Power grid resource association mining early warning method, system and device for novel power system, and storage medium
By constructing a fault space model and using association rule mining methods, the problem of proactive early warning of potential grid resource hazards in new power systems was solved, improving the accuracy and efficiency of early warning and enabling a reasonable evaluation of the operating status of grid resources.
Patent Information
- Application Number
- CN202410958520.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-17
- Publication Date
- 2024-11-08
AI Technical Summary
Existing technologies lack the ability to proactively warn of potential grid resource risks in new power systems. Traditional association rule mining methods fail to effectively handle unrecorded fault characteristics, resulting in low early warning efficiency.
By constructing a fault space model, calculating the correlation factors between the fault ontology and one-dimensional and two-dimensional fault features, eliminating relationships with small correlation factors, and using the multidimensional fault feature cohesion algorithm and the Apriori algorithm to mine association rules, a fused binary association early warning frequent set is constructed to achieve real-time feature similarity judgment.
It enables a reasonable evaluation of the operating status of power grid resources, improves the accuracy and efficiency of early warning, and can promptly detect anomalies and potential hazards in power grid resources.
Smart Images

Figure CN118916846A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of power grid digitization, and in particular to a power grid resource association mining early warning method, system, equipment and storage medium for a new power system. Background Art
[0002] In recent years, with the continuous advancement of the construction of new power systems, the power supply structure and grid structure have been continuously adjusted, the scale of grid resources has become increasingly large, the operating characteristics of the power system have undergone major changes, and the system security situation has become more severe. Therefore, higher standards and requirements have been put forward for the daily management of massive power grid resources, but the current power grid resource operation and maintenance system has only realized post-alarm and real-time alarm, and has not realized active early warning. It lacks the ability to warn when various types of power grid resources in the new power system have hidden dangers and abnormalities, lacks the ability to solve hidden dangers, and lacks a supporting management system. In addition, with the construction and development of new power systems, a large amount of data is continuously generated and accumulated, but at present there are only some of the most basic analysis methods, which have failed to realize the value of power grid resource data. When there are hidden dangers in power grid resources, the rapidity and accuracy of early warning and judgment and the timeliness of hidden danger handling not only affect the user's power experience and interests, but also the operating performance and service level of the power supply company.
[0003] The analysis of association rules of fault feature quantities of power grid resources is not accurate enough. The traditional association rule mining method does not take into account the situation that unrecorded fault feature quantities appear in the data, which leads to missed judgments in the association mining warning of power grid resources. On the basis of calculating the association factor of fault feature quantities, when the association factor between fault feature quantities is very small, it means that the probability of causing a fault when the unrecorded fault feature quantity is associated with the recorded fault feature quantity is small. At this time, if the association relationship with a small association factor is not eliminated, the efficiency of association mining warning will be low. Therefore, the mutation capture warning method needs to take into account the situation that unrecorded fault feature quantities appear in the data, and eliminate the association relationship with a small association factor to improve the accuracy and efficiency of association rule analysis. Summary of the invention
[0004] Purpose of the invention: The purpose of the present invention is to provide a grid resource association mining early warning method, system, device and storage medium for a new type of power system to solve the problems existing in the background technology.
[0005] Technical solution: The present invention discloses a method for early warning of power grid resource association mining for a new type of power system, comprising the following steps: aggregating data from multi-source and multi-modal data sources of power grid resources, and extracting aggregated abnormal data of power grid resources; constructing a fault space model based on the abnormal data of power grid resources; calculating the correlation factor between the fault entity and the one-dimensional fault feature quantity based on the fault space model; constructing a multi-dimensional fault feature quantity space from the fault feature quantities corresponding to the power grid resources in the fault space; calculating the two-dimensional fault feature quantity, the two-dimensional fault feature quantity association factor threshold, the correlation degree between the fault entity and the multi-dimensional fault feature quantity, and constructing association rules based on the multi-dimensional fault feature quantity space; obtaining a fused binary association early warning frequent set based on the two-dimensional fault feature quantity association factor, the two-dimensional fault feature quantity association factor threshold, the correlation degree between the fault entity and the multi-dimensional fault feature quantity, and the association rules; calculating the real-time feature quantity similarity based on the fused binary association early warning frequent set, and obtaining a comprehensive early warning result.
[0006] Furthermore, the data sources of the multi-source and multi-modal data of power grid resources include: a new generation of dispatching technical support system, a new generation of centralized control station system, distribution automation system, substation automation system, electricity consumption information collection system, marketing business application system, power grid resource business middle platform, data middle platform and a new generation of equipment asset lean management system.
[0007] Furthermore, the fault entity and one-dimensional fault feature quantity are calculated according to the fault space model as follows:
[0008] According to the historical data of the power grid resources in the fault space, the transaction set within the time t0-t before the fault occurs is obtained and the fault characteristics are recorded. After removing the data within the time t0, an index between each type of power grid resource and the fault characteristics is established; the influence factor algorithm is used to calculate the influence factor between each power grid resource and all its fault characteristics. The algorithm for the correlation factor between the fault entity and a single fault characteristic is:
[0009]
[0010] Among them, IF(d,i j ) is the jth fault characteristic quantity i j The correlation factor with the power grid resource d, n is the number of failures in the historical data of the power grid resource, m p is the fault characteristic quantity i in the pth fault j The number of times the fault occurs within t0-t before it occurs. t0 is configured by the user according to the actual situation. p is the time when the pth fault occurs, t′ q is the fault characteristic quantity i within the time t0-t before the fault occurs j The time when the event occurs, k is an artificial parameter and can be configured by the user according to the actual situation.
[0011] Furthermore, the multi-dimensional fault feature space is constructed as follows:
[0012] The fault feature quantities corresponding to the power grid resources in the fault space are projected into the fault feature quantity space. The cohesion degree is calculated by the cohesion degree and density of the fault feature quantity class to measure the importance of the fault feature quantity. The cohesion degree algorithm of the multi-dimensional fault feature quantity is:
[0013]
[0014] Among them, RD(C i ) represents the fault characteristic type C i The cohesion, C i represents the i-th fault feature class in the fault feature space, m i represents the i-th fault feature type C in the fault feature space i The number of fault features in the fault feature space, n represents the number of all fault features in the fault feature space, n′ represents the number of fault feature classes in the fault feature space, c j The vector representation of the jth fault feature in the fault feature class, v p is the clustering center of the fault feature class, and k is an artificial parameter that can be configured by the user according to actual conditions.
[0015] Furthermore, the calculation of the correlation factor of the two-dimensional fault feature quantity is as follows: the correlation factor between the existing fault feature quantity and the unrecorded fault feature quantity is calculated according to the index, and the formula is as follows:
[0016]
[0017] Among them, i1, i2 are fault features, U1 is the set of fault entities containing the fault feature i1, d1 is a fault information in U1, U2 is the set of fault entities containing the fault feature i2, d2 is a fault information in U2, U′ is the intersection of U1 and U2, that is, U′=U1∩U2, u, u′ is a fault information in U′, S u is the set of fault feature quantities recorded in fault information u, S u′ is the set of fault feature quantities recorded in the fault information u′, |S u |,|S u′ |represents the number of elements in the two sets, |S u ∩S u′ | represents the set S u With the set S u′ The number of elements in the union, α, is an artificial parameter and can be configured by the user according to actual conditions.
