Multi-modal edge fusion-based power fault diagnosis method and multi-modal edge fusion-based power fault diagnosis system

Abstract

The invention discloses a multi-modal edge fusion power fault diagnosis method and system, and relates to the technical field of power fault diagnosis. The method comprises the following steps: establishing a power equipment topology in a preset area, wherein the power equipment topology comprises a plurality of pieces of power equipment; respectively acquiring electrical quantity parameters, non-electrical quantity parameters and peripheral equipment response parameters of the power equipment, and calculating multi-type fault probabilities of the power equipment; and generating a plurality of fault probability distributions according to the multi-type fault probabilities of the power devices, and determining a target fault device and a target fault type based on distribution characteristics of the plurality of fault probability distributions. According to the invention, the accuracy and anti-interference capability of power fault diagnosis are effectively improved.

Description

Technical Field

[0001] This invention relates to the field of power fault diagnosis technology, specifically to a power fault diagnosis method and system based on multimodal edge fusion. Background Technology

[0002] As power systems develop towards larger capacity and higher intelligence, the power grid topology is becoming increasingly complex, making the safe and stable operation of power equipment a core element in ensuring energy supply. Existing power fault diagnosis technologies mostly rely on single electrical quantity parameters such as voltage, current, and power factor, using threshold judgments or simple rule matching to achieve fault detection.

[0003] However, traditional power fault diagnosis methods rely on a single dimension of diagnostic data. They are unable to cope with interference from complex scenarios such as electromagnetic interference and equipment aging by relying solely on electrical quantity parameters, which easily leads to false detections and missed detections. Furthermore, the use of fixed weights cannot adapt to the characteristic differences of different fault types, resulting in limited accuracy of diagnostic results and difficulty in accurately identifying faulty equipment and specific fault types. Summary of the Invention

[0004] This invention provides a multimodal edge fusion power fault diagnosis method and system, aiming to solve the technical problems of insufficient accuracy and anti-interference capability in existing power fault diagnosis technologies.

[0005] In view of the above problems, the present invention provides a power fault diagnosis method and system based on multimodal edge fusion.

[0006] In a first aspect, the present invention provides a power fault diagnosis method based on multimodal edge fusion, comprising: Establish a power equipment topology within a preset area, wherein the power equipment topology includes multiple power devices; The electrical quantity parameters, non-electrical quantity parameters, and peripheral equipment response parameters of each power equipment are collected respectively, and the multi-type fault probabilities of each power equipment are calculated. Based on the multiple types of fault probabilities of each power device, multiple fault probability distributions are generated, and the target fault device and target fault type are determined based on the distribution characteristics of the multiple fault probability distributions.

[0007] Secondly, the present invention provides a multimodal edge fusion power fault diagnosis system, comprising: The device topology construction module is used to establish the power device topology within a preset area, wherein the power device topology includes multiple power devices; The fault probability calculation module is used to collect electrical quantity parameters, non-electrical quantity parameters and peripheral equipment response parameters of each of the power equipment, and calculate the multi-type fault probability of each of the power equipment. The fault location module is used to generate multiple fault probability distributions based on the multiple types of fault probabilities of each of the power equipment, and to determine the target fault equipment and the target fault type based on the distribution characteristics of the multiple fault probability distributions.

[0008] One or more technical solutions provided in this invention have at least the following technical effects or advantages: This invention provides a multimodal edge fusion-based power fault diagnosis method and system. By constructing power equipment topology and fusing three-dimensional data of electrical quantities, non-electrical quantities, and peripheral equipment responses, a multi-level integrated learning architecture fault probability assessment engine is built. An innovative dynamic weight adjustment mechanism based on matching degree is adopted, combined with a pre-trained fault location engine to mine topological fault propagation patterns, achieving intelligent full-process fault diagnosis. It breaks through the limitations of traditional single-parameter diagnosis, laying a solid data foundation for fault assessment through multimodal data cross-validation. The dynamic weight adjustment mechanism intelligently allocates weights based on the matching degree of electrical quantities and peripheral equipment responses, effectively improving the anti-interference capability and accuracy of diagnostic results. The location engine based on topological fault propagation patterns solves the problem of fuzzy location in traditional methods, accurately identifying target faulty equipment and types. The multi-level integrated learning design can adapt to complex and changing operating conditions. Ultimately, a complete solution integrating fault detection, location, and identification is formed, effectively improving the operation and maintenance efficiency and safe operation level of power systems. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 A flowchart illustrating the multimodal edge fusion power fault diagnosis method provided in this embodiment of the invention; Figure 2 This is a schematic diagram of the structure of the multimodal edge fusion power fault diagnosis system provided in an embodiment of the present invention; The components represented by each number in the attached diagram are explained below: Equipment topology construction module 11, fault probability calculation module 12, fault location module 13. Detailed Implementation

[0011] This invention provides a multimodal edge fusion power fault diagnosis method and system to address the technical problems of insufficient accuracy and anti-interference capability in existing power fault diagnosis technologies.

[0012] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0013] It should be noted that the terms "comprising" and "having" are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or modules that are not explicitly listed or that are inherent to these processes, methods, products, or devices.

[0014] Example 1, as Figure 1 As shown, this invention provides a power fault diagnosis method based on multimodal edge fusion, the method comprising: S100: Establish a power equipment topology within a preset area, wherein the power equipment topology includes multiple power equipment.

[0015] In this embodiment of the invention, a power equipment topology is established within a preset area, which includes multiple power devices. Various power devices in a power system do not operate in isolation, but rather form an interconnected organic whole through components such as lines and switches. The power equipment topology serves as a carrier describing the physical connections and electrical relationships between devices. In traditional power fault diagnosis processes, without clear equipment topology support, isolated analysis of parameters for individual devices is insufficient to determine the fault propagation path based on the relationships between devices, easily leading to fault location errors. Subsequent multimodal parameter acquisition and fault probability calculation lack clear spatial scope and object boundaries, resulting in data redundancy or omissions. Therefore, before conducting multimodal edge fusion-based power fault diagnosis, it is necessary to first establish a power equipment topology within a preset area. This defines a clear analysis scope and clarifies the relationship logic between devices for subsequent diagnostic processes, serving as a prerequisite and foundation for accurate fault diagnosis.

[0016] First, determine the boundaries and scope of the pre-defined area. The pre-defined area refers to a specific power network area designated based on the power system's operation and maintenance needs, where fault diagnosis is required. This area can be divided according to voltage level, geographical location, or functional units. For example, a pre-defined area might be a 10kV distribution network zone in an industrial park. The power supply boundary of this zone is: the incoming line connects to the 10kV outgoing line cabinet of a 110kV substation in the city power grid; the outgoing line covers the distribution terminals of three production workshops and two office buildings within the park; and the area does not contain grid-connected equipment of other voltage levels.

[0017] Secondly, a list of electrical equipment within the pre-defined area was compiled. Using a combination of on-site surveys and data verification with the power GIS system, the type, model, serial number, and operating status of all electrical equipment within the pre-defined area were statistically analyzed to form a complete equipment list. For example, the list shows that the electrical equipment in the 10kV distribution network area of ​​this industrial park includes: 1 ring main unit (numbered H-01), 3 distribution transformers (numbered B-01 / B-02 / B-03, corresponding to 3 production workshops respectively), 5 switchgear units (numbered K-01~K-05, covering workshop and office building terminals), and 12 sets of load monitoring terminals (numbered F-01~F-12).

[0018] Furthermore, by reviewing electrical wiring diagrams and verifying cable routing and switch switching logic on-site, the connection relationship data between the collected equipment was clarified to understand the physical connection relationship and electrical correlation logic between various power devices. Emphasis was placed on recording the power supply relationships between upstream and downstream units and the parallel operation relationships. For example, after verification, the connection relationship in this area is as follows: 110kV substation outgoing line cabinet - ring main unit H-01; ring main unit H-01 has three outgoing lines, connecting to distribution transformers B-01, B-02, and B-03 respectively; the outgoing lines of distribution transformer B-01 connect to switch cabinets K-01~K-02, corresponding to production workshop No. 1 and office building No. 1; the outgoing lines of distribution transformers B-02 and B-03 correspond to the remaining workshops and office building respectively, and all switch cabinets are connected to the corresponding load monitoring terminals.

[0019] Finally, the power equipment topology is constructed and verified. Power equipment topology refers to a structured model of all power equipment and their connections within a pre-defined area. It includes not only the physical location information of the equipment but also the electrical connection logic between equipment, such as feeder connections and upstream / downstream power supply relationships. Using topology modeling tools such as power system-specific topology analysis software and visual modeling platforms, the equipment list and connection relationship data are entered into the system to construct a visual topology model with equipment as nodes and connections as edges. Subsequently, maintenance personnel conduct on-site wiring verification to correct any discrepancies between the topology model and the actual network, ensuring the accuracy of the topology model. For example, using a power GIS topology modeling tool, ring main unit H-01, distribution transformers B-01~B-03, etc., are used as nodes, and the power supply lines between the equipment are used as edges to generate a topology map of the 10kV distribution network area in the industrial park. After on-site verification, the incorrect connection relationship between distribution transformer B-02 and switchgear K-03 in the original drawing was corrected, ultimately forming a power equipment topology model consistent with reality.

