A new energy distribution network fault research and decision-making method

Abstract

The application discloses a new energy distribution network fault research and decision-making method, and relates to the technical field of intelligent operation and maintenance of distribution networks, which comprises the following steps: extracting the multi-source features of original alarms to construct a trusted fusion vector; merging the trusted fusion vector into a plurality of potential root cause events to obtain a soft distribution coefficient; calculating the root cause confidence of each potential root cause event and screening out effective root cause events; calculating the graph reasoning score of each effective root cause event for each strategy in the distribution network fault knowledge graph and correcting it to obtain a corrected strategy score, and screening out the candidate disposal strategies of each effective root cause event; calculating the comprehensive utility score of each candidate disposal strategy, combining binary integer programming and preset constraints, and taking the maximization of the global disposal benefit function as the target to solve the optimal work order combination and generate the optimal target work order for each effective root cause event, thereby solving the problem that the prior art is difficult to adapt to the fault root cause research and operation and maintenance decision-making under the background of multi-source heterogeneous alarms.

Description

Technical Field

[0001] This invention relates to the field of intelligent operation and maintenance technology for power distribution networks, and more specifically, to a method for fault assessment and auxiliary decision-making in new energy power distribution networks. Background Technology

[0002] With the large-scale integration of new energy sources and new loads such as distributed photovoltaics, energy storage, charging facilities, and flexible loads into the distribution network, the network's operation is characterized by increased volatility on both the source and load sides, diversified equipment types, frequent switching of operating modes, and complex fault propagation chains. Compared with traditional distribution networks, new energy distribution networks generate more frequent, complex, and correlated alarm information during operation. These alarms include structured signals such as protection actions, switch changes, over-limit alarms, communication interruptions, and abnormal status, as well as semi-structured or unstructured text information from monitoring platforms, edge terminals, and equipment logs. How to quickly identify real faults, locate key root causes, and provide further handling suggestions from these massive heterogeneous alarms has become a crucial issue in the current intelligent operation and maintenance of distribution networks.

[0003] In existing technologies, the following technical solutions are commonly used for alarm processing and fault analysis in power distribution networks:

[0004] Alarm handling schemes based on threshold judgment and rule triggering: This type of method typically involves the monitoring system pre-setting fixed thresholds for monitored quantities such as voltage, current, temperature, communication status, and switch position. When the real-time monitored value exceeds the set range, an alarm is triggered. Alternatively, maintenance personnel pre-compile alarm rules, and when a certain rule condition is met, the alarm is marked as a specific fault type. The implementation process of this type of scheme usually includes: collecting monitored quantities, judging whether the limit is exceeded, triggering the alarm, and displaying or simply reporting according to the preset level. It is simple to implement, easy to deploy, and has certain practicality in scenarios with single devices, single variables, and clear rules. However, this type of scheme is essentially a static rule-driven mode, which is difficult to adapt to the massive multi-source heterogeneous alarms of new energy distribution networks.

[0005] Event merging schemes based on alarm association rules: To reduce the flooding of multi-source alarms, some existing systems merge alarms with similar timing, origin, and category. For example, they use fixed time windows, fixed equipment ranges, and / or fixed logic templates to merge multiple alarms generated by the same feeder, the same area, and / or similar equipment within a certain time period into a single event record, thereby reducing the number of alarms and improving the efficiency of dispatchers. These methods typically perform association based on manually defined rules such as time proximity, equipment consistency, and / or alarm category matching. For example, if two alarms occur within a preset time window, originate from the same or adjacent equipment, and belong to a predefined mergeable type, they are considered the same event. While this approach can reduce alarms to some extent, its merging logic relies heavily on manual experience and fixed templates, resulting in extremely low efficiency in handling multi-source heterogeneous alarms in new energy distribution networks.

[0006] Fault identification schemes based on keyword matching or simple classification models: For alarm texts, existing technologies often employ methods such as keyword recognition, template matching, and shallow machine learning classification to map alarm information to several preset fault types. For example, by matching keywords such as "trip," "loss of voltage," "communication interruption," and "protection action," the text is classified into categories such as line fault, communication anomaly, and equipment failure to operate; or features are manually extracted and input into a traditional classifier to predict the alarm category. Compared with simple thresholding methods, these schemes have increased semantic processing capabilities to some extent, but most still remain at the level of surface-level text feature recognition and are difficult to handle problems such as polysemous expressions, contextual dependencies, and heterogeneous semantics across systems.

[0007] The manual assessment and work order dispatching scheme based on expert experience: In the existing operation and maintenance system, the most common practice is still: after the monitoring system generates an alarm, the dispatcher or operation and maintenance personnel conduct a manual assessment based on their operating experience, equipment ledger, line topology diagram and historical handling records, and then manually select the handling measures and dispatch the operation and maintenance work order. This scheme is highly flexible in handling complex scenarios, but it is highly dependent on personnel experience, and the processing efficiency and consistency are difficult to guarantee. Especially in the case of massive concurrent alarms, it is easy to miss, misjudge and delay the response. Summary of the Invention

[0008] The purpose of this invention is to provide a method for fault assessment and auxiliary decision-making in new energy distribution networks, so as to solve the problem that the existing technology mentioned in the background art is difficult to adapt to the fault root cause assessment and operation and maintenance decision-making under the background of multi-source heterogeneous alarms in new energy distribution networks.

[0009] To achieve the above objectives, the present invention provides the following technical solution:

[0010] A method for fault assessment and auxiliary decision-making in new energy distribution networks includes the following steps:

[0011] Acquire several original alarm messages from the power distribution network, extract alarm features, text semantic features, and confidence values ​​for each original alarm message, and construct a trusted fusion vector corresponding to each original alarm message based on the alarm features, text semantic features, and confidence values.

[0012] Based on a soft clustering strategy, all the trusted fusion vectors are grouped into several potential root cause events, generating an event center vector corresponding to each potential root cause event, and a soft assignment coefficient for each original alarm message belonging to each potential root cause event.

[0013] Based on the confidence value and soft allocation coefficient of each original alarm message, the root cause confidence of each potential root cause event is calculated, and potential root cause events with a root cause confidence greater than or equal to a preset confidence threshold are selected to obtain several valid root cause events.

[0014] A distribution network fault knowledge graph is constructed, comprising several fault-to-phenomenon-to-strategy paths. Based on multi-path reasoning of the knowledge graph, the graph reasoning score for each effective root cause event on each strategy entity is calculated. The graph reasoning score is corrected according to a preset correction factor to obtain the corrected strategy score for each effective root cause event on each strategy entity. For each effective root cause event, strategy entities with corrected strategy scores greater than or equal to a preset strategy score threshold are selected to obtain a set of candidate handling strategies for each effective root cause event.

