Method and system for path optimization and operation mode adjustment of important user power supply

Abstract

This invention provides a method and system for path optimization and operation mode adjustment for ensuring power supply to critical users, relating to the field of resource planning and allocation technology. By constructing a complete technical closed loop from multi-dimensional risk perception and dynamic hierarchical attribution to adaptive robust optimization, it provides a method for path optimization and operation mode adjustment that improves the accuracy, interpretability, and system robustness of power supply guarantee decisions for critical users. Through precise risk tracing, it achieves a shift from passive response to proactive elimination of risk root causes, improving the transparency and traceability of decision-making. By introducing dynamic risk transmission factors, it makes risk assessment results closer to the physical reality of the power grid, enhancing the accuracy of decision-making. Through adaptive multi-objective fusion optimization, it enhances the stability and reliability of the decision-making system under complex operating conditions and uncertain data environments.

Description

Technical Field

[0001] This invention relates to the field of resource planning and allocation technology, specifically to a method and system for optimizing power supply paths and adjusting operation modes for important users. Background Technology

[0002] In modern society, the stable operation of critical infrastructure is fundamental to ensuring the normal functioning of economic and social activities. Particularly in the energy sector, with the increasing demands for power quality and continuity from critical users such as data centers, advanced manufacturing, and medical institutions, how to plan power grid resources and optimize operation modes to ensure uninterrupted power supply has become a core issue in the fields of power grid dispatch automation and resource planning and allocation. Utilizing technologies such as big data analytics and knowledge graphs to improve the intelligence level of power supply guarantee decisions is an important development direction in this field.

[0003] Existing technologies face several challenges in addressing power supply issues for critical users, primarily in the following areas: The disconnect between quantifying risk and understanding its logical causes: One type of technical solution focuses on quantifying the risk assessment of potential power outages, such as calculating economic losses. However, the assessment results are often scalar values, making it difficult to reveal the specific transmission paths and root causes of risks within the power grid topology and business dependencies. Another type of solution utilizes technologies such as knowledge graphs for logical reasoning, which can identify conflict paths between power grid status and user needs. However, it lacks a unified quantitative scale for the severity of risks corresponding to different conflict paths, resulting in a lack of precise perception of the urgency of risks in decision-making.

[0004] Staticization of Risk Attribution Models: Even when some technologies attempt to attribute risks, the transmission models or weighting coefficients they rely on are usually pre-set based on offline analysis, exhibiting static or quasi-static characteristics. Such models struggle to reflect in real time the dynamic risk transmission capabilities of power grid components caused by factors such as operating status, environmental changes, or aging and wear, leading to discrepancies between risk attribution results and physical reality, thus affecting the accuracy of subsequent optimization decisions.

[0005] Vulnerability of Decision-Making Systems to Data Quality: Some existing optimization systems, such as those disclosed in Chinese Patent CN119647644A, integrate multiple modules including big data analysis, resource optimization, and decision support. However, their optimization decision-making process typically relies on the premise of intact input data quality. In actual operation, when sensor failures, communication delays, or model failures lead to a decrease in the reliability of input data (quantified risk values ​​or topological relationships), these systems lack smooth and robust degradation and adaptive mechanisms. They may generate drastically fluctuating or even erroneous control commands due to reliance on unreliable inputs, posing a potential threat to the stable operation of the power grid. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for optimizing the power supply path and adjusting the operation mode for important users, so as to at least solve or improve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: In a first aspect, the present invention provides a method for optimizing power supply paths and adjusting operating modes for critical users, comprising: Obtain quantitative risk data characterizing the potential service interruption losses of the important users under a preset power grid operation mode; Based on a pre-built knowledge graph containing power grid physical entities and user business activity entities, semantic conflict path data representing the logical conflict between the power grid physical state and user business needs is obtained. The quantified risk data, along with the semantic conflict path data, is backpropagated and attributed on the knowledge graph to generate a risk contribution distribution signal that characterizes the contribution of each entity in the knowledge graph to the quantified risk data. Based on the risk contribution distribution signal, an optimized control command is generated to adjust the preset power grid operation mode; Based on the risk contribution distribution signal and the semantic conflict path data, a decision support visualization signal is generated to present the risk tracing results.

[0008] In a second aspect, the present invention provides a path optimization and operation mode adjustment system for ensuring power supply to critical users, comprising: The first acquisition module is used to acquire quantitative risk data representing potential business interruption losses of important users under a preset power grid operation mode; The second acquisition module is used to acquire semantic conflict path data that characterizes the logical conflict between the physical state of the power grid and the needs of users' business activities, based on a pre-set knowledge graph containing power grid physical entities and user business activity entities. The attribution processing module is used to perform backpropagation attribution processing on the knowledge graph along the semantic conflict path data of the quantified risk data, so as to generate a risk contribution distribution signal that represents the contribution of each entity in the knowledge graph to the quantified risk data. The instruction generation module is used to generate optimized control instructions for adjusting the preset power grid operation mode based on the risk contribution distribution signal. The visualization generation module is used to generate a decision support visualization signal to present the risk tracing results based on the risk contribution distribution signal and the semantic conflict path data.

[0009] Compared with the prior art, the beneficial effects of the present invention are: By constructing a multidimensional risk vector that incorporates economic impact and time urgency, the essence of risk can be more comprehensively characterized. Utilizing semantic conflict paths in a knowledge graph as the carrier of information transmission, a backpropagation attribution mechanism is proposed. This mechanism interpretably and traceably distributes the multidimensional risk vector from the end-user side to various root physical entities in the power grid topology, forming a refined risk contribution distribution.

[0010] The attribution mechanism introduces a dynamic risk transmission factor calculated based on the real-time operating status of power grid components and historical fault data, and constructs a two-stage hierarchical attribution model. This model first performs causal attribution along the direct power supply path, and then analyzes the associated impact caused by shared resources, thereby achieving a dynamic and accurate characterization of risk distribution.

[0011] An adaptive, multi-objective fusion optimization method based on multi-dimensional confidence assessment is proposed. This method continuously evaluates the confidence level of various input data and, based on the confidence level, dynamically weights and fuses the primary optimization objective (minimizing risk contribution) with multiple backup optimization objectives (minimizing the number of logical conflicts) through a continuous weight mapping function. This ensures that decisions can smoothly and proportionally adapt to gradual changes in data quality, fundamentally eliminating control command chattering. Simultaneously, combined with a hierarchical command generation framework, an optimal balance is achieved between ensuring power supply security and controlling economic costs. Attached Figure Description

[0012] Figure 1 This is a schematic diagram of a method for optimizing the path and adjusting the operation mode of power supply for important users according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the execution logic of steps S1 to S3 in an embodiment of the present invention; Figure 3 This is a schematic diagram of the execution logic of steps S4 to S5 in an embodiment of the present invention; Figure 4 This is a block diagram of a system for optimizing the power supply path and adjusting the operation mode for important users, as described in an embodiment of the present invention. Detailed Implementation

[0013] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0014] It is understood that the terms first, second, etc., used herein may be used to describe various elements, but unless otherwise stated, these elements are not limited by these terms. These terms are used only to distinguish one element from another.

[0015] Example 1: Please see Figures 1 to 3 This invention provides a technical solution: a method for optimizing the power supply path and adjusting the operation mode for important users, comprising the following steps: S1: Obtain quantitative risk data characterizing potential service interruption losses for important users under preset power grid operation modes; S2: Based on a pre-built knowledge graph containing power grid physical entities and user business activity entities, obtain semantic conflict path data that represents the logical conflict between the power grid physical state and user business needs; S3: Quantify the risk data and perform backpropagation attribution processing on the knowledge graph along the semantic conflict path data to generate a risk contribution distribution signal that represents the contribution of each entity in the knowledge graph to the quantified risk data. S4: Based on the risk contribution distribution signal, generate optimized control commands for adjusting the preset power grid operation mode; S5: Based on the risk contribution distribution signal and semantic conflict path data, generate a decision support visualization signal to present the risk tracing results.

[0016] This invention provides a method for optimizing power supply for critical users by deeply integrating quantitative risk and logical reasoning. Its core technical features include: firstly, a multi-dimensional risk vector that simultaneously incorporates economic impact and time urgency to more comprehensively characterize the nature of the risk; secondly, after identifying the logical conflict paths between the physical state of the power grid and user business needs, a hierarchical and dynamic risk attribution mechanism is proposed. This mechanism calculates a dynamic risk transmission factor in real time based on the real-time operating status of power grid components and historical fault records. Through a two-stage hierarchical attribution model, causal attribution is first performed along the direct power supply path, and then the associated impacts caused by shared resources are analyzed, thereby accurately and interpretably allocating the multi-dimensional risk vector to each root entity in the knowledge graph.