[0018] Furthermore, the threshold of the correlation factor of the two-dimensional fault feature quantity is calculated as follows: the correlation relationship with a small correlation factor is eliminated through threshold screening, and the formula is as follows:
[0019]
[0020] Where E represents the threshold of the correlation factor of the two-dimensional fault feature quantity, IF(D,i p ) represents the pth fault characteristic i p The impact factor of the fault set D, RF(item i ,item k ) represents the fault feature item i ,item k The correlation factor between them, n is the number of fault characteristics, Indicates the fault feature item k The smallest correlation factor with all other fault characteristics.
[0021] 7. According to the method of power grid resource association mining and early warning for new power systems described in claim 1, it is characterized in that the correlation between the fault entity and the multidimensional fault feature quantity is calculated as follows: the maximum number of connected fault feature quantities is recorded as K, that is, there are at most K fault feature quantities that are strongly correlated with each other; the correlation between m strongly correlated fault feature quantities is calculated using the fault entity and multidimensional fault feature quantity correlation algorithm, where 2≤m≤K; the formula is:
[0022]
[0023] Among them, S i represents the i-th strongly correlated fault feature class, which contains m fault features, s1, s2, …, s m represents the fault characteristic quantity, f i Represents the strongly correlated fault feature class S i The corresponding fault entity, P(f i ) represents the fault entity f i The probability of occurrence, P(s1|f i ) indicates that the fault body f i P(s2|s1) represents the probability that the fault characteristic quantity s1 will occur at the same time when the fault characteristic quantity s1 occurs, and P(s2|s1) represents the probability that the fault characteristic quantity s2 will occur at the same time when the fault characteristic quantity s1 occurs, and so on until P(s m |s m-1 ). m represents the weight when there are m strongly correlated fault features, n m It indicates the number of m strongly correlated fault feature quantities in the big data statistics of historical data.
[0024] Furthermore, the association rules are constructed as follows: the association rules of the ontology fault are mined using the branch screening optimization strategy and the Apriori algorithm, the missing association rules are found, and the k-item frequent set containing the fault ontology data is obtained.
[0025] Furthermore, the association rule mining based on the branch screening optimization strategy is as follows:
[0026] Filter the event set, assuming that the set of frequent items is A1, A2, ...A n , all events contained in the set are a1, a2, ...a m , the number of times these events occur is k1, k2, ...k m , the frequency of each event is determined, the formula is as follows:
[0027]
[0028] Filter out the events that pass the judgment, which are b1, b2, ...b x , through the events screened out, the number of occurrences of these events corresponds to q1, q2, ...q x , find the set that contains half or more of the events as the judgment passing events. The found set is recorded as B1, B2, ...B y ;
[0029] The event set is pruned, and then the event is judged again. The formula is as follows:
[0030]
[0031] Filter out the events that pass the judgment, which are c1, c2, ...c z , B1, B2, ...B y Central Africa c1, c2, ... c z The event removal results in a set of C1, C2, ...C y ; The event sets C1, C2, ...C y Traverse to form an FP tree, starting from each leaf node of the tree, find the candidate set of two frequent sets, the candidate set of three frequent sets, the candidate set of four frequent sets, the candidate set of five frequent sets... until the calculation of the candidate set is completed;
[0032] Confidence judgment of the found frequent item sets: For any n-item frequent set: calculate the support of each event in all frequent sets and sort them by size. Assume that an event in an n-item frequent set is sorted by support and recorded as t1, t2, ..., t n
[0033]
[0034] The frequent sets that pass the judgment are retained, and those that fail are removed.
[0035] Furthermore, the formula for the fused binary association warning frequent set is as follows:
[0036] A rule =A S02 ∪A S06 ∪A S07
[0037]
[0038] Among them, A rule A represents the association rule set after the fusion of the binary association warning frequent set. S02 Indicates the single association rule set of the fault ontology, A S06 represents the 2-K association rule set of the fault ontology, F represents the fault ontology, represents the single association rule set of the fault ontology obtained by the first method of the fault ontology and the multi-dimensional fault feature quantity association strategy, …, represents the K-item association rule set of the fault ontology obtained by the first method of the fault ontology and the multi-dimensional fault feature quantity association strategy, represents the single association rule set of the fault ontology obtained by the second method based on the branch screening optimization strategy and the Apriori algorithm, …, A represents the K-item association rule set of the fault ontology obtained by the second method of the fault ontology and the multi-dimensional fault feature quantity association strategy, S07 It indicates that the 1-K association rule set of the fault ontology is obtained, and the non-existent set is an empty set.
[0039] Furthermore, the real-time feature similarity formula is as follows:
[0040]
[0041] Among them, i j is the jth feature in the associated warning feature database, i j ′ is the same as i j similar feature quantities, m is the number of similar associated warning feature quantities. E is the two-dimensional fault feature quantity correlation factor threshold in step S05, n is the number of all fault feature quantities in the fault information space in step S03, n′ is the number of fault feature quantity classes in the fault information space, n q represents the number of fault features in the qth fault feature class in the fault feature space, v q is the cluster center of the fault feature class, c p It is the pth fault feature in this fault feature class.
[0042] The present invention provides a power grid resource association mining early warning system for a new power system, comprising:
[0043] Summarization module: used to summarize the data from multi-source and multi-modal data sources of power grid resources and extract the summarized abnormal data of power grid resources;
[0044] Fault space module: used to build a fault space model based on abnormal data of power grid resources;
[0045] Correlation factor module: used to calculate the correlation factor between the fault entity and the one-dimensional fault feature quantity according to the fault space model;
[0046] Multi-dimensional fault feature module: used to construct a multi-dimensional fault feature space from the fault feature quantities corresponding to the power grid resources in the fault space;
[0047] Association rule module: used to calculate two-dimensional fault feature quantity, two-dimensional fault feature quantity correlation factor threshold, calculate the correlation degree between fault entity and multi-dimensional fault feature quantity and construct association rules according to multi-dimensional fault feature quantity space;
[0048] Fusion module: used to obtain a fused binary correlation warning frequent set based on the correlation factor of the two-dimensional fault feature quantity, the threshold of the two-dimensional fault feature quantity correlation factor, the calculation of the correlation degree between the fault entity and the multi-dimensional fault feature quantity, and the construction of the association rules;
[0049] Comprehensive warning module: used to calculate the similarity of real-time feature quantities based on the fused binary correlation warning frequent set to obtain comprehensive warning results.
[0050] Furthermore, in the summary module, the data sources of the multi-source and multi-modal data of power grid resources include: a new generation of dispatching technical support system, a new generation of centralized control station system, distribution automation system, substation automation system, electricity consumption information collection system, marketing business application system, power grid resource business middle platform, data middle platform and a new generation of equipment asset lean management system.
[0051] Furthermore, in the correlation factor module, the fault entity and the one-dimensional fault feature quantity are calculated according to the fault space model as follows:
[0052] According to the historical data of the power grid resources in the fault space, the transaction set within the time t0-t before the fault occurs is obtained and the fault characteristics are recorded. After removing the data within the time t0, an index between each type of power grid resource and the fault characteristics is established; the influence factor algorithm is used to calculate the influence factor between each power grid resource and all its fault characteristics. The algorithm for the correlation factor between the fault entity and a single fault characteristic is:
[0053]
[0054] Among them, IF(d,ij ) is the jth fault characteristic quantity i j The correlation factor with the power grid resource d, n is the number of failures in the historical data of the power grid resource, m p is the fault characteristic quantity i in the pth fault j The number of times the fault occurs within t0-t before it occurs. t0 is configured by the user according to the actual situation. p is the time when the pth fault occurs, t′ q is the fault characteristic quantity i within the time t0-t before the fault occurs j The time when the event occurs, k is an artificial parameter and can be configured by the user according to the actual situation.
