Power distribution network power failure probability analysis and resource scheduling optimization method based on knowledge graph

Abstract

The invention relates to the technical field of smart power grids, and provides a power distribution network power failure probability analysis and resource scheduling optimization method based on a knowledge graph, which realizes minute-level node association weight updating through a streaming updating mechanism, and remarkably improves the power failure risk identification accuracy in extreme weather by combining with dynamic adjustment of a meteorological risk function. A topology / environment double-attention architecture is adopted to fuse multi-source features, so that the fusion efficiency of equipment node features and environment node features is improved; and a fault deduction model with dynamic weight constraint is introduced, so that the secondary fault early warning capability is enhanced. In a resource scheduling link, a mixed integer programming model of a carbon emission factor and island power balance constraint is integrated, and Pareto optimization of first-aid repair efficiency and an environmental protection target is realized. Through transfer learning, knowledge graph cross-region adaptation is realized, and the novel transformer area access error is controlled to be reduced in a distribution network reconstruction scene. The method supports minute-level topology reconstruction and second-level weight updating, has low delay and high expansion capability, and is suitable for provincial power grid real-time deployment.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent operation and maintenance of power systems, and in particular to a method for power outage probability analysis and resource scheduling optimization in distribution networks based on knowledge graphs. Background Art

[0002] With the continuous advancement of the construction of new power systems, the distribution network has gradually shown characteristics such as a high proportion of distributed new energy access and a rapid growth in electric vehicle charging load. The overall structure and operation mode are becoming increasingly complex. Currently, the management model based on manual experience or static rules is still widely used in operation and maintenance work. When faced with a dynamically changing operating environment, the following problems exist:

[0003] Difficulty integrating multi-source data and weak dynamic modeling capabilities: Data from multiple sources, such as distribution network equipment status, meteorological conditions, and user loads, is highly heterogeneous, making it difficult for existing methods to achieve real-time dynamic integration. For example, equipment records, fault logs, and sensor data are often scattered across independent systems, creating information silos. This results in outage risk prediction relying on offline historical data, making it unable to adapt to dynamic changes such as network reconfiguration and extreme weather.

[0004] Fault identification relies on fixed assumptions and has limited accuracy: Traditional methods such as impedance and traveling wave methods rely on fixed topological assumptions and lack a dynamic characterization of fault propagation mechanisms within the network. This results in insufficient accuracy for complex or hidden faults and is unable to effectively capture hidden fault propagation paths within complex networks. While the recently introduced knowledge graph method can assist in building relationships between devices, its ability to model spatiotemporal features is limited, and the accuracy of entity recognition and relationship reasoning is only around 90%, making it difficult to support high-precision power outage probability assessment.

[0005] Dispatch strategies are limited in scope: Existing operations and maintenance decisions often focus on optimizing repair timelines, ignoring constraints such as carbon emissions and skill matching. For example, emergency dispatch models often prioritize the shortest path, failing to consider the carbon efficiency differences between electric and fuel-powered vehicles or the island support capabilities of distributed power sources. This makes it difficult to balance resource utilization with environmental benefits.

[0006] Weak algorithm generalization capabilities: AI models based on statistical learning (such as traditional GCNs) require large amounts of labeled data. However, distribution network fault samples are scarce and unevenly distributed, making generalization performance degraded, especially in scenarios with topological changes. Furthermore, existing knowledge graph update mechanisms lag and cannot achieve minute-level dynamic reconstruction, limiting the model's adaptability to real-time operating conditions.

[0007] Therefore, in response to the above problems, there is an urgent need for a distribution network outage probability analysis and resource scheduling optimization method based on knowledge graph to meet the efficient operation and maintenance needs of the distribution network under the new power system. Summary of the Invention

[0008] The purpose of this invention is to address the deficiencies of the existing technology and provide a distribution network outage probability analysis and resource scheduling optimization method based on knowledge graph to meet the efficient operation and maintenance needs of the distribution network under the new power system.

[0009] In order to achieve the above object, the present invention adopts the following technical solutions:

[0010] A method for power outage probability analysis and resource scheduling optimization in a distribution network based on a knowledge graph includes the following steps:

[0011] S1. Build a dynamic knowledge graph: Map distribution network devices into graph nodes, map the connection relationships between distribution network devices into weighted edges, set edge weights, build a dynamic spatiotemporal association matrix, and define dynamic association weights between nodes;

[0012] S2. Identify weak links in the distribution network based on a graph neural network: Based on a graph neural network model, perform feature extraction and reasoning analysis on the dynamic knowledge graph in S1, evaluate the comprehensive power outage probability of each node, and identify weak links in the distribution network accordingly;

[0013] S3. Based on the identification results in S2, perform multimodal resource scheduling optimization, establish a resource map of the power supply station, dynamically calculate the arrival time of the emergency repair team, build a multi-objective mixed integer programming model, and use a distributed solution algorithm to optimize the scheduling plan.

[0014] Preferably, the distribution network equipment attributes in S1 include capacity C i , operating life T i , real-time load rate L i (t).

[0015] Preferably, the dynamic spatiotemporal association matrix association weight in S1 is calculated by the following formula:

[0016]

[0017] in, is the geographic-electrical coupling term; d ij is the geographical distance between node i and device j; θ ij is the electrical phase angle difference between nodes i and j, in radians; F i , F j are the historical failure frequencies of nodes i and j respectively; γ·Corr(F i , F j ) is the historical fault correlation; Corr(F i , F j ) is the Pearson correlation coefficient; μ·WeatherRisk(i, j, t) is the meteorological risk term; λ d∈[0.1, 0.5] is the distance attenuation coefficient; WeatherRisk(i, j, t) is the meteorological risk value calculated based on lightning density, wind speed and precipitation intensity; λ d ∈[0.1, 0.5], γ∈[0.2, 0.8], μ∈[0.3, 1.0] are adjustable parameters.