[0015] Based on the preset work order utility function, the comprehensive utility score of each strategy entity in each candidate disposal strategy set is calculated. Based on all the comprehensive utility scores and binary integer programming, a global disposal benefit function is constructed. Based on the preset constraints, the optimal work order combination is obtained by maximizing the global disposal benefit function as the optimization objective. Based on the optimal work order combination, an optimal target work order is generated for each effective root cause event.

[0016] The principle of this invention, a fault assessment and auxiliary decision-making method for new energy distribution networks, is as follows: By extracting alarm features, textual semantic features, and confidence values ​​from the original alarm information, a reliable fusion vector is constructed. This transforms structured electrical signals and semi-structured and unstructured text alarms into a standardized vector representation, solving the problem of inconsistent formats of multi-source heterogeneous alarm data, which makes collaborative processing difficult. Simultaneously, confidence values ​​of different alarm information are incorporated when constructing the standardized reliable fusion vector to reduce interference from low-reliability alarms, providing a more accurate data foundation for subsequent analysis. A soft clustering strategy is used to merge the reliable fusion vector, generating event center vectors for potential root cause events and soft alarm allocation coefficients. Unlike the traditional fixed-rule hard partitioning method, the soft clustering strategy adapts to the complexity of alarm information in new energy distribution networks, and the correlation and coupling characteristics between different alarms enable flexible merging of different alarms at the event level. Furthermore, combined with... The system calculates root cause confidence using a confidence value and a soft allocation coefficient, and filters out valid root cause events, while filtering out redundant and / or false alarm merging events. This addresses the challenges of ambiguous root cause localization or misjudgment / missed judgment under massive alarms. It constructs a knowledge graph from fault to phenomenon to strategy and performs multi-path reasoning. Combining domain knowledge, it accurately matches root cause events with handling strategies. Simultaneously, it corrects the graph reasoning score using a preset correction factor to select candidate handling strategies that balance safety and feasibility, avoiding detachment of simple graph reasoning from actual on-site operations and maintenance. The system calculates the comprehensive utility score of strategies using a work order utility function, constructs a global handling benefit function using binary integer programming, and solves for the optimal work order combination with the goal of maximizing benefit. This overcomes the limitations of local decision-making for single faults, achieving globally optimal allocation of operation and maintenance resources and work orders in multi-fault concurrent scenarios. This effectively solves the technical problems of low fault assessment efficiency and disordered operation and maintenance decision-making in complex alarm scenarios of new energy distribution networks.

[0017] Preferably, to improve the comprehensiveness of the trusted fusion vector's representation of alarm information and provide more accurate data support for subsequent alarm attribution, the construction method of the trusted fusion vector specifically includes:

[0018] Each of the original alarm messages is processed in a standardized manner to obtain the alarm characteristics of each original alarm message;

[0019] Calculate the confidence value of each original alarm message based on the alarm source benchmark confidence level and preset penalty items;

[0020] Based on the alarm features and the confidence value, construct a reliable alarm vector for each of the original alarm messages;

[0021] Extract the textual semantic features of each original alarm message according to a preset text encoding function;

[0022] Based on the topology graph theory model of the distribution network, the nodes to which the original alarm information belongs are topologically embedded to obtain the topology embedding vector of each original alarm information.

[0023] By fusing the trusted alarm vector, the text semantic features, and the topological embedding vector, a multimodal fusion vector is obtained;

[0024] Based on the confidence value and the multimodal fusion vector, calculate the confidence fusion vector for each of the original alarm messages.

[0025] Preferably, the step of grouping all the trusted fusion vectors into several potential root cause events based on the soft clustering strategy specifically includes:

[0026] Calculate the combined correlation strength of any two of the aforementioned trusted fusion vectors;

[0027] Based on the trusted fusion vector and the comprehensive correlation strength, an alarm merging objective function is constructed, which includes: an alarm semantic space constraint term, an entropy regularization constraint term, and an alarm correlation constraint term.

[0028] With the goal of minimizing the alarm merging objective function as the optimization objective and the constraint that each original alarm message must be assigned to a certain potential root cause event as the constraint, the soft merging matrix and the event center vectors of several potential root cause events are obtained. Based on the soft merging matrix, the soft assignment coefficient of each original alarm message belonging to each potential root cause event is obtained.

[0029] Existing conventional clustering algorithms lack specific constraints that match the characteristics of power grid alarms, while commonly used hard clustering methods cannot reflect the ambiguity of alarm information membership, resulting in overly coarse event merging results. This invention solves for the soft allocation coefficients and event center vectors of different alarm information based on a multi-constraint objective function, which can achieve flexible merging of massive alarm information and adapt to the actual characteristics of the coupling and correlation of distribution network fault alarms, thereby improving the accuracy of event-level merging of massive alarms.

[0030] Preferably, the comprehensive correlation strength includes: the cosine similarity of the two trusted fusion vectors, the temporal proximity of the two trusted fusion vectors, the electrical coupling of the two alarm devices corresponding to the two trusted fusion vectors, and the prior causal correlation strength between the two alarm categories corresponding to the two trusted fusion vectors. By extracting and fusing the correlation of at least four dimensions of any two trusted fusion vectors to form the comprehensive correlation strength, the comprehensiveness of the quantitative data on the intrinsic correlation between different alarms can be improved, thereby improving the accuracy of subsequent alarm merging.

[0031] Preferably, to achieve quantitative classification of potential root cause events, thereby screening out effective root cause events that fit different alarm information, and to ensure the accuracy and reliability of the root cause event screening results, the formula for calculating the root cause confidence level is as follows:

[0032] ;

[0033] In the formula, Indexed by the sequence number of the potential root cause event. For the first Root cause confidence of a potential root cause event For the distribution network in the first The number of raw alarm messages received within the time window. and This is the sequence number index of the original alarm information. For the first The original alarm message belongs to the first The soft assignment coefficient of each potential root cause event. For the first The original alarm message belongs to the first The soft assignment coefficient of each potential root cause event. For the first The confidence value of the original alarm message. The correlation enhancement coefficient, For the first Article and Section The overall correlation strength between the trusted fusion vectors corresponding to each of the original alarm messages.

[0034] Preferably, to improve the reliability of knowledge graph path reasoning results, the calculation method for the graph reasoning score includes:

[0035] Based on the event center vector of each valid root cause event and the embedding vector of each fault entity, calculate the fault matching score between each valid root cause event and each fault entity.