[0055] Furthermore, in the multi-dimensional fault feature module, the multi-dimensional fault feature space is constructed as follows:
[0056] The fault feature quantities corresponding to the power grid resources in the fault space are projected into the fault feature quantity space. The cohesion degree is calculated by the cohesion degree and density of the fault feature quantity class to measure the importance of the fault feature quantity. The cohesion degree algorithm of the multi-dimensional fault feature quantity is:
[0057]
[0058] Among them, RD(C i ) represents the fault characteristic type C i The cohesion, C i represents the i-th fault feature class in the fault feature space, m i represents the i-th fault feature type C in the fault feature space i The number of fault features in the fault feature space, n represents the number of all fault features in the fault feature space, n′ represents the number of fault feature classes in the fault feature space, c j The vector representation of the jth fault feature in the fault feature class, v p is the clustering center of the fault feature class, and k is an artificial parameter that can be configured by the user according to actual conditions.
[0059] Furthermore, in the association rule module, the two-dimensional fault feature correlation factor is calculated as follows: the correlation factor between the existing fault feature and the unrecorded fault feature is calculated according to the index, and the formula is as follows:
[0060]
[0061] Among them, i1, i2 are fault features, U1 is the set of fault entities containing the fault feature i1, d1 is a fault information in U1, U2 is the set of fault entities containing the fault feature i2, d2 is a fault information in U2, U′ is the intersection of U1 and U2, that is, U′=U1∩U2, u, u′ is a fault information in U′, S u is the set of fault feature quantities recorded in fault information u, S u′ is the set of fault feature quantities recorded in the fault information u′, |S u |,|S u′ |represents the number of elements in the two sets, |S u ∩S u′ | represents the set S u With the set S u′ The number of elements in the union, α, is an artificial parameter and can be configured by the user according to actual conditions.
[0062] 17. A power grid resource association mining early warning system for a new power system according to claim 12, characterized in that, in the association rule module, the threshold of the association factor of the two-dimensional fault feature quantity is calculated as follows: the association relationship with a small association factor is eliminated by threshold screening, and the formula is as follows:
[0063]
[0064] Where E represents the threshold of the correlation factor of the two-dimensional fault feature quantity, IF(D,i p ) represents the pth fault characteristic i p The impact factor of the fault set D, RF(item i ,item k ) represents the fault feature item i ,item k The correlation factor between them, n is the number of fault characteristics, Indicates the fault feature item k The smallest correlation factor with all other fault characteristics.
[0065] Furthermore, in the association rule module, the correlation between the fault ontology and the multidimensional fault feature quantity is calculated as follows: the maximum number of connected fault feature quantities is recorded as K, that is, there are at most K fault feature quantities that are strongly correlated with each other; the correlation between the fault ontology and the multidimensional fault feature quantity correlation algorithm is used to calculate the correlation between m strongly correlated fault feature quantities, where 2≤m≤K; the formula is:
[0066]
[0067] Among them, S irepresents the i-th strongly correlated fault feature class, which contains m fault features, s1, s2, …, s m represents the fault characteristic quantity, f i Represents the strongly correlated fault feature class S i The corresponding fault entity, P(f i ) represents the fault entity f i The probability of occurrence, P(s1|f i ) indicates that the fault body f i P(s2|s1) represents the probability that the fault characteristic quantity s1 will occur at the same time when the fault characteristic quantity s1 occurs, and P(s2|s1) represents the probability that the fault characteristic quantity s2 will occur at the same time when the fault characteristic quantity s1 occurs, and so on until P(s m |s m-1 ). β m represents the weight when there are m strongly correlated fault features, n m It indicates the number of m strongly correlated fault feature quantities in the big data statistics of historical data.
[0068] Furthermore, in the association rule module, the association rules are constructed as follows: the association rules of the ontology fault are mined using the branch screening optimization strategy and the Apriori algorithm, the missing association rules are found, and the k-item frequent set containing the fault ontology data is obtained.
[0069] Furthermore, in the association rule module, the association rule mining based on the branch screening optimization strategy is as follows:
[0070] Filter the event set, assuming that the set of frequent items is A1, A2, ...A n , all events contained in the set are a1, a2, ...a m , the number of times these events occur is k1, k2, ...k m , the frequency of each event is determined, the formula is as follows:
[0071]
[0072] Filter out the events that pass the judgment, which are b1, b2, ...b x , through the events screened out, the number of occurrences of these events corresponds to q1, q2, ...q x , find the set that contains half or more of the events as the judgment passing events. The found set is recorded as B1, B2, ...B y ;
[0073] The event set is pruned, and then the event is judged again. The formula is as follows:
[0074]
[0075] Filter out the events that pass the judgment, which are c1, c2, ...c z , B1, B2, ...B y Central Africa c1, c2, ... c z The event removal results in a set of C1, C2, ...C y ; The event sets C1, C2, ...C y Traverse to form an FP tree, starting from each leaf node of the tree, find the candidate set of two frequent sets, the candidate set of three frequent sets, the candidate set of four frequent sets, the candidate set of five frequent sets... until the calculation of the candidate set is completed;
[0076] Confidence judgment of the found frequent item sets: For any n-item frequent set: calculate the support of each event in all frequent sets and sort them by size. Assume that an event in an n-item frequent set is sorted by support and recorded as t1, t2, ..., t n
[0077]
[0078] The frequent sets that pass the judgment are retained, and those that fail are removed.
[0079] Furthermore, in the fusion module, the formula for the fused binary association warning frequent set is as follows:
[0080] A rule =A S02 ∪A S06 ∪A S07
[0081]
[0082] Among them, A rule A represents the association rule set after the fusion of the binary association warning frequent set. S02 Indicates the single association rule set of the fault ontology, A S06 represents the 2-K association rule set of the fault ontology, F represents the fault ontology, represents the single association rule set of the fault ontology obtained by the first method of the fault ontology and the multi-dimensional fault feature quantity association strategy, …, represents the K-item association rule set of the fault ontology obtained by the first method of the fault ontology and the multi-dimensional fault feature quantity association strategy, represents the single association rule set of the fault ontology obtained by the second method based on the branch screening optimization strategy and the Apriori algorithm, …, A represents the K-item association rule set of the fault ontology obtained by the second method of the fault ontology and the multi-dimensional fault feature quantity association strategy, S07 It indicates that the 1-K association rule set of the fault ontology is obtained, and the non-existent set is an empty set.
[0083] Furthermore, in the comprehensive early warning module, the real-time feature quantity similarity formula is as follows:
[0084]
[0085] Among them, i j is the jth feature in the associated warning feature database, i j ′ is the same as i j similar feature quantities, m is the number of similar associated warning feature quantities. E is the two-dimensional fault feature quantity correlation factor threshold in step S05, n is the number of all fault feature quantities in the fault information space in step S03, n′ is the number of fault feature quantity classes in the fault information space, n q represents the number of fault features in the qth fault feature class in the fault feature space, v q is the cluster center of the fault feature class, c p It is the pth fault feature in this fault feature class.
[0086] An electronic device described in the present invention includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and is characterized in that when the computer program is loaded into the processor, it implements any one of the methods for associated mining and early warning of power grid resources for a new power system.
[0087] A storage medium according to the present invention stores a computer program, and when the computer program is executed by a processor, any one of the methods for mining and early warning of power grid resources for a new power system is implemented.
[0088] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: by summarizing the grid resource data, a complete grid resource status evaluation system is established to achieve a reasonable evaluation of the grid resource operation status. BRIEF DESCRIPTION OF THE DRAWINGS
[0089] Figure 1 It is a schematic diagram of the process of the present invention;
[0090] Figure 2 It is a flow chart of the multi-dimensional fault feature correlation strategy of the present invention;
[0091] Figure 3 It is a schematic diagram of association rule mining of the present invention. DETAILED DESCRIPTION
[0092] The technical solution of the present invention is further described below in conjunction with the accompanying drawings.