[0018] Preferably, the calculation formula of WeatherRisk(i, j, t) is:

[0019]

[0020] Among them, Lightning t Real-time lightning density, unit: times / square kilometer hour;

[0021] Wind t is the real-time wind speed, unit: m / s;

[0022] S is the meteorological risk adjustment threshold, with a default value of 1.2;

[0023] When Lightning t When the frequency is >1010 times / km² / hour, the insulation margin detection of distribution network equipment will be automatically triggered.

[0024] Preferably, the S1 also includes a streaming update mechanism to update the real-time load rate L of the distribution network equipment. i (t) is updated, and the update formula is as follows:

[0025]

[0026] Among them, T window =5 minutes, is the sliding window size, ∈=0.2 is the burst load adjustment factor.

[0027] Preferably, S2 includes the following sub-steps:

[0028] S2.1. Construct a heterogeneous graph, add the environment node as a virtual node, and establish an association edge between the virtual node and the power distribution equipment node;

[0029] The distribution network equipment nodes are:

[0030] h i =[DHI(i), L i (t), WeatherRisk(i)];

[0031] Wherein, DHI(i) is the distribution network equipment health index;

[0032] The health index DHI(i) is calculated by the following formula:

[0033]

[0034] Among them, β = 0.05 is the health sensitivity coefficient. The larger the value, the more significant the impact of MTBF deviation from the mean; MTBF i is the mean time between failures of the equipment;

[0035] Environmental nodes are weather stations and user repair hotspots, and are connected by edge weights W env (i) = 1 - e -0.2·ComplaintRate(i) Connected to the distribution network equipment node;

[0036] S2.1. Construct a heterogeneous graph, add the environment node as a virtual node, and establish an association edge between the virtual node and the power distribution equipment node;

[0037] The distribution network equipment nodes are:

[0038] h i =[DHI(i), L i (t), WeatherRisk(i)];

[0039] Wherein, DHI(i) is the distribution network equipment health index;

[0040] The health index DHI(i) is calculated by the following formula:

[0041]

[0042] Among them, β = 0.05 is the health sensitivity coefficient. The larger the value, the more significant the impact of MTBF deviation from the mean; MTBF i is the mean time between failures of the equipment;

[0043] Environmental nodes are weather stations and user repair hotspots, and are connected by edge weights W env (i) = 1 - e -0.2·ComplaintRate(i) Connected to the distribution network equipment node;

[0044] S2.2. Construct a graph neural network, whose structure includes:

[0045] The first layer is the topological attention layer, which models the electrical-geographic relationship. The topological attention coefficient is:

[0046]

[0047] in, is the topological attention coefficient of device node i to neighbor node j; W1 is the trainable weight matrix of the topological attention layer; h i is the characteristic vector of the device node; h j is the feature vector of neighbor node j, with the same structure as h i Consistent; / / represents vector concatenation operation;

[0048] The second layer is the environmental attention layer, which integrates the influence of environmental nodes. The environmental attention coefficient is:

[0049]

[0050] in, is the environmental attention coefficient of device node i to environment node k; W2 is the trainable weight matrix of the environmental attention layer, independent of W1; E(i) is the set of environmental nodes associated with device node i; h k is the feature vector of environment node k associated with device node i, and h in the denominator m is the feature vector of any environment node m in the set E(i);

[0051] S2.3. Calculate the comprehensive power outage probability of each device node. The specific formula is:

[0052]

[0053] Among them, h j , h k is the node feature vector; P outage (i) is the comprehensive power outage probability of device i; σ is the Sigmoid activation function; W3 and W4 are trainable weight matrices; b is the bias term; N(i) is the set of directly electrically connected neighbor nodes of device i; E(i) is the environmental node associated with device i.

[0054] Preferably, S2 also includes fault propagation deduction based on the comprehensive power outage probability and load pressure to identify secondary risk nodes, and the formula is as follows:

[0055]

[0056] Among them, P cascade is the fault propagation probability of node i to node j, which indicates the probability of causing a chain power outage at node j when a fault occurs at node i; σ is the Sigmoid activation function; w p , w q is the dynamic weight coefficient; w p is the power outage probability impact weight; w q is the load pressure influence weight; P outage (j) is the comprehensive power outage probability of node j; L j (t) is the real-time load rate of node j, which is dynamically updated through the sliding window; c j is the rated capacity of node j, which is the inherent attribute parameter of the equipment and is used to calculate the ratio of load pressure to capacity; when P cascade When >0.7, node j is marked as a secondary failure risk point.

