Bidirectional research and judgment method for power distribution network fault in combination with incidence matrix and dynamic Bayesian

Abstract

The invention discloses a bidirectional research and judgment method for power distribution network faults in combination with an incidence matrix and dynamic Bayesian, and the method comprises the following steps: constructing the incidence matrix through an actual topological structure of a power distribution network, and constructing a fault research and judgment matrix in combination with a fault information matrix; fault criteria under the conditions of node fault information missing report and fault information false report are established respectively, so that fault feeder identification from top to bottom is realized; constructing a dynamic Bayesian network according to an actual topological structure of the power distribution network and a'station-line-transformer-household 'relationship, and reasoning the fault probability of each region; constructing a membership function and a fuzzy inference rule to obtain a fuzzy inference system for adjusting the fault probability; and finally inferring the most probable fault area, and realizing fault area research and judgment from bottom to top. According to the method, on the basis of a traditional matrix algorithm, the fault criterion is improved by constructing the causal incidence matrix, the feeder line identification problem under the conditions of fault information missing report and fault information false report is effectively solved, the calculation amount is reduced, and the fault positioning accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the field of distribution network fault detection, and in particular to a distribution network fault bidirectional analysis method combining association matrix and dynamic Bayesian. Background Art

[0002] As the key link between users and the power system, the safe operation of the distribution network is crucial to the reliability of users' electricity use. The topology of the low-voltage distribution network is complex and changeable, and it is difficult to obtain its accurate topology. Even if a fault signal is received, it is difficult to accurately locate the power outage area. In addition, the amount of data generated by the distribution secondary terminal is huge, and due to its complex operating environment, information is lost during the upload process, which further increases the difficulty of fault area analysis.

[0003] There are already a large number of literatures on fault area analysis at home and abroad. The current feeder fault identification mostly uses matrix calculation. Although it simplifies the calculation process to a certain extent, the overall algorithm still has a certain degree of complexity, especially when dealing with large-scale distribution networks. It is still relatively complex and the calculation cost is still high. Summary of the invention

[0004] In order to solve the above technical problems, the present invention provides a distribution network fault bidirectional analysis method combining correlation matrix and dynamic Bayesian which has a simple algorithm and is less affected by fault information omission and false alarm.

[0005] The technical solution of the present invention to solve the above technical problems is: a two-way analysis method for distribution network faults combining correlation matrix and dynamic Bayesian, comprising the following steps:

[0006] Step 1: Construct a correlation matrix based on the actual topological structure of the distribution network, and construct a fault analysis matrix based on the fault information matrix;

[0007] Step 2: Establish fault judgment criteria for missed and false alarms of node fault information, so as to realize top-down fault feeder identification;

[0008] Step 3: Build a dynamic Bayesian network based on the actual topological structure of the distribution network and the “station-line-transformer-user” relationship to infer the fault probability of each area;

[0009] Step 4: Construct membership function and fuzzy reasoning rules to obtain a fuzzy reasoning system to adjust the fault probability;

[0010] Step 5: Finally, the most likely fault area is inferred to achieve bottom-up fault area analysis.

[0011] The above-mentioned two-way analysis method for distribution network faults combining correlation matrix and dynamic Bayesian, the specific steps of step 1 are:

[0012] (1-1) When a fault occurs on a feeder in the distribution network, the fault indicator connected to the head end of the fault feeder will monitor whether a fault current exceeding the set value flows through the fault indicator, and upload the monitored signal to the cloud platform. The fault indicator installed at the end of the fault feeder does not send a signal. The cloud platform generates a row vector matrix with the same number of columns as D′ as the fault information matrix G based on the uploaded information. D′ is the causal association matrix between the feeder section and the switch node;

[0013] (1-2) The inverse matrices of G and D′ are multiplied to obtain the fault analysis matrix Q as follows:

[0014]

[0015] In the formula, when the element in the pth row of Q is 1, it means that the pth feeder has a fault, and when the element in the pth row is 0, it means that there is no fault in the pth feeder; when the element in the qth row of G is 1, it means that a fault current exceeding the set value flows through the fault indicator on the qth node, and when the element in the qth row is 0, it means that no fault current exceeding the set value flows through the fault indicator on the qth node; the element d in the pth row and qth column of D′ is pq When it is 1, it means that the fault of the pth feeder affects the qth fault indicator, d pq When it is 0, it means that the pth feeder is not associated with the qth fault indicator.

[0016] In the above-mentioned two-way fault analysis method for distribution network combining association matrix and dynamic Bayesian, in step 2, when node fault information is missed, it is manifested that the fault information uploaded by the feeder terminal unit FTU at the node is empty. When processing missed information, the following two analysis rules are constructed:

[0017] Rule 1: When the fault information of the node's front and rear adjacent nodes are the same, the missed information is set to the same fault information as the front and rear adjacent nodes;

[0018] Rule 2: When the fault information of the adjacent nodes before and after a node is different, the node is removed as an invalid node and is not considered in D and G, and a new assessment is performed.

