for rapid adaptation and knowledge enhancement, and their effectiveness and secondary risk control effect are verified in step S5 through execution receipts and reconciliation closure.

[0043] Step S4: Rapid Adaptation and Knowledge Enhancement Without affecting production, this step enables the judgment criteria and handling strategies to quickly absorb new operating conditions and out-of-distribution characteristics, and improves root cause convergence and maintenance guidance capabilities with structured knowledge. Based on the uncertainty and modal consistency of step S2, the system aggregates uncertain samples, boundary samples, and conservatively downgraded samples into a queue to be adapted, sorted by information gain and representativeness, taking into account both temporal continuity and cross-device diversity, organizing small-batch labeling and recording participating roles and criterion descriptions. After obtaining a small number of high-value labels, a lightweight parameter module is used for minor adjustments without altering the main model structure, affecting only newly added feature pathways or environmental conditional branches, avoiding regression of existing capabilities. The fine-tuned version enters the shadow channel for parallel inference with the existing version, comparing indicators such as false positives, false negatives, handling triggers, secondary risks, work order volume, and closed-loop duration; when multiple indicator constraints are met and there are no significant negative impacts, it is released in stages by site, feeder, or equipment family, setting a minimum observation period and rollback threshold; anomalies are rolled back while retaining the cause code and data snapshot. The adaptive adjustment of the threshold follows the same grayscale process to ensure that parameter drift is controllable and the evolution of the caliber is traceable.

[0044] Knowledge enhancement maintains a knowledge graph with equipment, components, symptoms, operating conditions, and treatments as its framework. Equipment and component nodes originate from ledgers and equipment profiles; symptom nodes originate from event summaries and diagnostic conclusions from maintenance receipts; operating condition nodes originate from environmental profiles and primary topology states; and treatment nodes originate from the treatment level and expected trajectory abstraction in step S3. For out-of-distribution or multimodal conflict events, the graph retrieves similar patterns and calculates the correlation strength, outputs ordered root cause candidates, and generates a verification list and suggested measurement points, clearly defining the temperature rise curve, pulse texture, electrical transients, and environmental response that should be observed. The verification list is issued by the work order system, and after the receipt is archived, the graph is written back to update the edge weights of symptom-root cause and root cause-treatment, correcting inapplicable correlations and improving the priority and interpretation consistency of new patterns in subsequent judgments. In cases where the expected-actual deviation or strategy regression occurs during treatment execution, it is also written into the "treatment effectiveness" edge in the form of a ticket, reducing the priority of the treatment in similar situations. All adaptations and knowledge updates use event summaries, device profiles, and environment profiles as index keys, candidate events, evidence chains, handling levels, expected trajectories, and execution receipts as evidence sources, and shadow reports and grayscale records as the basis for going online, forming a complete record chain as the data and knowledge foundation for step S5.

[0045] Step S5: Linked Execution and Closed-Loop Evolution This step implements coordinated execution based on the expected trajectory verified by digital twins, and establishes a closed loop from policy issuance to deviation management with execution receipt reconciliation as the core. Simultaneously, it performs long-term measurement and auditing of indicators and drift during operation, supporting the continuous adaptation and knowledge enhancement of step S4. Under a unified clock, the platform packages the expected trajectory into a single issuance transaction, including a version signature, idempotent key, and time validity window. The edge controller receives and executes the transaction after verifying the signature and window. Actions with preconditions and mutual exclusion relationships are checked locally on the edge side, and timing is constrained by the station-end time base. In the event of network jitter or short-term disconnection, the edge side maintains low-risk autonomy without exceeding conservative boundaries, retaining the right to execute conservative actions such as rate limiting, load reduction, dehumidification, and encrypted temperature measurement. Execution logs and quality markers are retransmitted after the link is restored. Strong actions involving topology changes, protection coordination, and potential secondary risks are strictly executed according to the expected trajectory, and no relaxation or expansion is allowed.