[0057] Preferably, the step S3 includes the following sub-steps:

[0058] S3.1. Establish a resource map for power stations, with node attributes including personnel skills, vehicle types, and tool inventory;

[0059] S3.2. Evaluate resource accessibility and update arrival time based on real-time traffic conditions. The evaluation formula is:

[0060]

[0061] Among them, T reach (m) is the estimated total time for resource m to reach the failure point; d m is the geographical distance from the current location of resource m to the fault point; v m is the benchmark driving speed of resource m, representing the theoretical average speed under the vehicle type or road speed limit constraint; ρ m is the real-time road congestion rate on the route of resource m, which quantifies the degree of traffic flow obstruction and has a value range of 0≤ρ m ≤1, where ρ m =0 means no congestion, ρ m =1 means complete blockage; δ 拥堵 The detour flag, 0 means no detour, 1 means detour; t detour The additional time increment caused by the detour reflects the combined impact of the mileage and speed of the alternative route;

[0062] S3.3. Construct a multi-objective optimization model as follows:

[0063]

[0064] Among them, T total (m) is the total scheduling time of resource m; C(m) is the scheduling cost of resource m; x m is the scheduling decision variable, 0 means no scheduling, 1 means scheduling; w1 and w2 are the weight coefficients of emergency repair timeliness and carbon emissions, which must satisfy w1+w2=1;

[0065] (Skill matching constraints);

[0066] Among them, s m is the skill matching matrix of resource m, with the dimension of [skill 1, skill 2, ...], where the value is 1 if the skill is available, and 0 otherwise; r 需求 =[1,0,1] means high voltage certificate and relay protection debugging skills are required;

[0067] (Island power balancing);

[0068] Among them, P gis the rated output of distributed power source g; y g is a binary variable, indicating whether the distributed power supply g is enabled; ΔP load is the load shortage caused by the fault; δ island ∈{0, 1} is the island operation flag, where 1 indicates island mode;

[0069]

[0070]

[0071] Among them, C total is carbon emissions; α m Carbon emission factor of vehicle type, fuel vehicle α = 2.3 kg / km, electric vehicle α = 0; β is the carbon emission coefficient of tool use (unit: kg / hour); t repair，m is the on-site maintenance time of resource m; x m is a binary scheduling decision variable, indicating whether to schedule resource m; C max The carbon emission cap for a single dispatch;

[0072] S3.4. Use distributed solution algorithm to optimize the scheduling plan:

[0073] Use K-means to cluster fault points and generate regionalized scheduling sub-problems;

[0074] Each subproblem is solved using the branch and bound method, and the main problem coordinates the global constraints through Lagrangian relaxation;

[0075] The iteration is terminated when the global carbon emission change converges to the threshold ε

[0076] Preferably, the S3 further includes collaborative scheduling of virtual power plants:

[0077] Step a: Define the distributed power supply support capability index:

[0078]

[0079] Among them, η g is the charging and discharging efficiency factor of the distributed power source g, which represents the energy conversion efficiency of the energy storage system; E bat (g) is the real-time available energy storage capacity of distributed power source g; is the rated energy storage capacity of distributed power source g; DHI(g) is the health index of distributed power source g;

[0080] Step b: When SI(g)>0.6, the distributed generation is allowed to participate in the black start task;

[0081] Step c: In the island operation scenario, prioritize energy storage systems with high SI values ​​to delay the emergency repair window and ensure power supply to critical loads.

[0082] The present invention discloses a distribution network power outage probability analysis and resource scheduling optimization method based on knowledge graph, which has the following beneficial effects.

[0083] The present invention constructs a dynamic spatiotemporal association matrix through a streaming update mechanism, realizing minute-level real-time update of device node association weights, which improves the response speed compared with traditional offline evaluation models; combined with the dynamic adjustment mechanism of the meteorological risk function, the meteorological factor weight μ can be adaptively increased to 1.0 under extreme weather conditions.

[0084] The present invention adopts a topology / environment dual attention layer architecture, and improves the fusion efficiency of device node features and environment node features through feature splicing operations and collaborative training of independent parameter vectors; the fault propagation deduction model introduces dynamic weight constraints to improve the accuracy of secondary fault warning.

[0085] The present invention integrates carbon emission factors and island power balance constraints into a multi-objective mixed integer programming model to achieve Pareto optimization of emergency repair timeliness and environmental protection indicators: the average emergency repair response time is shortened; the carbon emissions of a single task are reduced; and when the distributed power supply support capability index SI(g)>0.6, the power supply guarantee time for critical loads is extended.

[0086] The present invention realizes cross-regional adaptation of the knowledge graph topology structure through a transfer learning framework. In the distribution network reconstruction scenario: the dynamic attention head adaptive adjustment mechanism improves the model convergence speed.

[0087] This invention builds a full-chain management system of "dynamic assessment → fault simulation → resource scheduling → carbon efficiency monitoring": a real-time data pipeline that supports minute-level topology reconstruction and second-level weight updates; a decision-making engine that integrates multi-dimensional data such as GIS path planning and personnel skill maps; and a single-node computing delay of less than 200ms during system expansion, meeting provincial-level grid-level deployment requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0088] Figure 1 This is a flow chart of the method for power outage probability analysis and resource scheduling optimization of distribution network based on knowledge graph of the present invention;

[0089] Figure 2 This is a flow chart for constructing a dynamic knowledge graph provided by the present invention;

[0090] Figure 3 It is a structural diagram of the multi-level graph neural network model provided by the present invention;

[0091] Figure 4 This is a flow chart of the multimodal resource scheduling optimization decision engine provided by the present invention;

[0092] Figure 5 This is a schematic diagram of collaborative scheduling of virtual power plants provided by the present invention. DETAILED DESCRIPTION

[0093] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0094] References to "embodiments" herein mean that the specific features, structures, or characteristics described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the word "embodiment" in various places in the specification does not necessarily refer to the same embodiment, nor does it particularly limit its independence or relevance to other embodiments. In principle, in this application, as long as there are no technical contradictions or conflicts, the various technical features mentioned in the embodiments can be combined in any manner to form a corresponding implementable technical solution.

