A substation multi-device fault association rule mining method based on FP-growth

Abstract

The application is suitable for the technical field of substation equipment fault diagnosis, and provides a substation multi-device fault association rule mining method based on FP-Growth, which comprises the following steps: collecting multi-source monitoring data through online monitoring devices arranged on key primary devices and secondary devices in a substation, standardizing the collected multi-source monitoring data to establish a standardized fault transaction database, mining frequent item sets based on the standardized fault transaction database to obtain all frequent item sets satisfying a minimum support threshold, generating fault association rules based on the mined frequent item sets and screening the generated association rules through multi-dimensional evaluation indexes to reserve association rules satisfying preset conditions, and storing the screened association rules and corresponding evaluation indexes into a fault association rule knowledge base for root cause tracing of substation device faults. The mining efficiency under large-scale sparse monitoring data and the accuracy and practicability of the mining results are improved.

Description

Technical Field

[0001] This invention relates to the field of substation equipment fault diagnosis technology, and in particular to a method for mining fault association rules of multiple equipment in substations based on the Frequent Pattern Growth (FP-Growth) algorithm. Background Technology

[0002] In substation operation and maintenance management, equipment fault analysis is typically based on various status data collected by online monitoring devices deployed on critical equipment such as transformers, combined electrical appliances, and transmission lines. Existing technologies analyze this single-source monitoring data, such as monitoring parameters like the content of characteristic gases in transformer oil chromatograms, the amplitude of partial discharge signals, or the operating characteristics of circuit breakers. When a parameter exceeds a preset threshold, an alarm is triggered, prompting maintenance personnel to focus on or repair the equipment. Furthermore, some systems employ association rule mining algorithms to analyze historical fault data. By scanning the fault record database, they calculate the co-occurrence frequency between fault events, identify simultaneous or sequential fault patterns, and output these patterns in rule form to assist maintenance personnel in fault diagnosis.

[0003] However, when performing substation fault association rule mining, existing technologies rely on traditional algorithms to process the massive and sparse monitoring data of substations. This requires multiple database scans and generates a large number of candidate options, resulting in low mining efficiency and difficulty in handling large-scale data scenarios. Furthermore, the directly output association rules lack an effective filtering process, with many rules lacking practical guidance filling the results, leading to insufficient accuracy and practicality of the mining results and making it difficult to effectively support root cause tracing of substation equipment faults. Summary of the Invention

[0004] This invention provides a method for mining fault association rules of substation multi-equipment based on FP-Growth, which is used to solve the problem that the mining results are not accurate and practical, and it is difficult to effectively support the root cause tracing of substation equipment faults.

[0005] This invention provides a method for mining fault association rules of substations based on FP-Growth, including: Multi-source monitoring data is collected through online monitoring devices deployed on key primary and secondary equipment within substations; The collected multi-source monitoring data are standardized and a standardized fault transaction database is established. Based on the standardized fault transaction database, the FP-Growth algorithm is used to mine frequent itemsets to obtain all frequent itemsets that meet the minimum support threshold. Fault association rules are generated based on the frequent itemsets obtained from mining. The generated association rules are filtered through multi-dimensional evaluation indicators, and association rules that meet the preset conditions are retained. The fault association rules are represented as the implication relationship between the antecedent fault code and the consequent fault code, and the antecedent and consequent together constitute the fault propagation link. The filtered association rules and their corresponding evaluation indicators are stored in the fault association rule knowledge base for root cause tracing of substation equipment faults.

[0006] Furthermore, the standardization processing of the collected multi-source monitoring data to establish a standardized fault transaction database includes: Map fault events in the raw monitoring data to a unified fault code; A sliding time window is used to segment consecutively occurring fault events, and the fault codes appearing in each time window are deduplicated to form a transaction item set; All transaction item sets are aggregated to form a standardized fault transaction database.

[0007] Furthermore, based on the standardized fault transaction database, the FP-Growth algorithm is used to mine frequent itemsets, obtaining all frequent itemsets that satisfy the minimum support threshold, including: The standardized fault transaction database is scanned for the first time to calculate the support of each fault code. Fault codes with support reaching the minimum support threshold are identified as frequent items, and a header table is constructed. The header table records the fault code, support count, and pointers to the corresponding nodes in the FP-Tree. The frequent items are sorted in descending order of support to obtain a sorted list of frequent items; The standardized fault transaction database is scanned a second time. Only the frequent items of each transaction are retained. The retained items are sorted according to the order of the sorted list of frequent items. The sorted transaction item sequence is inserted into the FP-Tree as a path. The common prefix path is shared and the node count is increased to construct the FP-Tree. Starting with the most frequent item with the lowest support in the header table, each item is used as a suffix. All paths containing that suffix are backtracked in the FP-Tree to generate the conditional pattern base for that suffix. Each conditional pattern base is used as a new dataset. Conditional FP-Tree is recursively constructed, and the paths obtained during the recursion are combined with the suffix to generate all frequent itemsets that satisfy the minimum support threshold. After processing all frequent items in the header tables, output all frequent itemsets.