[0093] like Figure 1-3As shown, an embodiment of the present invention provides a power grid resource association mining early warning method for a new power system, comprising the following steps:
[0094] (1) Summarizing and processing the data from the multi-source and multi-modal data source of power grid resources to obtain the integrated data required for early warning and analysis of power grid resources; the data source of the multi-source and multi-modal data of power grid resources includes: a new generation dispatching technology support system, a new generation centralized control station system, a distribution automation system, a substation automation system, a power consumption information collection system, a marketing business application system, a power grid resource business middle platform, a data middle platform, and a new generation of equipment asset lean management system; wherein the multi-source and multi-modal data of power grid resources includes: (S1) the main primary power equipment of the transmission network, specifically including: Transformer, busbar, generator, transformer winding, circuit breaker, switch, grounding switch, load, shunt reactor, shunt capacitor, series reactor, series compensation device, shunt compensation device, current transformer, voltage transformer, lightning arrester, wave trap, combined filter, filter capacitor, AC filter, arc extinguishing device, grounding resistor, smoothing reactor, DC filter, converter valve, converter transformer, DC voltage transformer, DC current transformer, DC circuit breaker, DC disconnector, DC grounding switch, DC lightning arrester, grounding electrode, DC wave trap, DC busbar. (S2) Main overhead primary power equipment of distribution network, including: pole transformer, pole circuit breaker, pole load switch, pole disconnector, pole recloser, pole drop-out fuse, line arrester, line fault indicator, pole capacitor, pole voltage transformer, pole current transformer, pole sectionalizer. (S3) The main sub-power equipment in the distribution network, including: distribution transformers, transformers, grounding transformers, circuit breaker disconnectors, load switches, fuses, jumper wires, busbars, reactors, current transformers, voltage transformers, power capacitors, lightning arresters, grounding resistors, fault indicators, cables in the station, and AC filters. (S4) The main low-voltage sub-power equipment in the distribution network, including: low-voltage pole switches, low-voltage pole fuses, low-voltage fuses, low-voltage pole capacitors, low-voltage reactive power compensation devices, low-voltage pole arresters, low-voltage meter boxes, low-voltage fuse boxes, wall brackets, low-voltage distribution boxes, low-voltage residual current protectors, low-voltage cable junction boxes, low-voltage cable terminal boxes, low-voltage capacitors, low-voltage switches, low-voltage busbars, low-voltage distribution boards, low-voltage switch cabinets, low-voltage power electronic equipment, low-voltage energy storage devices, and anti-islanding devices. (S5) Main control systems, relay protection and safety automatic devices of power grid, including: converter station control system, fault recorder, protection fault information system substation, relay, converter station protection system, converter station control protection system, AC protection device, DC system protection device, safety automatic device. (S6) Main online detection equipment of power grid, including: tower monitoring, conductor monitoring, meteorological environment monitoring, transmission video / image monitoring, transformer / reactor / converter monitoring, capacitor type equipment monitoring, metal oxide arrester monitoring, circuit breaker / GIS monitoring, substation video / image monitoring, cable body monitoring.(S7) Main automation equipment of the power grid, including: automation system, remote terminal unit (RTU), remote terminal unit (FTU), remote terminal unit (DTU), remote terminal unit (TTU), electric energy remote terminal, phase measurement device, secondary system safety protection equipment, clock synchronization device, computer equipment, storage equipment, power supply equipment, remote acquisition equipment, electric energy acquisition equipment, dispatching simulation screen, large screen. (S8) Main communication equipment of the power grid, including: communication room, communication cabinet, communication optical cable, communication optical cable segment, communication optical path, communication room, communication power supply, transmission system, PCM access system, terminal communication access system, power carrier, data network, switching network, communication network management system. (S9) Other power grid resources, including historical data and power grid resource model.
[0095] S02. Constructing the fault space
[0096] The abnormal data and the marked operation abnormal data are clustered again to construct the fault space and obtain the power grid resource fault space. The distance between the fault vectors of the same type is close.
[0097] S03. Calculate the correlation factor between the fault entity and the one-dimensional fault feature quantity
[0098] According to the historical data of the power grid resources in the fault space, the transaction set within the time t0-t before the fault occurs is taken and the fault characteristics are recorded. The information within the time t0 before the fault occurs will bring a large error. After removing the data within this time, an index between each type of power grid resource and the fault characteristics is established. The influence factor algorithm is used to calculate the influence factor between each power grid resource and all its fault characteristics. The algorithm for the correlation factor between the fault entity and a single fault characteristic is:
[0099]
[0100] Among them, IF(d,i j ) is the jth fault characteristic quantity i j The correlation factor with the power grid resource d, n is the number of failures in the historical data of the power grid resource, m p is the fault characteristic quantity i in the pth fault j The number of times the fault occurs within t0-t before it occurs. t0 is configured by the user according to the actual situation. p is the time when the pth fault occurs, t′ q is the fault characteristic quantity i within the time t0-t before the fault occurs j The time when the event occurs, k is an artificial parameter and can be configured by the user according to the actual situation.
[0101] S04. Constructing a multi-dimensional fault feature space
[0102] The fault feature quantities corresponding to the power grid resources in the fault space are projected into the fault feature quantity space. Multiple fault feature quantities of the same type of fault are close in the fault feature quantity space, and the fault feature quantities of multiple similar faults are close in the fault feature quantity space, which is easy to form clusters. The cohesion degree is calculated by the cohesion degree and density of the fault feature quantity class to measure the importance of the fault feature quantity. The higher the repetition, the more important the fault feature quantity class is. The cohesion degree algorithm of multi-dimensional fault feature quantity is:
[0103]
[0104] Among them, RD(C i ) represents the fault characteristic type C i The cohesion, C i represents the i-th fault feature class in the fault feature space, m i represents the i-th fault feature type C in the fault feature space i The number of fault features in the fault feature space, n represents the number of all fault features in the fault feature space, n′ represents the number of fault feature classes in the fault feature space, c j The vector representation of the jth fault feature in the fault feature class, v p is the clustering center of the fault feature class, and k is an artificial parameter that can be configured by the user according to actual conditions.
[0105] S05. Calculate the correlation factor between two-dimensional fault feature quantities
[0106] A fault entity does not contain all fault feature quantities. When a fault occurs in a power grid resource, a fault feature quantity that is not recorded in its own historical data appears. It is necessary to calculate the impact of the unrecorded fault feature quantity on the fault entity through the correlation degree between the fault feature quantities. Therefore, the correlation factor between the existing fault feature quantity and the unrecorded fault feature quantity is calculated according to the index. The algorithm for the correlation factor of the two-dimensional fault feature quantity is:
[0107]
[0108] Among them, i1, i2 are fault features, U1 is the set of fault entities containing the fault feature i1, d1 is a fault information in U1, U2 is the set of fault entities containing the fault feature i2, d2 is a fault information in U2, U′ is the intersection of U1 and U2, that is, U′=U1∩U2, u, u′ is a fault information in U′, S u is the set of fault feature quantities recorded in fault information u, S u′ is the set of fault feature quantities recorded in the fault information u′, |S u |,|S u′|represents the number of elements in the two sets, |S u ∩S u′ | represents the set S u With the set S u′ The number of elements in the union, α, is an artificial parameter and can be configured by the user according to actual conditions.