[0095] Unless otherwise defined, the technical terms used herein have the same meanings as those generally understood by those skilled in the art to which this application belongs; the use of relevant terms herein is only for describing specific embodiments and is not intended to limit this application.

[0096] In the description of this application, the term "and / or" is used to describe a logical relationship between objects, indicating that three relationships can exist. For example, A and / or B means: A exists, B exists, and both A and B exist. In addition, the character " / " in this document generally indicates that the objects before and after are in a logical "or" relationship.

[0097] Unless otherwise specified or limited, in the description of the embodiments of this application, the terms "installed" and "connected" are used.

[0098] Terms such as "connect," "fixed," and "set" should be understood in a broad sense. For example, the "connection" can be a fixed connection, a detachable connection, or an integrated setting; it can be a mechanical connection, an electrical connection, or a communication connection; it can be a direct connection or an indirect connection through an intermediate medium; it can be the internal connection between two components or the interaction between two components. For those skilled in the art of the present application, the specific meanings of the above terms in the embodiments of the present application can be understood according to the specific circumstances.

[0099] Example 1

[0100] like Figure 1 As shown, a method for power outage probability analysis and resource scheduling optimization of distribution network based on knowledge graph includes the following steps:

[0101] S1. Build a dynamic knowledge graph: Map distribution network devices into graph nodes, map the connection relationships between distribution network devices into weighted edges, set edge weights, build a dynamic spatiotemporal association matrix, and define dynamic association weights between nodes;

[0102] like Figure 2 As shown in the figure, it shows the process of building a dynamic knowledge graph. Its core is to integrate distribution network equipment, attributes, connection relationships and multi-source data into a dynamically updated knowledge network. First, topology modeling maps transformers, circuit breakers, feeders and other equipment into graph nodes and establishes connection relationships between devices. The connection weights include electrical impedance, geographical distance and historical common cause failure probability.

[0103] Subsequently, the spatiotemporal correlation matrix constructs the dynamic correlation weights between nodes, taking into account factors such as geographic-electrical coupling, historical fault correlation, and meteorological risks.

[0104] Ultimately, a dynamic knowledge graph containing device attributes, connection relationships, and dynamic association weights is generated, providing a data basis for subsequent power outage probability analysis and operation and maintenance resource evaluation.

[0105] As a preference, in this embodiment, the distribution network equipment in S1 includes transformers, circuit breakers, feeders, etc., and the attributes of the distribution network equipment in S1 include capacity C i , operating life T i , real-time load rate L i (t).

[0106] Preferably, in this embodiment, the dynamic spatiotemporal correlation matrix correlation weight in S1 is calculated by the following formula:

[0107]

[0108] in, is the geographic-electrical coupling term; d ij is the geographical distance between node i and device j; θ ij is the electrical phase angle difference between nodes i and j, in radians; F i , F j are the historical failure frequencies of nodes i and j respectively; γ·Corr(F i , F j ) is the historical fault correlation; Corr(F i , F j ) is the Pearson correlation coefficient; μ·WeatherRisk(i, j, t) is the meteorological risk term; λ d ∈[0.1, 0.5] is the distance attenuation coefficient; WeatherRisk(i, j, t) is the meteorological risk value calculated based on lightning density, wind speed and precipitation intensity; λd ∈[0.1, 0.5], γ∈[0.2, 0.8], μ∈[0.3, 1.0] are adjustable parameters.

[0109] Preferably, in this embodiment, the calculation formula of WeatherRisk(i, j, t) is:

[0110]

[0111] Among them, Lightning t Real-time lightning density, unit: times / square kilometer hour;

[0112] Wind t is the real-time wind speed, unit: m / s;

[0113] S is the meteorological risk adjustment threshold, with a default value of 1.2;

[0114] When Lightning t When the frequency is >1010 times / km² / hour, the insulation margin detection of distribution network equipment will be automatically triggered.

[0115] S2. Identify weak links in the distribution network based on graph neural network: Based on the graph neural network model, feature extraction and reasoning analysis are performed on the dynamic knowledge graph in S1 to evaluate the comprehensive power outage probability of each node and identify weak links in the distribution network accordingly.

[0116] Preferably, in this embodiment, S2 includes the following sub-steps:

[0117] S2.1. Construct a heterogeneous graph, add the environment node as a virtual node, and establish an association edge between the virtual node and the power distribution equipment node;

[0118] The distribution network equipment nodes are:

[0119] h i =[DHI(i), L i (t), WeatherRisk(i)];

[0120] Wherein, DHI(i) is the distribution network equipment health index;

[0121] The health index DHI(i) is calculated by the following formula:

[0122]

[0123] Among them, β = 0.05 is the health sensitivity coefficient. The larger the value, the more significant the impact of MTBF deviation from the mean; MTBF i is the mean time between failures of the equipment;

[0124] Environmental nodes are weather stations and user repair hotspots, and are connected by edge weights W env (i) = 1 - e -0.2·ComplaintRate(i) Connected to the distribution network equipment node;

[0125] S2.2, build a graph neural network, such as Figure 3 As shown in the figure, it shows the structure of a multi-level graph neural network model. Its purpose is to use graph neural network technology to mine the correlation characteristics between device attributes, topological structures and environmental factors, so as to achieve accurate assessment of the probability of power outage of equipment.