[0008] Furthermore, the first scan of the standardized fault transaction database calculates the support of each fault code, and fault codes whose support reaches the minimum support threshold are designated as frequent terms, including: Calculate each fault code using the following formula support : in: To standardize the fault transaction database, For the total number of transactions, For the first One transaction, For including fault codes The number of transactions; Will satisfy The fault codes are treated as frequent items, forming a frequent itemset. And record the support count for each frequent item, where This is the preset minimum support threshold.

[0009] Furthermore, the step of sorting the frequent items in descending order of support to obtain a sorted list of frequent items includes: Frequent itemsets The items in the list are sorted from highest to lowest according to their support count, generating an ordered list. ,satisfy ,in These are the frequent items after sorting.

[0010] Furthermore, the second scan of the standardized fault transaction database retains only the frequent items for each transaction, and sorts the retained items according to the order of the sorted frequent item list. The sorted transaction item sequence is then inserted into the FP-Tree as a path, sharing a common prefix path and increasing the node count to construct the FP-Tree, which includes: Filter out the infrequent items in each transaction, and rearrange the remaining items according to the order in the ordered list to obtain an ordered sequence of items; The ordered sequence of items is inserted into the FP-Tree as a path. Starting from the root node, the current node is checked in sequence to see if there is a corresponding child node. If there is, the count of the child node is incremented. If there is no child node, a new child node is created and the count is initialized to 1. At the same time, the pointer of the corresponding item in the header table is pointed to the new node, or the new node is added to the linked list through node linking. After all transactions are processed, a complete FP-Tree is obtained.

[0011] Furthermore, the step of starting with the most frequent item with the lowest support in the header table and sequentially using it as a suffix, backtracking through all paths containing that suffix in the FP-Tree to generate a conditional pattern base for that suffix includes: Select an unprocessed frequent item from the header table as a suffix, and traverse all nodes in the FP-Tree that contain the suffix along the node linked list corresponding to the frequent item in the header table; For each node traversed, backtrack the path from the root node to the parent node of the node, excluding the node itself, to obtain a prefix path, and set the count of the prefix path as the count of the node. All backtracked prefix paths and their corresponding counts are summarized to form the conditional pattern base of the suffix.

[0012] Furthermore, the conditional pattern base is used as a new dataset to recursively construct a conditional FP-Tree, and the paths obtained during the recursion are combined with the suffix to generate all frequent itemsets that satisfy the minimum support threshold, including: Treat the conditional pattern base as a new transaction dataset, scan all items in it, calculate the local support of each item, and retain only the items whose local support reaches the minimum support threshold. Based on the retained items, the conditional FP-Tree of the suffix is ​​recursively constructed in the same way as the FP-Tree is constructed; During the recursive construction of the conditional FP-Tree, the suffix is ​​combined with the frequent itemsets mined from the conditional FP-Tree to obtain a new frequent itemset; Repeat the above recursive process until there are no more items to retain in the conditional pattern base, thus completing the frequent itemset mining of the suffix.

[0013] Furthermore, the process of generating fault association rules based on the frequently mined itemsets is further refined by using multi-dimensional evaluation metrics to filter the generated association rules and retain those that meet preset conditions, including: If each frequent itemset obtained by mining contains at least two fault codes, then all non-empty proper subsets of the frequent itemset are enumerated as candidate antecedents, and the fault codes remaining in the frequent itemset after removing the candidate antecedents are taken as the corresponding candidate consequents, and candidate association rules are generated. Calculate the confidence level of each candidate association rule, whereby the confidence level measures the probability that the candidate consequent will occur simultaneously given that the candidate antecedent has occurred. Candidate association rules with a confidence level greater than or equal to the preset minimum confidence threshold are selected as initial association rules; Calculate the lift of each initial association rule, whereby the lift measures the degree to which the occurrence of the candidate antecedent increases the probability of the occurrence of the candidate consequent. Initially selected association rules with a lift greater than 1 are used as fault association rules that meet the preset conditions.

