Electric power product potential safety hazard troubleshooting intelligent management system based on industrial internet

Abstract

The invention relates to the technical field of industrial internet of things, in particular to an intelligent management system for troubleshooting potential safety hazards of electric power products based on industrial internet, which comprises a fault parameter acquisition module, a regular expression generation module, a potential safety hazard data analysis module, a knowledge graph construction module, a knowledge graph reasoning module and a management database, according to the system, historical fault reports are associated, regular expressions corresponding to hidden danger data are generated, then entities are extracted to form an entity relation pair list, a graph network is constructed according to the entity relation pair list, fault propagation paths are obtained through fault tracing, a knowledge graph is constructed, then rule matching reasoning is conducted on the three-dimensional roll iron core, and the three-dimensional roll iron core is obtained. And the high-frequency fault type of the three-dimensional roll iron core is output, so that a comprehensive fault knowledge system can be formed, the most likely fault type of the three-dimensional roll iron core can be quickly positioned, the operation reliability of equipment is improved, and the maintenance cost and the downtime when the fault occurs are reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial Internet of Things, and specifically, to an intelligent management system for troubleshooting safety hazards of power products based on the industrial Internet. Background Art

[0002] With the rapid development of the global economy and the increasing dependence of society on electricity, the safe and stable operation of the power system, as the "lifeline" of modern society, is of vital importance. Power products, as a key component of the power system, cover a wide range from power generation equipment, transmission lines to distribution equipment and various types of power terminal equipment. The reliable operation of these power products is directly related to the efficiency of the entire power system and the normal production and life of users.

[0003] Against the backdrop of the current booming power industry, power infrastructure construction continues to advance, the number and types of power products continue to increase, and the operating environment becomes increasingly complex. On the one hand, a large number of new power products are put into use. While new technologies and processes improve product performance, they also bring new types of safety hazards and challenges. Although the existing power product safety hazard investigation technology and management system have ensured the safe operation of the power system to a certain extent, there are still many shortcomings.

[0004] For example, the existing Chinese patent with application number 202010343739.8 discloses a method and system for managing user equipment safety hazards. The solution promptly follows up on the safety hazards of user equipment by generating user equipment safety hazard management information based on the single-line diagram and the user equipment safety hazard related information imported into the single-line diagram, and replaces the old mode of manual tracking by generating user equipment safety management information. It can avoid the occurrence of power safety accidents caused by the omission of user equipment safety hazards due to negligence of management personnel, and can improve the efficiency and management effect of user equipment safety hazard management.

[0005] However, this solution has the following shortcomings: it only generates management information based on the single-line diagram and imported information related to user equipment safety hazards. The mining of hazard data is relatively shallow, and historical data is not fully utilized for in-depth correlation analysis. When faced with complex power safety hazard scenarios, it is difficult to extract key features from past similar hazard cases to assist in the judgment and handling of current hazards, and it is easy to miss potential hazard factors.

[0006] For example, the existing Chinese patent with application number 202210534241.9 discloses a method for generating safety hazard information codes for violation portraits and safety portraits of power members. The scheme calculates the violation portraits according to the number of violation incidents and violation operation attributes of all power members in the team through the collection terminal, and generates a fused safety hazard information code corresponding to the team according to the number of safety incidents and safety hazard attributes of all power members in the team. The server determines the number of training nodes in the initial training path and the training duration of each training node according to the fused safety hazard information code to generate an actual training path, divides the training nodes in the actual training path, and obtains violation training sub-paths and safety training sub-paths, so that the training process is targeted.

[0007] However, this scheme has the following shortcomings: the scheme simply divides the training nodes in the actual training path according to the training duration of each training node, which is too simple and mechanical, and does not consider the relevance and logic of the training content, making the sub-path content confusing. In the case of power product safety hazard inspection, if the inspection path lacks logic, for example, in the inspection process, the upstream propagation node is not determined according to the fault node to trace the fault source, but the inspection is carried out in disorder, which will make the inspection work chaotic and disorderly. Summary of the invention

[0008] In order to overcome the shortcomings in the background technology, the embodiments of the present invention provide an intelligent management system for checking safety hazards of power products based on the industrial Internet, which can effectively solve the problems involved in the above-mentioned background technology.

[0009] The purpose of the present invention can be achieved through the following technical solutions: The present invention provides an intelligent management system for detecting safety hazards of power products based on the industrial Internet, including: a fault parameter acquisition module, used to filter hidden danger data from the operating parameters of the three-dimensional wound iron core, and obtain first-category hidden danger data and second-category hidden danger data in turn.

[0010] The regular expression generation module is used to associate the first type of hidden danger data and the second type of hidden danger data with the historical fault report, and generate regular expressions corresponding to the first type of hidden danger data and the second type of hidden danger data.

[0011] The hidden danger data analysis module is used to extract entities in the first category hidden danger data and the second category hidden danger data according to regular expressions, form a preliminary entity relationship pair list, determine the dependency relationship in the preliminary entity relationship pair list, and update it into the entity relationship pair list.