[0126] First, the device node feature extraction node extracts device node features, including device health, real-time load rate, and meteorological risk. Subsequently, the hierarchical GAT model calculates the electrical-geographical connections between devices through the topological attention layer and integrates the influence of environmental nodes through the environmental attention layer, thereby comprehensively considering the impact of the device's own status and environmental factors on the power outage probability. Finally, the comprehensive power outage probability calculation node calculates the comprehensive power outage probability of the device based on the output of the hierarchical GAT model, identifies potential weak links, and provides a basis for subsequent operation and maintenance resource scheduling. Its structure includes:

[0127] The first layer is the topological attention layer, which models the electrical-geographic relationship. The attention coefficient is:

[0128]

[0129] in, is the topological attention coefficient of device node i to neighbor node j, which is used to model the strength of electrical-geographical correlation. The weighted features after LeakyReLU activation are normalized by the Softmax function; a1 is the parameter vector of the topological attention mechanism, which is used to calculate the attention weights between nodes and is optimized through training; W1 is the trainable weight matrix of the topological attention layer, which is used to map node features to the attention space; h i It is the characteristic vector of the device node, including the health index (DHI(i)), real-time load rate (L i (t)), weather risk value (WeatherRisk(i)); h j is the feature vector of neighbor node j, with the same structure as h i Consistent; / / indicates vector concatenation operation, W1h i / / W1h j Indicates concatenating the mapping features of nodes i and j.

[0130] The second layer is the environmental attention layer, which integrates the influence of environmental nodes. The environmental attention coefficient is:

[0131]

[0132] in, is the environmental attention coefficient of device node i to environmental node k, reflecting the weight of the impact of environmental factors on the device. It is calculated by exponential normalization; W2 is the trainable weight matrix of the environmental attention layer, independent of W1; E(i) is the set of environmental nodes associated with device node i (such as weather stations, user repair points); h k is the characteristic vector of the environmental node k, including meteorological data (lightning density, wind speed) or user repair hotspot information; h in the denominator m is the feature vector of any environment node m in the set E(i);

[0133] S2.3. Calculate the comprehensive power outage probability of each device node. The specific formula is:

[0134]

[0135] Among them, P outage (i) is the comprehensive power outage probability of device i; σ is the Sigmoid activation function, which is used to map the comprehensive features to the interval [0, 1], representing the power outage probability. The expression is Convert the nonlinear output of the linear combination into a probability value. is the topological attention coefficient, which describes the electrical-geographical connection strength of node i to its neighbor node j. is the environmental attention coefficient, which represents the influence weight of the environment node k on the device node i. W3, W4 are trainable weight matrices, corresponding to the linear transformation parameters of the topological features and environmental features respectively. If the node feature dimension is d, then W3, W4∈R d×d .h j The node feature vector is h j =[DHI(j), L j (t), WeatherRisk(j)], including the health index, real-time load rate, and weather risk value of device j. k is the feature vector of an environmental node, containing meteorological data (e.g., lightning density) or user repair hotspot information. b is a bias term used to adjust the baseline offset of the model output. N(i) is the set of directly electrically connected neighboring nodes of device i (based on the distribution network topology). E(i) is the set of environmental nodes associated with device i (e.g., weather stations, user repair hotspots).

[0136] S3. Based on the identification results in S2, perform multimodal resource scheduling optimization, establish a resource map of the power supply station, dynamically calculate the arrival time of the emergency repair team, build a multi-objective mixed integer programming model, and use a distributed solution algorithm to optimize the scheduling plan.

[0137] Preferably, in this embodiment, S3 includes the following sub-steps:

[0138] S3.1. Establish a resource map for power stations, with node attributes including personnel skills, vehicle types, and tool inventory;

[0139] S3.2. Evaluate resource accessibility and update arrival time based on real-time traffic conditions. The evaluation formula is:

[0140]

[0141] Among them, T reach (m) is the estimated total time for resource m (repair team or vehicle) to reach the fault point, expressed in hours (h), determined by road distance and speed units. This dynamically reflects the efficiency of repair resource movement under real-time traffic conditions and is used to calculate timeliness objectives in scheduling optimization models.

[0142] d m is the geographic distance from the current location of resource m to the fault point. The unit is kilometers (km). This is calculated using GIS or real-time positioning data, combined with an electronic map path planning algorithm to determine the optimal driving path.

[0143] v m is the baseline speed of resource m, representing the theoretical average speed under the constraints of vehicle type or road speed limit, in kilometers per hour (km / h). The constraint is that the baseline speeds of fuel vehicles and electric vehicles are different. In special weather conditions (such as typhoons), dynamic adjustments must be made based on meteorological warnings.

[0144] ρ m is the real-time road congestion rate on the route of resource m, which quantifies the degree of traffic flow obstruction and has a value range of 0≤ρ m ≤1, where ρ m =0 means no congestion, ρ m =1 indicates complete blockage, and the data source is real-time synchronization through the data interface of the traffic management department.