[0014] Furthermore, the step of selecting initial association rules with a lift greater than 1 as fault association rules that meet preset conditions includes: If the lift of the initial association rule is greater than 1, it is determined that there is a positive correlation between the candidate antecedent and the candidate consequent of the initial association rule, and the initial association rule is identified as a fault association rule. If the lifting degree of the initial association rule is equal to 1, then the candidate antecedent and candidate consequent of the initial association rule are determined to be independent of each other, and the initial association rule is eliminated. If the lift of the initial association rule is less than 1, it is determined that there is a negative correlation between the candidate antecedent and the candidate consequent of the initial association rule, and the initial association rule is removed.

[0015] As can be seen from the above technical solutions, the present invention has the following advantages: This invention collects multi-source monitoring data through online monitoring devices deployed on key primary and secondary equipment in substations. The collected multi-source monitoring data is standardized to establish a standardized fault transaction database. Based on the standardized fault transaction database, the FP-Growth algorithm is used to mine frequent itemsets to obtain all frequent itemsets that meet the minimum support threshold. Fault association rules are generated based on the mined frequent itemsets, and the generated association rules are filtered and retained by multi-dimensional evaluation indicators. The filtered association rules and their corresponding evaluation indicators are stored in a fault association rule knowledge base for root cause tracing of substation equipment faults. This invention introduces the FP-Growth algorithm, which requires only two database scans to complete frequent itemset mining, significantly improving mining efficiency under large-scale sparse monitoring data. Simultaneously, by constructing multi-dimensional evaluation indicators to filter the generated association rules, only rules that meet preset conditions such as confidence and lift are retained, eliminating a large number of redundant rules without practical guidance, effectively improving the accuracy and practicality of the mining results. Finally, the stored fault association rule knowledge base clearly presents the implicit relationships between fault codes and the fault propagation chain, providing reliable data support for root cause tracing of substation equipment faults. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of an embodiment of a method for mining fault association rules of substations based on FP-Growth in this invention; Figure 2This is a flowchart of the first scan and frequent item sorting in this invention; Figure 3 This is a flowchart of the FP-Tree construction process in this invention; Figure 4 This is a flowchart of the recursive mining process for frequent itemsets in this invention; Figure 5 This is a flowchart of the association rule generation and filtering process in this invention. Detailed Implementation

[0017] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “corresponding to,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0018] Example 1 The implementation method in this embodiment can be implemented in a system, on a server, or on a terminal; no specific limitation is made. The method in this application will be described below from the perspective of system implementation. Please refer to... Figure 1 The method provided in this application includes the following steps: S1. Collect multi-source monitoring data through online monitoring devices deployed on key primary and secondary equipment in substations; The primary equipment here includes main system equipment that directly participates in power generation and distribution, such as transformers, gas-insulated switchgear, circuit breakers, disconnect switches, current transformers, voltage transformers, surge arresters, and transmission lines. Secondary equipment includes auxiliary equipment that monitors, controls, regulates, and protects the primary equipment, such as protection devices, measurement and control devices, automatic devices, fault recorders, and remote control devices. In this embodiment, multi-source monitoring data includes online monitoring data, fault recorder data, protection action information, and maintenance records. The data equipment types include transformers, switchgear, and transmission lines. The online monitoring data for transformers includes oil chromatography online monitoring data, partial discharge monitoring data, winding temperature, top oil temperature, core grounding current, on-load tap changer vibration signal, and cooler operating status. Specifically, the oil chromatography online monitoring data includes the content and gas production rate of characteristic gases such as hydrogen, carbon monoxide, methane, and acetylene. Partial discharge monitoring data includes discharge amplitude, phase, and frequency. The online monitoring data for combined electrical equipment includes high-frequency partial discharge signals, ultrasonic partial discharge signals, sulfur hexafluoride gas pressure, sulfur hexafluoride gas density, trace moisture content, circuit breaker opening and closing coil current waveforms, energy storage motor current, and mechanical characteristic parameters. Mechanical characteristic parameters include opening and closing time and speed. The online monitoring data for transmission lines includes fault recording data, traveling wave ranging data, surge arrester leakage current and number of operations, insulator pollution monitoring data, and conductor temperature and sag. Fault recording data includes voltage transient waveforms and current transient waveforms.

[0019] S2. Standardize the collected multi-source monitoring data and establish a standardized fault transaction database; Based on the raw monitoring data collected through the above steps, this data is converted into a standardized database that can be directly processed by the FP-Growth algorithm, including the following: 1. Map fault events in the raw monitoring data to a unified fault code; 2. A sliding time window is used to segment consecutively occurring fault events, and the fault codes appearing in each time window are deduplicated to form a transaction item set; 3. Summarize all transaction item sets to form a standardized fault transaction database.