[0012] The knowledge graph construction module is used to construct a graph network corresponding to the entity relationship based on the entity relationship list, trace the fault source based on the graph network, obtain the propagation path of each fault, and construct a knowledge graph.

[0013] The knowledge graph reasoning module is used to perform rule matching reasoning on the three-dimensional wound iron core through the knowledge graph and output the high-frequency fault types of the three-dimensional wound iron core.

[0014] Compared with the prior art, the present invention has the following beneficial effects: 1. The present invention associates the first-category hidden danger data and the second-category hidden danger data obtained from the operating parameters of the three-dimensional wound core with the historical fault reports to generate corresponding regular expressions, and then extracts entities to form an entity relationship pair list. The characteristics of previous faults can be accurately associated with the current hidden danger data. The generation of regular expressions helps to accurately extract useful entity information from complex data, and the formation of an entity relationship pair list helps to sort out the logical relationship between data.

[0015] 2. The present invention constructs a graph network based on the list of entity relationships, obtains the propagation paths of each fault through fault tracing, constructs a knowledge graph, and then performs rule matching reasoning on the three-dimensional wound iron core, and outputs the high-frequency fault types of the three-dimensional wound iron core. The construction of the knowledge graph can integrate various fault knowledge to form a comprehensive fault knowledge system, which is helpful to quickly obtain relevant information from the knowledge graph for analysis when facing new fault situations. The rule matching reasoning output high-frequency fault type can quickly locate the most likely fault type of the three-dimensional wound iron core, which is convenient for taking targeted maintenance measures in advance, improving the operating reliability of the equipment, and reducing maintenance costs and downtime when faults occur. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for describing the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0017] Figure 1 This is a system module connection diagram of the present invention.

[0018] Figure 2 for Figure 1 Flowchart of the regular expression generation module.

[0019] Figure 3 for Figure 2 Flow chart of step S3 in FIG. DETAILED DESCRIPTION

[0020] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0021] See also Figure 1 As shown, the intelligent management system for checking safety hazards of power products based on the industrial Internet includes a fault parameter acquisition module, a regular expression generation module, a hidden danger data analysis module, a knowledge graph construction module, a knowledge graph reasoning module, and a management database.

[0022] The management database is connected to the regular expression generation module, the hidden danger data analysis module, the knowledge graph construction module, and the knowledge graph reasoning module; the knowledge graph reasoning module is connected to the hidden danger data analysis module and the knowledge graph construction module; the regular expression generation module is connected to the fault parameter acquisition module, the hidden danger data analysis module, and the knowledge graph construction module.

[0023] The fault parameter acquisition module is used to screen the hidden danger data from the operating parameters of the three-dimensional wound core, and sequentially obtain the first type of hidden danger data and the second type of hidden danger data.

[0024] The operating parameters of the three-dimensional wound core include current, voltage and power.

[0025] The implementation method of the fault parameter acquisition module also includes: setting upper and lower limit thresholds for the operating parameters of the three-dimensional wound iron core. When the operating parameters of the three-dimensional wound iron core are not within the set upper and lower limit thresholds, a fault alarm is triggered, and the operating parameters of the three-dimensional wound iron core at this time are classified and recorded; different types of hidden danger data can reflect abnormal conditions in different aspects of the equipment, providing a targeted data basis for in-depth exploration of the root cause of the fault.

[0026] If the monitored equipment operating parameters are less than the set lower limit threshold of the equipment operating parameters, the equipment operating data at this time will be recorded as the first type of hidden danger data; if the monitored equipment operating parameters are greater than the set upper limit threshold of the equipment operating parameters, the equipment operating data at this time will be recorded as the second type of hidden danger data; different categories of hidden danger data may correspond to different failure modes or causes of failure, which helps to focus on the key of the problem and improve the efficiency of troubleshooting.

[0027] It should be noted that the lower limit threshold of the equipment operating parameters is set according to the distribution of the data. If the data presents a normal distribution, it can be set according to 3 In principle, the mean minus a certain multiple of the standard deviation is used as a reference for the lower threshold. For example, if the mean of the temperature data of a device is 50°C and the standard deviation is 5°C, the lower threshold is obtained by subtracting 1.5 times the standard deviation from the mean. In this case, the lower threshold is set to 50-1.5×5=42.5°C. The upper threshold is determined by adding 1.5 times the standard deviation to the mean. The upper threshold is 50+1.5×5=57.5°C.

[0028] 3 The specific content of the principle is: if the data obeys the normal distribution, that is, ,in is the mean, is the standard deviation. According to the properties of normal distribution, about 68.27% of the data will fall within Within the range, about 95.45% of the data will fall within Within the range, about 99.73% of the data will fall within Within the range.

[0029] The regular expression generation module is used to associate the first type of hidden danger data and the second type of hidden danger data with the historical fault report, and generate regular expressions corresponding to the first type of hidden danger data and the second type of hidden danger data.