[0020] In this embodiment of the invention, by establishing a power equipment topology within a preset area, a clear object range is defined for subsequent multimodal parameter acquisition, avoiding redundant acquisition of data from irrelevant equipment areas and improving the efficiency of edge data acquisition; the power supply and connection relationships between equipment are clarified, providing a basis for subsequent analysis of fault propagation patterns between equipment and fault location based on topological features; the constructed visual topology model can intuitively present the equipment association logic, providing a clear reference framework for operation and maintenance personnel to understand diagnostic results and carry out fault handling.

[0021] S200: Collect electrical quantity parameters, non-electrical quantity parameters and peripheral equipment response parameters of each power device respectively, and calculate the multi-type fault probability of each power device.

[0022] In this embodiment of the invention, electrical quantity parameters, non-electrical quantity parameters, and peripheral device response parameters of each power device are collected to calculate the multi-type fault probabilities of each power device. Different fault types of power devices exhibit differentiated characteristic signals: short-circuit faults are mainly reflected in electrical quantities, overheating faults are more pronounced in non-electrical quantities, and fault propagation is reflected in the response parameters of associated peripheral devices. Traditional fault diagnosis relies only on a single electrical quantity parameter, which cannot comprehensively cover the characteristic dimensions of various faults, easily leading to the omission of fault features or misjudgment; at the same time, it lacks a probability assessment system customized for equipment, making it difficult to accurately quantify the occurrence probability of different fault types. Therefore, this step, by collecting multi-dimensional parameters and using a customized fault probability assessment engine to calculate the fault probability, is the core link in realizing multimodal fusion diagnosis, providing accurate probabilistic basis for subsequent fault location.

[0023] Step S200 in the method provided in this embodiment of the invention includes: From the plurality of power devices, a first power device is determined, and the first electrical quantity parameter, the first non-electrical quantity parameter, and the first peripheral device response parameter of the first power device are extracted; The fault probability assessment engine bound to the first power equipment is invoked. The fault probability assessment engine includes a first evaluator, a second evaluator, and a third evaluator. Each evaluator includes assessment units for multiple fault types. Based on the first evaluator, the second evaluator and the third evaluator respectively, the first electrical quantity parameter, the first non-electrical quantity parameter and the first peripheral equipment response parameter are processed to obtain the multi-type fault probability of the first power equipment; The multi-type fault probabilities of the remaining power equipment are obtained by obtaining the multi-type fault probabilities of the first power equipment in the same way as obtaining the multi-type fault probabilities of the first power equipment.

[0024] First, a first power device is identified from the plurality of power devices, and its first electrical quantity parameters, first non-electrical quantity parameters, and first peripheral device response parameters are extracted. The first power device refers to the first power device selected from the preset area power device topology to be diagnosed and analyzed; it is the first object for subsequent parameter acquisition and probability calculation. The first electrical quantity parameters refer to the electrical data generated during the operation of the first power device itself, reflecting the electrical operating status of the device. The first non-electrical quantity parameters refer to the physical state data of the first power device, reflecting its physical operating status. The first peripheral device response parameters refer to the electrical quantity parameters of peripheral devices that have a direct electrical connection with the first power device, used to reflect the propagation characteristics of faults between devices. From the existing power equipment topology, select the equipment to be analyzed first as the first power equipment; deploy corresponding sensors to collect equipment parameters: deploy electrical quantity acquisition sensors: at the incoming / outgoing terminals of the first power equipment, deploy voltage transformers, current transformers, and power factor transducers, and connect them to the data acquisition unit through the secondary wiring ports of the equipment; deploy non-electrical quantity acquisition sensors: according to the type of non-electrical quantity, deploy dedicated sensors such as temperature sensors mounted on the surface or embedded in the first power equipment, acoustic sensors fixed on the equipment casing, and electromagnetic radiation sensors deployed around the equipment, and connect the sensors to the data acquisition unit wirelessly or via wired means; obtain peripheral equipment response parameters: using the existing electrical quantity monitoring sensors of the peripheral equipment, synchronously retrieve their real-time electrical quantity data through power communication networks such as power line carrier or 5G, as response parameters.

[0025] For example, in a 10kV distribution network area of ​​an industrial park, distribution transformer B-01 is selected as the first power equipment: First electrical quantity parameters are collected: a voltage transformer is deployed at the 10kV incoming terminal of B-01, a current transformer is deployed at the outgoing terminal, and a power factor transducer is deployed on the secondary side; ultimately, the following parameters are collected: real-time voltage of B-01 is 10.2kV, load current is 120A, and power factor is 0.85; First non-electrical quantity parameters are collected: a PT100 temperature sensor is embedded inside the winding of B-01, and the winding temperature is collected to be 78℃; a vibration and noise collector is fixed on the top of the equipment casing, and the operating acoustic signal is collected to be 120dB, including abnormal high-frequency noise; First peripheral equipment response parameters are collected: using the current transformer already deployed in switchgear K-01 connected to B-01, the real-time current data is synchronously retrieved via power line carrier communication and is 118A, which is 15% higher than the reference value.

[0026] Secondly, the fault probability assessment engine bound to the first power equipment is invoked. The fault probability assessment engine includes a first evaluator, a second evaluator, and a third evaluator, each of which includes assessment units for multiple fault types.

[0027] The construction steps of the fault probability assessment engine bound to the first power equipment include: Obtain the historical monitoring records of the first power equipment. The historical monitoring records include multiple sets of historical monitoring data. Each set of historical monitoring data includes historical electrical quantity parameters, historical non-electrical quantity parameters, historical peripheral equipment response parameters, and corresponding fault identification parameters. Based on multiple sets of historical monitoring data, a sample electrical quantity parameter set, a sample non-electrical quantity parameter set, and a sample peripheral equipment response parameter set are constructed. Based on the fault identification parameters and the multiple fault types, multiple sample fault label sets are constructed, and the multiple fault types correspond one-to-one with the multiple sample fault label sets; Based on the sample electrical quantity parameter set and the multiple sample fault label sets, a first evaluator is constructed; Based on the sample non-electrical quantity parameter set and the multiple sample fault label sets, a second evaluator is constructed; Based on the sample peripheral device response parameter set and the multiple sample fault label sets, a third evaluator is constructed; The first evaluator, the second evaluator, and the third evaluator are integrated to obtain the fault probability evaluation engine for the first power equipment.

[0028] First, the historical monitoring records of the first power equipment are obtained. These records include multiple sets of historical monitoring data, each set containing historical electrical parameters, historical non-electrical parameters, historical peripheral equipment response parameters, and corresponding fault identification parameters. The historical monitoring records refer to the comprehensive monitoring data set of the first power equipment over its past operating cycles, including the equipment's status parameters and corresponding operating status identifiers. Fault identification parameters are status tags bound to each set of historical monitoring data, used to identify whether the first power equipment is operating normally or experiencing a specific fault type, such as a short-circuit fault or overheating fault. Historical data of the first power equipment is retrieved from the long-term monitoring database and maintenance fault record system. Records with ≥95% data integrity are filtered to ensure that each set of records simultaneously includes historical electrical parameters, historical non-electrical parameters, and historical peripheral equipment response parameters. The equipment status is confirmed through maintenance work orders and fault alarm logs, and the corresponding fault identification parameters for the first power equipment are matched to each set of records.

[0029] For example, retrieve the historical monitoring records of B-01, totaling 1200 sets of valid data: each set of records includes: historical electrical quantity parameters (B-01 itself): voltage 10.2kV, load current 120A; historical non-electrical quantity parameters (B-01 itself): winding temperature 78℃, operating acoustic signal 120dB; historical peripheral equipment response parameters (associated switchgear K-01): K-01 real-time current 118A. Example of fault identification parameters: Record 150: B-01 arc fault; Record 380: B-01 overheat fault; Record 620: B-01 insulation fault; Record 850: B-01 mechanical fault; the remaining 1196 records: B-01 normal operation.

[0030] Secondly, based on multiple sets of historical monitoring data, sample electrical quantity parameter sets, sample non-electrical quantity parameter sets, and sample peripheral equipment response parameter sets are constructed. The sample electrical quantity parameter set refers to a structured sample set containing only the historical electrical quantity parameters of the first power equipment itself, serving as input features for training the electrical quantity estimator. The sample non-electrical quantity parameter set refers to a structured sample set containing only the historical non-electrical quantity parameters of the first power equipment itself, serving as input features for training the non-electrical quantity estimator. The sample peripheral equipment response parameter set refers to a structured sample set containing only the historical electrical quantity parameters of the peripheral equipment associated with the first power equipment, serving as input features for training the peripheral response estimator.

[0031] For example, the historical monitoring records of the first power equipment after screening are classified and split, and the historical electrical quantity parameters, historical non-electrical quantity parameters, and historical response parameters of the associated peripheral equipment are extracted from each group of records. Data preprocessing is performed on the split parameters, including removing outliers exceeding the rated operating range of the equipment, filling missing values ​​with the mean of parameters under the same operating conditions, and standardizing the parameters to a preset value range, such as 0-1. The preprocessed parameters are then organized into independent structured sample sets, with each sample set stored in a format where rows correspond to a single group of monitoring data and columns correspond to a single parameter dimension. For example, based on the 1200 records of the distribution transformer mentioned above, the following are constructed: Sample electrical quantity parameter set: 1200 rows × 2 columns, with columns corresponding to transformer voltage and transformer load current; Sample non-electrical quantity parameter set: 1200 rows × 2 columns, with columns corresponding to transformer winding temperature and transformer acoustic signal strength; Sample peripheral equipment response parameter set: 1200 rows × 1 column, with a column corresponding to the current of the associated switchgear.