[0020] Specifically, when mapping fault events in the raw monitoring data to unified fault codes, a fault type coding dictionary is first established. This dictionary assigns a unique code identifier to each identifiable fault event based on the equipment type and fault phenomenon. For example, a transformer oil chromatography anomaly event can be assigned the code "TR-OIL-001"; a combined electrical appliance sulfur hexafluoride gas pressure low event can be assigned the code "GIS-SF6-002"; and a transmission line single-phase ground fault can be assigned the code "LINE-SC-003". Through this mapping method, all equipment and various types of fault events are transformed into a unified fault code form, facilitating subsequent algorithm processing. When using a sliding time window to segment continuously occurring fault events, a fixed value between fifteen minutes and two hours is set for the window width, and the step size is set to one-quarter to one-half of the window width. Starting from the beginning of each time window, the fault codes corresponding to all fault events occurring within the window are extracted. Duplicate fault codes appearing repeatedly within the same window are deduplicated to ensure that each fault code appears at most once within a transaction. The resulting set of fault codes constitutes a transaction itemset. Simultaneously, metadata information such as the start and end times of the time window and the list of devices involved within the window is appended to each transaction itemset. Finally, all transaction itemsets corresponding to all time windows are aggregated to form a standardized fault transaction database. Each record in this database represents a combination of fault events occurring within a time window, and each record consists of several fault codes.

[0021] S3. Based on the standardized fault transaction database, the FP-Growth algorithm is used to mine frequent itemsets to obtain all frequent itemsets that meet the minimum support threshold; The minimum support threshold is preset based on the sparsity of the substation's historical fault data and the mining objective. The specific value is set through experiments or experience based on the actual application scenario; in this embodiment, it is set to a value between 0.01 and 0.05. This step uses the FP-Growth algorithm to efficiently mine the standardized fault transaction database, discovering frequently occurring fault code combinations, providing a foundation for subsequent association rule generation. Please refer to [link to relevant documentation]. Figures 2 to 4 Specifically, it includes the following: 101. The first scan of the standardized fault transaction database is performed to calculate the support of each fault code. Fault codes whose support reaches the minimum support threshold are taken as frequent items, and a header table is constructed. The header table records the fault code, support count, and pointers to the corresponding nodes in the Frequent Pattern Tree (FP-Tree). Specifically, the first scan of the standardized fault transaction database involves traversing all transactions in the database and counting the total number of occurrences of each fault code. Each fault code is calculated using the following formula. support : in: To standardize the fault transaction database, For the total number of transactions, For the first One transaction, For including fault codes The number of transactions; Will satisfy The fault codes are treated as frequent items, forming a frequent itemset. And record the support count for each frequent item, where This is the preset minimum support threshold.

[0022] Simultaneously, a header table is constructed, which records three pieces of information for each frequent item: fault code, support count of the fault code, and a pointer. The pointer is initially empty and is used to point to the first node in the tree where the fault code appears when the FP-Tree is constructed later.

[0023] 102. Sort the frequent items in descending order of support to obtain the sorted list of frequent items; Frequent itemsets The items in the list are sorted from highest to lowest according to their support count, generating an ordered list. ,satisfy ,in These are the frequently occurring items after sorting. This sorting process aims to ensure that, during subsequent FP-Tree construction, frequently occurring fault codes can share prefix paths as much as possible, thereby achieving a high degree of tree structure compression.

[0024] 103. The second scan of the standardized fault transaction database is performed. Only frequent items are retained for each transaction. The retained items are sorted according to the order of the sorted list of frequent items. The sorted transaction item sequence is inserted into the FP-Tree as a path. The common prefix path is shared and the node count is increased to construct the FP-Tree. The second scan of the standardized faulty transaction database refers to, after completing the header table construction and frequent item sorting, re-traversing each transaction in the database, filtering and rearranging the transactions based on the sorted frequent item list, and constructing an FP-Tree based on the processed ordered item sequence. Specifically, this includes the following: 1. Filter out the infrequent items in each transaction, and rearrange the remaining items in the order of the ordered list to obtain an ordered sequence of items; 2. Insert the ordered sequence of items into the FP-Tree as a path; starting from the root node, check in order whether the current node has a corresponding child node. If it exists, increment the count of the child node. If it does not exist, create a new child node and initialize the count to 1. At the same time, point the pointer of the corresponding item in the head table to the new node, or add the new node to the linked list through node linking. 3. After all transactions are processed, a complete FP-Tree is obtained.