[0030] See also Figure 2 As shown, the specific operation method of the regular expression generation module is: S1. Taking the historical fault report of the three-dimensional wound iron core as an example, establish the abnormal association text between the abnormal parameters under each fault state and the first type of hidden danger data and the second type of hidden danger data; it can be clearly seen which abnormal parameters and hidden danger data are related under different fault states, thereby providing a reference for preventing similar faults.

[0031] S2. Create a conversion table and extract the sample features corresponding to each example based on the mapping relationship of the abnormal associated text; extracting sample features can simplify data representation and extract key features from a large number of examples, which helps to focus on important information in the data and remove unnecessary interference information.

[0032] The example features are abstract expressions that can represent the key information of the original examples after conversion.

[0033] S3. Count the occurrence probability of each example feature, filter out each target example feature according to the set probability threshold, obtain the longest common subsequence in each target example feature and mark the variable slot, extract its corresponding substring, generate each coarse-grained regular expression from each substring, combine them into sub-regular expressions, and finally generate a regular expression; the acquisition of the longest common subsequence and the generation of regular expressions help to describe the data pattern more accurately. Regular expressions can match and extract data that conforms to a specific pattern. When analyzing the fault data of the three-dimensional wound core, the data that conforms to the target pattern can be quickly located, providing a powerful tool for fault diagnosis.

[0034] It should be noted that the specific method for filtering out each target example feature according to the set probability threshold is: setting the probability threshold, and filtering out the original examples corresponding to the example features whose probability of appearing in the cumulative probability distribution diagram is lower than the set threshold.

[0035] It should be noted that if the probability distribution of the occurrence of example features in the data is relatively uniform, without obvious high and low differences, a moderate probability threshold of 0.5 can be selected to retain more features while removing some obviously less relevant features; if the data distribution is extremely uneven, with a few high-probability features and a large number of low-probability features, a relatively low threshold that can distinguish the main features can be set based on the probability of the high-probability features according to the specific situation. For example, if the probability of the high-probability feature is above 0.8, the threshold can be set to 0.3 to screen out relatively important features.

[0036] The longest common subsequence is the longest character sequence that exists in multiple target example features, and the variable slot is a reserved position with a variable and replaceable part in the text of each target example feature, which can be filled with different values ​​or contents.

[0037] See also Figure 3 As shown, the step S3 includes: S31. Counting the occurrence probability of each example feature in all examples, sorting the example features in order of occurrence probability from high to low, accumulating them in sequence to obtain a cumulative probability distribution diagram, and screening the example features in the cumulative probability distribution diagram to obtain each target example feature; it helps to determine a reasonable screening standard, screen out more representative and important target example features, reduce data redundancy, and focus on the most critical data part.

[0038] S32. Obtain the longest common subsequence from each target example feature, mark the parts of the longest common subsequence that will change in different target example features as variable slots, and then extract specific substrings corresponding to the positions of these variable slots from each target example feature; marking the variable slots can identify the parts that change in different target example features, and extracting the substrings corresponding to the variable slots provides specific elements for the subsequent construction of regular expressions, which can accurately describe the characteristics of different parts of the data.

[0039] S33. For each substring set composed of specific substrings, generate each coarse-grained regular expression for each substring in each substring set, combine the coarse-grained regular expressions corresponding to each variable slot according to the structure of the longest common subsequence to form a sub-regular expression of each target example feature, and combine the sub-regular expressions of each target example feature to generate regular expressions corresponding to the first type of hidden danger data and the second type of hidden danger data; regular expressions can accurately identify, screen and analyze data related to the two types of hidden danger data in a large amount of data, thereby improving the degree of automation and accuracy of processing and analysis of data related to three-dimensional coil core faults.

[0040] It should be noted that, in a specific embodiment, there are the following fault reports, please refer to Table 1 for details, and Table 1 lists representative fault reports.

[0041] Table 1. Data from some of the collected fault reports

[0042]

[0043] Five sample texts were obtained, and the correspondence between abnormal parameters (current, voltage) and fault phenomena (core overheating, partial discharge, insulation damage) was established, and the hidden danger data categories (first category, second category) were marked.

[0044] Create a conversion table, see Table 2 for details.

[0045] Table 2. Concrete character-abstract character conversion table

[0046]

[0047] The example features are converted in this way, and the converted example features are: Example 1: "When I is A1 and V is B1, the three-dimensional wound core exhibits F1 phenomenon, which belongs to H1 hidden danger data."

[0048] Example 2: "When I is A2 and V is B1, the three-dimensional wound core exhibits F2 phenomenon, which belongs to H2 hidden danger data."

[0049] Example 3: "When I is A3 and V is B2, the three-dimensional wound core exhibits F1 phenomenon, which belongs to H1 hidden danger data."

[0050] Example 4: "When I is A4 and V is B3, the three-dimensional wound core exhibits F3 phenomenon, which belongs to H2 hidden danger data."

[0051] Example 5: "When I is A1 and V is B1, the three-dimensional wound core exhibits F1 phenomenon, which belongs to H1 hidden danger data."