[0032] Furthermore, based on the fault identification parameters and the multiple fault types, multiple sample fault label sets are constructed, with each fault type corresponding one-to-one with a sample fault label set. A sample fault label set is a set of labels that corresponds one-to-one with a specific operating state of the first power equipment, used to identify whether each set of data in the sample parameter set corresponds to that specific state of the first power equipment; a label value of 1 indicates a correspondence, and 0 indicates no correspondence.

[0033] For example, the target operating state type of the first power equipment is determined, including normal operating state and common fault types of the equipment such as arc fault, overheating fault, insulation fault, and mechanical fault. For each target operating state type, all data samples in the sample parameter set are iterated one by one: if the fault identification parameter corresponding to a sample is the target operating state type, the sample is labeled 1; otherwise, it is labeled 0. For each target operating state type, an independent sample fault label set is generated to achieve a one-to-one correspondence between the target operating state type and the sample fault label set. For example, for the arc fault state of the above-mentioned distribution transformer B-01, a corresponding sample fault label set is generated: 1200 sets of samples are traversed, and only samples with the fault identification of arc fault of the transformer are labeled 1, and all other samples are labeled 0. Similarly, for overheating fault, insulation fault, and other states, 1200 independent sample fault label sets are generated respectively.

[0034] Secondly, based on the sample electrical quantity parameter set and the multiple sample fault label sets, a first evaluator is constructed.

[0035] The first evaluator is constructed based on the sample electrical quantity parameter set and the multiple sample fault label sets, including: A first fault type is determined from the plurality of fault types, and a corresponding first sample fault label set is determined from the plurality of sample fault label sets; Construct a multi-first evaluation sub-unit architecture; Based on the sample electrical quantity parameter set and the first sample fault label set, the architecture of the plurality of first evaluation sub-units is trained until convergence, thereby obtaining a plurality of first evaluation sub-units; The plurality of first evaluation subunits are integrated to obtain a first evaluation unit; Following the method of obtaining the first evaluation unit for the first fault type, the evaluation units for the remaining fault types are trained to obtain multiple evaluation units; The multiple evaluation units are integrated to obtain the first evaluator.

[0036] First, a first fault type is determined from the multiple fault types, and a corresponding first sample fault label set is determined from multiple sample fault label sets. The first fault type refers to the fault type of the first evaluation unit to be constructed selected from the multiple fault types of the first power equipment. The first sample fault label set is a set of sample fault labels that corresponds one-to-one with the first fault type, used to identify whether the first power equipment corresponding to the sample electrical quantity parameter set is in the first fault type. From all the preset fault types of the first power equipment, any fault type is selected as the first fault type; from multiple sample fault label sets, the first sample fault label set uniquely corresponding to the first fault type is matched. For example, taking B-01 as an example, the preset fault types include arc fault, overheating fault, insulation fault, and mechanical fault. Arc fault is selected as the first fault type, and the corresponding distribution transformer arc fault sample fault label set is matched. Label 1 indicates that the sample corresponds to an arc fault, and 0 indicates no correspondence.

[0037] Secondly, multiple first evaluation sub-unit architectures are constructed. The first evaluation sub-unit architecture is the basic model architecture used to process sample electrical quantity parameters and output binary judgment results for the first fault type, and it is constructed using a classification model. Models suitable for electrical quantity parameter classification, such as random forest and support vector machine (SVM), are selected as the basic models for the first evaluation sub-units. Independent model parameters are configured for each sub-unit to ensure sub-unit diversity, and multiple first evaluation sub-unit architectures with consistent structure but different parameters are constructed. For example, 10 first evaluation sub-unit architectures are constructed, all using binary classification models. The 10 sub-units use different model types and different parameter configurations to ensure the diversity of judgment dimensions. For example: Sub-unit 1 uses a random forest classifier, with 100 decision trees and a maximum depth of 5 per tree; Sub-unit 2 uses a random forest classifier, with 80 decision trees and a maximum depth of 6 per tree; Sub-unit 3 uses an SVM classifier, with a kernel function equal to radial basis function and a regularization parameter of 1.0.

[0038] Further, the multiple first evaluation sub-unit architectures are trained until convergence based on the sample electrical quantity parameter set and the first sample fault label set, resulting in multiple first evaluation sub-units. Convergence means that the validation set accuracy of the evaluation sub-unit model remains stable within a preset range for multiple consecutive rounds, and the model output no longer fluctuates significantly. A first evaluation sub-unit refers to a trained model that can output a binary judgment result of the first fault type based on the input electrical quantity parameters. The sample electrical quantity parameter set is preprocessed, including outlier removal, missing value filling, and standardization, and divided into a training set and a validation set according to a preset ratio. The preprocessed sample electrical quantity parameter set is used as input, and the first sample fault label set is used as supervision labels, which are then input into the multiple first evaluation sub-unit architectures for training. The validation set accuracy is calculated after each round of training until the model's validation set accuracy remains stable at ≥90% for 5 consecutive rounds, at which point the model training is considered converged, resulting in multiple trained first evaluation sub-units.

[0039] For example, after preprocessing the sample electrical quantity parameter set, it is divided into a training set and a validation set in an 8:2 ratio. Ten first evaluation sub-unit architectures are input for training: Sub-unit 1 converges after 12 rounds of training, with a validation set accuracy of 92%; Sub-unit 2 converges after 10 rounds of training, with a validation set accuracy of 91%; Sub-unit 3 converges after 15 rounds of training, with a validation set accuracy of 90%; the remaining sub-units are trained in the same way, and finally 10 first evaluation sub-units that can output the binary judgment result of arc fault are obtained.

[0040] Then, the multiple first evaluation subunits are integrated to obtain a first evaluation unit. The proportion of 1-valued outputs is calculated as the integration method for the final fault probability by statistically analyzing the binary output results of multiple evaluation subunits. A first evaluation unit refers to a model unit that, corresponding to a first fault type, can output the probability that the first power equipment is in that fault type. The output rule is determined as follows: after receiving electrical quantity parameters, the first evaluation subunit outputs a binary result of 1 or 0; the output results of multiple first evaluation subunits are statistically analyzed, and the proportion of subunits outputting 1 is calculated out of the total number of subunits; this proportion is the final fault probability of the first evaluation unit for the first fault type, and the integrated units form the first evaluation unit. For example, integrating 10 first evaluation subunits yields the first evaluation unit, namely, the arc fault evaluation unit.

[0041] Then, following the same method as obtaining the first evaluation unit for the first fault type, evaluation units for the remaining fault types are trained to obtain multiple evaluation units. Evaluation units for the remaining fault types refer to model units with the same structure as the first evaluation unit, corresponding to other fault types of the first power equipment besides the first fault type. Each of the remaining fault types of the first power equipment is selected sequentially; the above operations are repeated: matching the corresponding sample fault label set, constructing the sub-unit architecture, training the sub-units until convergence, and integrating the output layer to obtain the evaluation unit corresponding to each fault type; the evaluation units for all fault types are then aggregated to obtain multiple evaluation units.

[0042] For example, repeat the above steps to construct evaluation units for overheating faults, insulation faults, and mechanical faults of distribution transformers: Overheating fault evaluation unit: integrates 10 sub-units and outputs the overheating fault probability; similarly, insulation fault and mechanical fault evaluation units are obtained; finally, four evaluation units corresponding to different fault types are obtained, each of which contains 10 sub-units.

[0043] Finally, the multiple evaluation units are integrated to obtain the first evaluator. The first evaluator is a set of models that integrate all fault type evaluation units and can output the probabilities of multiple fault types in parallel based on a single electrical quantity parameter. Input channel configuration: The electrical quantity parameter input port of the first power equipment is simultaneously associated with the input terminals of multiple evaluation units; Output rules are determined: When the electrical quantity parameter is input to the first evaluator, the parameter is simultaneously input to all evaluation units, and each evaluation unit calculates the probability of the corresponding fault type; The output results of all evaluation units are summarized and encapsulated to obtain the first evaluator.

[0044] Next, based on the sample non-electrical quantity parameter set and the multiple sample fault label sets, a second evaluator is constructed. The second evaluator is a set of models specifically designed to process the non-electrical quantity parameters of the first power equipment and output the probabilities of each fault type of the equipment. Its structure is consistent with the first evaluator, and it is constructed by integrating sub-units at the output layer and integrating evaluation units at the input layer. A fault type of the first power equipment is selected and matched with the corresponding second sample fault label set; multiple binary classification sub-unit architectures are constructed, with different parameters configured to ensure diversity; the sample non-electrical quantity parameter set is preprocessed to remove outliers, fill missing values, and standardize; the sub-units are trained in conjunction with the label set until convergence, i.e., the validation set accuracy is ≥90%; using the output layer integrated sub-unit, the proportion of 1-value outputs is statistically analyzed as the fault probability to obtain the evaluation unit for the corresponding fault type; the above process is repeated to obtain evaluation units for all fault types, and then all evaluation units are integrated through the input layer to encapsulate and obtain the second evaluator.

[0045] For example, taking the first power equipment as an example: Sample non-electrical quantity parameter set: containing 1200 sets of data including transformer winding temperature and acoustic signal; construct multiple sub-units, the specific training process is the same as the second evaluator, after training, integrate the sub-units to obtain the evaluation units for each fault type, after the input layer is integrated, the second evaluator can receive temperature and acoustic signals, and output the probabilities of arc faults, overheating faults, etc. in parallel.