[0036] Based on the merging results and soft allocation coefficients of the original alarm information obtained by the soft clustering strategy, the support score of each effective root cause event for each phenomenon entity is calculated.

[0037] Based on the fault matching score, the support score, and the knowledge weight of each fault-to-phenomenon-to-policy path, the maximum path reasoning score of each effective root cause event for each policy entity is calculated to obtain the graph reasoning score.

[0038] Preferably, the preset correction factors include: a strategy conflict factor, a security risk penalty factor, and an executability factor; the calculation formula for the correction strategy score is:

[0039] ;

[0040] In the formula, and These are the sequence index of valid root cause events and the strategy index in the distribution network fault knowledge graph, respectively. For the first One effective root cause event for the strategy The score of the corrected strategy, For the first One effective root cause event for the strategy The graph reasoning score, For the first One effective root cause event and strategy Strategic conflict factors To adopt a strategy Disposal of the first A valid root cause event's security risk penalty factor To adopt a strategy Disposal of the first The executability factor of a valid root cause event;

[0041] The graph reasoning score obtained solely from the reasoning results of the knowledge graph does not take into account constraints such as operation and maintenance conflicts, security risks, and on-site executability. As a result, the recommended strategies are difficult to adapt to the on-site management and control needs of the power distribution network. This invention introduces multi-dimensional correction factors to constrain and correct the graph reasoning score, and selects candidate disposal strategies based on the corrected strategy scores. This can filter out low-quality strategies and improve the security and executability of candidate disposal strategies.

[0042] Preferably, to improve the comprehensiveness of the work order utility function in quantifying the utility of candidate strategies, thereby providing a high-quality data foundation for subsequent solutions to the optimal work order combination, the preset work order utility function includes: a modified strategy score item, an event severity item, a resource matching degree item, an execution cost item, an execution duration item, and an operational risk item. The expression of the preset work order utility function is as follows:

[0043] ;

[0044] In the formula, and These are the candidate disposal strategy index and the resource unit index, respectively. To make resource units Assigned to the Identify valid root cause events and implement candidate treatment strategies. The overall utility score, For the first One effective root cause event for candidate handling strategies The score of the corrected strategy, For the first The severity of each valid root cause event. For resource units With candidate handling strategy Resource matching degree , and Respectively utilizing resource units Implement candidate disposal strategy The execution cost, execution time, and operational risks, ~ The weights are respectively the score of the correction strategy, the severity of the event, the resource matching degree, the execution cost, the execution time, and the operational risk.

[0045] Preferably, the expression for the global disposal benefit function is:

[0046] ;

[0047] In the formula, For the global disposal benefit function value, For the first A set of candidate handling strategies for each valid root cause event; Let be the binary decision variables in the binary integer programming problem. Used to indicate whether resource units are selected. Assigned to the Identify valid root cause events and implement candidate treatment strategies. ; To make resource units Assigned to the Identify valid root cause events and implement candidate treatment strategies. The overall utility score;

[0048] The handling strategy recommended based on a single root cause event has the limitation of local optima and it is difficult to achieve global optimization for the overall fault handling of the distribution network in the case of multiple concurrent faults. This invention builds a global benefit calculation model based on binary integer programming and configures global work orders with the goal of overall benefit optimization. This can ensure the global optimization of work order combination configuration in the case of multiple concurrent faults, thereby improving the resource allocation efficiency and effectiveness of fault operation and maintenance.

[0049] Preferably, to achieve closed-loop adaptive optimization of the decision-making model and improve the decision-making accuracy of this method for long-term fault assessment, the new energy distribution network fault assessment and auxiliary decision-making method further includes:

[0050] Collect the handling feedback information of the distribution network after performing fault handling according to each of the optimal target work orders. The handling feedback information includes: alarm suppression rate, actual alarm handling time, number of repeated alarms and fault handling execution quality score.

[0051] The feedback reward value for each of the optimal target work orders is calculated based on the alarm suppression rate, the actual alarm handling time, the number of repeated alarms, and the fault handling execution quality score.

[0052] Based on the feedback reward value, fault matching score, support level score and preset graph learning rate corresponding to each optimal target work order, update the knowledge weight of the fault to phenomenon to strategy path corresponding to each optimal target work order;

[0053] Based on the alarm suppression rate and the preset confidence update coefficient, update the alarm source baseline confidence corresponding to each optimal target work order.

[0054] One or more technical solutions provided by this invention have at least the following technical effects or advantages:

[0055] 1. By using trusted vector modeling, soft clustering root cause merging, knowledge graph reasoning, and integer programming to optimize work order arrangement, we can achieve event-level root cause merging of massive, multi-source, heterogeneous, and strongly coupled complex alarms in the new energy distribution network, as well as globally optimal operation and maintenance decisions that take into account safety, efficiency, and cost, thereby improving the response speed and decision consistency of the distribution network fault events.

[0056] 2. By jointly utilizing the temporal proximity, spatial adjacency, semantic similarity, and device correlation among alarms, massive alarms can be coupled and merged in terms of time, space, and semantics. Representative root cause events can be extracted from multiple apparent alarms, realizing the transformation from alarm-level processing to event-level processing, and improving the accuracy and stability of power distribution network fault identification in complex scenarios.

[0057] 3. By constructing a distribution network fault knowledge graph containing several paths from fault to phenomenon to strategy, the root cause events are matched and reasoned with the knowledge graph to automatically generate candidate handling strategies. Furthermore, work orders are optimized and arranged by combining event severity, resource availability, execution cost, operational risk, and working time constraints. This enables the direct generation of executable operation and maintenance work orders from fault assessment results, significantly reducing the burden of manual analysis and improving the speed of response and the consistency of decision-making.

[0058] 4. By collecting feedback information such as alarm clearance status, processing time, rework status, and manual confirmation results after work order execution, and dynamically updating the path weights and alarm source credibility in the knowledge graph based on the feedback results, the system can use the actual processing results in reverse for subsequent analysis and decision optimization, gradually improving the accuracy of fault identification, the effectiveness of strategy recommendation, and the rationality of work order arrangement. It has strong adaptability and long-term application value. Attached Figure Description

[0059] The accompanying drawings, which are provided to further illustrate embodiments of the invention and constitute a part of this invention, are not intended to limit the scope of the invention.

[0060] Figure 1 This is a flowchart illustrating a method for fault assessment and auxiliary decision-making in a new energy distribution network according to the present invention.

[0061] Figure 2 This refers to the confidence value of the alarm source and the distribution of the original alarm information after weighted processing in Embodiment 2 of the present invention.