[0017] right Figure 1 The following explanation is provided: Risk perception and conflict identification correspond to steps S1 and S2.

[0018] Risk perception corresponds to S1: representing the acquisition of quantitative risk data; it is a risk value with economic significance that is directly linked to the business of important users.

[0019] Conflict identification corresponds to S2: extracting semantic conflict path data from a pre-built knowledge graph. This represents the logical incompatibility between the physical state of the power grid and user business requirements.

[0020] The knowledge graph reverse attribution corresponds to step S3; the backpropagation attribution processing involves tracing the quantified risks perceived in the first stage back along the semantic conflict paths identified in the first stage on the knowledge graph. It introduces a dynamic risk transmission factor, which, based on the real-time status and historical data of each power grid component, unevenly distributes the risks of end users to various root physical entities in the power grid topology. This ultimately generates a risk contribution distribution signal.

[0021] The core decision-making process of step S4 in adaptive multi-objective optimization is reflected in two aspects: Multi-objective optimization: The decision-making objective is not simply risk minimization, but rather seeking the optimal balance among multiple objectives, such as ensuring power supply security and controlling economic costs. It continuously evaluates the confidence level of various input data. When the quality of the data source deteriorates, it does not rigidly switch decision logic, but rather smoothly and proportionally reduces reliance on that data source through a continuous weighting mapping function, while simultaneously increasing the reliance on backup optimization objectives. This eliminates control command chattering caused by data fluctuations, ensuring the robustness and stability of the decisions.

[0022] The risk tracing and instruction generation correspond to the final output of step S4 and the entire content of step S5. This is the result output and human-computer interaction stage of the present invention. Two key outputs are generated: The instruction generation corresponds to the S4 output: transforming the optimized decision scheme into a set of specific, executable optimized control instructions. This includes two levels: regular continuous instructions and discrete instructions for handling severe risks.

[0023] Risk tracing corresponds to S5: generating visual signals to support decision-making. The entire analysis process, from user risk to conflict paths, then to the root cause of risk, and finally to optimization instructions, is presented to scheduling and operation personnel in a visual and traceable manner.

[0024] Figure 1The power plant / substation icon on the left represents the energy supply side. It is the starting point of the entire physical power grid and the resource foundation for all power supply guarantee strategies. The icons on the right are important user icons (data centers / factories / hospitals): representing the energy demand side. The use of three different, highly recognizable icons is intended to emphasize that the application scenario of this invention is not ordinary users, but key facilities with significant socio-economic value that have extreme requirements for power supply continuity. The solid line represents the optimal power supply path; the thin dashed line represents the suboptimal / high-risk path: representing alternative power supply paths that have not been optimized or have been rejected by this invention. Its shape represents lower priority, and the dashed line symbolizes instability, potential interruption, or logical conflict. The interruption / lightning symbol placed along this path in the figure clearly marks it as a path with known risks, thus forming a strong contrast with the optimal path. The knowledge graph network in the form of a neural network represents the logical center of this invention. It is used to emphasize that it processes logical relationships, business rules, and semantic information that transcend physical connections. The reverse attribution arrow is the most vivid visual translation of the core algorithm (S3) of this invention. The arrow's flow path starts from important users, injects knowledge graph downwards, and after processing, points precisely upwards to specific components in the physical power grid, metaphorically representing the entire process of backpropagation attribution.

[0025] Further explanation: In S2, the steps for obtaining semantic conflict path data include: In the knowledge graph, the graph reasoning engine identifies dependency paths from grid-side entities to user-side business activity entities that have state conflicts or constraint violations under the preset power grid operation mode, and these paths are used as semantic conflict path data.

[0026] It should be noted that the knowledge graph is used to construct a structured description of the power grid system and its operating rules. The power grid-related nodes in the knowledge graph are divided into an ontology structure with a clear hierarchical relationship. Specifically, all nodes related to the power grid are collectively referred to as power grid-side entities; power grid-side entities are further divided into at least the following two subclasses: Grid-side physical entities: This subclass refers to nodes corresponding to hardware devices in the real world. Each grid-side physical entity represents a specific, tangible electrical device, such as, but not limited to, generators, transformers, transmission lines, circuit breakers, or busbars. The key characteristic of this type of entity is that it is a direct source of real-time telemetry data (voltage, current, active / reactive power) and is associated with historical fault and maintenance data records.

[0027] Grid-side logical entities: This subclass refers to abstract concepts, events, plans, or rules without physical form involved in grid operation. Examples include, but are not limited to, a predetermined maintenance plan, a reserve capacity call-up strategy, a grid stability control plan, or a set of relay protection settings. The key characteristic of these entities is that they define the constraints, objectives, or sequences of operations for system operation, and are an important component of semantic conflict reasoning, but they do not directly generate physical measurement data.

[0028] Further explanation: The quantitative risk data is a multi-dimensional risk vector, which includes at least: the potential economic impact representing potential economic losses, and the time urgency index representing the timeliness requirements of business interruption.

[0029] Further explanation: Before performing backpropagation attribution processing, dynamic risk transmission factors are calculated in real time based on the real-time operating status data and historical fault data of the physical entities on the power grid side in the knowledge graph; backpropagation attribution processing is performed along semantic conflict path data based on dynamic risk transmission factors. Further explanation: Backpropagation attribution processing is a hierarchical attribution process. When this process is performed on semantically conflicting path data, it includes: First attribution stage: Along the direct power supply dependencies in the semantic conflict path data, perform the first round of backpropagation attribution processing to obtain the direct causal risk contribution of each entity; The second attribution stage: Based on the shared resource dependencies defined in the knowledge graph, the direct causal risk contribution of multiple entities with shared upstream dependencies identified in the first attribution stage is corrected for related impacts to obtain the final risk contribution distribution signal.

[0030] The following is a detailed implementation description of the above content: In this embodiment, a dynamic, multi-dimensional risk assessment and hierarchical attribution method is provided. The technical problem to be solved is that when conducting risk assessment, using a single-dimensional indicator cannot distinguish the nature of the risk (whether it is a high risk with greater economic loss or a more urgent timeframe); at the same time, when tracing the source of risk, using static, preset logical relationships cannot reflect the impact of the dynamic operating characteristics of the power grid on the risk transmission capacity, leading to a deviation between the source tracing results and the physical reality. The calculation process in this embodiment involves the following key parameters: 3.1) The multidimensional risk vector, denoted as MRV, is a two-dimensional vector consisting of the following two components: 3.11) Economic Impact Potential Energy (EIP): This is a normalized measure of direct and indirect economic losses caused by service interruptions. It is a dimensionless value between 0 and 1. It is determined as follows: The economic value of service interruption per unit time is obtained from the user's self-reported value or as agreed upon through a service level agreement; the current service duration of the user is obtained; the economic value of service interruption per unit time is multiplied by the service duration to obtain a total estimated economic loss; this estimated economic loss is then divided by the estimated maximum possible economic loss for all important users to complete the normalization process. In a power grid containing important users with various service types, the magnitudes of their estimated economic losses differ. If a stable benchmark is not used for normalization, the EIP value will fluctuate drastically when users are added or removed from the system, affecting the stability of the algorithm. This embodiment uses the concept of quantiles from statistics to determine this maximum benchmark value. Its core principle is to select the 95th percentile in the historical or estimated data distribution as a benchmark to cover the loss magnitude of all users in most cases, while effectively eliminating the interference of extreme outliers on the overall measurement system, thereby obtaining a more statistically robust benchmark.

[0031] The steps for determining the maximum economic loss estimate are as follows: Obtain historical business data for all connected key users, or obtain estimated business data for future scheduling cycles based on their business plans. For each user, perform a multiplication operation based on the economic value of business interruption per unit time and the estimated maximum business duration to obtain the user's individual economic loss estimate. All these individual estimates constitute a data sample set. Arrange all values ​​in the aforementioned data sample set in ascending order. Obtain a percentage parameter to define high quantiles. This parameter defines the statistical boundary for excluding extreme outliers. In this embodiment, the percentage parameter is set to 95%. Based on the aforementioned percentage parameter, calculate the corresponding position in the sorted data sample set and extract the value at that position as the benchmark value for the final estimated maximum economic loss for all key users in the system.

[0032] 3.12) Time Urgency Index (TUI): Its logical meaning is the urgency of restoring power after a service interruption. It is a dimensionless value between 0 and 1. It is calculated by obtaining the maximum allowable interruption time corresponding to the user's service type; dividing the preset maximum system response time benchmark of 30 minutes by this maximum allowable interruption time to obtain a ratio; finally, a logical function is applied to this ratio to ensure that its value smoothly falls within the 0-1 range. The shorter the interruption time, the closer the index value is to 1.