[0046] Furthermore, a third evaluator is constructed based on the sample peripheral device response parameter set and the multiple sample fault label sets. The third evaluator is a set of models specifically designed to process the response parameters of the first power equipment associated with its peripheral devices, outputting the probabilities of each fault type. Its structure is consistent with the first and second evaluators. A fault type of the first power equipment is selected and matched with the corresponding third sample fault label set; multiple binary classification sub-unit architectures are constructed with different parameters configured; the sample peripheral device response parameter set is preprocessed, and the sub-units are trained to convergence using the label sets; the output layer integrates the sub-units to obtain the evaluation unit corresponding to the fault type; this process is repeated to obtain evaluation units for all fault types, which are then integrated through the input layer and encapsulated to obtain the third evaluator.

[0047] For example, taking the first power equipment as an example: Sample peripheral equipment response parameter set: contains 1200 sets of data on the current of the switchgear connected to the transformer; construct multiple sub-units, the specific training process is the same as the second evaluator, after training integration, the third evaluator can receive the switchgear current parameters and output the probability of each fault type of the transformer in parallel.

[0048] Finally, the first evaluator, the second evaluator, and the third evaluator are integrated to obtain the fault probability assessment engine for the first power equipment. The fault probability assessment engine is a dedicated model module uniquely bound to the first power equipment. Integrating the first, second, and third evaluators, it can receive three types of parameter inputs and output the comprehensive probability results for each fault type of the equipment. The unique equipment identifier refers to the exclusive information used to bind the assessment engine to the first power equipment, such as the equipment number and factory serial number, ensuring exclusive engine compatibility. Dedicated input channels are configured for the three evaluators: the first evaluator is bound to the equipment's own electrical quantity parameter input port, the second evaluator is bound to the equipment's own non-electrical quantity parameter input port, and the third evaluator is bound to the peripheral equipment response parameter input port; the output format of the three evaluators is unified, all outputting in the form of fault type-probability value key-value pairs; the assessment engine is bound to the unique identifier of the first power equipment, establishing a dedicated association; the three evaluators are encapsulated as independent functional modules, and integrated to form the fault probability assessment engine.

[0049] For example, taking B-01 as an example: Configure input channels: voltage / current input to the first evaluator, temperature / acoustic signal input to the second evaluator, and switchgear current input to the third evaluator; Unify output format: such as arc fault: 65%, overheat fault: 28%; Bind device number B-01, after encapsulation, a transformer-specific fault probability evaluation engine is obtained, which can simultaneously receive three types of parameters and output the probability of each fault type.

[0050] Furthermore, based on the first evaluator, the second evaluator, and the third evaluator, the first electrical quantity parameter, the first non-electrical quantity parameter, and the first peripheral equipment response parameter are processed respectively to obtain the multi-type fault probabilities of the first power equipment.

[0051] Specifically, based on the processing of the first electrical quantity parameter, the first non-electrical quantity parameter, and the first peripheral equipment response parameter by the first evaluator, the second evaluator, and the third evaluator, respectively, the multi-type fault probabilities of the first power equipment are obtained, including: Based on the first electrical quantity parameter, the first fault probability of each fault type is obtained by each evaluation unit of the first evaluator; Based on the first non-electrical quantity parameter, the second fault probability of each fault type is obtained by each evaluation unit of the second evaluator; Based on the response parameters of the first peripheral device, the third fault probability of each fault type is obtained by each evaluation unit of the third evaluator. Based on the first and third failure probabilities of each failure type, the first and second weights of each failure type are obtained. By weighting the first and second fault probabilities of each fault type using the first and second weights, the multi-type fault probabilities of the first power equipment are obtained.

[0052] First, based on the first electrical quantity parameter, the first fault probability of each fault type is obtained by each evaluation unit of the first evaluator.

[0053] Specifically, based on the first electrical quantity parameter, the first fault probability of each fault type is obtained by each evaluation unit of the first evaluator, including: Input the first electrical quantity parameter into the first evaluator; The first evaluator simultaneously distributes the first electrical quantity parameter to multiple internal evaluation units. Each evaluation unit processes the first electrical quantity parameter according to multiple internal evaluation sub-units to obtain multiple evaluation result sets. Based on the multiple evaluation result sets, the first failure probability of each failure type is obtained.

[0054] First, the first electrical quantity parameter is input into the first evaluator. The first electrical quantity parameter refers to the real-time electrical operating parameters of the first power equipment, such as voltage, current, and power factor, which are the core input data for evaluating electrical faults of the equipment. The first electrical quantity parameter needs to be preprocessed to ensure that its parameter format and numerical range are consistent with those used during the training of the first evaluator, so as to ensure that the model can recognize it correctly. The preprocessing of the collected first electrical quantity parameter includes removing outliers that exceed the rated operating range of the equipment, filling in missing data, and standardizing to the [0,1] interval; the preprocessed first electrical quantity parameter is then input into the model through the dedicated input port of the first evaluator to complete the parameter input operation.

[0055] For example, taking the first power equipment as an example: the original first electrical quantity parameters collected are: voltage 10.2kV and load current 120A; preprocessing: the voltage is standardized to 0.7 within the range of 10kV±5%, and the current is standardized to 0.6 within the rated 200A; the standardized two-dimensional parameters [0.7,0.6] are input into the first evaluator.

[0056] Secondly, the first evaluator simultaneously distributes the first electrical quantity parameter to multiple internal evaluation units. Each evaluation unit processes the first electrical quantity parameter according to its multiple internal evaluation sub-units, obtaining multiple evaluation result sets. Input layer integration and distribution is the core operating mechanism of the first evaluator, meaning that the input electrical quantity parameter is synchronously distributed to all internal evaluation units. Each evaluation unit corresponds to a fault type, realizing parallel evaluation of multiple fault types. The evaluation result set is the output set of multiple evaluation sub-units within a single evaluation unit, with each sub-unit outputting a binary result of 1 or 0. The first evaluator, through the input layer integration mechanism, simultaneously distributes the preprocessed first electrical quantity parameter to each internal fault type evaluation unit; after receiving the parameter, each evaluation unit distributes it to multiple internal evaluation sub-units; each evaluation sub-unit independently judges based on the trained model parameters, outputting a binary result of 1 or 0, and the output results of all sub-units of a single evaluation unit are summarized into an evaluation result set.

[0057] For example, the first evaluator contains four evaluation units, such as arc fault and overheat fault, and each unit contains 10 sub-units: the first evaluator synchronously distributes the parameter [0.7, 0.6] to the four evaluation units; the 10 sub-units of the arc fault evaluation unit independently judge and output the evaluation result set of the unit as [1,0,1,0,1,1,0,1,1,0]; the other evaluation units similarly output their respective evaluation result sets.

[0058] Further, based on the multiple evaluation result sets, the first fault probability for each fault type is obtained. The first fault probability is calculated based on the evaluation result set of a single evaluation unit, reflecting the probability value of the first power equipment being in the corresponding fault type of that evaluation unit. The calculation logic is an output layer integration method. Output layer integration calculation refers to the proportion of the number of sub-units with output 1 in the evaluation result set to the total number of sub-units. This proportion is the first fault probability for the corresponding fault type. For the evaluation result set of each evaluation unit, the number of sub-units with output 1 is counted; the first fault probability is calculated according to the formula: First fault probability = (Number of 1s in the result set / Total number of sub-units in this unit) × 100%; the first fault probabilities corresponding to all evaluation units are calculated sequentially, and the first fault probability set for each fault type is obtained by summarizing them.

[0059] For example, the result sets of each evaluation unit are calculated as follows: Arc fault evaluation unit result set [1,0,1,0,1,1,0,1,1,0], the number of 1s is 6, the total number of sub-units is 10, and the first fault probability = (6 / 10) × 100% = 60%; Overheating fault evaluation unit: result set [0,0,0,1,0,0,0,0,0,0], the number of 1s is 1, and the first fault probability = (1 / 10) × 100% = 10%; The other fault type evaluation units are calculated in the same way, and finally the first fault probability set of each fault type is obtained: arc fault probability 60%, overheating fault probability 10%, insulation fault probability 20%, and mechanical fault probability 10%.

[0060] Secondly, based on the first non-electrical quantity parameters, the second fault probability for each fault type is obtained by each evaluation unit of the second evaluator. The first non-electrical quantity parameters refer to the non-electrical quantity parameters of the first power equipment collected in real time, such as winding temperature and acoustic signals, which are preprocessed to meet the input requirements of the second evaluator. The second fault probability refers to the probability value of the first power equipment being in the corresponding fault type, output by each evaluation unit of the second evaluator based on the first non-electrical quantity parameters. The collected first non-electrical quantity parameters are preprocessed to remove outliers and standardize; the preprocessed first non-electrical quantity parameters are input into the second evaluator and simultaneously transmitted to the evaluation units of each fault type; each evaluation unit independently calculates and outputs the probability of the corresponding fault type, which is the second fault probability for each fault type, and the calculation logic is the same as that of the first fault probability. For example, the first non-electrical quantity parameters are: preprocessed winding temperature 80℃, acoustic signal 125dB; after being input into the second evaluator, each evaluation unit outputs the second fault probability: arc fault 30%, overheating fault 60%, insulation fault 20%, and mechanical fault 10%.