[0062] Figure 3 This is a distribution diagram of the adaptive fusion weights and weighted trusted fusion vectors of the first 20 alarms in Embodiment 2 of the present invention;

[0063] Figure 4 This is a heatmap of the overall correlation strength between different alarms in Embodiment 2 of the present invention, and a distribution diagram of the root cause event clustering results for each alarm.

[0064] Figure 5 This refers to the candidate strategy screening results and the modified strategy score of each candidate strategy in Embodiment 2 of the present invention.

[0065] Figure 6 This is the optimal work order arrangement result in Embodiment 2 of the present invention;

[0066] Figure 7 This is a comparison of the results before and after the update of the alarm source credibility and the knowledge weight of the path in Embodiment 2 of the present invention. Detailed Implementation

[0067] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, where there is no conflict, the embodiments of the present invention and the features thereof can be combined with each other.

[0068] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0069] Example 1

[0070] Please refer to Figure 1 The present invention provides a method for fault assessment and auxiliary decision-making in a new energy distribution network, comprising the following steps:

[0071] A number of original alarm messages from the power distribution network are acquired, and alarm features, text semantic features, and confidence values ​​are extracted from each original alarm message. Based on the alarm features, text semantic features, and confidence values, a trusted fusion vector is constructed for each original alarm message. The original alarm messages include: alarm occurrence timestamp, alarm device identifier, alarm category, alarm text, measurement value of the alarm device, and alarm source channel identifier.

[0072] Based on a soft clustering strategy, all the trusted fusion vectors are grouped into several potential root cause events, generating an event center vector corresponding to each potential root cause event, and a soft assignment coefficient for each original alarm message belonging to each potential root cause event.

[0073] Based on the confidence value and soft allocation coefficient of each original alarm message, the root cause confidence of each potential root cause event is calculated, and potential root cause events with a root cause confidence greater than a preset confidence threshold are selected to obtain several valid root cause events.

[0074] A distribution network fault knowledge graph is constructed, comprising several fault-to-phenomenon-to-strategy paths. Based on multi-path reasoning of the knowledge graph, the graph reasoning score for each effective root cause event on each strategy entity is calculated. The graph reasoning score is corrected according to a preset correction factor to obtain the corrected strategy score for each effective root cause event on each strategy entity. For each effective root cause event, strategy entities with corrected strategy scores greater than or equal to a preset strategy score threshold are selected to obtain a set of candidate handling strategies for each effective root cause event.

[0075] Based on the preset work order utility function, the comprehensive utility score of each strategy entity in each candidate disposal strategy set is calculated. Based on all the comprehensive utility scores and binary integer programming, a global disposal benefit function is constructed. Based on the preset constraints, the optimal work order combination is obtained by maximizing the global disposal benefit function as the optimization objective. Based on the optimal work order combination, an optimal target work order is generated for each effective root cause event.

[0076] The specific methods for constructing the trusted fusion vector include:

[0077] Each of the original alarm messages is processed in a standardized manner to obtain the alarm characteristics of each original alarm message;

[0078] Calculate the confidence value of each original alarm message based on the alarm source benchmark confidence level and preset penalty items;

[0079] Based on the alarm features and the confidence value, construct a reliable alarm vector for each of the original alarm messages;

[0080] Extract the textual semantic features of each original alarm message according to a preset text encoding function;

[0081] Based on the topology graph theory model of the distribution network, the nodes to which the original alarm information belongs are topologically embedded to obtain the topology embedding vector of each original alarm information.

[0082] By fusing the trusted alarm vector, the text semantic features, and the topological embedding vector, a multimodal fusion vector is obtained;

[0083] Based on the confidence value and the multimodal fusion vector, calculate the confidence fusion vector for each of the original alarm messages;

[0084] Specifically, in this embodiment, the distribution network will be in the first... A single original alarm message received within the time window is denoted as:

[0085] , ;

[0086] In the formula, This is the sequence number index of the original alarm information. For the first Vector representation of each original alarm message. , , , , and The first The original alarm includes the alarm occurrence timestamp, alarm device identifier code, alarm category code, alarm text, measured value of the alarm device, and alarm source channel identifier code. For the distribution network in the first Total number of alarms received within the time window;

[0087] Because different systems or devices send alarms with different formats, units, and precision, each original alarm message is standardized to obtain the following vector representation of alarm characteristics:

[0088] ;

[0089] In the formula, For the first Vector representation of alarm features of each original alarm message. This is a function that maps alarm timestamps. For alarm device identifier encoding mapping function, Encode a mapping function for alarm categories. A standardization mapping function for the measured values. For the alarm source channel identifier encoding mapping function, This is a transpose operation; the five mapping functions mentioned above can be normalized mappings, or feature extraction functions can be customized for different categories of information; in this embodiment, to specifically extract the alarm features of each type of original alarm information, the following are respectively... , , and Take it as the following expression:

[0090] ;

[0091] ;

[0092] ;

[0093] ;

[0094] In the formula, and The first The minimum and maximum timestamps of all original alarm messages received within the time window. The 24-hour time value corresponding to the timestamp when the alarm occurred. To prevent extremely small positive numbers with a denominator of 0, this embodiment takes... ; This represents a one-hot encoded mapping. For alarm devices The number of layers in the topological hierarchy. For the first The maximum number of topology layers for all alarm devices within the time window. For alarm devices Connectivity in the distribution network topology diagram For the first The maximum connectivity of all alarm devices within the time window. For alarm devices The electrical distance to its power source or the beginning of the feeder. For the first The maximum electrical distance of all alarm devices within the time window;

[0095] Since a single device may report multiple types of measurement values, this embodiment will... Recorded as In the formula, , ,……and They represent the first The first, second, ... and the first of the original alarm messages Class measurement value, For alarm devices The total number of categories of measurement values ​​that can be reported; then this embodiment will Take it as the following expression:

[0096] ;

[0097] In the formula, , ,……and They represent the first The first, second, ... and the first of the original alarm messages Normalized values ​​of class measures , ,……and They represent the first The first, second, ... and the first of the original alarm messages Missing marker values ​​for class measurements, among which, for The formulas for normalizing and marking missing measurements for each type of measurement are as follows:

[0098] ;

[0099] ;

[0100] In the formula, For measurement value category index, For the first The first original alarm message Normalized values ​​of class measures and Alarm devices The The mean and standard deviation of the class measure in a historical sample. To prevent extremely small positive numbers with a denominator of 0, this embodiment takes... , For the first The first original alarm message Missing marker values ​​for class measurements;

[0101] The formula for calculating the reliability value of the original alarm information is as follows:

[0102] ;

[0103] In the formula, For the first The confidence value of the original alarm message. For the first The baseline confidence level of the alarm source for each original alarm message. and The first The degree of information loss and noise disturbance intensity of the original alarm information For the first The drift in the statistical distribution of original alarm messages and historical alarm messages from the same source. , and These are penalty coefficients for the degree of information loss, the intensity of noise disturbance, and the amount of drift, respectively. During the initial deployment phase, they are taken as... , , ;

[0104] The trusted alarm vector is obtained by weighting the vector representation of alarm features with trusted values, specifically as follows:

[0105] ;

[0106] In the formula, For the first A reliable alarm vector for each original alarm message;

[0107] The preset text encoding function can be implemented using word embedding, Transformer encoder, or domain semantic encoder. In this embodiment, a domain semantic encoder is selected to extract the text semantic features of each original alarm message.