[0052] Count the number of times a character appears in each example feature. For example, "F1" appears 3 times, "F2" appears 1 time, and "F3" appears 1 time. Calculate the probability of occurrence. For example, the probability of occurrence of "F1" is 3 / 5=0.6, the probability of occurrence of "F2" is 1 / 5=0.2, and the probability of occurrence of "F3" is 1 / 5=0.2.

[0053] If the probability threshold is set to 0.25, the example features "F2" and "F3" that are less than the set probability threshold are screened out, and the remaining "F1" related example features are recorded as target example features: "When I is A1 and V is B1, the three-dimensional wound iron core exhibits F1 phenomenon, which belongs to H1 hidden danger data"; "When I is A3 and V is B2, the three-dimensional wound iron core exhibits F1 phenomenon, which belongs to H1 hidden danger data"; "When I is A1 and V is B1, the three-dimensional wound iron core exhibits F1 phenomenon, which belongs to H1 hidden danger data".

[0054] The longest common subsequence is "the three-dimensional wound core exhibits F1 phenomenon when I is and V is, which belongs to H1 hidden danger data", where the "" part is the variable slot, namely the current value (the value after I) and the voltage value (the value after V).

[0055] Extract the specific substrings corresponding to the variable slot positions: such as A1, A3, A1 (current values) and B1, B2, B1 (voltage values). For the substring set of current values ​​{A1, A3, A1}, assume that the coarse-grained regular expression generated by a specific connection method is "A1|A3" (indicating matching A1 or A3); for the substring set of voltage values ​​{B1, B2, B1}, the coarse-grained regular expression generated is "B1|B2".

[0056] The coarse-grained regular expressions corresponding to the variable slots are combined according to the structure of the longest common subsequence to form a sub-regular expression: "When I is (A1|A3) and V is (B1|B2), the three-dimensional coiled core exhibits F1 phenomenon, which belongs to H1 hidden danger data."

[0057] Because there is only one target example feature (about "F1"), the final regular expression is the sub-regular expression. If there are multiple target example features, sub-regular expressions are generated separately and combined according to logical relationships.

[0058] The hidden danger data analysis module is used to extract entities in the first category hidden danger data and the second category hidden danger data according to regular expressions, form a preliminary entity relationship pair list, determine the dependency relationship in the preliminary entity relationship pair list, and update it into the entity relationship pair list.

[0059] The specific operation method of the preliminary entity relationship pair list is: scan the first category hidden danger data and the second category hidden danger data, when there is content that conforms to the regular expression in the data corresponding to the first category hidden danger data and the second category hidden danger data, record the matching position and content, preset the grouping rules of the regular expression, extract the corresponding entity from the matching content, and based on the matching result of the regular expression, preliminarily judge the entity relationship and record it to form a preliminary entity relationship pair list; it is helpful to deeply understand the intrinsic connection between the two categories of hidden danger data, and finally analyze the cause and development trend of the fault.

[0060] It should be noted that the grouping rules of the regular expression are: based on semantic grouping: the parts with clear semantic associations are divided into different groups. For example, the regular expression (current)\s* of "the current is 10A" is \s*(\d+[A-Za-z]*) as an example, "current" representing the name of the physical quantity is placed in the first group, and the specific value and unit "10A" are placed in the second group.

[0061] Group by data category: For "fault phenomenon is iron core overheating", the regular expression (fault phenomenon)\s* is \s*(.*), and the "fault phenomenon" part representing the data category (the category describing the equipment fault condition) is taken as the first group, and the specific fault description "iron core overheating" is taken as the second group.

[0062] It should be noted that, in a specific embodiment, the following hidden danger data about the three-dimensional wound core is obtained from the database of the power system: the current is 10A, the voltage is 220V, the fault phenomenon is core overheating, and the hidden danger category is the first type of hidden danger data; these numerical values ​​are converted into string text, the preprocessed text is scanned line by line, and the regular expression is matched: (current)\s* is\s*(\d+[A-Za-z]*), for "voltage is 220V", it is set to (voltage)\s* is\s*(\d+[A-Za-z]*), for "fault phenomenon is core overheating", it is set to (fault phenomenon)\s* is\s*(.*), for "hidden danger category is the first type of hidden danger data", it is set to (hidden danger category)\s* is\s*(.*).

[0063] When "the current is 10A" is scanned, the matching position and content are recorded according to the regular expression matching. According to the preset grouping rules, the first group (current) extracts the entity "current", and the second group (\d+[A-Za-z]*) extracts the entity "10A". In the same way, for "the voltage is 220V", the entities "voltage" and "220V" are extracted, and for "the fault phenomenon is iron core overheating", the entities "fault phenomenon" and "iron core overheating" are extracted, and for "the hidden danger category is the first category of hidden danger data", the entities "hidden danger category" and "first category of hidden danger data" are extracted.

[0064] The entity relationship is preliminarily judged, and it is identified that "iron core overheating" is the description of "fault phenomenon" and "first type of hidden danger data" is the description of "hidden danger category", thus forming a preliminary entity relationship pair, such as (fault phenomenon, description, iron core overheating), (hidden danger category, description, first type of hidden danger data), forming a preliminary entity relationship pair list, see Table 3 for details.