[0145] t detour It is the additional time increment caused by detour, reflecting the combined impact of mileage and speed of the alternative path. The unit is the same as T reach (m) is consistent, and the calculation method is where d detour is the detour path distance, ρdetour is the detour path congestion rate; ρ m is the road congestion rate, ranging from 0 to 1; δ 拥堵 The detour flag, 0 does not require detour, 1 requires detour;

[0146] S3.3, build a multi-objective optimization model, such as Figure 4As shown in the figure, it shows the process of the multimodal resource scheduling optimization decision engine, whose goal is to formulate the optimal scheduling plan based on the fault point and the resource situation of the power supply station to minimize the emergency repair time and carbon emissions. First, resource portrait modeling constructs a resource map of the power supply station, including information such as personnel skills, vehicle types, and tool inventory. Subsequently, dynamic accessibility calculation calculates the arrival time of personnel and vehicles based on real-time traffic conditions to provide a basis for scheduling decisions. The multi-objective optimization model establishes an optimized scheduling plan with the goals of emergency repair time and carbon emissions, and takes into account constraints such as skill matching and island power balance. Finally, the distributed solution algorithm decomposes and solves the problem, and finally obtains an optimized scheduling plan to guide operation and maintenance personnel to carry out emergency repair work, as follows:

[0147]

[0148] Among them, T total (m) is the total scheduling time of resource m; C(m) is the scheduling cost of resource m; x m is the scheduling decision variable, 0 means no scheduling, 1 means scheduling; w1 and w2 are the weight coefficients of emergency repair timeliness and carbon emissions, which must satisfy w1+w2=1;

[0149] (Skill matching constraints);

[0150] Among them, s m is the skill matching matrix of resource m (repair team), with the dimension of [skill 1, skill 2, ...]. If the skill is available, the value is 1, otherwise it is 0.

[0151] r 需求 is a Boolean vector of skills required for fault handling, for example, rRequirement = [1, 0, 1] indicates that high-voltage operation certificate and relay protection debugging skills are required;

[0152] (Island power balancing);

[0153] Among them, P g is the rated output of distributed power source g (in kW);

[0154] y g is a binary variable, indicating whether distributed power generation g, y g ∈{0,1};ΔP load The unit of load shortage caused by the fault is kW;

[0155] δ island ∈{0, 1} is the island operation flag, δ island =1 means to enable island mode, δ island =0 means normal networking mode;

[0156]

[0157]

[0158] Among them, α m Carbon emission factor of vehicle type, fuel vehicle α = 2.3 kg / km, electric vehicle α = 0; β is the carbon emission coefficient of tool use (unit: kg / hour); t repair，m The on-site maintenance time of resource m, including the time spent on fault diagnosis, component replacement, and system debugging, is expressed in hours (h). It is dynamically corrected by weighting the historical MTTR (mean time to repair) of the equipment type (e.g., transformer, circuit breaker). m It is a binary scheduling decision variable, indicating whether to schedule resource m, and its value range is x m ∈{0, 1}, where x m =1 for scheduling resource m; x m =0 means resource m is not scheduled. max The upper limit of carbon emissions allowed for a single dispatch task is dynamically set by the grid carbon quota policy and is expressed in kilograms (kg). The dynamic adjustment mechanism is to decompose the regional carbon intensity target (such as the annual emission reduction rate) into a single task. The threshold can be temporarily relaxed in extreme weather (such as typhoons). max The carbon emission cap for a single dispatch;

[0159] S3.4. Use distributed solution algorithm to optimize the scheduling plan:

[0160] Use K-means to cluster fault points and generate regionalized scheduling sub-problems;

[0161] Each subproblem is solved using the branch and bound method, and the main problem coordinates the global constraints through Lagrangian relaxation;

[0162] The iteration is terminated when the global carbon emission change converges to the threshold ε

[0163] Example 2

[0164] As a preference, in this embodiment, S1 also includes a streaming update mechanism to update the real-time load rate L of the distribution network equipment. i (t) is updated, and the update formula is as follows:

[0165]

[0166] Among them, T window =5 minutes, is the sliding window size, ∈=0.2 is the burst load adjustment factor.

[0167] The streaming update mechanism integrates device sensor data through a time window sliding average algorithm to achieve dynamic updates, and can automatically adjust meteorological risk weights based on meteorological warning information to trigger pre-inspection instructions for equipment in high-risk areas.

[0168] Example 3

[0169] Preferably, in this embodiment, S2 further includes fault propagation deduction based on the comprehensive power outage probability and load pressure to identify secondary risk nodes, and the formula is as follows:

[0170]

[0171] Among them, P cascade is the probability of fault propagation from node i to node j, which indicates the probability of causing a cascading power outage at node j when a fault occurs at node i. cascade When >0.7, node j is marked as a secondary failure risk point.

[0172] σ is the Sigmoid activation function, which is used to map the comprehensive features to the interval [0, 1], representing the power outage probability, expressed as σ Convert the nonlinear output of the linear combination into a probability value.

[0173] w p , w q is the dynamic weight coefficient, which respectively represents the contribution of power outage probability and load pressure to secondary risk; p is the impact weight of power outage probability, and its value is related to the stability of power network topology; q is the load pressure impact weight, the value of which is related to the overload sensitivity of the equipment, and the constraint is that w p +w q =1, ensuring the normalization of risk superposition.

[0174] P outage (j) is the comprehensive power outage probability of node j, including equipment health, environmental risk and topology correlation characteristics. j (t) is the real-time load rate of node j, which is dynamically updated through the sliding window to reflect the current operating pressure of the equipment. j is the rated capacity of node j, which is the inherent attribute parameter of the device and is used to calculate the ratio of load pressure to capacity. i , corresponding to the distribution network equipment capacity defined in Example 1

[0175] Example 4

[0176] As a preference, in this embodiment, S3 also includes virtual power plant collaborative scheduling, such as Figure 5As shown in the figure, it demonstrates the coordinated dispatch of a virtual power plant. Its purpose is to leverage the support capacity of distributed generation (DGs) to provide power to critical loads, thereby delaying the emergency repair window and reducing power outage losses. First, the DG support capacity index calculation node calculates the support capacity of the DGs, taking into account factors such as their capacity, health, and charge and discharge status. Subsequently, the black start participation condition judgment node determines whether the DGs can participate in the black start based on the support capacity index. Finally, the coordinated dispatch node formulates a coordinated dispatch plan based on the support capacity of the DGs and load demand. For example, it can use energy storage systems to provide power to critical loads such as hospitals, delaying the emergency repair window until after the typhoon passes.