[0108] The multimodal fusion vector is obtained by gating the credible alarm vector, textual semantic features, and topological embedding vector of each original alarm message. Its expression is:

[0109] ;

[0110] In the formula, , and The first The multimodal fusion vector of the original alarm information, the vector representation of textual semantic features, and the topological embedding vector. for The structural feature mapping matrix, , and The first The gating fusion weights of the textual semantic feature vector, trusted alarm vector, and topological embedding vector of the original alarm information. , and The gating mechanism is adaptively provided, and the expression is:

[0111] ;

[0112] In the formula, For the gated weight matrix, For bias terms, This represents a vector concatenation operation. Used for protection , and The sum is 1;

[0113] The trusted fusion vector is obtained by weighting the multimodal fusion vector with trust values, specifically:

[0114] ;

[0115] In the formula, This is a reliable fusion vector.

[0116] Specifically, the step of grouping all the trusted fusion vectors into several potential root cause events based on the soft clustering strategy includes:

[0117] Calculate the comprehensive correlation strength of any two trusted fusion vectors; the comprehensive correlation strength includes: the cosine similarity of the two trusted fusion vectors, the temporal proximity of the two trusted fusion vectors, the electrical coupling degree of the two alarm devices corresponding to the two trusted fusion vectors respectively, and the prior causal correlation strength between the two alarm categories corresponding to the two trusted fusion vectors respectively; the formula for calculating the comprehensive correlation strength is:

[0118] ;

[0119] In the formula, For the first Article and Section The overall correlation strength between the trusted fusion vectors corresponding to each of the original alarm messages. The time decay constant, For alarm devices and Electrical coupling degree in distribution network topology graph theory model Alarm Category and The strength of the prior causal relationship between them , , and These are the weighting coefficients for cosine similarity, temporal proximity, electrical coupling, and prior causal correlation strength, respectively. In this embodiment, we take... , , , , ;

[0120] Based on the trusted fusion vector and the comprehensive correlation strength, an alarm merging objective function is constructed. This objective function includes: an alarm semantic space constraint term, an entropy regularization constraint term, and an alarm correlation constraint term. The expression for the alarm merging objective function is:

[0121] ;

[0122] In the formula, The alarm merging objective function, and This is the sequence number index of the original alarm information. Indexed by the sequence number of the potential root cause event. The number of potential root cause events determined adaptively. and The first Article and Section The original alarm message belongs to the first The soft assignment coefficient of each potential root cause event. For the first The event center vector of each potential root cause event. This is a semantic space constraint term for alarms, used to constrain alarms within the same cluster to be close in the semantic space. This is an entropy regularization constraint term, used to avoid premature hard clustering. This is an alarm correlation constraint used to strengthen the grouping of highly correlated alarms into the same underlying root cause event. and These are the adjustment coefficients for the entropy regularization constraint and the alarm association constraint, respectively. In this embodiment, we take... , ;

[0123] With the goal of minimizing the alarm merging objective function, and the constraint that each original alarm message must be assigned to a potential root cause event, a soft merging matrix and event center vectors of several potential root cause events are obtained. Based on the soft merging matrix, the soft assignment coefficient of each original alarm message belonging to each potential root cause event is obtained; that is, under the condition of satisfying... and Under the constraints, solve for the objective function that enables alarm merging. Minimize the soft merge matrix and event center collection And thus obtain The event center vector of the potential root cause event, and the event center vector of the potential root cause event. The original alarm message belongs to the first The soft allocation coefficient of each potential root cause event .

[0124] The formula for calculating the root cause confidence of each potential root cause event is as follows:

[0125] ;

[0126] In the formula, For the first Root cause confidence of a potential root cause event For the correlation enhancement coefficient, this embodiment takes... ; Set the root cause confidence level to be greater than or equal to the preset confidence threshold. All potential root cause events are denoted as the set of valid root cause events. In the formula, For the sequence number index of the valid root cause event, For the first The event center vector of a valid root cause event. For the first The root cause confidence of a valid root cause event is taken in this embodiment. .

[0127] The expression for the distribution network fault knowledge graph constructed in this embodiment is as follows:

[0128] ;

[0129] In the formula, Representing a knowledge graph of distribution network faults, and These are the graph entity set and the graph relation set, respectively. The graph entity set includes at least several fault entities, several phenomenon entities, and several strategy entities. The graph relation set includes at least several causes, representations, applications, constraints, and conflicts. The constructed distribution network fault knowledge graph contains several fault-to-phenomenon-to-strategy paths and several fault-to-strategy paths.

[0130] The calculation method for the graph reasoning score includes:

[0131] Based on the event center vector of each valid root cause event and the embedding vector of each fault entity, a fault matching score is calculated for each valid root cause event and each fault entity; the formula for calculating the fault matching score is as follows:

[0132] ;

[0133] In the formula, This serves as the index for the fault entities in the distribution network fault knowledge graph. For the first Effective root cause events and failures Fault matching score, For the Sigmoid mapping function, and The faults are respectively Embedding vectors and fault bias terms;

[0134] Based on the merging results and soft allocation coefficients of the original alarm information obtained by the soft clustering strategy, the support score of each valid root cause event for each phenomenon entity is calculated; the formula for calculating the support score is as follows:

[0135] ;

[0136] In the formula, This serves as an index of phenomena in the knowledge graph of distribution network faults. For the first Effective root cause events for phenomena Support score Let be an indicator function, representing the expression when the first... Alarm category of original alarm information With phenomenon If they are equal, use 1; otherwise, use 0.

[0137] Based on the fault matching score, the support score, and the knowledge weight of each fault-to-phenomenon-to-policy path, the maximum path reasoning score for each valid root cause event to each policy entity is calculated to obtain the graph reasoning score; the formula for calculating the graph reasoning score is:

[0138] ;

[0139] In the formula, For the first Effective root cause event strategy The graph reasoning score, Fault To phenomenon To Strategy Knowledge weight of the path.