[0065] Table 3. Preliminary entity-relationship pair list

[0066]

[0067] The specific analysis method of the dependency relationship is: perform entity extraction on the first category hidden danger data and the second category hidden danger data, obtain the marked part of speech and dependency relationship label of each entity, traverse the entity relationship pair list, match the entities therein with the words in the determined grammatical structure, and generate an entity relationship pair list according to the grammatical structure of each entity in the dependency relationship label; through entity extraction, key semantic units can be extracted from the hidden danger data, and obtaining the marked part of speech and dependency relationship label helps to understand the grammatical function and semantic role of these entities in the sentence in more detail, so as to deeply understand the information conveyed by the data; by matching the entities with the words in the determined grammatical structure, the original entity relationship pair list can be optimized, unreasonable relationships can be removed, and missing relationships can be supplemented, so that the list is more in line with the logic of the language and the actual situation of the data.

[0068] The parts of speech include nouns, verbs, adjectives, etc., and the dependency labels include subjects, objects, etc.

[0069] It should be noted that the entity relationship pair list is traversed, and the entities therein are matched with the words in the determined grammatical structure. According to the dependency relationship between the entities in the grammatical structure, it is determined that "core overheating" is the object, and "10A" and "220V" are the subjects, and it can be determined that there is an association relationship between them; the entity relationship obtained through the dependency relationship analysis is updated to the entity relationship pair list, see Table 4 for details.

[0070] Table 4. List of entity-relationship pairs

[0071]

[0072] The knowledge graph construction module is used to construct a graph network corresponding to the entity relationship based on the entity relationship list, trace the fault source based on the graph network, obtain the propagation path of each fault, and construct a knowledge graph.

[0073] The specific operation method of constructing the graph network corresponding to the entity relationship is as follows: reading the entity relationship pair list and creating each entity node.

[0074] According to the relationship between the entities recorded in the entity relationship pair list, relationship edges are created between the entity nodes; representing the entity relationship as relationship edges between nodes can clearly show the direct connection between different entities, help to intuitively understand the position and relationship of each entity in the entire system, and make complex data relationships clear at a glance.

[0075] Traverse each entity node, supplement the entity relationship connection context information that is missing key information, identify the repeated nodes between each entity relationship, superimpose the repeated nodes, and establish a graph network based on the superimposed entity relationships; the established graph network is the basic structure of the knowledge graph, which shows the relationship between entities in an intuitive graph form. On this basis, further enrich the semantic information, attribute information and other content of the nodes and edges to construct a complete knowledge graph to support more in-depth fault analysis and reasoning tasks.

[0076] It should be noted that, in a specific embodiment, there is a description of a power equipment failure, which mentions that "device A has an abnormality and the current value exceeds the normal range" and "the abnormality causes device A to be unable to work normally." The entity relationship of "abnormality" lacks key information (such as the specific abnormality type). From the context, it can be known that it is related to the previously mentioned "current value exceeds the normal range", and the key information can be supplemented as "current abnormality."

[0077] It should be noted that, in a specific embodiment, the entity identifier of the power equipment is "equipment name_equipment type_text paragraph number_order in the paragraph". There are two nodes in different entity relationships, one is "equipment A_transformer_paragraph 3_2", and the other is "equipment A_transformer_paragraph 5_1". By comparing key parts such as the equipment name and equipment type in the identifier, it is determined that they are repeated nodes. After the repeated nodes are identified, the entity relationships related to these repeated nodes are superimposed. When superimposing repeated nodes, the weights need to be adjusted accordingly. The weight of the relationship between the node "equipment A_transformer" and "overload" in the first entity relationship is 3 (indicating 3 occurrences), and the weight in the second entity relationship is 2. After superposition, the weight of the relationship is updated to 5.

[0078] The specific analysis method of the fault tracing is as follows: for the relationship edges between each entity node, the occurrence frequency of the relationship edges in the historical fault data is counted, and the occurrence frequency of the relationship edges between each entity node is converted into a weight through a normalization function, thereby obtaining the weight of the relationship edges between each entity node; the importance of the relationship edges between each entity node in the fault occurrence process can be quantified in numerical form, which helps to distinguish the different influences of different relationship edges on the fault.

[0079] It should be noted that in a specific embodiment, there are 100 historical fault records, the relationship edge "overload → oil temperature is too high" appears 30 times, "cooling fan failure → oil temperature is too high" appears 25 times, and "oil temperature is too high → winding short circuit" appears 15 times. Suppose the total frequency of all relationship edges is , No. The frequency of occurrence of the relationship edge is , Indicates The number of the relationship edge, , through the normalization function Calculate its weight. In the above, , for "load is too large → oil temperature is too high", its weight , for "Cooling fan failure → oil temperature is too high", its weight , for "Oil temperature is too high → winding short circuit", its weight .