[0025] Specifically, for each transaction in the standardized fault transaction database, firstly based on frequent itemsets... Filter out all infrequent items in the transaction, keeping only the frequent items. Then, sort the remaining frequent items according to... The order of items is rearranged to obtain an ordered sequence. This ordered sequence is then used as a path to insert into the FP-Tree: starting from the root node, each fault code in the ordered sequence is retrieved sequentially. The current node is checked to see if it has a child node corresponding to that fault code. If it does, the count of that child node is incremented by 1. If it doesn't, a new child node is created, its count is initialized to 1, and it is linked to the current node's child node chain. Simultaneously, the pointer corresponding to that fault code in the header table is set to the new node, or the new node is added to the node chain of that fault code in the header table via node linking. After processing all transactions sequentially, frequent item combinations in all transactions are compressed and stored in a tree, forming a complete FP-Tree. For example, for a transaction containing fault codes "TR-OIL-001", "LINE-SC-003", and "GIS-SF6-002", suppose... If the order is “LINE-SC-003”, “TR-OIL-001”, and “GIS-SF6-002”, then the insertion order of the path is as follows: “LINE-SC-003” is inserted first as a child node of the root node, “TR-OIL-001” is inserted as a child node of “LINE-SC-003”, and “GIS-SF6-002” is inserted as a child node of “TR-OIL-001”.

[0026] 104. Starting with the most frequent item with the lowest support in the header table, use each item as a suffix and backtrack through all paths containing that suffix in the FP-Tree to generate the conditional pattern base for that suffix; This step extracts the conditional pattern base corresponding to each frequent item from the FP-Tree, providing a data foundation for subsequent recursive mining. Specifically, it includes the following: 1. Select an unprocessed frequent item from the head table as the suffix, and traverse the linked list of nodes corresponding to the frequent items in the head table to find all nodes in the FP-Tree that contain the suffix; 2. For each node traversed, backtrack the path from the root node to the node's parent node, excluding the node itself, to obtain a prefix path, and set the count of the prefix path as the count of the node. 3. Summarize all the prefix paths obtained from backtracking and the counts corresponding to each prefix path to form the conditional pattern base of the suffix.

[0027] Specifically, the process begins by selecting the most frequent item with the lowest support from the head table and designating it as the current suffix. The process then traverses the linked list of nodes corresponding to this frequent item in the head table, iterating through all nodes in the FP-Tree that contain it. For each node encountered, the process backtracks upwards to the root node, but the backtracking path does not include the current node itself; it only records all nodes from the root node to the current node's parent node, forming a prefix path. The count of this prefix path is set as the count of the current node. The sum of the prefix paths obtained from backtracking all traversed nodes and the counts corresponding to each path constitutes the conditional pattern base for the current suffix. For example, given the suffix "GIS-SF6-002", when traversing the FP-Tree, we find two nodes containing this suffix. The first node has a count of 3, and its prefix path is "LINE-SC-003" → "TR-OIL-001"; the second node has a count of 2, and its prefix path is "LINE-SC-003". Therefore, the conditional schema base of "GIS-SF6-002" contains two paths: the path "LINE-SC-003" → "TR-OIL-001" has a count of 3, and the path "LINE-SC-003" has a count of 2.

[0028] 105. Treat each conditional pattern base as a new dataset, recursively construct a conditional FP-Tree, and combine the paths obtained during the recursion with the suffix to generate all frequent itemsets that satisfy the minimum support threshold; This step, based on the conditional pattern base of each suffix, recursively constructs a conditional FP-Tree to mine all frequent itemsets containing that suffix. Specifically, it includes the following: 1. Treat the conditional pattern base as a new transaction dataset, scan all items in it, calculate the local support of each item, and retain only the items whose local support reaches the minimum support threshold; 2. Based on the retained items, recursively construct the suffix conditional FP-Tree using the same method as constructing the FP-Tree; 3. During the recursive construction of the conditional FP-Tree, the suffix is ​​combined with the frequent itemsets mined from the conditional FP-Tree to obtain new frequent itemsets; 4. Repeat the above recursive process until there are no more items to retain in the conditional pattern base, thus completing the mining of frequent itemsets for suffixes.

[0029] Specifically, the conditional pattern base obtained in the previous step is treated as a new transaction dataset. All items in this dataset are scanned, and the total number of times each item appears in the conditional pattern base is counted. The local support of each item is calculated. The local support is calculated by dividing the total number of times the item appears by the total number of paths in the conditional pattern base. Only items whose local support reaches the minimum support threshold are retained. Based on these retained items, a conditional FP-Tree is constructed for the current suffix using the exact same method as in step 103. During the construction of the conditional FP-Tree, whenever a frequent itemset is mined from the conditional FP-Tree, it is combined with the current suffix to generate a new frequent itemset. Then, the above process is recursively executed on the conditional FP-Tree of the current suffix, that is, the conditional FP-Tree is used as a new FP-Tree, the item with the lowest support is selected as the new suffix, and the conditional pattern base is generated and the next layer of conditional FP-Tree is constructed until the local support of all items in a certain layer of conditional pattern base is lower than the minimum support threshold, and it is no longer possible to construct a conditional FP-Tree. At this point, all frequent itemsets containing the current suffix have been mined.