[0109] S06. Calculate the threshold of the correlation factor of the two-dimensional fault feature quantity
[0110] When the correlation factor between fault feature quantities is very small, it means that the probability of causing a fault when the unrecorded fault feature quantity is associated with the recorded fault feature quantity is small. Therefore, the correlation relationship with a small correlation factor is eliminated through threshold screening, reducing the correlation relationship between fault feature quantities, reducing data redundancy and reducing complexity. The threshold setting algorithm for the correlation factor of two-dimensional fault feature quantities is:
[0111]
[0112] Where E represents the threshold of the correlation factor of the two-dimensional fault feature quantity, IF(D,i p ) represents the pth fault characteristic i p The impact factor of the fault set D, RF(item i ,item k ) represents the fault feature item i ,item k The correlation factor between them, n is the number of fault characteristics, Indicates the fault feature item k The smallest correlation factor with all other fault characteristics.
[0113] S07. Calculate the correlation between the fault entity and the multi-dimensional fault feature quantity
[0114] When calculating the correlation factor in step S04, a graph of all fault feature quantities connected to each other is generated, in which there are many edges with correlation factors lower than the threshold. After the threshold is removed in S05, the remaining fault feature quantities have a high correlation. The maximum number of fault feature quantities that are still connected to each other at this time is recorded as K, that is, at most K fault feature quantities are strongly correlated with each other at this time. The correlation algorithm between the fault ontology and the multidimensional fault feature quantity is used to calculate the correlation between m (2≤m≤K) strongly correlated fault feature quantities. The algorithm formula is:
[0115]
[0116] Among them, S i represents the i-th strongly correlated fault feature class, which contains m fault features, s1, s2, …, s m represents the fault characteristic quantity, fi Represents the strongly correlated fault feature class S i The corresponding fault entity, P(f i ) represents the fault entity f i The probability of occurrence, P(s1|f i ) indicates that the fault body f i P(s2|s1) represents the probability that the fault characteristic quantity s1 will occur at the same time when the fault characteristic quantity s1 occurs, and P(s2|s1) represents the probability that the fault characteristic quantity s2 will occur at the same time when the fault characteristic quantity s1 occurs, and so on until P(s m |s m-1 ). β m represents the weight when there are m strongly correlated fault features, n m It indicates the number of m strongly correlated fault feature quantities in the big data statistics of historical data.
[0117] S08. Association rule mining
[0118] In order to improve the accuracy and coverage of association rule mining, this paper uses the branch screening optimization strategy and Apriori algorithm to mine association rules for ontology faults, mines out the association rules that may be missed in the above steps, and obtains the k-item frequent set containing fault ontology data.
[0119] Based on the branch screening optimization strategy and Apriori algorithm, the following steps are performed:
[0120] 1. Read each transaction item in the database in turn and generate a support calculation support array corresponding to each transaction item;
[0121] 2. Based on the Apriori algorithm, the branch screening optimization strategy is implemented to generate frequent item sets;
[0122] 3. Based on the frequent item sets and support calculation support array, the confidence calculation of the association rules is performed, and the association rules are obtained under the minimum confidence constraint.
[0123] The adjacency dictionary list constructed in this method has N parent nodes, and the calculation formula of N is:
[0124] N = f × (10 e +1)
[0125] Among them, f is the number of binomial frequent itemsets in the binomial frequent itemset, and e is the number of binomial frequent itemsets that ensure 10 e The smallest power greater than f.
[0126] 4. Determine the location of all data items in the candidate set of the binarized frequent set:
[0127] s(m,n)=(I m +I n×10 e )modN
[0128] Among them, I m Represents the index of the first item in the binomial frequent set, I n Represents the index of the second item in the binomial frequent set.
[0129] Each parent node corresponds to a node {N1, N2, Number, Flag}, where N1 and N2 are two indexes of data items in the binary frequent set, which correspond one-to-one to the Value of the parent node. Number is the number of times the {N1, N2} node appears, and Flag indicates that the initial value is 0. When the value of Number reaches the minimum support, the value of Flag is set to 1.
[0130] When the Flag value is 1, the two data items are marked as "selected" and will no longer participate in the calculation in all subsequent algorithm processes. This achieves dynamic deletion of invalid branches and greatly improves algorithm efficiency.
[0131] The process of the branch screening optimization strategy described in this method is:
[0132] Cartesian product is performed on the data item set to which the first transaction belongs to obtain the candidate set of binarized frequent sets, and an adjacency dictionary linked list is established to correspond the binarized frequent sets to the candidate sets one by one; binarized frequent sets of the 2nd to tth transactions are iteratively generated. During the execution of the algorithm, the support of the corresponding binarized frequent sets is calculated based on the support calculation support array, and useless data items are dynamically removed based on the minimum support to achieve the effect of branch screening; the details are as follows:
[0133] S1 filters the event set
[0134] Assume that the set of frequent items is A1, A2, ...A n .
[0135] All events contained in the set are a1, a2, ...a m , the number of times these events occur is k1, k2, ...k m , determine the frequency of each event
[0136]
[0137] Filter out the events that pass the judgment, which are b1, b2, ...b x , through the events screened out, the number of occurrences of these events corresponds to q1, q2, ...q x , find the set that contains half or more of the events as the judgment passing events. The found set is recorded as B1, B2, ...B y .
[0138] S2 prunes the event set
[0139] Then make another judgment on the event.
[0140]
[0141] Filter out the events that pass the judgment, which are c1, c2, ...c z , B1, B2, ...B y Central Africa c1, c2, ... c z Event removal. The resulting set is C1, C2, ...C y
[0142] The event set C1, C2, ...C y Traverse to form an FP tree, starting from each leaf node of the tree, find the candidate set of two-item frequent sets, the candidate set of three-item frequent sets, the candidate set of four-item frequent sets, the candidate set of five-item frequent sets... until the calculation of the candidate set is completed.
[0143] S3 makes confidence judgment on the found frequent item sets
[0144] Any n-item frequent set:
[0145] Calculate the support of each event in all frequent sets and sort them by size. Suppose there are n frequent events in a set, sorted by support and recorded as t1, t2, ..., t n
[0146]
[0147] The frequent sets that pass the judgment are retained, and those that fail are removed.
[0148] According to the idea of Apriori algorithm, based on the candidate set of two frequent sets, three frequent sets, four frequent sets, five frequent sets, etc. are generated until the calculation of the candidate set is completed.
[0149] S09. Comprehensive analysis of related warning feature quantities
[0150] The association warning adopts two methods, namely the strategy of correlation between fault ontology and multi-dimensional fault feature quantity and the method based on branch screening optimization strategy and Apriori algorithm, to obtain association rules respectively. Among them, the strategy of correlation between fault ontology and multi-dimensional fault feature quantity uses the fault ontology and single fault feature quantity correlation factor algorithm to obtain the single association rule of the fault ontology in step S03; in S07, the fault ontology and multi-dimensional fault feature quantity correlation algorithm is used to obtain the 2-K association rules of the fault ontology. S08 directly adopts the branch screening optimization strategy and Apriori algorithm to obtain the 1-K association rules of the fault ontology. In order to increase the number of rules in the association warning rule base, it is necessary to obtain the union of the rules obtained by the two methods. The binary association warning frequent set fusion algorithm is as follows:
[0151] A rule =A S02 ∪A S06 ∪A S07
[0152]
[0153] Among them, A rule A represents the association rule set after the fusion of the binary association warning frequent set. S02 The strategy for representing the association degree between the fault ontology and the multi-dimensional fault feature quantity uses the fault ontology and the single fault feature quantity association factor algorithm in step S03 to obtain a single association rule set of the fault ontology, A S06 In step S07, the fault ontology and the multi-dimensional fault feature quantity association algorithm are used to obtain a 2-K association rule set of the fault ontology, F represents the fault ontology, represents the single association rule set of the fault ontology obtained by the first method of the fault ontology and the multi-dimensional fault feature quantity association strategy, …, represents the K-item association rule set of the fault ontology obtained by the first method of the fault ontology and the multi-dimensional fault feature quantity association strategy, represents the single association rule set of the fault ontology obtained by the second method based on the branch screening optimization strategy and the Apriori algorithm, …, It represents the K-item association rule set of the fault ontology obtained by the second method of the fault ontology and the multi-dimensional fault feature quantity association strategy, and there is no empty set.