[0140] The formula for calculating the revised policy score for each policy entity based on each valid root cause event is as follows:

[0141] ;

[0142] In the formula, For the first One effective root cause event for the strategy The score of the corrected strategy, For the first One effective root cause event and strategy Strategic conflict factors To adopt a strategy Disposal of the first A valid root cause event's security risk penalty factor To adopt a strategy Disposal of the first The executability factors of a valid root cause event, among which,

[0143] ;

[0144] ;

[0145] ;

[0146] The score of the correction strategy corresponding to each valid root cause event is greater than or equal to the preset strategy score threshold. The strategy entity is denoted as In the formula, For the first A set of candidate handling strategies for each valid root cause event. As an index for candidate disposal strategies, this embodiment takes .

[0147] The preset work order utility function includes: a correction strategy score item, an event severity item, a resource matching degree item, an execution cost item, an execution duration item, and an operational risk item. The expression of the preset work order utility function is:

[0148] ;

[0149] In the formula, To make resource units Assigned to the Identify valid root cause events and implement candidate treatment strategies. The overall utility score, For the first One effective root cause event for candidate handling strategies The score of the corrected strategy, For the first The severity of each valid root cause event. For resource units With candidate handling strategy Resource matching degree , and Respectively utilizing resource units Implement candidate disposal strategy The execution cost, execution time, and operational risks, ~ The weights are respectively the correction strategy score, event severity, resource matching degree, execution cost, execution time, and operational risk. In this embodiment, we take... , , , , , ;in, Can be derived from the first The root cause confidence of the first valid root cause event, the first The impact level of distribution network equipment and the impact on user scale caused by each valid root cause event are calculated comprehensively. The calculation formula is:

[0150] ;

[0151] In the formula, , and For the first The normalized values ​​of the root cause confidence, device impact level, and user scale impact of each valid root cause event on the root cause confidence, device impact level, and user scale impact of all valid root cause events, respectively. , and These are the weights for root cause confidence, device impact level, and user scale impact, respectively. In this embodiment, we take... , , .

[0152] The expression for the global disposal benefit function is as follows:

[0153] ;

[0154] In the formula, For the global disposal benefit function value, Let the binary decision variables be binary integer programming variables. Used to indicate whether resource units are selected. Assigned to the Identify valid root cause events and implement candidate treatment strategies. This embodiment selects 0-1 programming as the binary integer programming approach. If resource units are selected... Assigned to the Identify valid root cause events and implement candidate treatment strategies. ,but ,otherwise, ;

[0155] Specifically, the method for generating the optimal target work order for each valid root cause event is as follows: under preset constraints, solve for all possible solutions that maximize the global disposal benefit function. The value of is determined based on all the values ​​obtained from the solution. The value of is used to generate the optimal target work order for each valid root cause event. In the formula, For the first The optimal target work order for each valid root cause event. Indicates the first One valid root cause event, For valid root cause events The optimal handling strategy This refers to the optimal resource unit matched according to the optimal disposal strategy. For suggested startup time, Work order priority; the system prioritizes each valid root cause event. Optimal target work order and Work order priority Allocate the order of handling each valid root cause event and schedule the corresponding resource units to perform fault handling work in the distribution network;

[0156] in, Based on the optimal target work order The corresponding overall utility score and execution time are calculated. The calculation formula is:

[0157] ;

[0158] In the formula, To prevent extremely small positive numbers with a denominator of 0, this embodiment takes... ;

[0159] The preset constraints include the following three constraints:

[0160] ;

[0161] ;

[0162] ;

[0163] In the formula, For resource units Total available working hours in the current scheduling period Candidate handling strategy Mutually exclusive candidate handling strategy index, The set of mutually exclusive candidate disposal strategies is defined by the first constraint, which ensures that each valid root cause event is allocated at most one disposal strategy within the current scheduling period. The second constraint ensures that resource units are used within the current scheduling period. The total working hours shall not exceed its total available working hours. The third constraint is used to restrict conflicting work orders that cannot be executed simultaneously from being issued at the same time.

[0164] The aforementioned method for fault assessment and auxiliary decision-making in new energy distribution networks further includes the following steps:

[0165] Collect the handling feedback information of the distribution network after performing fault handling according to each optimal target work order. The handling feedback information includes: alarm suppression rate, actual alarm handling time, number of repeated alarms and fault handling execution quality score.

[0166] Based on the alarm suppression rate, the actual alarm handling time, the number of repeated alarms, and the fault handling execution quality score, the feedback reward value for each optimal target work order is calculated; the formula for calculating the feedback reward value is:

[0167] ;

[0168] In the formula, For the optimal target work order Feedback reward value, , , , and These are the valid root cause events. The alarm suppression rate, fault handling execution quality score, number of repeated alarms, actual alarm handling time and standard handling time, ~ These are the alarm suppression rate reward coefficient, the fault handling execution quality score reward coefficient, the repeated alarm count penalty coefficient, and the alarm handling time penalty coefficient. In this embodiment, we take... , , , ;

[0169] Based on the feedback reward value, fault matching score, support level score, and preset graph learning rate corresponding to each optimal target work order, the knowledge weight of the fault-to-phenomenon-to-strategy path corresponding to each optimal target work order is updated; the update formula for the knowledge weight is:

[0170] ;

[0171] In the formula, and The faults before and after the update are respectively. To phenomenon To Strategy Knowledge weight of the path In this embodiment, the preset learning rate for the graph is taken as... ;

[0172] Based on the alarm suppression rate and the preset confidence update coefficient, the alarm source baseline confidence level corresponding to each optimal target work order is updated; the update formula for the alarm source baseline confidence level is:

[0173] ;

[0174] In the formula, and These are grouped into valid root cause events. The baseline confidence level of alarm sources before and after the original alarm information is updated. For the credibility update coefficient, this embodiment takes... .

[0175] Example 2

[0176] Based on Example 1, Example 2 will be described and illustrated with specific implementation cases.