[0061] Further, based on the first peripheral device response parameters, the third fault probability for each fault type is obtained by each evaluation unit of the third evaluator. The first peripheral device response parameters refer to the electrical quantity parameters collected in real time by the peripheral devices associated with the first power equipment, which are preprocessed to meet the input requirements of the third evaluator. The third fault probability refers to the probability value output by each evaluation unit of the third evaluator based on the first peripheral device response parameters, indicating that the first power equipment is in the corresponding fault type. The collected first peripheral device response parameters are preprocessed; the preprocessed parameters are input into the third evaluator and simultaneously transmitted to the evaluation units for each fault type; each evaluation unit independently calculates and outputs the probability of the corresponding fault type, which is the third fault probability for each fault type. The calculation logic is the same as that of the first fault probability. For example, the first peripheral device response parameters are: preprocessed associated switchgear current 118A; after inputting into the third evaluator, each evaluation unit outputs the third fault probability: arc fault 70%, overheating fault 20%, insulation fault 10%, and mechanical fault 10%.

[0062] Then, based on the first fault probability and the third fault probability of each fault type, the first weight and the second weight of each fault type are obtained.

[0063] Among them, based on the first fault probability and the third fault probability of each fault type, the first weight and the second weight of each fault type are obtained, including: Determine the first fault type from among multiple fault types; Calculate the matching degree between the first fault probability and the third fault probability corresponding to the first fault type; The matching degree is used as the first weight corresponding to the first fault type, and the second weight corresponding to the first fault type is determined based on the first weight. In accordance with the method of obtaining the first weight and the second weight of the first fault type, the first weight and the second weight corresponding to the other fault types are obtained respectively.

[0064] First, the primary fault type is determined from multiple fault types. The primary fault type refers to the first fault type selected from all preset fault types of the first power equipment for weight calculation; it is the starting point for weight calculation. Preset fault types are pre-defined fault categories for the first power equipment, such as arc faults, overheating faults, insulation faults, and mechanical faults, covering common fault modes of the equipment. All preset fault types of the first power equipment are compiled into a fault type list; one fault type is arbitrarily selected from this list and designated as the primary fault type, serving as the first object for weight calculation. For example, if the preset fault type list for the first power equipment B-01 is: arc fault, overheating fault, insulation fault, and mechanical fault, then arc fault is selected as the primary fault type for subsequent weight calculations.

[0065] Secondly, the matching degree between the first fault probability and the third fault probability corresponding to the first fault type is calculated. The matching degree is the core indicator for quantifying the consistency between the first fault probability and the third fault probability, and its value range is [0,1]. The physical meaning of the matching degree is to reflect the consistency of fault propagation in the power system. When a target device fails, its own electrical quantity will undergo abnormal changes. These changes will propagate to peripheral devices through electrical connections, triggering electrical quantity responses from the peripheral devices. The higher the matching degree between the two probabilities, the more the fault propagation conforms to the physical laws of the power system. The first fault probability P1 and the third fault probability P3 corresponding to the first fault type are retrieved; the matching degree is calculated using the formula: matching degree = 1 - |P1 - P3|; if the calculation result exceeds the [0,1] interval, it is corrected according to the boundary value, with a result <0 taken as 0 and a result >1 taken as 1.

[0066] For example, the first fault type is an arc fault: the probability of the first fault P1 is 60%, the probability of the third fault P3 is 70%, and the matching degree is 1 - |60% - 70%| = 0.9. The matching degree of 0.9 indicates that the electrical quantity abnormality caused by the transformer arc fault is mostly propagated to the peripheral equipment and generates corresponding responses. The consistency of fault propagation is high, which is in line with the physical law of fault propagation in the power system. There are only minor differences, which may be caused by detection errors of the peripheral equipment.

[0067] Further, the matching degree is used as the first weight corresponding to the first fault type, and a second weight corresponding to the first fault type is determined based on the first weight. The first weight corresponds to the weighting coefficient of the first fault probability, and its value is equal to the matching degree; its physical meaning is to adjust the credibility of electrical quantity detection results through fault propagation consistency. The higher the matching degree, the larger the first weight, and the higher the reference value of electrical quantity detection. The second weight corresponds to the weighting coefficient of the second fault probability, and the formula is: second weight = 1 - first weight; its physical meaning is to adjust the credibility of non-electrical quantity detection results. The lower the matching degree, the larger the second weight, and the higher the reference value of non-electrical quantity detection. The physical logic of weight allocation: a high matching degree indicates that the electrical quantity detection is confirmed by the external response, the credibility is high, and the first weight is large; a low matching degree indicates that the electrical quantity detection may be interfered with and depends on non-electrical quantity detection, and the second weight is large. The matching degree is assigned as the first weight of the first fault type. To avoid the influence of extreme values, it is limited to [0.1, 0.9] according to the actual engineering. The second weight = 1 - first weight. The first and second weights of this fault type are recorded.

[0068] For example, in the case of a continuous arc fault: the matching degree = 0.9; the second weight = 1 - 0.9 = 0.1. The matching degree is extremely high, the electrical quantity detection results are consistent with the peripheral equipment response, the peripheral response fully confirms the electrical quantity abnormality, and the credibility is extremely high. Therefore, a high first weight of 0.9 is assigned, and only the non-electrical quantity detection results are assigned a second weight of 0.1. When weighting, the electrical quantity detection results are the main factor.

[0069] Next, following the same method used to obtain the first and second weights for the first fault type, obtain the first and second weights for the remaining fault types. The remaining fault types refer to those in the fault type list other than the first fault type. Select each of the remaining fault types in sequence; repeat the above steps for each type: retrieve the corresponding first and third fault probabilities, calculate the matching degree, and determine the first and second weights; summarize the weights of all fault types to form a weight description.

[0070] For example, the weights of other fault types are calculated as follows: Overheating fault P1=10%, P3=20%, matching degree=1-|10%-20%|=0.9; first weight=0.9, second weight=1-0.9=0.1. The matching degree is high, the consistency of fault propagation is good, the correlation between electrical quantity detection and external response is strong, and the reliability is high. Therefore, a first weight of 0.9 is assigned, and non-electrical quantity detection is assigned a weight of 0.1. Insulation fault P1=20%, P3=10%, matching degree=1-|20%-10%|=0.9; first weight=0.9, second weight=0.1. The matching degree is high, the consistency between electrical quantity detection and external response is good, and the fault propagation conforms to physical laws. The results of electrical quantity detection are the main factor, and non-electrical quantity detection is the auxiliary factor. For mechanical faults P1=10% and P3=10%, the matching degree is 1 - |10% - 10%| = 1.0; the first weight is 0.9 (based on the actual engineering limit), and the second weight is 1 - 0.9 = 0.1. The matching degree is extremely high, the electrical quantity detection and the external response are completely consistent, the fault propagation consistency is extremely strong, and the reliability is the highest. The electrical quantity detection results are the primary factor. The final weights for each fault type are as follows: the first weight for arc faults is 0.9, and the second weight is 0.1; the first weight for overheating faults is 0.9, and the second weight is 0.1; the first weight for insulation faults is 0.9, and the second weight is 0.1; and the first weight for mechanical faults is 0.9, and the second weight is 0.1.

[0071] Based on this, the first and second fault probabilities of each fault type are weighted using the first and second weights to obtain the multi-type fault probabilities of the first power equipment. The multi-type fault probability is the final fault probability obtained by multiplying the first and second fault probabilities of each fault type of the first power equipment by their corresponding first and second weights, respectively, and then summing the results. This comprehensively reflects the true probability that the equipment is in that fault type. Weighting refers to combining the reliability weights of electrical quantity detection and non-electrical quantity detection, and improving the accuracy and robustness of fault probability assessment by weighted summation of the two types of detection results. The weighting formula is: Final probability of a fault type = First fault probability × First weight + Second fault probability × Second weight. The process involves: identifying all fault types of the first power equipment, their corresponding first and second fault probabilities, and weight parameters; calculating the final fault probability for each fault type using the weighting formula; summing the final probabilities of all fault types to form a multi-type fault probability set for the first power equipment; and verifying the reasonableness of the final probability results to ensure error-free calculation.

[0072] For example, taking the first electrical equipment B-01 as an example, the final probability is calculated based on the first weight and the second weight: Known parameters: Arc fault: first fault probability 60%, second fault probability 30%, first weight 0.9, second weight 0.1; Overheating fault: first fault probability 10%, second fault probability 60%, first weight 0.9, second weight 0.1; Insulation fault: first fault probability 20%, second fault probability 20%, first weight 0.9, second weight 0.1; Mechanical fault: first fault probability 10%, second fault probability 10%, first weight 0.9, second weight 0.1. The final probability is calculated for each category: Arc fault: 60%×0.9+30%×0.1=57%; Overheating fault: 10%×0.9+60%×0.1=15%; Insulation fault: 20%×0.9+20%×0.1=20%; Mechanical fault: 10%×0.9+10%×0.1=10%. The probability of various types of faults in the first power equipment B-01 is as follows: arc fault 57%, overheating fault 15%, insulation fault 20%, and mechanical fault 10%. The transformer is most likely to have an arc fault, followed by an insulation fault. The risk of arc faults should be given special attention, and insulation hazards should be investigated simultaneously.