[0046] Execution receipts are returned at the action granularity, including the actual action time, configuration or location feedback, relevant transient power levels, synchronous environmental measurements and quality markers, and references the strategy document and version. The platform verifies the deviations in actions, timing, and measurements by reconciling the expected and actual results. When necessary, it combines equipment profiles and environmental profiles for contextualized interpretation, distinguishing between apparent differences and substantive deviations. When a deviation exceeds the limit, correction or rollback is triggered immediately. Correction is executed according to the rollback conditions and alternative paths specified in the document. Rollback restores the system to a safe configuration, freezes the current strategy, and writes the reason code, the original receipt text, the correction or rollback result, and a timestamp. Even if low-risk actions in the conservative downgrade state deviate from the expected outcome, the handling is not escalated; only the review cycle is shortened and the measurements are encrypted.

[0047] Operational metrics are integrated throughout the execution cycle. The platform maintains indicators such as accuracy, false alarm rate, false negative rate, number of triggers, success rate of coordinated actions, closed-loop processing time, expected-to-actual deviation, number of secondary risk events, and loss reduction estimates across device, site, and time dimensions. It also monitors the drift of input distribution, conceptual error rate, and proportion of quality anomalies. When trigger conditions are met, adaptation and knowledge enhancement task tickets are generated and handed over to step S4 for minor adjustments and graph updates in the shadow channel. Common patterns across sites are handled through parameter sharing or federated aggregation within compliance boundaries, exchanging only necessary statistics or weight differences without transmitting raw measurements, thus improving broad-area generalization and protecting sensitive information. The key nodes of strategy generation, verification, distribution, execution, receipt, reconciliation, correction, and rollback form a record chain, including weight and threshold versions, contextual parameters, time base status, ticket signature, original receipt text and quality mark, deviation and reason code, correction or rollback path and result, indicator snapshot and drift alarm, triggered adaptation and knowledge enhancement tasks, etc., which can be retrieved by device, site, and time range. When equipment is replaced, configuration is adjusted, or channel diameter is changed, the access verification and freeze policy mechanism is automatically triggered. Objects that fail the verification will not participate in the linkage. When the station disconnects or the clock is abnormal and the preconditions for the expected trajectory verification cannot be met, the transaction is converted to delayed execution or manual review, and the record chain is completed after recovery.

[0048] Example 2 The design of the risk intelligent monitoring and control system of the present invention is based on the method in Embodiment 1, specifically as follows: Figure 2 The system includes the following modules: data access and profiling module, anomaly screening and out-of-distribution identification module, tiered handling and conservative degradation module, rapid adaptation and knowledge enhancement module, and coordinated execution and closed-loop evolution module. All modules work collaboratively under a unified time benchmark and versioning standards, as detailed below.

[0049] The data access and profiling module is used to clean, align, and verify the quality of operational data, equipment information, and environmental information under a unified millisecond clock, forming three types of read-only inputs: "equipment profile, environmental profile, and first-time topology snapshot." The module also uses version numbers to solidify steady-state baselines, threshold calibers, and hysteresis / residential parameters. The equipment profile records rated parameters, connection relationships, protection settings, and historical maintenance, and calculates node topology criticality. The environmental profile generates environmental stress indices and environmental level labels. The first-time topology snapshot is aligned with the equipment profile key values, outputting a PTDF / LODF sensitivity library and a selective time interval (CTIref) benchmark for subsequent protection coordination and power flow estimation. The module synchronously generates quality tags and processing reason codes, writing them into the record chain for downstream reference.

[0050] The anomaly screening and out-of-distribution identification module takes "event summary, equipment profile, and environmental profile" as input. Based on internal and external dual baselines, it quickly compares key quantities such as temperature rise, partial discharge texture, power fluctuation, imbalance, and harmonics to form candidate anomalies and merge homogeneous events within a short window. For samples crossing the robustness band, it performs open-set retrieval and rejection, outputs suspected out-of-distribution and boundary sample labels, and performs modal consistency verification on multimodal evidence, generating interval-based confidence and conflict source descriptions. The module structurally solidifies candidate events along with "uncertainty conclusions, model / threshold versions, quality labels, and timestamps," providing traceable input for subsequent tiered handling.