[0030] 106. After processing all frequent items in the header tables, output all frequent itemsets.

[0031] After all frequent items in the header table have been used as suffixes to complete the above-mentioned conditional pattern base generation and recursive mining, all frequent itemsets that meet the minimum support threshold have been obtained. These frequent itemsets consist of several fault codes, each representing a set of fault events that frequently occur simultaneously within a time window. For example, the frequent itemsets {“LINE-SC-003”, “TR-OIL-001”} indicate that a single-phase ground fault and a transformer lead fault frequently co-occur within the same time window. All the final output frequent itemsets will serve as input data for subsequent association rule generation.

[0032] S4. Based on the frequent itemsets obtained from mining, generate fault association rules. Use multi-dimensional evaluation indicators to filter the generated association rules and retain those that meet the preset conditions. Fault association rules are represented as the implication relationship between the antecedent fault code and the consequent fault code, and the antecedent and consequent together constitute the fault propagation link. This step transforms the frequent itemsets obtained in step S3 into interpretable fault association rules. These rules are then quantitatively evaluated and filtered using multi-dimensional evaluation metrics such as confidence and lift, eliminating unreliable or meaningless rules and retaining only those rules that accurately reflect the relationships between faults for subsequent root cause analysis. Please refer to [link to relevant documentation]. Figure 5 Specifically, it includes the following: 101. If each frequent itemset obtained by mining contains at least two fault codes, then enumerate all non-empty proper subsets of the frequent itemset as candidate antecedents, and take the remaining fault codes in the frequent itemset after removing the candidate antecedents as the corresponding candidate consequents, and generate candidate association rules. 102. Calculate the confidence score for each candidate association rule. The confidence score measures the probability that the candidate consequent will also occur given that the candidate antecedent has occurred. Candidate association rules ,in As a candidate antecedent, It is a candidate consequent, and , To generate the frequent itemset corresponding to this rule, the confidence score is calculated using the following formula: in: For frequent itemsets Support level, indicating the number of people who support it. The percentage of transactions with all fault codes out of the total number of transactions; For candidate predecessor Support level, indicating the number of people who support it. The proportion of transactions with all fault codes to the total number of transactions. The confidence level ranges from 0 to 1. The closer the value is to 1, the higher the probability that the candidate consequent will occur given that the candidate antecedent has occurred.

[0033] 103. Select candidate association rules with a confidence level greater than or equal to the preset minimum confidence threshold as initial association rules; the minimum confidence threshold can be set according to actual business needs. In this embodiment, it is set to a value between 0.6 and 0.8 to ensure that the retained rules have a high predictive strength.

[0034] 104. Calculate the lift of each initial association rule. The lift measures how much the occurrence of a candidate antecedent increases the probability of the occurrence of a candidate consequent. For the initial selection association rules The formula for calculating the improvement is: in: as candidate consequent Support level, indicating the number of people who support it. The lift is the percentage of transactions with all fault codes out of the total number of transactions. This lift reflects the candidate predecessor. The emergence of candidate successors The degree of gain in the probability of occurrence.

[0035] 105. Select the initial association rules with a lift greater than 1 as the fault association rules that meet the preset conditions.

[0036] Based on the lift value, the initial association rules are judged and filtered as follows: 1. If the lift of the initial association rule is greater than 1, it is determined that there is a positive correlation between the candidate antecedent and the candidate consequent of the initial association rule, and the initial association rule is identified as a faulty association rule. 2. If the lifting degree of the initial association rule is equal to 1, then the candidate antecedent and candidate consequent of the initial association rule are determined to be independent of each other, and the initial association rule is removed. 3. If the lifting degree of the initial association rule is less than 1, it is determined that there is a negative correlation between the candidate antecedent and the candidate consequent of the initial association rule, and the initial association rule is removed.

[0037] The fault association rules selected through the above steps, such as "single-phase ground fault → transformer lead fault" or "transformer winding fault → circuit breaker fault," are each presented as an implication relationship between fault codes. The antecedent and consequent of a rule together constitute a fault propagation link, revealing the transmission path between different equipment fault events. These rules and their corresponding evaluation indicators such as support, confidence, and lift will be stored in the fault association rule knowledge base for subsequent fault root cause tracing analysis.