[0033] The sensitivity of user services to power outage duration is not linear. The increase in urgency from a permissible interruption of 10 minutes to 5 minutes is far greater than the increase from 60 minutes to 55 minutes. A simple linear reciprocal relationship cannot accurately characterize this nonlinear characteristic of drastic changes in urgency near critical time points. This embodiment employs a mathematical model capable of simulating this threshold effect. Specifically, it introduces the logistic function, derived from the field of neural network activation functions, to construct the TUI calculation model. The S-shaped curve of this function can produce drastic numerical changes near a central point, while tending to flatten out at both ends far from the central point. This characteristic highly matches the changing pattern of urgency of service interruption time: in the region far from the maximum permissible interruption time, the change in urgency is slow; when the actual interruption time approaches this critical point, the urgency increases rapidly.

[0034] The calculation steps for the Time Urgency Index (TUI) are as follows: First, obtain the maximum allowable interruption time corresponding to the user's service type. Second, subtract the obtained maximum allowable interruption time from the preset urgency response center point parameter to obtain a center offset. This center point parameter defines the interruption time corresponding to the steep change region of the S-curve. In this embodiment, this parameter is set to 10 minutes. In this embodiment, 10 minutes is considered a critical dividing line between tolerable and severely impactful for most continuous services. Third, obtain the urgency curve steepness parameter. This parameter controls the slope of the S-curve; the larger the value, the more drastic the curve changes near the center point. In this embodiment, this parameter is set to 0.5. Fourth, multiply the obtained center offset by the steepness parameter, then take the negative value of the result, and finally calculate the natural constant e to the power of this value to obtain an intermediate transformation value. Fifth, add 1 to the aforementioned intermediate transformation value, and then divide 1 by this sum to obtain the final Time Urgency Index (TUI).

[0035] 3.2) The dynamic risk transmission factor is denoted as RCF: its physical meaning is the actual efficiency or amplification effect of risk transmission along specific relationships (edges) in the knowledge graph. It is a dimensionless value in the interval of 0 to 1. Its determination method is as follows: The dynamic risk transmission factor (RCF) is dynamically generated by weighted summation of two sub-factors.

[0036] Calculate the Operating State Factor (OSF). This sub-factor reflects the impact of the current load on the stability of a component. The real-time load rate of the power grid components (including transformers and lines) is obtained; this load rate is input into a preset piecewise linear function, defined as follows: when the load rate is below 50%, the output value is 0.5; when the load rate is between 50% and 90%, the output value linearly increases from 0.5 to 0.9; and when the load rate is above 90%, the output value is 1.

[0037] The historical fault factor (HFF) is calculated based on a nonlinear mapping model that considers boundary effects and diminishing marginal impact. The calculation logic is as follows: First, the original fault count is obtained. Then, the cumulative number of faults recorded by the power grid element within a preset statistical review period is obtained. In this embodiment, the statistical review period is set to the past 365 days. To handle the special case of zero fault counts and introduce a significant change from zero to one, 1 is added to the aforementioned cumulative fault count. Next, a natural logarithmic operation is performed on this sum to obtain a logarithmic transformation value. The technical consideration for choosing the natural logarithm is that it has the most natural growth rate representation in engineering and scientific calculations, and can smoothly and effectively reflect the diminishing marginal impact of fault counts. A preset historical fault baseline upper limit parameter is obtained. The physical meaning of this parameter is that when the logarithmic transformation value of a power grid element reaches this upper limit, its historical reliability is considered to have reached its worst level, and its historical fault factor (HFF) value should be saturated. In this embodiment, the value of this historical fault baseline upper limit parameter is set to 3. The technical consideration for this value is that the natural logarithm of 3 is approximately equal to 19 cumulative faults, which is considered a high-frequency fault situation for most power grid components. Setting it as the upper limit for evaluation is statistically reasonable. Dividing the obtained logarithmic transformation value by this historical fault baseline upper limit parameter yields an initial scaling factor. To prevent the scaling factor from exceeding 1 due to extreme fault counts, the smaller value between the obtained initial scaling factor and 1 is taken. This scale factor, subject to boundary constraints, is multiplied by a preset maximum contribution coefficient for the historical fault factor HFF. In this embodiment, the maximum contribution coefficient is set to 0.5, thus obtaining the final historical fault factor HFF, whose value range is limited to the interval between 0 and 0.5.

[0038] The weighting coefficients for the Operating State Factor (OSF) and the Historical Fault Factor (HFF) are preset. In this embodiment, both weighting coefficients are set to 0.5, reflecting the technical consideration of giving equal importance to the current state and historical performance. The OSF is multiplied by its weighting coefficient to obtain the first result. The HFF is multiplied by its weighting coefficient to obtain the second result. Finally, the first and second results are added to obtain the final Dynamic Risk Transmission Factor (RCF). In this invention, setting the weighting coefficients for both the OSF and HFF to 0.5 is based on a risk assessment that emphasizes both current and historical factors. The OSF reflects the immediate, physical vulnerability of power grid components, while the HFF reflects their long-term, statistical reliability deficiencies. When prior knowledge is lacking to determine which is more important, equal weighting is the most robust and unbiased choice, avoiding over-reliance on single-dimensional information, thereby improving the robustness of the RCF parameters.

[0039] The risk transmission capability of power grid components is not simply linearly positively correlated with load rate. In lightly loaded areas (below 50%), small changes in load have little impact on component stability; however, in heavily loaded or overloaded areas (above 90%), even small increases in load can severely compromise stability, reaching the upper limit of risk transmission capability. The rationale for using a piecewise linear function in this embodiment lies in its ability to more accurately fit this nonlinear relationship: by setting a low, flat slope (constantly 0.5 in this embodiment) in the lightly loaded area, a steep linear growth slope in the intermediate region, and a flat saturation value of 1.0 in the heavily loaded area, the risk transmission characteristics of power grid components under different load ranges are simulated with lower computational complexity.

[0040] In this embodiment, the impact of the number of faults on risk transmission capability is non-linear. From 0 faults to 1 fault, risk assessment needs to be improved; however, from 10 faults to 11 faults, the marginal impact should gradually weaken. This invention uses a logarithmic function for processing. The logarithmic function has the mathematical property of diminishing marginal utility. By taking the logarithm of the number of faults, the differences in the low fault count range can be effectively amplified, while the differences in the high fault count range can be compressed. This allows the final HFF value to more reasonably reflect the cumulative effect of historical faults, avoiding a disproportionate gap in HFF value between other components due to the high fault count of a few problematic grid components.

[0041] Furthermore, when a single line failure simultaneously impacts both a data center requiring high economic security and a hospital requiring high timeliness, a single risk value cannot guide differentiated response strategies. Moreover, for users located at two different substations, since they rely on the same communication network for control, their risks are actually related, making it difficult to reveal this correlation. To address these limitations of the existing regulations, this invention designs an innovative hierarchical attribution fusion mechanism based on multidimensional risk vectors.

[0042] The core idea of ​​this embodiment is to decompose the complex risk attribution problem into two orthogonal and progressive sub-problems for processing, specifically including the following steps: The first step is to initialize the risk vector: At the end of each semantic conflict path identified in step S2, i.e. the user-side business activity entity, a multi-dimensional risk vector MRV, which includes two components, the economic impact potential energy (EIP) and the time urgency index (TUI), is attached.

[0043] The second step executes the first attribution phase: the goal of this phase is to handle direct causal relationships along the power supply path. Only direct dependencies such as "power supply to" and "belong to" defined in semantically conflicting paths are considered. For each component (EIP and TUI) of the multidimensional risk vector, backpropagation attribution processing is performed independently and in parallel. The specific process is as follows: the component values ​​are used as the initial risks to be assigned, and the dynamic risk transmission factor (RCF) is calculated in real time based on the components connected by each relationship along the path. Multiplication and accumulation are then performed layer by layer to allocate the risk to upstream entities. After this phase is completed, each grid-side entity in the knowledge graph will obtain a two-dimensional direct causal risk contribution vector.

[0044] The third step is to execute the second attribution stage. This stage aims to quantify the risk coupling effect caused by shared resources. Its calculation logic is as follows: Traverse the knowledge graph to identify all downstream entities that received a non-zero risk contribution in the first attribution stage and point to the same upstream shared resource entity through shared...class relationships. These downstream entities constitute a set of associated entities. For each entity in the aforementioned set, perform the following operations to calculate the scalar magnitude of its risk: Obtain the two components of its two-dimensional risk contribution vector calculated in the first attribution stage, namely the Economic Impact Potential (EIP) component and the Time Urgency Index (TUI) component; secondly, calculate the square values ​​of these two components respectively; thirdly, add the two square values ​​together; finally, calculate the square root of this sum, and the result is the entity's comprehensive risk magnitude. Summate the comprehensive risk magnitudes calculated by all entities in the associated entity set to obtain a total associated risk magnitude. Perform vector addition on the two-dimensional risk contribution vectors of all entities in the associated entity set to obtain the total associated risk vector. By performing vector addition on the total associated risk vector and the risk contribution vector of the shared resource entity itself, all risks caused by downstream coupling effects are attributed to this common risk convergence point.