[0177] Implementation Case 1: Verifying the feasibility and effectiveness of the present invention's method for fault assessment and auxiliary decision-making in a new energy distribution network; please refer to... Figure 2-7 This embodiment constructs a simulated operation scenario of a new energy distribution network. This scenario includes multiple distribution nodes, branch connection relationships, and heterogeneous alarm data from various alarm sources such as main station monitoring, protection devices, edge sensing units, and distributed photovoltaic / energy storage monitoring units. By processing the multi-source alarms generated during the simulated operation, the invention's ability to visually verify fault root cause analysis and operation and maintenance decision-making under the background of multi-source heterogeneous alarms in a new energy distribution network is achieved. Specifically, it includes the following steps:

[0178] First, the raw alarm information from each alarm source is standardized and its features are extracted to obtain multiple alarm features. Then, considering the reliability of the alarm source, data loss, noise level, and drift, the confidence value of each alarm is calculated. Finally, the corresponding alarm features are weighted using the confidence values ​​of the alarm sources to obtain a reliable alarm vector for each raw alarm message. Figure 2 This represents the confidence value of the alarm source and the distribution of the original alarm information after weighted processing. Figure 2 (a) shows the baseline confidence level and confidence level calculation results for different alarm sources. It can be seen that there are significant differences in confidence level between alarms from different sources. Figure 2 (b) shows the distribution of all original alarm information after standardization. Each circle in the figure represents an alarm, and the size of the area of ​​different circles represents the confidence level of the alarm. Different colors represent different alarm sources. It can be seen that the confidence value weighting can effectively highlight high-quality alarm information, thereby providing a more reliable data foundation for subsequent analysis.

[0179] Then, the textual semantic features of each original alarm message are extracted using a domain semantic encoder, and the topological embedding of the node to which each alarm belongs is performed based on the topological graph theory model of the power distribution network to obtain the topological embedding vector of each alarm. The trusted alarm vector, textual semantic features and topological embedding vector of each alarm are adaptively fused using a gating mechanism to obtain the multimodal fusion vector of each alarm. The multimodal fusion vector is then weighted using the confidence value of the source of each alarm to finally obtain the trusted fusion vector of each alarm. Figure 3 (a) Taking the first 20 alarms as an example, the adaptive fusion weights of different modal features for each alarm are given. It can be seen that different alarms contribute differently to text, structure, topology and contextual information. Figure 3 (b) with Figure 3 Taking the first 20 alarms in (a) as an example, the trusted fusion vector of each alarm after adaptive fusion and trust weighting is given. It can be seen that the multimodal adaptive weighted fusion mechanism can extract the true features of the alarm more comprehensively, and thus better reflect the true meaning of the alarm.

[0180] Next, based on a soft clustering strategy, considering the temporal proximity, spatial adjacency, semantic similarity, and device correlation among various alarms, a large number of alarms were merged to identify representative potential root cause events. Then, based on the root cause confidence of each potential root cause event, valid root cause events were selected, resulting in the following... Figure 4 The diagram shows the overall correlation strength between different alarms and the root cause event clustering results for each alarm. Figure 4 (a) is the alarm correlation strength matrix composed of the comprehensive correlation strength between each two alarms. It can be seen that there are obvious differences in the correlation strength between different alarms. Figure 4 (b) shows the clustering results of effective root cause events obtained based on the alarm merging objective function and root cause confidence. Each circle in the figure represents an alarm, and circles of the same color represent alarms that have been merged into the same effective root cause event.

[0181] Subsequently, the obtained valid root cause events are mapped one by one to a pre-constructed knowledge graph. The graph reasoning score for each root cause event on each strategy is calculated. Based on strategy conflict factors, security risk penalty factors, and executability factors, the graph reasoning score is corrected to obtain a corrected strategy score. Strategy entities with corrected strategy scores greater than or equal to a preset strategy score threshold, along with their corresponding fault-to-phenomenon-to-strategy paths, are selected, resulting in... Figure 5 The candidate strategy screening results shown are as follows: Figure 5 (a) A distribution network fault knowledge graph representing multiple fault-to-phenomenon-to-strategy paths corresponding to each selected valid root cause event. Figure 5 (b) The normalized result of the modified policy score for each candidate policy for each valid root cause event;

[0182] Then, by combining event severity, resource accessibility, execution cost, operational risk, time constraints, and 0-1 integer programming, a global disposal benefit function is constructed. Solving this function under constraints yields the optimal work order combination, and an optimal target work order is generated for each valid root cause event, resulting in the following: Figure 6 The optimal work order arrangement result is shown below; Figure 6 (a) The optimal target work order and its overall utility score for each root cause event. Figure 6 (b) The work order scheduling results under different operation and maintenance resources are shown; the results show that the present invention can effectively realize the fault judgment and auxiliary decision-making functions in complex alarm scenarios of new energy distribution network through unified modeling of multi-source alarms, fusion coding, root cause event identification, knowledge graph reasoning and intelligent work order arrangement.

[0183] After the work order is completed, further feedback information is collected, including alarm clearance status, processing time, rework status, and manual confirmation results. Based on the feedback results, the baseline credibility of the alarm source and the knowledge weights of the fault-to-phenomenon-to-policy path in the knowledge graph are updated to obtain... Figure 7 The weight update results shown are as follows: Figure 7 (a) Shows the baseline confidence levels before and after updates for different alarm sources. Figure 7 (b) shows the knowledge weights before and after different path updates in the knowledge graph. It can be seen that the present invention can adapt the weight coefficients according to the feedback results of work order execution, realize the closed-loop self-learning of the system, and achieve long-term stable operation.

[0184] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0185] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for fault assessment and auxiliary decision-making in a new energy distribution network, characterized in that, Includes the following steps: Acquire several original alarm messages from the power distribution network, extract alarm features, text semantic features, and confidence values ​​for each original alarm message, and construct a trusted fusion vector corresponding to each original alarm message based on the alarm features, text semantic features, and confidence values. Based on a soft clustering strategy, all the trusted fusion vectors are grouped into several potential root cause events, generating an event center vector corresponding to each potential root cause event, and a soft assignment coefficient for each original alarm message belonging to each potential root cause event. Based on the confidence value and soft allocation coefficient of each original alarm message, the root cause confidence of each potential root cause event is calculated, and potential root cause events with a root cause confidence greater than or equal to a preset confidence threshold are selected to obtain several valid root cause events. Construct a distribution network fault knowledge graph that includes several fault-to-phenomenon-to-strategy paths, and calculate the graph reasoning score of each effective root cause event for each strategy entity based on multi-path reasoning of the knowledge graph. The graph reasoning score is corrected according to a preset correction factor to obtain the corrected policy score for each policy entity for each valid root cause event. For each valid root cause event, select policy entities whose correction policy scores are greater than or equal to a preset policy score threshold to obtain a set of candidate handling policies for each valid root cause event. Based on the preset work order utility function, calculate the comprehensive utility score of each strategy entity in each candidate disposal strategy set. Based on all the comprehensive utility scores and binary integer programming, construct a global disposal benefit function. Based on preset constraints, with the maximization of the global disposal benefit function as the optimization objective, solve to obtain the optimal work order combination. Based on the optimal work order combination, generate the optimal target work order for each effective root cause event.