[0080] When a fault occurs in the three-dimensional wound core, the known fault node is recorded as the termination node. The weights of the relationship edges connected to the termination node are compared, and the relationship edge weight threshold is set. The relationship edges that are greater than or equal to the set relationship edge weight threshold are screened out, and the corresponding upstream nodes are recorded as the upstream propagation nodes of each fault. This provides a basis for positioning from the fault occurrence point to the possible fault source, highlights the upstream nodes that are more closely related to the fault, reduces unnecessary troubleshooting nodes, and improves the efficiency of fault tracing.

[0081] Each determined upstream propagation node of the fault is used as a new analysis node, and the weights of each relationship edge connected to it are obtained again. The method is then used to trace back to the upstream nodes along the edges with higher weights until a node without other upstream nodes is reached. This node is the starting node of the fault, and each fault propagation path is constructed. The starting node, ending node, and intermediate nodes of each fault propagation path are obtained, as well as the relationship type and relationship attributes of the edges between the nodes, to construct a knowledge graph. Constructing a knowledge graph integrates various information related to fault propagation and presents it in a visual and structured manner, which helps to understand the mechanism of fault propagation as a whole and provide a comprehensive knowledge basis for subsequent fault analysis, prevention, and repair.

[0082] The knowledge graph reasoning module is used to perform rule matching reasoning on the three-dimensional wound iron core through the knowledge graph, and output the high-frequency fault types of the three-dimensional wound iron core; by performing rule matching reasoning on the three-dimensional wound iron core through the knowledge graph, it can infer the high-frequency fault types of the three-dimensional wound iron core based on the information in the constructed knowledge graph, such as node relationships, attributes, etc.

[0083] The specific method of the knowledge graph reasoning module is: use the reasoning engine to traverse the entity relationships in the knowledge graph, and match them with the loaded rules through the forward linking algorithm. If the rule matches successfully, the reasoning engine executes the corresponding action part according to the logic of the rule to generate the reasoning result; it can make full use of the entity relationship information in the knowledge graph, so that the reasoning process is closely integrated with the existing knowledge structure, thereby obtaining a reasoning result that is more in line with the actual situation.

[0084] It should be noted that the specific content of the forward chaining algorithm is: starting from known facts, through the premise of matching rules, new facts are continuously deduced until the goal is achieved or no new conclusions can be generated.

[0085] For complex reasoning tasks, after each reasoning is completed, it is determined whether there are newly generated facts. The facts are a combination of entities and relationships. If there are newly generated facts, rule matching is performed again to determine whether new rules can be triggered. The above reasoning process is repeated and a reasoning cycle is performed until no new facts are generated. Through continuous cyclic reasoning, the potential information in the knowledge graph can be deeply mined. The newly generated facts may reveal previously undiscovered relationships or situations, which helps to understand the problem more comprehensively.

[0086] It should be noted that, based on the entity relationship in Table 3, the list is analyzed for a specific embodiment, and the rule content is set: "Rule 1: If there are two different operating parameters that are associated with the same type of equipment fault potential, then the possibility of this type of fault potential is higher; Rule 2: If a certain type of fault potential belongs to a specific hidden danger data category and has a higher possibility of occurrence, then the fault potential under this category should be paid special attention to."

[0087] The reasoning engine traverses the entity relationships in the knowledge graph. When it traverses the association between "10A" and "iron core overheating" and the association between "220V" and "iron core overheating", it matches them with Rule 1 through the forward chaining algorithm.

[0088] Since the condition in Rule 1 that "two different operating parameters (10A represents current, 220V represents voltage) are both associated with the same type of equipment failure potential (core overheating)" is met, the inference engine executes the action part according to the logic of Rule 1 and generates the inference result: "The possibility of the core overheating failure potential occurring is high", and this new fact is generated at this time.

[0089] Because there are newly generated facts, the reasoning engine performs rule matching again and finds that the newly generated facts match the condition in Rule 2: "A certain type of fault hidden danger (core overheating) belongs to a specific hidden danger data category (the first type of hidden danger data) and has a higher possibility of occurrence (just inferred)". The reasoning engine executes the action part of Rule 2 and generates the reasoning result: "We should focus on the core overheating fault hidden danger in the first type of hidden danger data."

[0090] The rule matching is performed again. No new rules can be triggered by the newly generated facts in this round, and the reasoning cycle ends.

[0091] Obtain all inference results generated by the inference engine and determine whether each inference result has a unique identifier. If no unique identifier exists, perform a deduplication operation on the inference results, classify the deduplication results into fault types, and count the number of inference results of each fault type. , Indicates The number of the fault type, , calculate the frequency of occurrence of each fault type , the formula is ,in It represents the total number of inference results, sets the frequency threshold, and records the fault types with a probability greater than the set frequency threshold as high-frequency fault types. The deduplication operation can avoid repeated inference results from interfering with subsequent analysis and ensure the uniqueness of the results. By calculating the frequency of occurrence of fault types, the possibility of occurrence of different fault types can be quantitatively analyzed, which helps to quickly determine the high-frequency fault types among many possible fault types.