[0038] S5. Store the filtered association rules and corresponding evaluation indicators in the fault association rule knowledge base for root cause tracing of substation equipment faults.

[0039] Specifically, each fault association rule determined in step S4, along with its corresponding support, confidence, and lift metrics, is formatted and stored to construct a fault association rule knowledge base. This knowledge base can be stored in a relational database or a graph database. In a relational database, each rule's antecedent fault code, consequent fault code, support value, confidence value, lift value, and rule discovery time are stored in a table format. If a graph database is used, each fault code is treated as a node, and the fault association rules are treated as directed edges between nodes, with the edges pointing from the antecedent to the consequent. The corresponding support, confidence, and lift are labeled on the edges as attribute information. When a new fault event occurs at the substation, maintenance personnel can query this knowledge base, using the currently occurring fault code as the antecedent for matching, to find possible consequent fault codes, thereby predicting the fault propagation path. Alternatively, they can perform a reverse query using the currently occurring fault code as the consequent to trace the antecedent fault codes that may have caused the fault, thus achieving root cause analysis of the fault. In this way, the fault association rule knowledge base provides direct data support and decision-making basis for tracing the root causes of substation equipment failures.

[0040] It is understood that those skilled in the art can combine various implementation methods in the above embodiments under the guidance of the above examples to obtain technical solutions with multiple implementation methods.

[0041] 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, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for mining fault association rules of substations based on FP-Growth, characterized in that, include: Multi-source monitoring data is collected through online monitoring devices deployed on key primary and secondary equipment within substations; The collected multi-source monitoring data are standardized and a standardized fault transaction database is established. Based on the standardized fault transaction database, the FP-Growth algorithm is used to mine frequent itemsets to obtain all frequent itemsets that meet the minimum support threshold. Fault association rules are generated based on the frequent itemsets obtained from mining. The generated association rules are filtered through multi-dimensional evaluation indicators, and association rules that meet the preset conditions are retained. The fault association rules are represented as the implication relationship between the antecedent fault code and the consequent fault code, and the antecedent and consequent together constitute the fault propagation link. The filtered association rules and their corresponding evaluation indicators are stored in the fault association rule knowledge base for root cause tracing of substation equipment faults.

2. The method for mining fault association rules of substation multi-equipment based on FP-Growth according to claim 1, characterized in that, The standardization process for the collected multi-source monitoring data, and the establishment of a standardized fault transaction database, includes: Map fault events in the raw monitoring data to a unified fault code; A sliding time window is used to segment consecutively occurring fault events, and the fault codes appearing in each time window are deduplicated to form a transaction item set; All transaction item sets are aggregated to form a standardized fault transaction database.

3. The method for mining fault association rules of substation multi-equipment based on FP-Growth according to claim 1, characterized in that, Based on the standardized fault transaction database, the FP-Growth algorithm is used to mine frequent itemsets, obtaining all frequent itemsets that satisfy the minimum support threshold, including: The standardized fault transaction database is scanned for the first time to calculate the support of each fault code. Fault codes with support reaching the minimum support threshold are identified as frequent items, and a header table is constructed. The header table records the fault code, support count, and pointers to the corresponding nodes in the FP-Tree. The frequent items are sorted in descending order of support to obtain a sorted list of frequent items; The standardized fault transaction database is scanned a second time. Only the frequent items of each transaction are retained. The retained items are sorted according to the order of the sorted list of frequent items. The sorted transaction item sequence is inserted into the FP-Tree as a path. The common prefix path is shared and the node count is increased to construct the FP-Tree. Starting with the most frequent item with the lowest support in the header table, each item is used as a suffix. All paths containing that suffix are backtracked in the FP-Tree to generate the conditional pattern base for that suffix. Each conditional pattern base is used as a new dataset. Conditional FP-Tree is recursively constructed, and the paths obtained during the recursion are combined with the suffix to generate all frequent itemsets that satisfy the minimum support threshold. After processing all frequent items in the header tables, output all frequent itemsets.

4. The method for mining fault association rules of substation multi-equipment based on FP-Growth according to claim 3, characterized in that, The first scan of the standardized fault transaction database calculates the support of each fault code, and fault codes whose support reaches the minimum support threshold are designated as frequent terms, including: Calculate each fault code using the following formula support : in: To standardize the fault transaction database, For the total number of transactions, For the first One transaction, For including fault codes The number of transactions; Will satisfy The fault codes are treated as frequent items, forming a frequent itemset. And record the support count for each frequent item, where This is the preset minimum support threshold.