[0154] The user can comprehensively judge whether the associated warning feature quantities are effective based on the judgment results of this method and refer to the feature quantities of experts, historical experience or other channels, eliminate some association rules according to the actual situation, and add the remaining effective associated warning feature quantities into the associated warning feature quantity library as the prerequisite for fault judgment.
[0155] S010. Real-time feature similarity analysis
[0156] During the actual operation of the associated warning feature database, there will be features similar to those in the associated warning feature database. At this time, it is necessary to use the real-time feature similarity judgment algorithm to determine whether the similarity of the feature will trigger an associated warning. When the following inequalities are established at the same time, it is considered that the similar feature is similar to the feature in the associated warning feature database, and an early warning is required. The real-time feature similarity algorithm is as follows:
[0157]
[0158] Among them, i j is the jth feature in the associated warning feature database, i j ′ is the same as i j similar feature quantities, m is the number of similar associated warning feature quantities. E is the two-dimensional fault feature quantity correlation factor threshold in step S05, n is the number of all fault feature quantities in the fault information space in step S03, n′ is the number of fault feature quantity classes in the fault information space, n q represents the number of fault features in the qth fault feature class in the fault feature space, v q is the cluster center of the fault feature class, c p It is the pth fault feature in this fault feature class.
[0159] S011: Comprehensive early warning analysis
[0160] An embodiment of the present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program, when loaded into the processor, implements any one of the methods for associated mining and early warning of power grid resources for a new power system.
[0161] An embodiment of the present invention further provides a storage medium storing a computer program, wherein the computer program, when executed by a processor, implements any one of the aforementioned methods for associated mining and early warning of power grid resources for a new power system.
Claims
1. A power grid resource association mining and early warning method for a new power system, characterized in that: The method comprises the following steps: summarizing data from multi-source and multi-modal data sources of power grid resources, and extracting the summarized abnormal data of power grid resources; constructing a fault space model according to the abnormal data of power grid resources; and calculating the correlation factor between the fault entity and the one-dimensional fault characteristic quantity according to the fault space model; Construct a multi-dimensional fault feature quantity space by using the fault feature quantities corresponding to the power grid resources in the fault space; According to the multidimensional fault feature space, the two-dimensional fault feature quantity and the two-dimensional fault feature quantity correlation factor threshold are calculated, the correlation between the fault entity and the multidimensional fault feature quantity is calculated, and the association rules are constructed; the k-item frequent set containing the fault entity data is obtained by using the branch screening optimization strategy and the Apriori algorithm; according to the two-dimensional fault feature quantity correlation factor, the two-dimensional fault feature quantity correlation factor threshold, the correlation between the fault entity and the multidimensional fault feature quantity is calculated, the association rules are constructed, and the fused binary correlation warning frequent set is obtained based on the branch screening optimization strategy and the Apriori algorithm; the real-time feature quantity similarity is calculated according to the fused binary correlation warning frequent set to obtain the comprehensive warning result.
2. A method for early warning of power grid resource association mining for new power systems according to claim 1, characterized in that: The data sources of the multi-source and multi-modal data of power grid resources include: a new generation of dispatching technology support system, a new generation of centralized control station system, distribution automation system, substation automation system, electricity consumption information collection system, marketing business application system, power grid resource business middle platform, data middle platform and a new generation of equipment asset lean management system.
3. The method for early warning of power grid resource association mining for new power systems according to claim 1 is characterized in that: The fault entity and one-dimensional fault characteristic quantity are calculated according to the fault space model as follows: According to the historical data of the power grid resources in the fault space, the transaction set within the time t0-t before the fault occurs is obtained and the fault characteristics are recorded. After removing the data within the time t0, an index between each type of power grid resource and the fault characteristics is established; the influence factor algorithm is used to calculate the influence factor between each power grid resource and all its fault characteristics. The algorithm for the correlation factor between the fault entity and a single fault characteristic is: Among them, IF(d,i j ) is the jth fault characteristic quantity i j The correlation factor with the power grid resource d, n is the number of failures in the historical data of the power grid resource, m p is the fault characteristic quantity i in the pth fault j The number of times the fault occurs within t0-t before it occurs. t0 is configured by the user according to the actual situation. p is the time when the pth fault occurs, t′ q is the fault characteristic quantity i within the time t0-t before the fault occurs j The time when the event occurs, k is an artificial parameter and can be configured by the user according to the actual situation.
4. The method for early warning of power grid resource association mining for new power systems according to claim 1, characterized in that: The construction of multi-dimensional fault feature space is as follows: The fault feature quantities corresponding to the power grid resources in the fault space are projected into the fault feature quantity space. The cohesion degree is calculated by the cohesion degree and density of the fault feature quantity class to measure the importance of the fault feature quantity. The cohesion degree algorithm of the multi-dimensional fault feature quantity is: Among them, RD(C i ) represents the fault characteristic type C i The cohesion, C i represents the i-th fault feature class in the fault feature space, m i represents the i-th fault feature type C in the fault feature space i The number of fault features in the fault feature space, n represents the number of all fault features in the fault feature space, n′ represents the number of fault feature classes in the fault feature space, c j The vector representation of the jth fault feature in the fault feature class, v p is the clustering center of the fault feature class, and k is an artificial parameter that can be configured by the user according to actual conditions.
5. The method for early warning of power grid resource association mining for new power systems according to claim 1 is characterized in that: The calculation of the correlation factor of the two-dimensional fault feature quantity is as follows: the correlation factor between the existing fault feature quantity and the unrecorded fault feature quantity is calculated according to the index, and the formula is as follows: Among them, i1, i2 are fault features, U1 is the set of fault entities containing the fault feature i1, d1 is a fault information in U1, U2 is the set of fault entities containing the fault feature i2, d2 is a fault information in U2, U′ is the intersection of U1 and U2, that is, U′=U1∩U2, u, u′ is a fault information in U′, S u is the set of fault feature quantities recorded in fault information u, S u′ is the set of fault feature quantities recorded in the fault information u′, |S u |,|S u′ |represents the number of elements in the two sets, |S u ∩S u′ | represents the set S u With the set S u′ The number of elements in the union, α, is an artificial parameter and can be configured by the user according to actual conditions.
6. The method for early warning of power grid resource association mining for new power systems according to claim 1, characterized in that: The calculation of the threshold of the correlation factor of the two-dimensional fault feature quantity is as follows: the correlation relationship with a small correlation factor is eliminated through threshold screening, and the formula is as follows: Where E represents the threshold of the correlation factor of the two-dimensional fault feature quantity, IF(D,i p ) represents the pth fault characteristic i p The impact factor of the fault set D, RF(item i ,item k ) represents the fault feature item i ,item k The correlation factor between them, n is the number of fault characteristics, Indicates the fault feature item k The smallest correlation factor with all other fault characteristics.