[0045] Furthermore, this embodiment, through independent attribution of multidimensional risk vectors, can determine the risk contribution of a component, as well as whether the risk primarily stems from its impact on downstream high-economic-value users or on users with high timeliness requirements, thus providing a basis for differentiated decision-making.

[0046] The second stage of correlation effect correction can reveal the common root cause behind two seemingly unrelated faults. An example of a common root cause is a shared control module fault, which improves the depth and accuracy of fault diagnosis.

[0047] The overall calculation process in this embodiment is as follows: All operating parameters are loaded from an external configuration file. For each important user, its multidimensional risk vector (MRV) is calculated in parallel, containing two components: Economic Impact Potential (EIP) and Time Urgency Index (TUI). Based on the knowledge graph, the graph reasoning engine identifies all existing semantic conflict paths. The operating status and historical fault data of relevant power grid components are acquired in real time, and the current dynamic risk transmission factor (RCF) is calculated for each edge in the knowledge graph.

[0048] Initialize multidimensional risk vectors at the end of all semantically conflicting paths. Perform backpropagation in parallel and independently for both EIP and TUI channels, allocating risk contributions to upstream entities along the direct power supply path.

[0049] The results of the first phase were analyzed to identify high-risk entities with shared resource dependencies, and the associated impact was corrected by adding the associated risks to the shared resource entities.

[0050] After two stages of processing, the final two-dimensional risk contribution vector set formed by all entities in the knowledge graph is output as the final risk contribution distribution signal for use in subsequent optimization and visualization steps S4 and S5.

[0051] Furthermore, to decouple the core algorithm of this invention from specific application strategies and to ensure the configurability and ease of debugging of the technical solution, in the specific implementation path of this invention, all configurable operating parameters are predefined and stored in a structured external data carrier. Preferably, this data carrier is a stored local spreadsheet file. The content structure of this file is described in plain text as follows: the first line of the file serves as the header line, defining the field names of each parameter, such as parameter name, parameter value, unit, and remarks; starting from the second line, each line represents a configurable parameter item.

[0052] The core output of this invention is the final risk contribution distribution signal, in which the contribution value of each entity is jointly determined by the intermediate key parameter multidimensional risk vector MRV and dynamic risk transmission factor RCF.

[0053] Analysis of the output range of the multidimensional risk vector MRV: When the output value of Economic Impact Potential (EIP) approaches 1, it indicates that if the important user experiences a business interruption, the potential economic loss is closer to the highest level assessed by the system; when its output value approaches 0, it indicates that the potential economic loss is negligible in the system's assessment framework.

[0054] When the output value of the Time Urgency Index (TUI) is closer to 1, it indicates that the important user's service has a higher timeliness requirement for power supply continuity, a shorter maximum allowable interruption time, and needs to be responded to with a higher priority; when its output value is closer to 0, it indicates that the user's service has a higher tolerance for interruption duration.

[0055] Analysis of the output range of the dynamic risk transmission factor RCF: When the output value of the Dynamic Risk Transmission Factor (RCF) is closer to 1, it indicates that the physical entity (component or line) of the power grid is more likely to be in a highly vulnerable or unstable state, and its ability to transmit risk upstream is assessed as higher, with less attenuation of risk value; when its output value is closer to 0, it indicates that the entity's operating state is more stable and reliable, and it plays a greater buffering or damping role in the risk transmission path.

[0056] The Multidimensional Risk Vector (MRV) input parameters are: economic value of business interruption per unit time and duration of business interruption. When other parameters remain constant, an increase in either the economic value of business interruption per unit time or the duration of business interruption will lead to a monotonically increasing Economic Impact Potential (EIP). The economic value per unit time directly defines the unit cost of the risk, while the duration of business interruption defines the duration of the risk's impact; their product constitutes the total risk exposure. This positive correlation design ensures that users with higher economic value and the longest critical business cycles can be identified and prioritized for protection.

[0057] When other parameters remain constant, increasing the maximum allowable interruption time will lead to a monotonically decreasing result in the Time Urgency Index (TUI). Furthermore, due to the use of a logistic function model, this decreasing relationship is non-linear. This design reflects the urgency characteristics of business interruptions in the real world. For businesses with short allowable interruption times (including hospitals undergoing surgery), the urgency is high; while for businesses with allowable interruptions of several hours, the urgency is relatively moderate. The use of a non-linear logistic function ensures that the TUI has the highest rate of change near the critical interruption duration, while changing gradually at both short and long durations. This is highly consistent with the response patterns of decision-makers to different levels of urgency, ensuring that the quantification of urgency in this invention has higher fidelity.

[0058] The parameter impact analysis for the Dynamic Risk Transmission Factor (RCF) is as follows: The input parameter is the real-time load rate. When other parameters remain constant, an increase in the real-time load rate of a grid component will lead to a monotonically increasing Operating State Factor (OSF), which in turn leads to a piecewise nonlinear relationship in the calculated RCF. This design is based on the fundamental physical laws of power system stability. Components have sufficient margins under light loads and are insensitive to disturbances, resulting in a low level of risk transmission capability. As the load increases, the component gradually approaches its physical limits, stability decreases, and the risk transmission capability correspondingly increases. Using a piecewise linear model, the differentiated performance of the component's risk transmission capability under three different physical stages—normal, heavy load, and overload—can be simulated, making risk attribution more closely reflect physical reality.

[0059] When other parameters remain constant, an increase in the number of historical faults recorded for power grid components will lead to a monotonically increasing historical fault factor (HFF), which in turn will result in a logarithmic nonlinear relationship in the calculated dynamic risk transmission factor (RCF). Components that have historically experienced frequent faults have a higher probability of potential defects or aging problems, and therefore should be assigned a higher inherent risk weight in risk assessment. The use of a logarithmic function reflects the logic of diminishing marginal effects: the transition from no faults to faults represents a qualitative change, and the risk factor should increase significantly; while the transition from multiple faults to even more faults represents quantitative accumulation, and the increase in the risk factor should tend to level off. This design ensures that the attribution results are not dominated by a few problematic components, enhancing the balance and rationality of the overall assessment.

[0060] This embodiment addresses the direct power supply dependencies in semantic conflict paths and the shared resource dependencies defined in the knowledge graph. It employs a two-stage hierarchical attribution processing structure, enabling the final risk contribution distribution signal to distinguish and quantify two completely different sources of risk: causal risks arising from direct physical connections and associated risks arising from shared non-power resources (control systems, cooling facilities). Traditional single-stage attribution methods can only handle tree-like or linear causal chains, failing to reveal the complex network coupling relationships widely present in modern power grids, introduced by digitization and automation. For example, users on two different feeders might experience simultaneous power outages due to sharing the same faulty remote communication unit. The first attribution stage in this embodiment accurately locates direct power supply line problems, while the second attribution stage is specifically designed to identify such hidden common-cause fault patterns that transcend traditional physical topologies. This reveals deeper and more fundamental risk roots.

[0061] This embodiment provides a highly robust and adaptive method for optimizing and adjusting the operation mode for ensuring power supply to critical users. Its core technical feature lies in constructing an optimization model under normal operating conditions with the core objective of minimizing the risk contribution distribution signal generated by the preceding steps (S1-S3). A further improvement in this embodiment is the proposal of a multi-objective dynamic weighted fusion optimization mechanism based on multi-dimensional confidence assessment to address the potential decision failure or drastic fluctuations that may occur when the input data quality deteriorates in traditional methods. This mechanism continuously and in parallel evaluates the confidence scores of quantified risk data and semantically conflicting path data. Then, the main optimization objective (minimizing risk contribution) and two backup optimization objectives (minimizing logically conflicting paths and minimizing total risk) are dynamically weighted and fused to form a unified, smoothly changing composite optimization objective function. The weight of each objective is a non-linear continuous function of its corresponding input data confidence score, ensuring that the system decision-making can smoothly and proportionally adapt to gradual changes in data quality. A hierarchical instruction generation framework is also proposed. This framework generates a set of continuously adjustable control commands by solving the aforementioned composite optimization objective; it evaluates the residual risk level of the system after executing the set of commands, and only authorizes the generation of a set of discrete control commands with a wider impact range when the residual risk is still higher than a preset emergency intervention threshold, thus achieving the best balance between control costs and risk reduction effects.