5. The method for mining fault association rules of substation multi-equipment based on FP-Growth according to claim 4, characterized in that, The step of sorting the frequent items in descending order of support to obtain a sorted list of frequent items includes: Frequent itemsets The items in the list are sorted from highest to lowest according to their support count, generating an ordered list. ,satisfy ,in These are the frequent items after sorting.

6. The method for mining fault association rules of substation multi-equipment based on FP-Growth according to claim 5, characterized in that, The second scan of the standardized fault transaction database retains only the frequent items for each transaction, and sorts the retained items according to the order of the sorted frequent item list. The sorted transaction item sequence is then inserted into the FP-Tree as a path, sharing a common prefix path and increasing the node count to construct the FP-Tree, which includes: Filter out the infrequent items in each transaction, and rearrange the remaining items according to the order in the ordered list to obtain an ordered sequence of items; The ordered sequence of items is inserted into the FP-Tree as a path. Starting from the root node, the current node is checked in sequence to see if there is a corresponding child node. If there is, the count of the child node is incremented. If there is no child node, a new child node is created and the count is initialized to 1. At the same time, the pointer of the corresponding item in the header table is pointed to the new node, or the new node is added to the linked list through node linking. After all transactions are processed, a complete FP-Tree is obtained.

7. The method for mining fault association rules of substation multi-equipment based on FP-Growth according to claim 6, characterized in that, Starting with the most frequent item with the lowest support in the header table, and using each item as a suffix, the process involves backtracking through all paths in the FP-Tree that contain that suffix to generate a conditional pattern base for that suffix, including: Select an unprocessed frequent item from the header table as a suffix, and traverse all nodes in the FP-Tree that contain the suffix along the node linked list corresponding to the frequent item in the header table; For each node traversed, backtrack the path from the root node to the parent node of the node, excluding the node itself, to obtain a prefix path, and set the count of the prefix path as the count of the node. All backtracked prefix paths and their corresponding counts are summarized to form the conditional pattern base of the suffix.

8. The method for mining fault association rules of substation multi-equipment based on FP-Growth according to claim 7, characterized in that, The process involves using each conditional pattern base as a new dataset, recursively constructing a conditional FP-Tree, and combining the paths obtained during the recursion with the suffixes to generate all frequent itemsets that satisfy the minimum support threshold, including: Treat the conditional pattern base as a new transaction dataset, scan all items in it, calculate the local support of each item, and retain only the items whose local support reaches the minimum support threshold. Based on the retained items, the conditional FP-Tree of the suffix is ​​recursively constructed in the same way as the FP-Tree is constructed; During the recursive construction of the conditional FP-Tree, the suffix is ​​combined with the frequent itemsets mined from the conditional FP-Tree to obtain a new frequent itemset; Repeat the above recursive process until there are no more items to retain in the conditional pattern base, thus completing the frequent itemset mining of the suffix.

9. The method for mining fault association rules of substation multi-equipment based on FP-Growth according to claim 1, characterized in that, The process of generating fault association rules based on frequently accessed itemsets is further refined using multi-dimensional evaluation metrics. Association rules that meet preset conditions are retained, including: If each frequent itemset obtained by mining contains at least two fault codes, then all non-empty proper subsets of the frequent itemset are enumerated as candidate antecedents, and the fault codes remaining in the frequent itemset after removing the candidate antecedents are taken as the corresponding candidate consequents, and candidate association rules are generated. Calculate the confidence level of each candidate association rule, whereby the confidence level measures the probability that the candidate consequent will occur simultaneously given that the candidate antecedent has occurred. Candidate association rules with a confidence level greater than or equal to the preset minimum confidence threshold are selected as initial association rules; Calculate the lift of each initial association rule, whereby the lift measures the degree to which the occurrence of the candidate antecedent increases the probability of the occurrence of the candidate consequent. Initially selected association rules with a lift greater than 1 are used as fault association rules that meet the preset conditions.

10. The method for mining fault association rules of substation multi-equipment based on FP-Growth according to claim 9, characterized in that, The step of selecting initial association rules with a lift greater than 1 as fault association rules that meet preset conditions includes: If the lift of the initial association rule is greater than 1, it is determined that there is a positive correlation between the candidate antecedent and the candidate consequent of the initial association rule, and the initial association rule is identified as a fault association rule. If the lifting degree of the initial association rule is equal to 1, then the candidate antecedent and candidate consequent of the initial association rule are determined to be independent of each other, and the initial association rule is eliminated. If the lift of the initial association rule is less than 1, it is determined that there is a negative correlation between the candidate antecedent and the candidate consequent of the initial association rule, and the initial association rule is removed.

Images