7. The method for early warning of power grid resource association mining for new power systems according to claim 1, characterized in that: The calculation of the correlation between the fault ontology and the multidimensional fault feature quantity is as follows: the maximum number of connected fault feature quantities is recorded as K, that is, there are at most K fault feature quantities that are strongly correlated with each other; the correlation between the fault ontology and the multidimensional fault feature quantity correlation algorithm is used to calculate the correlation between m strongly correlated fault feature quantities, where 2≤m≤K; the formula is: Among them, S i represents the i-th strongly correlated fault feature quantity class, which contains m fault feature quantities, s1, s2, …, s m represents the fault characteristic quantity, f i Represents the strongly correlated fault feature class S i The corresponding fault entity, P(f i ) represents the fault entity f i The probability of occurrence, P(s1|f i ) indicates that the fault body f i P(s2|s1) represents the probability that the fault characteristic quantity s1 will occur at the same time when the fault characteristic quantity s1 occurs, and P(s2|s1) represents the probability that the fault characteristic quantity s2 will occur at the same time when the fault characteristic quantity s1 occurs, and so on until P(s m |s m-1 ). m represents the weight when there are m strongly correlated fault features, n m It indicates that there are m strongly correlated fault feature quantities in the big data statistics of historical data.
8. The method for early warning of power grid resource association mining for new power systems according to claim 1 is characterized in that: The specific construction of association rules is as follows: Use the branch screening optimization strategy and Apriori algorithm to mine association rules for ontology faults, find out the missing association rules, and obtain the k-item frequent set containing fault ontology data.
9. A method for early warning of power grid resource association mining for new power systems according to claim 8, characterized in that: The specific details of association rule mining based on branch screening optimization strategy are as follows: Filter the event set, assuming that the set of frequent items is A1, A2, ...A n , all events contained in the set are a1, a2, ...a m , the number of times these events occur is k1, k2, ...k m , the frequency of each event is determined, the formula is as follows: Filter out the events that pass the judgment, which are b1, b2, ...b x , through the events screened out, the number of occurrences of these events corresponds to q1, q2, ...q x , find the set that contains half or more of the events as the judgment passing events. The found set is recorded as B1, B2, ...B y ; The event set is pruned, and then the event is judged again. The formula is as follows: Filter out the events that pass the judgment, which are c1, c2, ...c z , B1, B2, ...B y Central Africa c1, c2, ... c z The event removal results in a set of C1, C2, ...C y ; The event sets C1, C2, ...C y Traverse to form an FP tree, starting from each leaf node of the tree, find the candidate set of two frequent sets, the candidate set of three frequent sets, the candidate set of four frequent sets, the candidate set of five frequent sets, etc. until the calculation of the candidate set is completed; Confidence judgment of the found frequent item sets: For any n-item frequent set: calculate the support of each event in all frequent sets and sort them by size. Assume that an event in an n-item frequent set is sorted by support and recorded as t1, t2, ..., t n The frequent sets that pass the judgment are retained, and those that fail are removed.
10. The method for early warning of power grid resource association mining for new power systems according to claim 1, characterized in that: The formula for the fused binary association warning frequent set is as follows: A rule =A S02 ∪A S06 ∪A S07 Among them, A rule A represents the association rule set after the fusion of the binary association warning frequent set. S02 Indicates the single association rule set of the fault ontology, A S06 represents the 2-K association rule set of the fault ontology, F represents the fault ontology, represents the single association rule set of the fault ontology obtained by the first method of the fault ontology and the multi-dimensional fault feature quantity association strategy, …, represents the K-item association rule set of the fault ontology obtained by the first method of the fault ontology and the multi-dimensional fault feature quantity association strategy, represents the single association rule set of the fault ontology obtained by the second method based on the branch screening optimization strategy and the Apriori algorithm, …, A represents the K-item association rule set of the fault ontology obtained by the second method of the fault ontology and the multi-dimensional fault feature quantity association strategy, S07 It indicates that the 1-K association rule set of the fault ontology is obtained, and the non-existent set is an empty set.
11. The method for early warning of power grid resource association mining for new power systems according to claim 1, characterized in that: The real-time feature similarity formula is as follows: Among them, i j is the jth feature in the associated warning feature database, i j ′ is the same as i j similar feature quantities, m is the number of similar associated warning feature quantities. E is the two-dimensional fault feature quantity correlation factor threshold in step S05, n is the number of all fault feature quantities in the fault information space in step S03, n′ is the number of fault feature quantity classes in the fault information space, n q represents the number of fault features in the qth fault feature class in the fault feature space, v q is the cluster center of the fault feature class, c p It is the pth fault feature in this fault feature class.
12. A power grid resource association mining early warning system for new power systems, characterized in that: include: Summarization module: used to summarize the data from multi-source and multi-modal data sources of power grid resources and extract the summarized abnormal data of power grid resources; Fault space module: used to build a fault space model based on abnormal data of power grid resources; Correlation factor module: used to calculate the correlation factor between the fault entity and the one-dimensional fault feature quantity according to the fault space model; Multi-dimensional fault feature module: used to construct a multi-dimensional fault feature space from the fault feature quantities corresponding to the power grid resources in the fault space; Association rule module: used to calculate two-dimensional fault feature quantity, two-dimensional fault feature quantity correlation factor threshold, calculate the correlation degree between fault entity and multi-dimensional fault feature quantity and construct association rules according to multi-dimensional fault feature quantity space; Fusion module: used to obtain a fused binary correlation warning frequent set based on the correlation factor of the two-dimensional fault feature quantity, the threshold of the two-dimensional fault feature quantity correlation factor, the calculation of the correlation degree between the fault entity and the multi-dimensional fault feature quantity, and the construction of the association rules; Comprehensive warning module: used to calculate the similarity of real-time feature quantities based on the fused binary correlation warning frequent set to obtain comprehensive warning results.
13. A power grid resource association mining early warning system for a new power system according to claim 12, characterized in that: In the summary module, the data sources of the multi-source and multi-modal data of power grid resources include: a new generation of dispatching technology support system, a new generation of centralized control station system, distribution automation system, substation automation system, electricity consumption information collection system, marketing business application system, power grid resource business middle platform, data middle platform and a new generation of equipment asset lean management system.
14. The grid resource association mining early warning system for new power systems according to claim 12, characterized in that: In the correlation factor module, the fault entity and one-dimensional fault feature quantity are calculated according to the fault space model as follows: According to the historical data of the power grid resources in the fault space, the transaction set within the time t0-t before the fault occurs is obtained and the fault characteristics are recorded. After removing the data within the time t0, an index between each type of power grid resource and the fault characteristics is established; the influence factor algorithm is used to calculate the influence factor between each power grid resource and all its fault characteristics. The algorithm for the correlation factor between the fault entity and a single fault characteristic is: Among them, IF(d,i j ) is the jth fault characteristic quantity i j The correlation factor with the power grid resource d, n is the number of failures in the historical data of the power grid resource, m p is the fault characteristic quantity i in the pth fault j The number of times the fault occurs within t0-t before it occurs. t0 is configured by the user according to the actual situation. p is the time when the pth fault occurs, t′ q is the fault characteristic quantity i within the time t0-t before the fault occurs j The time when the event occurs, k is an artificial parameter and can be configured by the user according to the actual situation.