[0062] Further explanation: In S4, the steps for generating optimized control instructions include: A mathematical model is constructed with the optimization objective of minimizing the sum of the contributions of each entity in the risk contribution distribution signal, and the mathematical model is solved to generate optimized control commands.

[0063] Further explanation: Obtain the first confidence score characterizing the credibility of the quantitative risk data, and the second confidence score characterizing the credibility of the semantic conflict path data; If either the first confidence score or the second confidence score is lower than a preset confidence threshold, the mathematical model for the optimization objective is automatically switched. The switching steps include: First, determine whether the first confidence score is lower than the confidence threshold. If so, switch the optimization objective to minimizing the number of semantically conflicting paths. Otherwise, further determine whether the second confidence score is lower than the confidence threshold. If so, switch the optimization objective to minimizing the sum of quantitative risk data.

[0064] Further explanation: The mathematical model for automatically switching optimization objectives specifically includes: Based on the first confidence score and the second confidence score, a set of target weights is calculated through a preset continuous weight mapping function. The set of target weights includes: the main target weight, the first backup target weight, and the second backup target weight. The weighted main objective is obtained by multiplying the sum of the risk contribution distribution signals to minimize the total risk contribution distribution signals by the weight of the main objective. The objective of minimizing the number of semantically conflicting paths is multiplied by the weight of the first backup objective to obtain the first weighted backup objective; The objective of minimizing the sum of quantitative risk data is multiplied by the weight of the second backup objective to obtain the second weighted backup objective; The weighted primary objective, the first weighted backup objective, and the second weighted backup objective are summed to construct a unified composite optimization objective function, so as to achieve smooth switching between different optimization objectives through the dynamic change of a set of objective weights.

[0065] Further explanation: The step of solving the mathematical model to generate optimized control instructions is a hierarchical instruction generation process, which includes: First instruction generation stage: Under the constraint that only continuous power grid control variables can be adjusted, solve the mathematical model with the composite optimization objective function as the objective to generate a set of continuous optimization control instructions; Risk assessment phase: Predict and evaluate the residual risk contribution distribution signal after executing continuous optimization control instructions, and analyze and process it to obtain the total residual risk value; Second instruction generation stage: Determine whether the total residual risk value is higher than the preset emergency intervention threshold. If the determination is yes, then under the constraint of allowing adjustment of discrete power grid control variables, a set of discrete optimization control instructions is further generated.

[0066] The following is a detailed implementation description of the above content: In the power supply guarantee decision support mechanism of modern power grids, the core challenge is how to address the unavoidable data quality issues in the real world while ensuring the accuracy of decision-making. Although the preliminary steps (S1-S3) can generate accurate risk contribution distribution signals, they rely on upstream quantitative risk assessment (S1) and knowledge graph reasoning (S2). Once the upstream data source suffers from data quality degradation due to sensor failure, communication interruption, or model failure, optimization decisions based on erroneous inputs (S4) not only fail to reduce risk but also trigger secondary disasters. Existing technologies employ a degradation strategy of hard switching by setting a fixed threshold, that is, once the data confidence falls below a certain value, the current optimization objective is completely abandoned, and another backup objective is switched to. The fundamental flaw of this approach is that when the data confidence fluctuates slightly around the threshold, it can cause the system optimization objective to jump drastically and frequently between different modes, resulting in discontinuous or even contradictory control commands, threatening the stable operation of the power grid. To solve this stability problem, this embodiment is based on an adaptive, multi-objective fusion optimization method of multi-dimensional confidence assessment and a hierarchical command generation framework. The computational process in this embodiment involves the following key parameters. All of these parameters are configured by reading external spreadsheet files, achieving technical decoupling between the core algorithm logic and specific application strategies.

[0067] The first confidence score, denoted as CSQ, is used to quantitatively characterize the current credibility of the quantitative risk data generated in step S1. It is a dimensionless value between 0 and 1, where 1 indicates complete credibility and 0 indicates complete uncredibility. The determination of the first confidence score stems from the assessment of data freshness. Specifically, the current system timestamp and the generation timestamp of each piece of quantitative risk data are obtained. The time difference between the two timestamps is calculated. This time difference is input into a preset, inverse logistic function for mapping. This function is configured such that: when the time difference is 0, the first confidence score outputs 1; as the time difference increases, the function output smoothly and non-linearly decreases; when the time difference exceeds 300 seconds (as indicated by the preset data validity period), the first confidence score output approaches 0.

[0068] The second confidence score, denoted as CSP, is used to quantify the current credibility of the semantic conflict path data generated in step S2. It is a dimensionless value between 0 and 1. The idea behind determining the second confidence score stems from the evaluation of the stability of the knowledge graph reasoning process. Specifically, the computation time consumed by the graph reasoning engine in generating all paths this time is recorded and obtained. This computation time is compared with the standard reasoning time benchmark value of 500 milliseconds obtained through historical statistics. The actual computation time is divided by the standard reasoning time benchmark value to obtain the timeout ratio. The timeout ratio is input into the inverse logistic function, which is similar to the confidence score mentioned above. This function is set such that when the timeout ratio is 1 or less, the second confidence score outputs 1; as the timeout ratio increases, the function output smoothly decreases.

[0069] It should be further explained that the core technical issue in evaluating data credibility in this embodiment is how to smoothly and non-linearly map physical quantities (time delay, computation time) to a standardized [0,1] interval representing credibility. A simple linear mapping cannot reflect the real-world logic of high sensitivity near key inflection points and low sensitivity at the extreme ends. For data delays of 1 second and 2 seconds, the decrease in credibility should be insignificant; however, for delays of 299 seconds and 300 seconds, with 300 seconds being the failure threshold, the credibility should decrease drastically. To address this issue, this embodiment employs a logistic function model parameterized as follows: The logistic function has the property of mapping any real number input to the interval (0,1), and its S-shaped curve can perfectly simulate the aforementioned nonlinear mapping requirements. By adjusting the midpoint position and the steepness of the curve, the transition point and transition rate of the mapping can be precisely controlled. In this embodiment, the calculation logic for determining the confidence score further includes the following steps: For the first confidence score (CSQ): obtain the time difference and divide it by the configuration parameter data validity period of 300 seconds to obtain the standardized time difference ratio.

[0070] For the second confidence score CSP: obtain the timeout ratio. The purpose of this step is to unify the different physical inputs into dimensionless standardized input values ​​so that a unified function model can be used subsequently.

[0071] A confidence decay factor is set to define the steepness of the function curve. This parameter controls the rate at which the confidence score decreases as the standardized input value increases. In this embodiment, the confidence decay factor is set to 10. The technical consideration for setting this value is to ensure that when the standardized input value is close to the failure boundary represented by the value 1, the slope of the function curve is large, which can produce a significant confidence discrimination effect; while when the input value is much smaller than 1, the curve is relatively flat, allowing for a certain tolerance to small performance fluctuations. The preferred value of the confidence decay factor is determined through the following experimental calibration method: a critical transition interval is defined. In this embodiment, this interval is set to the range of standardized input values ​​from 0.75 to 0.95. The technical meaning of this interval is that when the data quality deteriorates to this level, the most sensitive response is required. Secondly, a minimum discriminative slope is set, requiring that the absolute value of the average rate of change of the confidence function within this critical transition interval is not less than 5.0, to ensure effective discrimination. Subsequently, through numerical calculation experiments, starting from the integer 5, the value of the attenuation factor was gradually increased, and the average rate of change of the function within the critical transition interval was calculated for each value. Finally, the minimum integer value satisfying the aforementioned minimum distinguishability slope requirement was determined to be 10. Therefore, in this embodiment, the preferred value of the confidence attenuation factor is set to 10. The preferred range of the confidence attenuation factor is between 8 and 15. Finally, the obtained standardized input value is multiplied by the set confidence attenuation factor. The product result from the previous step is then inversely multiplied. The power of the product with the natural constant e as the base and the result from the previous step as the exponent is calculated. The sum of the power obtained in the previous step and the value of 1 is divided by the value of 1.

[0072] The emergency intervention threshold, denoted as Th1, is a dimensionless numerical value used to determine whether discrete, high-impact control commands need to be initiated. In this embodiment, the emergency intervention threshold is set to 0.8. The technical consideration behind this value is that 0.8 is a critical point determined by expert evaluation based on historical power grid operation data analysis. Below this risk level, it is considered that continuous measures such as power flow adjustment can still effectively control the situation; while above this level, it indicates that the risk has become more severe, and stronger intervention measures such as canceling maintenance plans must be considered to achieve a technical balance between risk and operating costs.