[0051] The tiered handling and conservative degradation module reads the equipment profiles, environmental profiles, and a topology snapshot from the previous two modules. Combining this with real-time operating points, it calculates protection coordination disturbance indices and power flow redistribution sensitivity based on preset standard handling actions (local current limiting, segmented isolation, load transfer, etc.), fusing them to obtain handling coordination sensitivity. This sensitivity is then mapped to the polar coordinate domain along with the node topology criticality. The module calculates the handling magnitude and angle gating, and performs L0-L3 tiered judgment and jitter suppression based on radial thresholds and asymmetric hysteresis / minimum dwell time τmin. Under the constraints of protection coordination and thermal stability limits, the module generates a handling list and expected trajectory. When boundary conditions such as clock anomalies, unverified topology, or substandard quality are detected, conservative degradation is forcibly implemented, allowing only L1 and lower conservative actions and shortening the review cycle. Control commands are packaged and issued as a single transaction, accompanied by a version signature, idempotent key, and time validity window.

[0052] The rapid adaptation and knowledge enhancement module, without affecting production, aggregates uncertain samples, boundary samples, and conservatively downgraded samples into an adaptation queue, organizes small-batch annotation, and performs minor adjustments to the shadow channel using a lightweight parameter module without changing the main model's structure. It then gradually releases and rolls back the module using a gray-scale strategy. The module maintains a knowledge graph with equipment—component—symptom—operating condition—treatment as its framework. For out-of-distribution or modal conflict events, it outputs ordered root cause candidates, verification lists, and suggested measurement points. The effectiveness of treatment and the deviation between expected and actual outcomes are corrected through document write-back, forming an auditable record of the evolution of the scope.

[0053] The coordinated execution and closed-loop evolution module implements coordinated control based on the expected trajectory verified by digital twins. At the edge, signature verification and local consistency checks are performed, and low-risk autonomy is maintained within conservative boundaries during link jitter. The module sends back execution receipts at the action granularity (time, configuration / position feedback, transient measurements, environmental measurements, and quality markings), completing the expected-to-actual reconciliation, deviation correction, or rollback. It also records indicators such as accuracy, false alarms / missed alarms, action triggering, closed-loop duration, secondary risks, and drift monitoring, triggering adaptation and knowledge enhancement tasks to support long-term system evolution. Key versions, scenario parameters, and evidence chains of all the above modules are retrievably written into the record chain by device, site, and time range, ensuring consistency and traceability before and after actions.

[0054] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0055] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0056] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0057] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0058] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for intelligent risk monitoring and control, characterized in that, Includes the following steps: The operating data, equipment information and environmental information are cleaned, aligned and quality verified under a unified clock to form three types of read-only inputs: equipment profile, environmental profile and a topology snapshot; Anomaly screening is performed based on internal and external bistable baselines. Open set identification and rejection are performed on samples that cross the robustness band. Modality consistency verification is performed on multimodal evidence and interval confidence is output to generate structured candidate event records. A graded treatment determination is made by combining the node topological criticality and the treatment coordination sensitivity, where the treatment coordination sensitivity is obtained by fusing the protection coordination disturbance index and the power flow redistribution sensitivity. The node topological criticality and treatment coordination sensitivity are mapped radially by threshold and angle by gating in the polar coordinate domain to obtain the treatment levels of L0-L3, and the jitter in the critical interval is suppressed by asymmetric hysteresis and minimum dwell time.

2. The risk intelligent monitoring and control method according to claim 1, characterized in that: The sensitivity of coordinated response is obtained by fusing the protection coordination disturbance index and the power flow redistribution sensitivity. After obtaining the protection coordination disturbance index and the power flow redistribution sensitivity, a monotonically continuous and saturated smooth compression mapping is applied to each candidate response action to obtain the action-level sensitivity. The action-level sensitivity of all candidate actions at the same time is robustly aggregated in quantiles and combined with saturation pruning and interval reduction to obtain the coordinated response sensitivity. The slope parameter of the mapping is positive, and the equilibrium point parameter is used to determine the neutral equilibrium position.