[0092] It should be noted that the derivation of the frequency formula of each fault type is based on the basic definition of frequency: frequency refers to the proportion of a certain event in the total number of events. It is The number of times a fault type occurs. is the sum of the occurrences of all fault types, so It means the The frequency of a fault type appearing in all inference results.

[0093] It should be noted that the specific method of the deduplication operation is: obtain the inference results generated by the inference engine, and regard each inference result as a "whole" containing multiple aspects of information. For example, in the inference results of fault analysis, this "whole" includes the specific phenomenon of the fault (such as core overheating), the possibility of the fault (such as a high possibility), the hidden danger data category to which the fault belongs (such as the first category of hidden danger data), and related attention level prompts (such as should be paid special attention to), etc.

[0094] Check each inference result one by one. For the first inference result, treat it as non-duplicate and put it into a new set used to store unique results. For the second inference result, compare the key information it contains, such as fault phenomenon, hidden danger data category, etc., with the corresponding key information of the first inference result already in the set. If all the key information is exactly the same, it is judged that the two inference results are essentially duplicates, and the second inference result is not included in the set. If the key information is not exactly the same, the second inference result is included in the set.

[0095] In the same way, each inference result is checked and compared in turn, and the inference results with different key information are retained in the set, while the repeated inference results with the same key information are excluded, thereby completing the deduplication operation.

[0096] It should be noted that, in a specific embodiment, the inference engine generates two inference results: "The possibility of the core overheating fault hidden danger occurring is high" and "The core overheating fault hidden danger in the first category of hidden danger data should be paid special attention to." These two inference results do not have a unique identifier, and deduplication operations are performed (there are no duplicate results here).

[0097] After deduplication, the inference results are analyzed to determine that the fault type is "iron core overheating" and it belongs to the first category of hidden danger data. The number of inference results of the fault type "iron core overheating" is 2 (here only these two related results are inferred from the given entity relationship), and the total number of inference results is 2. The frequency of occurrence of the iron core overheating fault type is calculated. According to the formula, it can be obtained that: frequency of occurrence = 2 / 2 = 1.

[0098] The occurrence frequency threshold is set to 0.5. Since the occurrence frequency of the core overheat fault type is 1, which is greater than the set occurrence frequency threshold 0.5, "core overheat" is recorded as a high-frequency fault type.

[0099] Management database for storing historical fault reports of three-dimensional wound cores.

[0100] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations on the present invention. A person skilled in the art may make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention and they are still covered by the protection scope of the present invention.

Claims

1. The intelligent management system for checking safety hazards of power products based on the industrial Internet is characterized by: The system specifically includes the following modules: A fault parameter acquisition module is used to filter hidden danger data from the operating parameters of the three-dimensional wound core, and sequentially obtain the first type of hidden danger data and the second type of hidden danger data; A regular expression generation module, used for associating the first-category hidden danger data and the second-category hidden danger data with the historical fault report, and generating regular expressions corresponding to the first-category hidden danger data and the second-category hidden danger data; The hidden danger data analysis module is used to extract entities in the first category hidden danger data and the second category hidden danger data according to regular expressions, form a preliminary entity relationship pair list, determine the dependency relationship in the preliminary entity relationship pair list, and update it into the entity relationship pair list; The knowledge graph construction module is used to construct a graph network corresponding to the entity relationship based on the entity relationship list, trace the fault source based on the graph network, obtain the propagation path of each fault, and construct a knowledge graph; The knowledge graph reasoning module is used to perform rule matching reasoning on the three-dimensional wound iron core through the knowledge graph and output the high-frequency fault types of the three-dimensional wound iron core.

2. According to the industrial Internet-based intelligent management system for troubleshooting safety hazards of power products according to claim 1, it is characterized in that: The operating parameters of the three-dimensional wound core include current, voltage and power.

3. According to the industrial Internet-based intelligent management system for troubleshooting safety hazards of power products according to claim 1, it is characterized in that: The implementation of the fault parameter acquisition module also includes: Set upper and lower limit thresholds for the operating parameters of the three-dimensional wound core. When the operating parameters of the three-dimensional wound core are not within the set upper and lower limit thresholds, a fault alarm is triggered, and the operating parameters of the three-dimensional wound core at this time are classified and recorded; If the monitored equipment operating parameters are less than the set lower limit threshold of the equipment operating parameters, the equipment operating data at this time will be recorded as the first type of hidden danger data. If the monitored equipment operating parameters are greater than the set upper limit threshold of the equipment operating parameters, the equipment operating data at this time will be recorded as the second type of hidden danger data.

4. The intelligent management system for troubleshooting safety hazards of power products based on the industrial Internet according to claim 1 is characterized in that: The specific operation method of the regular expression generation module is: S1. Taking the historical fault report of the three-dimensional coil core as an example, establish the abnormal parameters under each fault state and the abnormal association text of the first type of hidden danger data and the second type of hidden danger data; S2. Create a conversion table and extract the sample features corresponding to each sample according to the mapping relationship of the abnormal associated text; S3. Count the occurrence probability of each example feature, filter out each target example feature according to the set probability threshold, obtain the longest common subsequence in each target example feature and mark the variable slot, extract its corresponding substring, generate each coarse-grained regular expression from each substring, combine them into sub-regular expressions, and finally generate a regular expression.