[0177] Specifically, in this embodiment, the coordinated scheduling of virtual power plants includes the following steps:

[0178] Step a: Define the distributed power supply support capability index:

[0179]

[0180] Among them, η g is the charge and discharge efficiency factor of the distributed power source g, which characterizes the energy conversion efficiency of the energy storage system and has a value range of 0≤η g ≤1. A larger value indicates a higher energy conversion efficiency. The source is determined based on the measured efficiency of the energy storage device type.

[0181] E bat (g) is the real-time available energy storage capacity of distributed power supply g (in kWh), reflecting the remaining energy of the energy storage device at the current moment. It is dynamically updated based on real-time monitoring by the battery management system (BMS) and calculated in combination with the state of charge (SOC).

[0182] is the rated energy storage capacity of the distributed power supply g (in kWh), is the nominal design parameter of the equipment, and is constrained to satisfy And dynamically correct it according to the battery aging coefficient.

[0183] DHI(g) is the health index of distributed generation g, which quantifies the degree of equipment performance degradation.

[0184] The calculation method is Among them, β=0.05 is the health sensitivity coefficient, MTBF g The mean time between failures of the equipment (in hours).

[0185] Step b: When SI(g)>0.6, the distributed generation is allowed to participate in the black start task;

[0186] Step c: In the island operation scenario, prioritize energy storage systems with high SI values ​​to delay the emergency repair window and ensure power supply to critical loads.

[0187] The above are merely preferred specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Substitutions may be partial structures, devices, or method steps, or they may be complete technical solutions. Any equivalent replacements or modifications based on the technical solution and inventive concept of the present invention shall be covered by the scope of protection of the present invention.

Claims

1. A method for power outage probability analysis and resource scheduling optimization in distribution networks based on knowledge graph, characterized in that: The following steps are involved: S1. Build a dynamic knowledge graph: Map distribution network devices into graph nodes, map the connection relationships between distribution network devices into weighted edges, set edge weights, build a dynamic spatiotemporal association matrix, and define dynamic association weights between nodes; S2. Identify weak links in the distribution network based on a graph neural network: Based on a graph neural network model, perform feature extraction and reasoning analysis on the dynamic knowledge graph in S1, evaluate the comprehensive power outage probability of each node, and identify weak links in the distribution network accordingly; S3. Based on the identification results in S2, perform multimodal resource scheduling optimization, establish a resource map of the power supply station, dynamically calculate the arrival time of the emergency repair team, build a multi-objective mixed integer programming model, and use a distributed solution algorithm to optimize the scheduling plan.

2. The method for power outage probability analysis and resource scheduling optimization based on knowledge graph of distribution network according to claim 1, characterized in that: The distribution network equipment attributes in S1 include capacity C i , operating life T i , real-time load rate L i (t).

3. The method for power outage probability analysis and resource scheduling optimization based on knowledge graph of distribution network according to claim 1, characterized in that: The dynamic spatiotemporal association matrix association weight in S1 is calculated by the following formula: in, is the geographic-electrical coupling term; d ij is the geographical distance between node i and device j; θ ij is the electrical phase angle difference between nodes i and j, in radians; F i , F j are the historical failure frequencies of nodes i and j respectively; γ·Corr(F i , F j ) is the historical fault correlation; Corr(F i , F j ) is the Pearson correlation coefficient; μ·WeatherRisk(i, j, t) is the meteorological risk term; λ d ∈[0.1, 0.5] is the distance attenuation coefficient; WeatherRisk(i, j, t) is the meteorological risk value calculated based on lightning density, wind speed and precipitation intensity; γ∈[0.2, 0.8], μ∈[0.3, 1.0] are adjustable parameters.

4. The method for power outage probability analysis and resource scheduling optimization based on knowledge graph of distribution network according to claim 3, characterized in that: The calculation formula of WeatherRisk(i, j, t) is: Among them, Lightning t Real-time lightning density, unit: times / square kilometer hour; Wind t is the real-time wind speed, unit: m / s; S is the meteorological risk adjustment threshold, with a default value of 1.2; When Lightning t When the frequency is >1010 times / km² / hour, the insulation margin detection of distribution network equipment will be automatically triggered.

5. The method for power outage probability analysis and resource scheduling optimization based on knowledge graph of distribution network according to claim 1, characterized in that: The S1 also includes a streaming update mechanism to update the real-time load rate L of the distribution network equipment. i (t) is updated, and the update formula is as follows: Among them, T window =5 minutes, is the sliding window size, ∈=0.2 is the burst load adjustment factor.

6. The method for power outage probability analysis and resource scheduling optimization based on knowledge graph of distribution network according to claim 1, characterized in that: The S2 includes the following sub-steps: S2.

1. Construct a heterogeneous graph, add the environment node as a virtual node, and establish an association edge between the virtual node and the power distribution equipment node; The distribution network equipment nodes are: h i =[DHI(i),L i (t),WeatherRisk(i)]; Wherein, DHI(i) is the distribution network equipment health index; The health index DHI(i) is calculated by the following formula: Among them, β = 0.05 is the health sensitivity coefficient. The larger the value, the more significant the impact of MTBF deviation from the mean; MTBF i is the mean time between failures of the equipment; Environmental nodes are weather stations and user repair hotspots, and are connected by edge weights W env (i) = 1 - e -0.2·ComplaintRate(i) Connected to the distribution network equipment node; S2.