[0073] Finally, following the same method used to obtain the multi-type fault probabilities of the first power equipment, the multi-type fault probabilities of the remaining power equipment are obtained, resulting in the multi-type fault probabilities of each power equipment. The remaining power equipment refers to other power equipment of the same type, voltage level, and operating scenario as the first power equipment, with fault types and evaluation logic completely consistent with the first power equipment. All remaining power equipment employs the same evaluator construction-weight calculation-weighted fusion process as the first power equipment, ensuring that the fault probability evaluation standards for all equipment are unified and the results are comparable. Based on the power equipment topology, a list of remaining power equipment to be evaluated is compiled, and the unique identifier of each equipment is clearly defined. For each remaining power equipment, the above complete process is repeated: collecting the equipment's own electrical quantity parameters, non-electrical quantity parameters, and peripheral equipment response parameters; obtaining the first, second, and third fault probabilities for each fault type through the first, second, and third evaluators respectively; calculating the first and second weights for each fault type based on the first and third fault probabilities; weighted fusion of the first and second fault probabilities to obtain the multi-type fault probabilities of the equipment; summarizing the multi-type fault probabilities of the first power equipment and the remaining power equipment to form a summary table of fault probabilities for all power equipment; and unifying the output format to ensure that the fault type and corresponding probability of each equipment are clearly traceable.

[0074] For example, taking two other 10kV distribution transformers B-02 and B-03 in the same substation as the remaining power equipment, the calculation is performed using the same process, and the results are as follows: Calculation process for the remaining power equipment B-02: First fault probability: arc fault 55%, overheating fault 15%, insulation fault 25%, mechanical fault 5%; Second fault probability: arc fault 25%, overheating fault 55%, insulation fault 20%, mechanical fault 8%; Weight calculation: First weight for each fault type 0.9, second weight... 0.1; Final multi-type fault probabilities: Arc fault: 55%×0.9+25%×0.1=52%; Overheating fault: 15%×0.9+55%×0.1=19%; Insulation fault: 25%×0.9+20%×0.1=24.5%; Mechanical fault: 5%×0.9+8%×0.1=5.3%; Other power equipment B-03 multi-type fault probabilities: Arc fault 62%; Overheating fault 12.2%; Insulation fault 15.7%; Mechanical fault 11.8%. All three distribution transformers have a high risk of arc faults, with B-03 having the highest risk. Arc fault testing should be prioritized for B-03. Insulation and overheating fault risks are relatively prominent for B-02, requiring simultaneous hazard investigation.

[0075] In this embodiment of the invention, three types of evaluators—electrical quantities, non-electrical quantities, and peripheral device responses—are constructed using a dual-layer integrated architecture of input layer + output layer. Weights are dynamically calculated based on the consistent logic of fault propagation in the power system. Multi-source detection data is then weighted and fused to obtain the probability of various types of equipment faults. This can be standardized and extended to other power equipment of the same type. This not only effectively improves the accuracy and robustness of fault probability assessment and enhances the physical rationality of the assessment results, but also accurately identifies high-risk fault types for each device. This provides scientific and readily applicable data support for identifying potential faults in power equipment and for risk classification and management.

[0076] S300: Generate multiple fault probability distributions based on the multiple types of fault probabilities of each power equipment, and determine the target fault equipment and target fault type based on the distribution characteristics of the multiple fault probability distributions.

[0077] Step S300 in the method provided in this embodiment of the invention includes: Based on the multi-type fault probabilities of each power equipment, multiple fault probability distributions are generated according to multiple fault types, and each fault probability distribution corresponds to a fault type. The multiple fault probability distributions are input into a pre-trained fault location engine, which outputs the target faulty device and the target fault type.

[0078] First, based on the multi-type fault probabilities of each power device, multiple fault probability distributions are generated for each fault type, with each distribution corresponding to a specific fault type. A fault probability distribution is a dataset comprised of the fault probabilities of all power devices to be evaluated for a single fault type, reflecting the probability distribution characteristics of that fault type within the device group. Generating by fault type means using fault type as the classification dimension, rather than the device dimension. Each fault type corresponds to an independent probability distribution, facilitating horizontal comparison of the risk differences of the same fault on different devices. Based on the power device topology, a list of all power devices to be evaluated and their corresponding multi-type fault probability data are compiled; the fault type classification dimension is determined; for each fault type, the fault probability of all power devices for that type is extracted and aggregated to form a fault probability distribution specific to that fault type; each distribution is then processed, and the unique identifier of the corresponding device is labeled to ensure a one-to-one correspondence between the distribution and the device.

[0079] For example, based on the multi-type fault probability data of B-01, B-02, and B-03, four fault probability distributions are generated according to fault type: Arc fault probability distribution: containing arc fault probabilities of B-01 (57%), B-02 (52%), and B-03 (62%), with a distribution dataset of [57%, 52%, 62%]; Overheat fault probability distribution: containing overheat fault probabilities of B-01 (15%), B-02 (19%), and B-03 (12.2%), with a distribution dataset of [57%, 52%, 62%]; Data set is [15%, 19%, 12.2%]; Insulation fault probability distribution: Insulation fault probabilities including B-01 (20%), B-02 (24.5%), B-03 (15.7%), the distribution data set is [20%, 24.5%, 15.7%]; Mechanical fault probability distribution: Mechanical fault probabilities including B-01 (10%), B-02 (5.3%), B-03 (11.8%), the distribution data set is [10%, 5.3%, 11.8%].

[0080] Secondly, the multiple fault probability distributions are input into the pre-trained fault location engine, which outputs the target faulty device and the target fault type.

[0081] The training methods for the fault location engine include: Obtain the topology historical monitoring records of the power equipment topology, and construct a sample probability distribution set based on the topology historical monitoring records. The sample probability distribution set includes multiple sets of sample probability distributions, and each set of sample probability distributions includes multiple sample fault probability distributions. Based on the historical monitoring records of the topology, fault labels are made on the probability distribution of each group of samples to obtain a sample fault label set; The fault localization engine is trained and generated using the sample probability distribution set as input features and the sample fault label set as supervision labels.

[0082] First, the historical monitoring records of the power equipment topology are obtained. Based on these records, a sample probability distribution set is constructed. This set includes multiple sample probability distributions, each containing multiple sample fault probability distributions. The historical monitoring records are complete monitoring data covering the power equipment topology throughout its historical operation. Each record corresponds to one historical monitoring event and includes historical electrical quantity parameters, historical non-electrical quantity parameters, historical peripheral equipment response parameters, and the actual status identifier corresponding to that monitoring event for all devices within the topology. The sample probability distribution set is a training dataset composed of multiple sample probability distributions. Each sample probability distribution corresponds to one historical monitoring record and contains multiple sample fault probability distributions categorized by fault type. Collect all historical monitoring records of the target power equipment topology, covering both fault and normal samples to ensure sample diversity. For each historical monitoring record, calculate the historical multi-type fault probabilities of each power device within the topology according to the aforementioned technical solutions S100-S200. Classify and integrate the historical multi-type fault probabilities according to the fault type dimension, that is, for each fault type, extract the historical fault probability of that type of all devices within the topology to form a sample fault probability distribution. Integrate all sample fault probability distributions corresponding to a historical monitoring record into a set of sample probability distributions. After summing up the multiple sets of sample probability distributions corresponding to multiple records, a sample probability distribution set is obtained.

[0083] For example, taking a power equipment topology containing three distribution transformers B-01, B-02, and B-03 as an example: 1000 historical monitoring records of this topology are collected, including 800 normal samples and 200 fault samples, covering arcing, overheating, insulation, and mechanical faults of B-01 / B-02 / B-03; for a certain fault monitoring record, the probabilities of multiple fault types are calculated as follows: B-01 [50%, 12%, 18%, 8%], B-02 [48%, 15%, 20%, 7%], and B-03 [65%, 10%]. [12%, 9%]; Classified by fault type, four sample fault probability distributions are generated: arc fault sample probability distribution: [50%, 48%, 65%], overheating fault sample probability distribution: [12%, 15%, 10%], insulation fault sample probability distribution: [18%, 20%, 12%], and mechanical fault sample probability distribution: [8%, 7%, 9%]. The above four sample fault probability distributions are integrated into a set of sample probability distributions. 1000 historical records correspond to 1000 sets of sample probability distributions, which together constitute the sample probability distribution set.

[0084] Secondly, based on the historical monitoring records of the topology, fault labeling is performed on the probability distributions of each group of samples to obtain a sample fault label set. Fault labeling refers to adding corresponding supervision labels to each group of sample probability distributions in the sample probability distribution set by comparing them with the real state identifiers in the historical monitoring records. The labels must be consistent with the real state in the historical monitoring records. The sample fault label set is a dataset composed of supervision labels corresponding to all sample probability distributions in the sample probability distribution set, and it corresponds one-to-one with the sample probability distribution set. It serves as the supervision basis for training the fault localization engine. The real state identifiers corresponding to each record in the historical monitoring records of the topology are extracted to form an original label list; each group of sample probability distributions in the sample probability distribution set is traversed, and the real state identifiers in the original label list are matched according to their corresponding historical monitoring records; supervision labels are added to each group of sample probability distributions: if the corresponding historical record is a fault state, the label is faulty device + fault type; if the corresponding historical record is a normal state, the label is normal state; all labeled sample probability distributions are summarized to generate the sample fault label set.

[0085] For example, extract the real status identifiers of 1000 historical monitoring records, such as the first record being labeled B-03 arc fault and the second record being labeled normal status; add the label B-03 arc fault to the sample probability distribution corresponding to the first record; add the label normal status to the sample probability distribution corresponding to the second record; complete the label matching of 1000 sets of sample probability distributions in this way, and finally form a sample fault label set containing fault labels and normal labels.