[0073] It should be further explained that the core objective of this calibration experiment is to find the optimal emergency intervention threshold through statistical analysis of historical power grid operation data. This emergency intervention threshold can most accurately distinguish between two historical scenarios: one is a risk event that can be successfully mitigated by continuous control measures (adjusting generator output); the other is a serious risk event that ultimately requires discrete intervention measures (including load shedding and canceling maintenance) to ensure power grid safety. Specifically, all recorded risk events that triggered power supply emergency responses within the past five years are selected from the power grid historical database (SCADA / EMS system). For each event, a structured data record is extracted, which must include: the sum of the risk contribution distribution signals calculated at the time of the event, recorded as the initial total risk value, and the type of final handling measure for the event, marked as either successful continuous control or discrete intervention. All data records are divided into two datasets based on their handling measure type: a continuous control group and a discrete intervention group. Within the range of 0 to 1, a test sequence containing 100 candidate thresholds (i.e., 0.01, 0.02, ..., 1.00) is generated with a step size of 0.01. For each candidate threshold, the following evaluation logic is executed: Sub-step 1: Calculate the True Positive Rate (TPR). In the discrete intervention group, count the number of records where the initial total risk value is greater than the current candidate threshold, and then divide this number by the total number of records in the discrete intervention group to obtain the True Positive Rate. This metric measures the model's ability to correctly identify events requiring strong intervention.

[0074] Sub-step 2: Calculate the False Positive Rate (FPR). In the continuous control group, count the number of records where the initial total risk value is greater than the current candidate threshold, and then divide this number by the total number of records in the continuous control group to obtain the false positive rate. This metric measures the degree to which the model misclassifies events that could have been handled mildly as requiring strong intervention. In this embodiment, the emergency intervention threshold is the point that achieves the optimal balance between the true positive rate (benefit) and the false positive rate (cost). For each candidate threshold, calculate a performance metric that equals the true positive rate minus the false positive rate. After iterating through all candidate thresholds, find the candidate threshold that maximizes this performance metric as the emergency intervention threshold.

[0075] Furthermore, this embodiment dynamically weights the three independent optimization objectives using confidence scores, fusing them into a single, continuously changing composite optimization objective function. The optimization objective is no longer a choice between three discrete options, but rather a continuous pursuit of a dynamic balance among these three objectives. Specifically: Obtain the first confidence score CSQ and the second confidence score CSP obtained from the aforementioned calculation.

[0076] Calculate the main target weight Wmain: Perform a multiplication operation between CSQ and CSP to obtain a comprehensive confidence level.

[0077] Calculate the weight Wfb1 of the first backup target: Subtract CSQ from the value 1 to obtain its confidence level. Then, multiply this confidence level with CSP. This demonstrates the activation of the first backup target, primarily in scenarios where quantitative risk data is unreliable, but logical path data is still reliable.

[0078] Calculate the second backup target weight Wfb2: Subtract CSP from the value 1 to obtain its confidence level. Then, multiply this confidence level with CSQ. This logic corresponds to scenarios where the logical path is unreliable, but the quantified risk is still credible.

[0079] A composite optimization objective is constructed as follows: the sum of the risk contribution distribution signals output by S3 is multiplied by Wmain; the number of semantically conflicting paths output by S2 is multiplied by Wfb1; and the sum of the quantified risk data output by S1 is multiplied by Wfb2. Finally, the results of these three multiplications are summed to obtain the final, unified composite optimization objective function value. This composite optimization objective function will be input into the subsequent solver for minimization. When the confidence level of all data is 1, Wmain is 1, and the remaining weights are zero, focusing on the primary optimization objective. As the confidence level of the data source smoothly decreases, the weight of its corresponding backup objective smoothly increases, while the weight of the primary objective smoothly decreases. This allows for a seamless and stable shift in the system's decision-making focus, eliminating decision jitter and enhancing the system's stability and adaptability to changes in data quality.

[0080] Furthermore, the calculation logic for steps S4 and S5 in this embodiment is as follows: Quantitative risk data from S1, semantic conflict path data from S2, and risk contribution distribution signal from S3 are received in parallel. Simultaneously, the first confidence score CSQ and the second confidence score CSP are calculated in real time.

[0081] A multi-objective dynamic weighted fusion algorithm is executed to dynamically weight and fuse three potential optimization objectives into a single, optimal composite optimization objective function under the current operating conditions using CSQ and CSP. An optimization solver based on simulated annealing is then invoked. All continuously adjustable control variables in the power grid (generator active power output adjustment, transformer tap position) are used as optimization variables, and their safe adjustment ranges are set. Discrete control variables (switching states) are locked at this stage and cannot be changed. The solver iteratively optimizes under the set constraints, aiming to minimize the constructed composite optimization objective function, ultimately generating a set of optimal continuous optimization control commands. A built-in fast simulation module predicts the power flow distribution and operating state of the power grid after executing the generated continuous commands. Based on this predicted state, the residual risk contribution distribution signal is recalculated.

[0082] The contribution values ​​of each entity in the residual risk contribution distribution signal are summed to obtain a total residual risk value. This total value is then compared with the emergency intervention threshold Th1. If the total residual risk value is higher than Th1, the second-level decision logic is initiated. This logic selects the discrete optimization control command that can most effectively reduce the risk of the current highest risk contribution point from a preset high-impact operation library (including canceling maintenance on line A, disconnecting unnecessary loads from user B, etc.), and merges it with the command from the first stage. If the value is lower than the emergency intervention threshold Th1, no additional command is generated in this stage.

[0083] The high-impact operation library is a structured collection of parameters used to define all optional discrete power grid control variables. Preferably, the library is stored as a database table or a JSON file, where each record represents an independent, executable high-impact operation and contains the following standardized data fields: Operation ID: A unique string used for program calls.

[0084] Operation description: Natural language description used for interface display, such as canceling the scheduled maintenance of Line 8.

[0085] Target entity set: An array containing the IDs of the grid-side entities directly affected by this operation.

[0086] Expected Risk Reduction Coefficient: A floating-point number between [0,1] representing the proportion of residual risk expected to be eliminated after performing this operation. This coefficient value is pre-calibrated based on offline simulation calculations and expert experience.

[0087] Execution cost: A standardized value that comprehensively reflects the economic cost of performing the operation, the negative impact on users, and other factors.

[0088] The high-impact operation library is compiled by power grid experts based on operating procedures and dispatch plans. During the second instruction generation phase, the solver of the mathematical model reads this high-impact operation library. Its optimization objective is to select a combination of one or more operations from the library, while ensuring that the sum of the final predicted residual risks is lower than the emergency intervention threshold, and to minimize the total execution cost of the selected operations.

[0089] Furthermore, the finalized optimized control commands, including continuous and potentially discrete types, the risk contribution distribution signal before execution, the residual risk contribution distribution signal after execution, and the semantic conflict path data that plays a key role in this decision-making process, are packaged and integrated into a structured decision support visualization signal for front-end interface display, and then output. This visualization signal allows the dispatcher to clearly see where the risks lie, how the system recommends responding, and the expected effects of the responses, thus completing the decision-making loop. The specific processing flow for generating the decision support visualization signal is further explained below: The system receives two input data: the first is a risk contribution distribution signal, which is a list indexed by the unique identifier of the power grid-side entity and containing the risk contribution value; the second is semantic conflict path data, which is a set of multiple conflict paths, where each path records at least two conflicting entity identifiers and conflicting text descriptions.

[0090] The processing flow executes the following core calculation steps: The risk contribution distribution signal list is sorted in descending order, and the five grid-side entities with the highest risk contribution values ​​are selected and marked as core risk sources. These five core risk source entities are traversed, and using each entity's unique identifier as an index, a search is performed in the semantic conflict path data set to identify all conflict paths related to that core risk source entity. A hierarchical data structure is constructed, with each core risk source entity and its risk contribution value as a parent node, and all associated conflict path information (including the identifier of the other party in the conflict and the conflict description) as its child nodes. The aforementioned hierarchical data structure is serialized into a JSON-formatted text string, serving as the final output decision support visualization signal. This structured signal is directly parsed by the front-end application and used to dynamically generate a risk source tracing topology map or hierarchical list on the graphical user interface, thereby providing operators with intuitive and clear decision support.

[0091] To achieve technical decoupling between the core algorithm and specific application strategies of this invention, and to ensure the configurability and ease of debugging of the technical solution, the operational configuration parameters defined in this invention, such as data validity period, standard inference time baseline value, and confidence decay factor, are all pre-configured and stored in a structured external data carrier. This data carrier can take various forms in different implementation environments. During the initialization of this invention or the entry into a new operational cycle, a parameter loading module is executed. The function of this module is to access the aforementioned external data carrier according to a preset path or address, parse the data content according to a predetermined format, retrieve and extract a set of currently effective operational configuration parameter values, and load these values ​​into the system's working memory for direct use by subsequent algorithm modules during computation.