3. The risk intelligent monitoring and control method according to claim 2, characterized in that: Under the assumption that the handling action is effective, the selective time interval margin of the critical protection pair is calculated along the protection coordination link; the selective time interval benchmark value is used to constrain the minimum action time difference between the upper and lower level protections, and is fixed and versioned according to the protection setting principles and setting table. Under this assumed operating condition, based on the protection settings and coordination rules and combined with the current operating point, the relative over-limit degree of the fault current and setting current of each protection element is determined to measure the pressure intensity on the setting; the time interval margin and the relative over-limit degree are dimensionlessly normalized and weighted and fused, and the fusion result is pruned according to the supremacy of the most unfavorable case of the protection pair and the [0,1] interval is pruned to form the protection coordination disturbance index.

4. The risk intelligent monitoring and control method according to claim 2, characterized in that: The acquisition of power flow redistribution sensitivity includes, under the assumption that the action is effective, calling the solidified PTDF / LODF sensitivity library or the verified linearized power flow approximation to quickly estimate the change in current carrying capacity of each line; comparing the line power before and after the action with thermal stability or long-term current carrying capacity as the caliber to obtain the over-limit intensity of each line and reflect the extent of the increase in load factor; performing upper quantile aggregation on the over-limit intensity of the entire network, and combining saturation clipping and interval reduction to normalize the results to [0,1] to obtain the power flow redistribution sensitivity.

5. The risk intelligent monitoring and control method according to claim 1, characterized in that: The minimum dwell time is determined by a weighted sum of two parameters and is constrained by upper and lower bounds. One parameter is the correlation time in the neighborhood of the threshold, which is obtained by autocorrelation analysis or equivalent first-order autoregression approximation of the radial composite index within a fixed-length proximal sliding window, and is used to characterize the time correlation scale of local dynamics. The other parameter is a site conservative constant, which is obtained by the environmental level according to a predetermined mapping.

6. The risk intelligent monitoring and control method according to claim 5, characterized in that: In the polar coordinate domain, the node topological criticality and the sensitivity to collaborative handling are mapped to a radius, which is the radial comprehensive index.

7. The risk intelligent monitoring and control method according to claim 1, characterized in that: After determining the handling level, a handling list and expected trajectory are generated based on the handling level; strategies involving strong actions are packaged and issued as a single transaction after digital twin verification, with version signature and idempotent key attached; when clock anomalies, unverified topology, or substandard input quality are detected, a conservative degradation channel is entered, allowing only L1 and below conservative actions and shortening the review cycle; for uncertain samples, boundary samples, and conservative degradation samples, shadow channel micro-adjustment and gray-scale release are implemented; during the linkage execution process, the expected and actual reconciliation and correction or rollback are completed based on the execution receipt, the handling effectiveness and drift monitoring indicators are recorded, and key versions and evidence chains are written into the searchable record chain.

8. The risk intelligent monitoring and control method according to claim 1, characterized in that: When classifying in polar coordinates, set a lower limit for angle gating; when the sensitivity of handling coordination is at a high level and the angle gating condition is not met, it is prohibited to directly upgrade from the current level to a higher level, and at most L1 level conservative handling is executed.

9. A risk intelligent monitoring and control system, characterized in that, The control system, based on the method described in any one of claims 1-8, includes the following modules: The data access and profiling modeling module is used to solidify steady-state baselines, threshold calibers, and related sensitivity libraries under a unified clock. The anomaly screening and out-of-distribution identification module is used to generate candidate events and output uncertainty and consistency conclusions. The graded treatment and conservative degradation module is used to determine the treatment level and perform hysteresis and dwell vibration suppression in the polar coordinate domain; The rapid adaptation and knowledge enhancement module is used for shadow channel fine-tuning, gray-scale release, and knowledge graph maintenance; The linkage execution and closed-loop evolution module is used for packaging and issuing a single transaction, reconciling execution receipts, correcting deviations or rolling back, and writing key versions and evidence chains into a searchable record chain.

Images