[0086] Further, the fault location engine is trained using the sample probability distribution set as input features and the sample fault label set as supervision labels. Specifically, the input features are feature vectors formed by concatenating the fault probabilities of each device according to the fault type within each sample probability distribution set. Multiple sample fault probability distributions contained in each sample probability distribution set together constitute the input dimension for engine training. Supervision labels include two types of identifiers: faulty device + fault type or normal state, used to guide the fault location engine in learning the correlation between sample probability distribution features and the actual state. The training objective of the fault location engine is to enable the engine to learn the baseline features of the probability distribution under normal operating conditions, as well as the abnormal probability distribution patterns of different fault types occurring on different devices, achieving the core functions of identifying fault state, locating faulty device, and determining fault type, while simultaneously distinguishing normal states to avoid false alarms. The sample probability distribution set and the corresponding sample fault label set are divided into training subset and validation subset according to a preset ratio. A suitable model architecture, such as a deep learning model or a gradient boosting tree model, is selected as the base model of the fault localization engine. The sample probability distribution set of the training subset is used as the input feature, and the sample fault label set of the training subset is used as the supervision label. The model is then trained by inputting these features into the base model. The validation subset is used to verify the model's performance during training, and the model's fault localization accuracy, normal state recognition rate, false alarm rate, and other indicators are evaluated. The model parameters, such as the learning rate and the number of iterations, are adjusted according to the validation results. The model is iterated repeatedly until the model indicators meet the preset requirements. After the model converges, the final fault localization engine is obtained.

[0087] For example, 1000 sets of sample probability distributions and corresponding labels are divided into 800 training sets and 200 validation sets in an 8:2 ratio. A deep neural network (DNN) is selected as the basic model architecture. The model input is a feature vector composed of the probability distribution of each sample according to the fault type, such as the concatenated vector of the probability distributions of arc, overheating, insulation, and mechanical faults of three transformers. The model output is the predicted label corresponding to the input sample, specifically the classification result of faulty equipment + fault type or normal state. The input training set data is used for iterative training, and the model gradually learns the rules through learning: under normal state, each set of equipment... The probability distribution of various faults in the equipment is relatively uniform and the values ​​are low. When an arc fault occurs in B-03, its arc fault probability is significantly higher than that of other equipment in the topology, forming an obvious abnormal peak. Validated with a validation set, the initial model had a low accuracy rate in identifying fault samples with minor anomalies. After adjusting the model learning rate, the fault location accuracy increased to 98%, the false alarm rate dropped to below 1%, and the normal state recognition rate met the standard. Moreover, the fluctuation range of these three core indicators was ≤0.5% in multiple consecutive iterations, reaching the preset convergence standard. The model converged, and a fault location engine that can accurately identify faulty equipment and types was generated.

[0088] Finally, the multiple fault probability distributions are input into the pre-trained fault location engine, which outputs the target faulty device and the target fault type. The target faulty device refers to a specific power device with a high fault risk, whose abnormal fault probability is not due to detection error and has clear physical meaning. The target fault type refers to the high-risk fault category corresponding to the target faulty device, i.e., the fault type whose fault probability far exceeds the group average and highly matches the abnormal distribution pattern learned by the engine. Data preprocessing: The generated multiple fault probability distributions are standardized to be consistent with the input format during model training. The fault probabilities of each device are concatenated into a one-dimensional feature vector according to the fault type, and the unique identifier of the device corresponding to each probability value in the vector is labeled. Model input: The preprocessed feature vector is input into the pre-trained fault location engine, ensuring that the input dimension and data format are consistent with those during training to avoid inference errors caused by format mismatch. Result output: The engine outputs the inference result, clearly labeling the unique identifier of the target faulty device and the corresponding target fault type. If all device probability distributions conform to the normal baseline, the output indicates no target faulty device and all devices are operating normally.

[0089] For example, the fault probability distribution is standardized, and feature vectors are concatenated in the order of arc fault - overheating fault - insulation fault - mechanical fault, and the equipment is labeled. The final input vector is [57%, 52%, 62%, 15%, 19%, 12.2%, 20%, 24.5%, 15.7%, 10%, 5.3%, 11.8%]; the fault location engine outputs: target fault equipment: distribution transformer B-03; target fault type: arc fault.

[0090] In this embodiment of the invention, a fault probability distribution covering all topology power equipment is generated according to the fault type dimension. This distribution is then input into a fault location engine pre-trained based on historical topology monitoring data. By utilizing the baseline features of the normal probability distribution and the abnormal fault distribution patterns learned by the engine, abnormal equipment whose probability deviates significantly from the group baseline can be accurately identified. This enables rapid location of the target faulty equipment and its corresponding fault type, while effectively distinguishing normal operating states and reducing false alarm rates. This process can be batch adapted to the fault diagnosis needs of topology power equipment, improving the accuracy and efficiency of power equipment fault investigation, and providing scientific and reliable decision support for equipment risk classification management and preventive maintenance.

[0091] Through the specific implementation methods described above, the embodiments of the present invention achieve the following technical effects: This invention provides a multimodal edge fusion-based power fault diagnosis method and system. It constructs three types of evaluators—electrical quantities, non-electrical quantities, and peripheral device responses—with an input layer + output layer integrated architecture. Weights are dynamically calculated based on the consistency law of power system fault propagation, and multi-source detection data is weighted and fused to obtain the probability of multiple fault types for a single device. This can be standardized and extended to all power devices within the topology. Furthermore, a fault location engine is trained using historical topology monitoring records to learn the baseline of normal probability distribution and abnormal fault patterns, achieving intelligent analysis and precise location of fault probability distribution across the entire topology. The overall technical solution forms a complete closed loop of multi-source data fusion evaluation—dynamic weight adjustment—topology-level fault location. This not only improves the accuracy and robustness of fault probability assessment and effectively reduces the false alarm rate, but also accurately identifies high-risk fault devices and their corresponding types. It provides a scientific and feasible decision-making basis for risk classification management and preventive maintenance of power equipment, improving the intelligence and refinement of power system operation and maintenance.

[0092] Example 2, as Figure 2 As shown, the present invention provides a power fault diagnosis system based on multimodal edge fusion, the system comprising: The device topology construction module 11 is used to establish a power device topology within a preset area, wherein the power device topology includes multiple power devices; The fault probability calculation module 12 is used to collect electrical quantity parameters, non-electrical quantity parameters and peripheral equipment response parameters of each of the power equipment, and calculate the multi-type fault probability of each of the power equipment. The fault location module 13 is used to generate multiple fault probability distributions based on the multiple types of fault probabilities of each of the power equipment, and to determine the target fault equipment and the target fault type based on the distribution characteristics of the multiple fault probability distributions.

[0093] In one embodiment, the fault probability calculation module 12 is further configured to: From the plurality of power devices, a first power device is determined, and the first electrical quantity parameter, the first non-electrical quantity parameter, and the first peripheral device response parameter of the first power device are extracted; The fault probability assessment engine bound to the first power equipment is invoked. The fault probability assessment engine includes a first evaluator, a second evaluator, and a third evaluator. Each evaluator includes assessment units for multiple fault types. Based on the first evaluator, the second evaluator and the third evaluator respectively, the first electrical quantity parameter, the first non-electrical quantity parameter and the first peripheral equipment response parameter are processed to obtain the multi-type fault probability of the first power equipment; The multi-type fault probabilities of the remaining power equipment are obtained by obtaining the multi-type fault probabilities of the first power equipment in the same way as obtaining the multi-type fault probabilities of the first power equipment.

[0094] Specifically, based on the processing of the first electrical quantity parameter, the first non-electrical quantity parameter, and the first peripheral equipment response parameter by the first evaluator, the second evaluator, and the third evaluator, respectively, the multi-type fault probabilities of the first power equipment are obtained, including: Based on the first electrical quantity parameter, the first fault probability of each fault type is obtained by each evaluation unit of the first evaluator; Based on the first non-electrical quantity parameter, the second fault probability of each fault type is obtained by each evaluation unit of the second evaluator; Based on the response parameters of the first peripheral device, the third fault probability of each fault type is obtained by each evaluation unit of the third evaluator. Based on the first and third failure probabilities of each failure type, the first and second weights of each failure type are obtained. By weighting the first and second fault probabilities of each fault type using the first and second weights, the multi-type fault probabilities of the first power equipment are obtained.

[0095] Specifically, based on the first electrical quantity parameter, the first fault probability of each fault type is obtained by each evaluation unit of the first evaluator, including: Input the first electrical quantity parameter into the first evaluator; The first evaluator simultaneously distributes the first electrical quantity parameter to multiple internal evaluation units. Each evaluation unit processes the first electrical quantity parameter according to multiple internal evaluation sub-units to obtain multiple evaluation result sets. Based on the multiple evaluation result sets, the first failure probability of each failure type is obtained.

[0096] Among them, based on the first fault probability and the third fault probability of each fault type, the first weight and the second weight of each fault type are obtained, including: Determine the first fault type from among multiple fault types; Calculate the matching degree between the first fault probability and the third fault probability corresponding to the first fault type; The matching degree is used as the first weight corresponding to the first fault type, and the second weight corresponding to the first fault type is determined based on the first weight. In accordance with the method of obtaining the first weight and the second weight of the first fault type, the first weight and the second weight corresponding to the other fault types are obtained respectively.