2. Construct a graph neural network, whose structure includes: The first layer is the topological attention layer, which models the electrical-geographic relationship. The topological attention coefficient is: in, is the topological attention coefficient of device node i to neighbor node j; W1 is the trainable weight matrix of the topological attention layer; h i is the characteristic vector of the device node; h j is the feature vector of neighbor node j, with the same structure as h i Consistent; / / represents vector concatenation operation; The second layer is the environmental attention layer, which integrates the influence of environmental nodes. The environmental attention coefficient is: in, is the environmental attention coefficient of device node i to environment node k; W2 is the trainable weight matrix of the environmental attention layer, independent of W1; E(i) is the set of environmental nodes associated with device node i; h k is the feature vector of environment node k associated with device node i, and h in the denominator m is the feature vector of any environment node m in the set E(i); S2.

3. Calculate the comprehensive power outage probability of each device node. The specific formula is: Among them, h j , h k is the node feature vector; P outage (i) is the comprehensive power outage probability of device i; σ is the Sigmoid activation function; W3 and W4 are trainable weight matrices; b is the bias term; N(i) is the set of directly electrically connected neighbor nodes of device i; E(i) is the environmental node associated with device i.

7. The method for power outage probability analysis and resource scheduling optimization based on knowledge graph of distribution network according to claim 6, characterized in that: S2 also includes fault propagation deduction based on the comprehensive power outage probability and load pressure to identify secondary risk nodes, and its formula is as follows: Among them, P cascade is the fault propagation probability of node i to node j, which indicates the probability of causing a chain power outage at node j when a fault occurs at node i; σ is the Sigmoid activation function; w p , w q is the dynamic weight coefficient; w p is the power outage probability impact weight; w q is the load pressure influence weight; P outage (j) is the comprehensive power outage probability of node j; L j (t) is the real-time load rate of node j, which is dynamically updated through the sliding window; c j is the rated capacity of node j, which is the inherent attribute parameter of the equipment and is used to calculate the ratio of load pressure to capacity; when P cascade When >0.7, node j is marked as a secondary failure risk point.

8. The method for power outage probability analysis and resource scheduling optimization based on knowledge graph of distribution network according to claim 1, characterized in that: The S3 includes the following sub-steps: S3.

1. Establish a resource map for power stations, with node attributes including personnel skills, vehicle types, and tool inventory; S3.

2. Evaluate resource accessibility and update arrival time based on real-time traffic conditions. The evaluation formula is: Among them, T reach (m) is the estimated total time for resource m to reach the failure point; d m is the geographical distance from the current location of resource m to the fault point; v m is the benchmark driving speed of resource m, representing the theoretical average speed under the vehicle type or road speed limit constraint; ρ m is the real-time road congestion rate on the route of resource m, which quantifies the degree of traffic flow obstruction and has a value range of 0≤ρ m ≤1, where ρ m =0 means no congestion, ρ m =1 means complete blockage; δ 拥堵 The detour flag, 0 does not require detour, 1 requires detour; t detour The additional time increment caused by the detour reflects the combined impact of the mileage and speed of the alternative route; S3.

3. Construct a multi-objective optimization model as follows: Among them, T total (m) is the total scheduling time of resource m; C(m) is the scheduling cost of resource m; x m is the scheduling decision variable, 0 means no scheduling, 1 means scheduling; w1 and w2 are the weight coefficients of emergency repair timeliness and carbon emissions, which must satisfy w1+w2=1; Among them, s m is the skill matching matrix of resource m, with the dimension of [skill 1, skill 2, ...], where the value is 1 if the skill is available, and 0 otherwise; r 需求 =[1,0,1] means high voltage certificate and relay protection debugging skills are required; Among them, P g is the rated output of distributed power source g; y g is a binary variable, indicating whether distributed generation g is enabled; ΔP load is the load shortage caused by the fault; δ island ∈{0, 1} is the island operation flag, where 1 indicates island mode; Among them, C total is carbon emissions; α m Carbon emission factor of vehicle type, fuel vehicle α = 2.3 kg / km, electric vehicle α = 0; β is the carbon emission coefficient of tool use (unit: kg / hour); t repair，m is the on-site maintenance time of resource m; x m is a binary scheduling decision variable, indicating whether to schedule resource m; C max The carbon emission cap for a single dispatch; S3.

4. Use distributed solution algorithm to optimize the scheduling plan: Use K-means to cluster fault points and generate regionalized scheduling sub-problems; Each subproblem is solved using the branch and bound method, and the main problem coordinates the global constraints through Lagrangian relaxation; The iteration is terminated when the global carbon emission change converges to the threshold ε 9. The method for power outage probability analysis and resource scheduling optimization based on knowledge graph of distribution network according to claim 8, characterized in that: The S3 also includes virtual power plant collaborative scheduling: Step a: Define the distributed power supply support capability index: Among them, η g is the charging and discharging efficiency factor of the distributed power source g, which represents the energy conversion efficiency of the energy storage system; E bat (g) is the real-time available energy storage capacity of distributed power source g; is the rated energy storage capacity of distributed power source g; DHI(g) is the health index of distributed power source g; Step b: When SI(g)>0.6, the distributed generation is allowed to participate in the black start task; Step c: In the island operation scenario, prioritize energy storage systems with high SI values ​​to delay the emergency repair window and ensure power supply to critical loads.

Images