15. The grid resource association mining early warning system for new power systems according to claim 12, characterized in that: In the multi-dimensional fault feature module, the multi-dimensional fault feature space is constructed as follows: The fault feature quantities corresponding to the power grid resources in the fault space are projected into the fault feature quantity space. The cohesion degree is calculated by the cohesion degree and density of the fault feature quantity class to measure the importance of the fault feature quantity. The cohesion degree algorithm of the multi-dimensional fault feature quantity is: Among them, RD(C i ) represents the fault characteristic type C i The cohesion, C i represents the i-th fault feature class in the fault feature space, m i represents the i-th fault feature type C in the fault feature space i The number of fault features in the fault feature space, n represents the number of all fault features in the fault feature space, n′ represents the number of fault feature classes in the fault feature space, c j The vector representation of the jth fault feature in the fault feature class, v p is the clustering center of the fault feature class, and k is an artificial parameter that can be configured by the user according to actual conditions.
16. The grid resource association mining early warning system for new power systems according to claim 12, characterized in that: In the association rule module, the calculation of the two-dimensional fault feature correlation factor is as follows: the correlation factor between the existing fault feature and the unrecorded fault feature is calculated according to the index, and the formula is as follows: Among them, i1, i2 are fault features, U1 is the set of fault entities containing the fault feature i1, d1 is a fault information in U1, U2 is the set of fault entities containing the fault feature i2, d2 is a fault information in U2, U′ is the intersection of U1 and U2, that is, U′=U1∩U2, u, u′ is a fault information in U′, S u is the set of fault feature quantities recorded in fault information u, S u′ is the set of fault feature quantities recorded in the fault information u′, |S u |,|S u′ |represents the number of elements in the two sets, |S u ∩S u′ | represents the set S u With the set S u′ The number of elements in the union, α, is an artificial parameter and can be configured by the user according to actual conditions.
17. The grid resource association mining early warning system for new power systems according to claim 12, characterized in that: In the association rule module, the calculation of the two-dimensional fault feature value correlation factor threshold is as follows: the correlation relationship with a small correlation factor is eliminated through threshold screening, and the formula is as follows: Where E represents the threshold of the correlation factor of the two-dimensional fault feature quantity, IF(D,i p ) represents the pth fault characteristic i p The impact factor of the fault set D, RF(item i ,item k ) represents the fault feature item i ,item k The correlation factor between them, n is the number of fault characteristics, Indicates the fault feature item k The smallest correlation factor with all other fault characteristics.
18. The grid resource association mining early warning system for new power systems according to claim 12, characterized in that: In the association rule module, the correlation between the fault ontology and the multidimensional fault feature quantity is calculated as follows: the maximum number of connected fault feature quantities is recorded as K, that is, there are at most K fault feature quantities that are strongly correlated with each other; the correlation between the fault ontology and the multidimensional fault feature quantity correlation algorithm is used to calculate the correlation between m strongly correlated fault feature quantities, where 2≤m≤K; the formula is: Among them, S i represents the i-th strongly correlated fault feature quantity class, which contains m fault feature quantities, s1, s2, …, s m represents the fault characteristic quantity, f i Represents the strongly correlated fault feature class S i The corresponding fault entity, P(f i ) represents the fault entity f i The probability of occurrence, P(s1|f i ) indicates that the fault body f i P(s2|s1) represents the probability that the fault characteristic quantity s1 will occur at the same time when the fault characteristic quantity s1 occurs, and P(s2|s1) represents the probability that the fault characteristic quantity s2 will occur at the same time when the fault characteristic quantity s1 occurs, and so on until P(s m |s m-1 ). m represents the weight when there are m strongly correlated fault features, n m It indicates that there are m strongly correlated fault feature quantities in the big data statistics of historical data.
19. The grid resource association mining early warning system for new power systems according to claim 12, characterized in that: In the association rule module, the association rules are constructed as follows: the association rules of the ontology fault are mined using the branch screening optimization strategy and the Apriori algorithm to find the missing association rules and obtain the k-item frequent set containing the fault ontology data.
20. The grid resource association mining early warning system for new power systems according to claim 19, characterized in that: In the association rule module, the association rule mining based on the branch screening optimization strategy is as follows: Filter the event set, assuming that the set of frequent items is A1, A2, ...A n , all events contained in the set are a1, a2, ...a m , the number of times these events occur is k1, k2, ...k m , the frequency of each event is determined, the formula is as follows: Filter out the events that pass the judgment, which are b1, b2, ...b x , through the events screened out, the number of occurrences of these events corresponds to q1, q2, ...q x , find the set that contains half or more of the events as the judgment passing events. The found set is recorded as B1, B2, ...B y ; The event set is pruned, and then the event is judged again. The formula is as follows: Filter out the events that pass the judgment, which are c1, c2, ...c z , B1, B2, ...B y Central Africa c1, c2, ... c z The event removal results in a set of C1, C2, ...C y ; The event sets C1, C2, ...C y Traverse to form an FP tree, starting from each leaf node of the tree, find the candidate set of two frequent sets, the candidate set of three frequent sets, the candidate set of four frequent sets, the candidate set of five frequent sets, etc. until the calculation of the candidate set is completed; Confidence judgment of the found frequent item sets: For any n-item frequent set: calculate the support of each event in all frequent sets and sort them by size. Assume that an event in an n-item frequent set is sorted by support and recorded as t1, t2, ..., t n The frequent sets that pass the judgment are retained, and those that fail are removed.
21. A method for early warning of power grid resource association mining for new power systems according to claim 12, characterized in that: In the fusion module, the formula for the fused binary association warning frequent set is as follows: A rule =A S02 ∪A S06 ∪A S07 Among them, A rule A represents the association rule set after the fusion of the binary association warning frequent set. S02 Indicates the single association rule set of the fault ontology, A S06 represents the 2-K association rule set of the fault ontology, F represents the fault ontology, represents the single association rule set of the fault ontology obtained by the first method of the fault ontology and the multi-dimensional fault feature quantity association strategy, …, represents the K-item association rule set of the fault ontology obtained by the first method of the fault ontology and the multi-dimensional fault feature quantity association strategy, represents the single association rule set of the fault ontology obtained by the second method based on the branch screening optimization strategy and the Apriori algorithm, …, A represents the K-item association rule set of the fault ontology obtained by the second method of the fault ontology and the multi-dimensional fault feature quantity association strategy, S07 It indicates that the 1-K association rule set of the fault ontology is obtained, and the non-existent set is an empty set.
22. The grid resource association mining early warning system for new power systems according to claim 12, characterized in that: In the comprehensive early warning module, the real-time feature quantity similarity formula is as follows: Among them, i j is the jth feature in the associated warning feature database, i′ j is with i j similar feature quantities, m is the number of similar associated warning feature quantities. E is the two-dimensional fault feature quantity correlation factor threshold in step S05, n is the number of all fault feature quantities in the fault information space in step S03, n′ is the number of fault feature quantity classes in the fault information space, n q represents the number of fault features in the qth fault feature class in the fault feature space, v q is the cluster center of the fault feature class, c p It is the pth fault feature in this fault feature class.
23. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the computer program is loaded into the processor, it implements a power grid resource association mining and early warning method for a new power system according to any one of claims 1 to 12.
24. A storage medium storing a computer program, characterized in that: When the computer program is executed by the processor, it implements a power grid resource association mining and early warning method for a new power system according to any one of claims 1 to 12.
Citation Information
Patent Citations
Quantitative configuration method of on-line monitoring device of converter station power equipment
CN111612296A
Power distribution network load prediction and line loss cause analysis method based on big data
CN113487447A
Meteorological disaster prediction-based power transmission line operation and maintenance investment optimization method and system
CN114021844A
Energy power safety dynamic early warning method and device
CN115471080A
Cited By
Power transformation equipment operation state evaluation system and method based on fault recording analysis
CN119199365A
Optical cable fault positioning method and system based on accurate image recognition
CN119963955A