5. The intelligent management system for troubleshooting safety hazards of power products based on the industrial Internet according to claim 4 is characterized in that: The step S3 comprises: S31. Count the occurrence probability of each example feature in all examples, sort the example features in descending order of occurrence probability, add them up in sequence to obtain a cumulative probability distribution map, and screen the example features in the cumulative probability distribution map to obtain each target example feature; S32. Obtain the longest common subsequence from each target example feature, mark the part of the longest common subsequence that will change in different target example features as a variable slot, and then extract the specific substring corresponding to the position of these variable slots from each target example feature; S33. For each substring set composed of specific substrings, generate coarse-grained regular expressions for each substring in each substring set, combine the coarse-grained regular expressions corresponding to each variable slot according to the structure of the longest common subsequence to form sub-regular expressions of each target example feature, and combine the sub-regular expressions of each target example feature to generate regular expressions corresponding to the first category of hidden danger data and the second category of hidden danger data.

6. The intelligent management system for troubleshooting safety hazards of power products based on the industrial Internet according to claim 5 is characterized in that: The specific operation method of the preliminary entity relationship pair list is: Scan the first-category hidden danger data and the second-category hidden danger data. When the data corresponding to the first-category hidden danger data and the second-category hidden danger data contain content that matches the regular expression, record the matching position and content, preset the regular expression grouping rules, extract the corresponding entities from the matching content, and based on the matching results of the regular expression, preliminarily judge the entity relationship and record it to form a preliminary entity relationship pair list.

7. The intelligent management system for troubleshooting safety hazards of power products based on the industrial Internet according to claim 1 is characterized in that: The specific analysis method of the dependency relationship is as follows: Perform entity extraction on the first and second hidden danger data, obtain the annotated part of speech and dependency tag of each entity, traverse the entity relationship pair list, match the entities with the words in the determined grammatical structure, and generate an entity relationship pair list according to the grammatical structure of each entity in the dependency tag.

8. The intelligent management system for troubleshooting safety hazards of power products based on the industrial Internet according to claim 1 is characterized in that: The specific operation method of constructing the graph network corresponding to the entity relationship is as follows: Read the entity relationship pair list and create each entity node; According to the relationship between the entities recorded in the entity-relationship pair list, a relationship edge is created between the entity nodes; Traverse each entity node, supplement the entity relationship connection context information that lacks key information, identify the repeated nodes between each entity relationship, superimpose the repeated nodes, and establish a graph network based on the superimposed entity relationships.

9. The intelligent management system for troubleshooting safety hazards of power products based on the industrial Internet according to claim 8 is characterized in that: The specific analysis method of the fault tracing is as follows: For the relationship edge between each entity node, its occurrence frequency in the historical fault data is counted, and the occurrence frequency of the relationship edge between each entity node is converted into a weight through a normalization function, thereby obtaining the weight of the relationship edge between each entity node; When a fault occurs in the three-dimensional wound core, the known fault node is recorded as the termination node, the weights of the relationship edges connected to the termination node are compared, the relationship edge weight threshold is set, and the relationship edges greater than or equal to the set relationship edge weight threshold are screened out, and the corresponding upstream nodes are recorded as the upstream propagation nodes of each fault; Take each determined upstream fault propagation node as a new analysis node, and obtain the weights of each relationship edge connected to it again, so as to trace back to the upstream node along the edge with higher weight until tracing back to a node without other upstream nodes. This node is the starting node of the fault, and thus construct each fault propagation path. From this, the starting node, end node, and intermediate nodes of each fault propagation path are obtained, and the relationship type and relationship attributes of the edges between nodes are constructed to construct a knowledge graph.

10. The intelligent management system for troubleshooting safety hazards of power products based on the industrial Internet according to claim 9 is characterized in that: The specific method of the knowledge graph reasoning module is: Use the inference engine to traverse the entity relationships in the knowledge graph and match them with the loaded rules through the forward linking algorithm. If the rule matches successfully, the inference engine executes the corresponding action part according to the logic of the rule and generates the inference result; For complex reasoning tasks, after each reasoning is completed, it is determined whether there are new facts generated. The facts are a combination of entities and relationships. If there are new facts, rule matching is performed again to determine whether new rules can be triggered. The above reasoning process is repeated and the reasoning cycle is performed until no new facts are generated. Obtain all inference results generated by the inference engine and determine whether each inference result has a unique identifier. If no unique identifier exists, perform a deduplication operation on the inference results, classify the deduplication results into fault types, and count the number of inference results of each fault type. , Indicates The number of the fault type, , calculate the frequency of occurrence of each fault type , the formula is ,in It represents the total number of inference results, sets the occurrence frequency threshold, and records the fault types with a probability greater than the set occurrence frequency threshold as high-frequency fault types.

Images