[0092] The following detailed implementation instructions apply to the above content: In this technical solution, the final optimization effect is quantitatively characterized by the total residual risk value. This value is a dimensionless value normalized to the [0,1] interval, obtained by predicting and evaluating the power grid state after executing the optimization control command.

[0093] The smaller the total residual risk, the more thorough the reduction effect of the generated optimized control command on the initial risk, and the more the power grid operating state is adjusted to a safer and more stable region. When the total residual risk value approaches 0, it indicates that the generated optimized control commands have a more thorough effect on reducing the initial risk, and the power grid operation state is adjusted to a safer and more stable region. Conversely, when the total residual risk value approaches 1, it indicates that although optimized control has been implemented, the system is still in a higher risk state, meaning that the severity of the initial risk is more likely to exceed the capacity limit of conventional continuous regulation methods, requiring a higher level of intervention.

[0094] The first confidence score (CSQ) and the second confidence score (CSP) together determine the composition of the composite optimization objective function, and their numerical changes have an indirect but decisive impact on the final total residual risk.

[0095] When other parameters remain constant, a decrease in the first confidence score (CSQ) will lead to a decrease in the weight of the primary weighted objective (minimizing risk contribution) in the composite optimization objective function, while simultaneously increasing the weight of the first weighted backup objective (minimizing semantic conflict paths). This is a negatively correlated adjustment. The underlying logic of this design is that CSQ characterizes the credibility of quantified risk data, and the primary optimization objective of risk contribution is calculated based on this data. When CSQ decreases, it means that the computational basis of the primary optimization objective is no longer reliable. At this point, the pursuit of the unreliable objective is automatically reduced, and greater emphasis is placed on the backup objective of semantic conflict paths generated from relatively reliable topological data. This ensures that when part of the input data source fails, the decision-making basis can be intelligently and smoothly migrated to a more reliable information dimension, avoiding the risk of making incorrect decisions based on erroneous data.

[0096] When other parameters remain constant, a decrease in the second confidence score (CSP) will also lead to a decrease in the weight of the weighted primary objective, but an increase in the weight of the second weighted backup objective (minimizing the sum of quantified risk data). CSP characterizes the credibility of semantically conflicting path data. When it decreases, the confidence in both the primary objective based on topological inference and the first backup objective decreases. At this point, it reverts to a more basic, albeit imprecise, but relatively robust optimization objective: directly minimizing the original sum of quantified risk data without complex attribution analysis. This constitutes a second layer of degradation protection, ensuring that even if topological inference fails, basic optimization actions aimed at reducing the overall risk level can still be performed.

[0097] To quantitatively verify the superiority of the technical solution of this invention, a series of technical effect verifications were conducted in a power grid model environment based on the IEEE-14-node standard test system. This section will specifically verify the two core innovations: the multi-objective dynamic weighted fusion optimization mechanism and the hierarchical instruction generation and processing. To verify the superiority of the multi-objective dynamic weighted fusion optimization mechanism based on multi-dimensional confidence assessment described in this invention, a control group (existing technology) was set up. This group adopted hard switching logic: when the confidence score is lower than a fixed threshold (0.5), the optimization objective is immediately and completely switched to the backup objective. The group of this invention adopts the dynamic weighted fusion mechanism described above, as shown in Table 1 below: Table 1: Performance Comparison and Verification Data of the Invention and Existing Technologies

[0098] The risk reduction rate, denoted as RRR, is calculated by obtaining the initial total risk value before optimization and the residual total risk value after optimization, and then performing the following calculations: Subtract the residual total risk value from the initial total risk value to obtain the absolute amount of risk reduction; then divide this absolute amount by the initial total risk value to obtain a ratio; finally, multiply this ratio by 100 and express it as a percentage. This parameter is used to quantitatively characterize the degree to which the optimized control command eliminates the initial risk faced by the system in a single execution. The higher the risk reduction rate, the more effective the technical solution of this invention.

[0099] Control command volatility, denoted as CCF, is a series of continuous control commands output by the system, including generator active power adjustment values, recorded in a scenario involving continuous time series. Specifically, the absolute value of the difference between any two consecutive commands is calculated; then, the standard deviation of this difference sequence is calculated. Control command volatility is used to quantitatively characterize the stationarity of the control system's output commands. A lower control command volatility indicates a smoother and more stable control process, less impact on power equipment, and stronger system robustness.

[0100] Scenario 2 (Quantitative Data Downgrading) Comparative Analysis: In the two experiments of Scenario 2, when the first confidence score (CSQ) decreased to 0.20 to 0.21, the risk reduction rate of the invention group reached 62.20% and 63.86%, while that of the control group was only 45.12% and 46.99%. The risk reduction effect of the invention group was 37.86% higher than that of the control group. This data strongly demonstrates that the dynamic weighting mechanism of the present invention can proportionally integrate backup optimization targets when unreliable quantitative data is detected, thereby achieving far superior risk control effects even in downgrading mode.

[0101] Scenario 3 (Confidence Fluctuation) Comparative Analysis: In Scenario 3, the CSQ fluctuates around the hard switching threshold of 0.5. Data shows that the control command fluctuation of the control group is as high as 1.85 and 1.92, while that of the present invention group is stable at 0.11 and 0.10. The control command stability of the present invention group is improved by 94% compared with the control group. This is because the continuous weight mapping function of the present invention ensures that even if the input confidence fluctuates slightly around the threshold, the change of its composite optimization objective is continuous and smooth, thereby completely eliminating system chattering and enhancing the stability and safety of the control system.

[0102] Furthermore, in the following scenario four, a severe N-2 cascading fault occurs in the power grid, causing the initial total risk value to far exceed the conventional level. Control group A (existing technology): employs hard switching logic and only has continuous control measures. Control group B (intermediate solution): employs the dynamic weighted fusion mechanism of this invention, but also only has continuous control measures and lacks a second instruction generation stage. This invention group employs complete dynamic weighted fusion and hierarchical instruction generation processing. See Table 2 below for details: Table 2: Performance Verification of the Hierarchical Instruction Generation Framework under Extreme Risk Scenarios

[0103] Scenario 4 (Extreme Risk Event) Comparative Analysis: Under extreme conditions where the initial risk was as high as 0.95, control groups A and B, which could only perform continuous control, had final residual risk values ​​as high as 0.86 and 0.83 respectively, failing to suppress the risk below the emergency intervention threshold (0.8). In contrast, after completing the first stage of continuous control, the risk assessment stage determined that the residual risk was still too high, and the second instruction generation stage was initiated. By generating discrete control instructions (including the removal of some secondary loads), the residual risk value was successfully reduced to 0.15.

[0104] Comparing the present invention group (residual risk 0.15) with the superior control group B (residual risk 0.83), the final risk level of the present invention group was reduced by 81.9% compared to control group B. The calculation logic for this reduction effect is as follows: obtain the difference between the residual risk value of control group B and the residual risk value of the present invention group; then, divide this difference by the residual risk value of control group B; finally, multiply the resulting quotient by 100. This data demonstrates that when facing severe risk events, optimizing continuous control variables alone is insufficient. The discrete intervention capability provided by hierarchical instruction generation processing is a key technical means to ensure that the power grid ultimately returns to a safe state under extreme operating conditions.

[0105] Furthermore, in order to transform the total residual risk value calculated by this invention into operations that can be directly understood and executed by dispatching and operation personnel, this embodiment defines the following three risk level ranges and their corresponding technical operations based on expert experience and the above-mentioned technical effect verification data.

[0106] Within a safe and controllable range, the total residual risk value ∈ [0, 0.3] indicates that the initial risk has been effectively suppressed through continuous optimization control instructions, the system is currently operating stably, and there is no imminent threat. No additional intervention is required. The entire process of this optimization event is automatically recorded, including the initial risk, the generated instructions, and the final residual risk value, serving as data accumulation for subsequent analysis and model optimization.

[0107] The key monitoring area, with a residual total risk value ∈ (0.3, 0.8], indicates that continuous control measures have been implemented to the best of their ability, but have failed to completely eliminate the risk, and the area remains in a state requiring vigilance. Although this state has not triggered the emergency intervention threshold, the potential risk cannot be ignored. Relevant risk sources and key users are highlighted on the decision support interface, and operators are advised on the recommended level of attention. Dispatchers should closely monitor the power grid operation status of this area based on the actual situation.