2. The method for fault assessment and auxiliary decision-making in a new energy distribution network according to claim 1, characterized in that, The specific methods for constructing the trusted fusion vector include: Each of the original alarm messages is processed in a standardized manner to obtain the alarm characteristics of each original alarm message; Calculate the confidence value of each original alarm message based on the alarm source benchmark confidence level and preset penalty items; Based on the alarm features and the confidence value, construct a reliable alarm vector for each of the original alarm messages; Extract the textual semantic features of each original alarm message according to a preset text encoding function; Based on the topology graph theory model of the distribution network, the nodes to which the original alarm information belongs are topologically embedded to obtain the topology embedding vector of each original alarm information. By fusing the trusted alarm vector, the text semantic features, and the topological embedding vector, a multimodal fusion vector is obtained; Based on the confidence value and the multimodal fusion vector, calculate the confidence fusion vector for each of the original alarm messages.

3. The method for fault assessment and auxiliary decision-making in a new energy distribution network according to claim 1, characterized in that, The soft clustering strategy, which groups all the trusted fusion vectors into several potential root cause events, specifically includes: Calculate the combined correlation strength of any two of the aforementioned trusted fusion vectors; Based on the trusted fusion vector and the comprehensive correlation strength, an alarm merging objective function is constructed, which includes: an alarm semantic space constraint term, an entropy regularization constraint term, and an alarm correlation constraint term. With the goal of minimizing the alarm merging objective function and the constraint that each original alarm message must be assigned to a potential root cause event, a soft merging matrix and event center vectors of several potential root cause events are obtained. Based on the soft merging matrix, the soft assignment coefficients of each original alarm message belonging to each potential root cause event are obtained.

4. The method for fault assessment and auxiliary decision-making in a new energy distribution network according to claim 3, characterized in that, The comprehensive correlation strength includes: the cosine similarity of the two trusted fusion vectors, the temporal proximity of the two trusted fusion vectors, the electrical coupling of the two alarm devices corresponding to the two trusted fusion vectors, and the prior causal correlation strength between the two alarm categories corresponding to the two trusted fusion vectors.

5. The method for fault assessment and auxiliary decision-making in a new energy distribution network according to claim 1, characterized in that, The formula for calculating the root cause confidence level is as follows: ； In the formula, Indexed by the sequence number of the potential root cause event. For the first Root cause confidence of a potential root cause event For the distribution network in the first The number of raw alarm messages received within the time window. and This is the sequence number index of the original alarm information. For the first The original alarm message belongs to the first The soft assignment coefficient of each potential root cause event. For the first The original alarm message belongs to the first The soft assignment coefficient of each potential root cause event. For the first The confidence value of the original alarm message. The correlation enhancement coefficient, For the first Article and Section The overall correlation strength between the trusted fusion vectors corresponding to each of the original alarm messages.

6. The method for fault assessment and auxiliary decision-making in a new energy distribution network according to claim 2, characterized in that, The calculation method for the graph reasoning score includes: Based on the event center vector of each valid root cause event and the embedding vector of each fault entity, calculate the fault matching score between each valid root cause event and each fault entity. Based on the merging results and soft allocation coefficients of the original alarm information obtained by the soft clustering strategy, the support score of each effective root cause event for each phenomenon entity is calculated. Based on the fault matching score, the support score, and the knowledge weight of each fault-to-phenomenon-to-policy path, the maximum path reasoning score of each effective root cause event for each policy entity is calculated to obtain the graph reasoning score.

7. The method for fault assessment and auxiliary decision-making in a new energy distribution network according to claim 1, characterized in that, The preset correction factors include: strategy conflict factor, security risk penalty factor, and executability factor; the calculation formula for the correction strategy score is: ； In the formula, and These are the sequence index of valid root cause events and the strategy index in the distribution network fault knowledge graph, respectively. For the first One effective root cause event for the strategy The score of the corrected strategy, For the first One effective root cause event for the strategy The graph reasoning score, For the first One effective root cause event and strategy Strategic conflict factors To adopt a strategy Disposal of the first A valid root cause event's security risk penalty factor To adopt a strategy Disposal of the first The executability factor of a valid root cause event.

8. The method for fault assessment and auxiliary decision-making in a new energy distribution network according to claim 1, characterized in that, The preset work order utility function includes: a correction strategy score item, an event severity item, a resource matching degree item, an execution cost item, an execution duration item, and an operational risk item. The expression of the preset work order utility function is: ； In the formula, , and These are the indexes for the sequence number of valid root cause events, the candidate disposal strategy index, and the resource unit index, respectively. To make resource units Assigned to the Identify valid root cause events and implement candidate treatment strategies. The overall utility score, For the first One effective root cause event for candidate handling strategies The score of the corrected strategy, For the first The severity of each valid root cause event. For resource units With candidate handling strategy Resource matching degree , and Respectively utilizing resource units Implement candidate disposal strategy The execution cost, execution time, and operational risks, ~ The weights are respectively the score of the correction strategy, the severity of the event, the resource matching degree, the execution cost, the execution time, and the operational risk.

9. The method for fault assessment and auxiliary decision-making in a new energy distribution network according to claim 1, characterized in that, The expression for the global disposal benefit function is: ； In the formula, For the global disposal benefit function value, For the sequence number index of the valid root cause event, For the first A set of candidate handling strategies for each valid root cause event. For candidate disposal strategy index, For resource unit indexing; Let be the binary decision variables in the binary integer programming problem. Used to indicate whether resource units are selected. Assigned to the Identify valid root cause events and implement candidate treatment strategies. ; To make resource units Assigned to the Identify valid root cause events and implement candidate treatment strategies. The overall utility score.

10. The method for fault assessment and auxiliary decision-making in a new energy distribution network according to claim 6, characterized in that, The method further includes: Collect the handling feedback information of the distribution network after performing fault handling according to each of the optimal target work orders. The handling feedback information includes: alarm suppression rate, actual alarm handling time, number of repeated alarms and fault handling execution quality score. The feedback reward value for each of the optimal target work orders is calculated based on the alarm suppression rate, the actual alarm handling time, the number of repeated alarms, and the fault handling execution quality score. Based on the feedback reward value, fault matching score, support level score and preset graph learning rate corresponding to each optimal target work order, update the knowledge weight of the fault to phenomenon to strategy path corresponding to each optimal target work order; Based on the alarm suppression rate and the preset confidence update coefficient, update the alarm source baseline confidence corresponding to each optimal target work order.

Images