[0097] The construction steps of the fault probability assessment engine bound to the first power equipment include: Obtain the historical monitoring records of the first power equipment. The historical monitoring records include multiple sets of historical monitoring data. Each set of historical monitoring data includes historical electrical quantity parameters, historical non-electrical quantity parameters, historical peripheral equipment response parameters, and corresponding fault identification parameters. Based on multiple sets of historical monitoring data, a sample electrical quantity parameter set, a sample non-electrical quantity parameter set, and a sample peripheral equipment response parameter set are constructed. Based on the fault identification parameters and the multiple fault types, multiple sample fault label sets are constructed, and the multiple fault types correspond one-to-one with the multiple sample fault label sets; Based on the sample electrical quantity parameter set and the multiple sample fault label sets, a first evaluator is constructed; Based on the sample non-electrical quantity parameter set and the multiple sample fault label sets, a second evaluator is constructed; Based on the sample peripheral device response parameter set and the multiple sample fault label sets, a third evaluator is constructed; The first evaluator, the second evaluator, and the third evaluator are integrated to obtain the fault probability evaluation engine for the first power equipment.

[0098] The first evaluator is constructed based on the sample electrical quantity parameter set and the multiple sample fault label sets, including: A first fault type is determined from the plurality of fault types, and a corresponding first sample fault label set is determined from the plurality of sample fault label sets; Construct a multi-first evaluation sub-unit architecture; Based on the sample electrical quantity parameter set and the first sample fault label set, the architecture of the plurality of first evaluation sub-units is trained until convergence, thereby obtaining a plurality of first evaluation sub-units; The plurality of first evaluation subunits are integrated to obtain a first evaluation unit; Following the method of obtaining the first evaluation unit for the first fault type, the evaluation units for the remaining fault types are trained to obtain multiple evaluation units; The multiple evaluation units are integrated to obtain the first evaluator.

[0099] In one embodiment, the fault location module 13 is further configured to: Based on the multi-type fault probabilities of each power equipment, multiple fault probability distributions are generated according to multiple fault types, and each fault probability distribution corresponds to a fault type. The multiple fault probability distributions are input into a pre-trained fault location engine, which outputs the target faulty device and the target fault type.

[0100] The training methods for the fault location engine include: Obtain the topology historical monitoring records of the power equipment topology, and construct a sample probability distribution set based on the topology historical monitoring records. The sample probability distribution set includes multiple sets of sample probability distributions, and each set of sample probability distributions includes multiple sample fault probability distributions. Based on the historical monitoring records of the topology, fault labels are made on the probability distribution of each group of samples to obtain a sample fault label set; The fault localization engine is trained and generated using the sample probability distribution set as input features and the sample fault label set as supervision labels.

[0101] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0102] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0103] This specification and accompanying drawings are merely illustrative examples of the invention and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of the invention. Clearly, those skilled in the art can make various alterations and modifications to the invention without departing from its scope. Therefore, if such modifications and modifications fall within the scope of the invention and its equivalents, the invention is intended to include these modifications and modifications.

Claims

1. A power fault diagnosis method based on multimodal edge fusion, characterized in that, include: Establish a power equipment topology within a preset area, wherein the power equipment topology includes multiple power devices; The electrical quantity parameters, non-electrical quantity parameters, and peripheral equipment response parameters of each power equipment are collected respectively, and the multi-type fault probabilities of each power equipment are calculated. Based on the multiple types of fault probabilities of each power device, multiple fault probability distributions are generated, and the target fault device and target fault type are determined based on the distribution characteristics of the multiple fault probability distributions.

2. The method according to claim 1, characterized in that, Electrical quantity parameters, non-electrical quantity parameters, and peripheral equipment response parameters of each power device are collected respectively, and the multi-type fault probabilities of each power device are calculated, including: From the plurality of power devices, a first power device is determined, and the first electrical quantity parameter, the first non-electrical quantity parameter, and the first peripheral device response parameter of the first power device are extracted; The fault probability assessment engine bound to the first power equipment is invoked. The fault probability assessment engine includes a first evaluator, a second evaluator, and a third evaluator. Each evaluator includes assessment units for multiple fault types. Based on the first evaluator, the second evaluator and the third evaluator respectively, the first electrical quantity parameter, the first non-electrical quantity parameter and the first peripheral equipment response parameter are processed to obtain the multi-type fault probability of the first power equipment; The multi-type fault probabilities of the remaining power equipment are obtained by obtaining the multi-type fault probabilities of the first power equipment in the same way as obtaining the multi-type fault probabilities of the first power equipment.

3. The method according to claim 2, characterized in that, Based on the processing of the first electrical quantity parameter, the first non-electrical quantity parameter, and the first peripheral equipment response parameter by the first evaluator, the second evaluator, and the third evaluator respectively, the multi-type fault probabilities of the first power equipment are obtained, including: Based on the first electrical quantity parameter, the first fault probability of each fault type is obtained by each evaluation unit of the first evaluator; Based on the first non-electrical quantity parameter, the second fault probability of each fault type is obtained by each evaluation unit of the second evaluator; Based on the response parameters of the first peripheral device, the third fault probability of each fault type is obtained by each evaluation unit of the third evaluator. Based on the first and third failure probabilities of each failure type, the first and second weights of each failure type are obtained. By weighting the first and second fault probabilities of each fault type using the first and second weights, the multi-type fault probabilities of the first power equipment are obtained.

4. The method according to claim 2, characterized in that, The construction steps of the fault probability assessment engine bound to the first power device include: Obtain the historical monitoring records of the first power equipment. The historical monitoring records include multiple sets of historical monitoring data. Each set of historical monitoring data includes historical electrical quantity parameters, historical non-electrical quantity parameters, historical peripheral equipment response parameters, and corresponding fault identification parameters. Based on multiple sets of historical monitoring data, a sample electrical quantity parameter set, a sample non-electrical quantity parameter set, and a sample peripheral equipment response parameter set are constructed. Based on the fault identification parameters and the multiple fault types, multiple sample fault label sets are constructed, and the multiple fault types correspond one-to-one with the multiple sample fault label sets; Based on the sample electrical quantity parameter set and the multiple sample fault label sets, a first evaluator is constructed; Based on the sample non-electrical quantity parameter set and the multiple sample fault label sets, a second evaluator is constructed; Based on the sample peripheral device response parameter set and the multiple sample fault label sets, a third evaluator is constructed; The first evaluator, the second evaluator, and the third evaluator are integrated to obtain the fault probability evaluation engine for the first power equipment.

5. The method according to claim 4, characterized in that, Based on the sample electrical quantity parameter set and the multiple sample fault label sets, a first evaluator is constructed, including: A first fault type is determined from the plurality of fault types, and a corresponding first sample fault label set is determined from the plurality of sample fault label sets; Construct a multi-first evaluation sub-unit architecture; Based on the sample electrical quantity parameter set and the first sample fault label set, the architecture of the plurality of first evaluation sub-units is trained until convergence, thereby obtaining a plurality of first evaluation sub-units; The plurality of first evaluation subunits are integrated to obtain a first evaluation unit; Following the method of obtaining the first evaluation unit for the first fault type, the evaluation units for the remaining fault types are trained to obtain multiple evaluation units; The multiple evaluation units are integrated to obtain the first evaluator.

6. The method according to claim 5, characterized in that, Based on the first electrical quantity parameter, the first fault probability of each fault type is obtained by each evaluation unit of the first evaluator, including: Input the first electrical quantity parameter into the first evaluator; The first evaluator simultaneously distributes the first electrical quantity parameter to multiple internal evaluation units. Each evaluation unit processes the first electrical quantity parameter according to multiple internal evaluation sub-units to obtain multiple evaluation result sets. Based on the multiple evaluation result sets, the first failure probability of each failure type is obtained.

7. The method according to claim 3, characterized in that, Based on the first and third failure probabilities of each failure type, a first weight and a second weight for each failure type are obtained, including: Determine the first fault type from among multiple fault types; Calculate the matching degree between the first fault probability and the third fault probability corresponding to the first fault type; The matching degree is used as the first weight corresponding to the first fault type, and the second weight corresponding to the first fault type is determined based on the first weight. In accordance with the method of obtaining the first weight and the second weight of the first fault type, the first weight and the second weight corresponding to the other fault types are obtained respectively.

8. The method according to claim 1, characterized in that, Based on the multi-type fault probabilities of each of the aforementioned power equipment, multiple fault probability distributions are generated, and the target fault equipment and target fault type are determined based on the distribution characteristics of the multiple fault probability distributions, including: Based on the multi-type fault probabilities of each power equipment, multiple fault probability distributions are generated according to multiple fault types, and each fault probability distribution corresponds to a fault type. The multiple fault probability distributions are input into a pre-trained fault location engine, which outputs the target faulty device and the target fault type.

9. The method according to claim 8, characterized in that, The training methods for the fault location engine include: Obtain the topology historical monitoring records of the power equipment topology, and construct a sample probability distribution set based on the topology historical monitoring records. The sample probability distribution set includes multiple sets of sample probability distributions, and each set of sample probability distributions includes multiple sample fault probability distributions. Based on the historical monitoring records of the topology, fault labels are made on the probability distribution of each group of samples to obtain a sample fault label set; The fault localization engine is trained and generated using the sample probability distribution set as input features and the sample fault label set as supervision labels.

10. A power fault diagnosis system based on multimodal edge fusion, characterized in that, The system is used to implement the power fault diagnosis method of multimodal edge fusion according to any one of claims 1-9, the system comprising: The device topology construction module is used to establish the power device topology within a preset area, wherein the power device topology includes multiple power devices; The fault probability calculation module is used to collect electrical quantity parameters, non-electrical quantity parameters and peripheral equipment response parameters of each of the power equipment, and calculate the multi-type fault probability of each of the power equipment. The fault location module is used to generate multiple fault probability distributions based on the multiple types of fault probabilities of each of the power equipment, and to determine the target fault equipment and the target fault type based on the distribution characteristics of the multiple fault probability distributions.

Images