[0108] In the emergency intervention zone, a residual total risk value > 0.8 indicates a severe initial risk that cannot be suppressed to an acceptable level by continuous control measures alone. This automatically triggers and generates a discrete optimized control command (cancel the maintenance task). The highest-level emergency intervention alarm is issued to operators, clearly displaying the generated discrete control command. This command requires final confirmation from operators before execution to ensure the safety of human-machine collaborative decision-making. Simultaneously, the system automatically generates a detailed event report, recording all relevant data for accident tracing.

[0109] The computational logic involved in this application can be constructed using algorithms such as regression analysis in machine learning, establishing a mathematical model by analyzing the inherent trends and interrelationships of the collected parameters. This process can be implemented using specialized computational tools (such as Python's Scikit-learn library or the R language environment). Throughout all calculations, to eliminate the influence of different physical dimensions and ensure that data is compared and analyzed on the same scale, the input parameters in each formula are dimensionless. The dimensionless techniques used include, but are not limited to, max-min normalization or Z-score standardization.

[0110] The algorithm of this invention is implemented as a Python script. Before executing the core logic, the program first executes a data loading module (e.g., using the widely used pandas library in Python) configured to read the aforementioned spreadsheet file and load its contents into the program's working memory (e.g., a DataFrame data structure). Subsequent algorithm steps will directly query and retrieve the required configuration parameters from this in-memory data structure.

[0111] It should be emphasized that the foregoing embodiments are merely illustrative of preferred implementations of the present invention and are not intended to limit the scope of protection of the present invention. This application also provides a computer-readable storage medium having computer program instructions stored thereon.

[0112] Example 2: Based on the same inventive concept as the embodiments, this embodiment 2 provides a path optimization and operation mode adjustment system for ensuring power supply to important users, such as... Figure 4 As shown, it includes: The first acquisition module is used to acquire quantitative risk data representing potential business interruption losses of important users under a preset power grid operation mode; The second acquisition module is used to acquire semantic conflict path data that characterizes the logical conflict between the physical state of the power grid and the needs of users' business activities, based on a pre-set knowledge graph containing power grid physical entities and user business activity entities. The attribution processing module is used to perform backpropagation attribution processing on the knowledge graph along the semantic conflict path data of the quantified risk data, so as to generate a risk contribution distribution signal that represents the contribution of each entity in the knowledge graph to the quantified risk data. The instruction generation module is used to generate optimized control instructions for adjusting the preset power grid operation mode based on the risk contribution distribution signal. The visualization generation module is used to generate a decision support visualization signal to present the risk tracing results based on the risk contribution distribution signal and the semantic conflict path data.

Claims

1. A method for optimizing power supply paths and adjusting operation modes for critical users, characterized in that, The specific steps include: Obtain quantitative risk data characterizing potential service interruption losses for key users under preset power grid operation modes; Based on a pre-built knowledge graph containing power grid physical entities and user business activity entities, semantic conflict path data representing the logical conflict between the power grid physical state and user business needs is obtained. The quantified risk data, along with the semantic conflict path data, is backpropagated and attributed on the knowledge graph to generate a risk contribution distribution signal that characterizes the contribution of each entity in the knowledge graph to the quantified risk data. Based on the risk contribution distribution signal, an optimized control command is generated to adjust the preset power grid operation mode; Based on the risk contribution distribution signal and the semantic conflict path data, a decision support visualization signal is generated to present the risk tracing results.

2. The method for optimizing the power supply path and adjusting the operation mode for important users according to claim 1, characterized in that, The steps for obtaining the semantic conflict path data include: In the knowledge graph, the graph reasoning engine identifies dependency paths from grid-side entities to user-side business activity entities that have state conflicts or constraint violations under the preset power grid operation mode, and these paths are used as semantic conflict path data. The quantitative risk data is a multidimensional risk vector, which includes at least: the potential economic impact representing potential economic losses, and the time urgency index representing the timeliness requirements of business interruption.

3. The method for optimizing the power supply path and adjusting the operation mode for important users according to claim 2, characterized in that, Before performing the backpropagation attribution process, a dynamic risk transmission factor is calculated in real time based on the real-time operating status data and historical fault data of the power grid-side physical entities in the knowledge graph. The backpropagation attribution processing is based on the dynamic risk transmission factor and is performed along the semantic conflict path data.

4. The method for optimizing the power supply path and adjusting the operation mode for important users according to claim 3, characterized in that, The backpropagation attribution process is a hierarchical attribution process. When the hierarchical attribution process is executed on the semantic conflict path data, it includes: First attribution stage: Along the direct power supply dependencies in the semantic conflict path data, perform the first round of backpropagation attribution processing to obtain the direct causal risk contribution of each entity; The second attribution stage: Based on the shared resource dependencies defined in the knowledge graph, the direct causal risk contribution of multiple entities with shared upstream dependencies identified in the first attribution stage is corrected for related impacts to obtain the final risk contribution distribution signal.

5. The method for optimizing the power supply path and adjusting the operation mode for important users according to claim 4, characterized in that, The steps for generating the optimized control instructions include: A mathematical model is constructed with the optimization objective of minimizing the sum of the contributions of each entity in the risk contribution distribution signal, and the mathematical model is solved to generate the optimized control command.

6. The method for optimizing the power supply path and adjusting the operation mode for important users according to claim 5, characterized in that, Obtain a first confidence score characterizing the credibility of the quantitative risk data, and a second confidence score characterizing the credibility of the semantic conflict path data; If either the first confidence score or the second confidence score is lower than a preset confidence threshold, the mathematical model of the optimization objective is automatically switched. The switching steps include: First, determine whether the first confidence score is lower than the confidence threshold. If so, switch the optimization objective to minimizing the number of semantically conflicting paths. Otherwise, it is further determined whether the second confidence score is lower than the confidence threshold. If so, the optimization objective is switched to minimizing the sum of quantitative risk data.

7. The method for optimizing the power supply path and adjusting the operation mode for important users according to claim 6, characterized in that, Automatically switching the mathematical model of the optimization objective specifically includes: Based on the first confidence score and the second confidence score, a set of target weights is calculated using a preset continuous weight mapping function. The set of target weights includes: the main target weight, the first backup target weight, and the second backup target weight. The weighted main objective is obtained by multiplying the sum of the risk contribution distribution signals that minimizes the risk contribution distribution by the weight of the main objective. The objective of minimizing the number of semantically conflicting paths is multiplied by the weight of the first backup objective to obtain the first weighted backup objective; The objective of minimizing the sum of quantified risk data is multiplied by the weight of the second backup objective to obtain the second weighted backup objective; The weighted primary objective, the first weighted backup objective, and the second weighted backup objective are summed to construct a unified composite optimization objective function, so as to achieve smooth switching between different optimization objectives through the dynamic change of a set of objective weights.

8. The method for optimizing the power supply path and adjusting the operation mode for important users according to claim 7, characterized in that, The steps of solving the mathematical model to generate optimized control instructions include a hierarchical instruction generation process, which includes: First instruction generation stage: Under the constraint that only continuous power grid control variables can be adjusted, solve the mathematical model with the composite optimization objective function as the objective to generate a set of continuous optimization control instructions; Risk assessment phase: Predict and evaluate the residual risk contribution distribution signal after executing the continuous optimization control command, and analyze and process it to obtain the total residual risk value; Second instruction generation stage: Determine whether the total residual risk value is higher than the preset emergency intervention threshold. If the determination is yes, then under the constraint of allowing adjustment of discrete power grid control variables, a set of discrete optimization control instructions is further generated.

9. The method for optimizing the power supply path and adjusting the operation mode for important users according to claim 8, characterized in that, The smaller the total residual risk, the more thorough the reduction effect of the generated optimized control commands on the initial risk, and the more the power grid operating state is adjusted to a safer and more stable region.

10. A path optimization and operation mode adjustment system for ensuring power supply to important users, characterized in that, include: The first acquisition module is used to acquire quantitative risk data representing potential business interruption losses of important users under a preset power grid operation mode; The second acquisition module is used to acquire semantic conflict path data that characterizes the logical conflict between the physical state of the power grid and the needs of users' business activities, based on a pre-set knowledge graph containing power grid physical entities and user business activity entities. The attribution processing module is used to perform backpropagation attribution processing on the knowledge graph along the semantic conflict path data of the quantified risk data, so as to generate a risk contribution distribution signal that represents the contribution of each entity in the knowledge graph to the quantified risk data. The instruction generation module is used to generate optimized control instructions for adjusting the preset power grid operation mode based on the risk contribution distribution signal. The visualization generation module is used to generate a decision support visualization signal to present the risk tracing results based on the risk contribution distribution signal and the semantic conflict path data.

Images