[0019] In the above-mentioned two-way fault analysis method for distribution network combining association matrix and dynamic Bayes, in step 2, when the fault information is misreported, it is manifested as the fault information uploaded by the FTU at the node is flipped, that is, the fault information is actually 1 but reported as 0, or actually 0 but reported as 1, and the following two analysis rules are constructed:

[0020] Rule 3: After uploading the false alarm fault information, the judgment result shows that multiple section faults occur on the same line. At this time, the error correction is performed based on the node adjacent information. If the fault information of a node is different from the fault information uploaded by the adjacent nodes, the fault information of the node is falsely alarmed. The false alarm information is modified to be the same as the adjacent nodes. If the value of an element in the judgment matrix is ​​-1, the value is set to 0.

[0021] Rule 4: After uploading the false alarm fault information, the analysis result is wrong, and the fault information needs to be collected again.

[0022] The above-mentioned two-way analysis method for distribution network faults combining correlation matrix and dynamic Bayesian, the specific steps of step three are:

[0023] (3-1) Integrate the fault data on each side of the distribution network (station-line-transformer-user), perform data cleaning and integration, build a complete data set containing fault information, and divide it into a training set and a test set;

[0024] (3-2) Taking the power outage events of users in the substation as the analysis basis, combined with the dependency and time series characteristics between fault data, a dynamic Bayesian network DBN is constructed; the DBN includes six key nodes: the power outage ratio of users in the substation, preliminary judgment of event type, complaint telephone number of users in the substation, event information of low-voltage branch line terminal, substation terminal information and final judgment event;

[0025] (3-3) The node parameters of DBN are learned through the maximum likelihood estimation method, and the probability of failure in different areas in the test set is calculated.

[0026] The above-mentioned two-way fault analysis method for distribution network combining correlation matrix and dynamic Bayesian is used for the i-th node x in the distribution Bayesian network. i , x i There are m power distribution events in the network. Assume that all events related to node x have been obtained. i The power distribution event E′ of the relevant node = {e 1 ,e 2 ,···,e n}, then x i The kth power distribution event e k The conditional probability of occurrence is as follows:

[0027]

[0028] T represents the number of time slices, represents the power distribution event e in the tth time slice k , Par represents the probability of power distribution event, P B is the transition probability;

[0029] When a fault occurs in the distribution network and causes a power outage, a fault event chain is represented by F(r)={t i│ x i,1 ,xi , 2,…x i,j …x i,n}, t i Represents x i Fault record, x i,n Represents x i The corresponding nth power distribution event, the meaning of this formula is that the fault event F(r) occurring on the line is due to many power distribution events x i,n The occurrence or non-occurrence of the fault event F(r) on the line at this time is inferred by the Bayesian formula. r ), calculated using the chain rule as shown below:

[0030] P(F r )=P(x 1 )P(x 2 ∣x 1 )...P(x i ∣x 1 x 2 ...x i-1 )

[0031] P(x 1 ) represents the node x 1 The probability of occurrence, P(x 2 ∣x 1 ) represents x 1 In case of x 2 Probability of occurrence;

[0032] For a network with T time slices, the power distribution events in the event chain include power outage events A for users in the substation area, preliminary judgment of power outage types B, complaint calls from users in the substation area C, low-voltage branch line terminal events D, substation terminal events E, and fault events F; where the fault events {F = l, l = 1, 2, 3, 4} represent user events, meter box events, low-voltage branch line events, and substation events, respectively. Within the time slice T, the posterior probability of line faults caused by these power distribution events is further expressed by the following formula:

[0033]

[0034] Where, P(F=l) T represents the probability of a fault event l occurring in the line in the Tth time slice, (·) 1,T Represents the observed values ​​of the set of elements contained in a certain characteristic factor on time slices 1 to T.

[0035] The above-mentioned two-way analysis method for distribution network faults combining correlation matrix and dynamic Bayesian, the specific steps of step 4 are:

[0036] (4-1) Based on the complete data set constructed in step (3-1), the interval of the input membership function is determined by using a fuzzy clustering algorithm, and fuzzy inference rules are formulated; according to the probability of occurrence of different fault events, the output membership function is divided into four levels: low, medium, high, and very high, and fuzzy weights are assigned to each level;

[0037] (4-2) constructing an output membership function to characterize the degree of fault according to the probability of power outage in each area inferred in step (3-3);

[0038] (4-3) Determine whether the power outage probability of each area obtained in step (3-3) needs to be corrected. If it needs to be corrected, the power outage probability is used as input to the fuzzy inference system, and a defuzzified value is obtained through a defuzzification process. The defuzzified value is used to correct the original maximum probability state to achieve adjustment and optimization of the power outage probability;

[0039] (4-4) The results obtained from the test set are compared with the actual recorded data to evaluate the performance of the model in power outage area assessment under conditions of complete information and incomplete information.

[0040] The beneficial effects of the present invention are:

[0041] 1. Based on the traditional matrix algorithm, the present invention combines the causal relationship between the faulty feeder and the node, constructs a causal correlation matrix, improves the fault judgment criteria, and effectively solves the feeder identification problem in the case of missed and false alarms of fault information. This method not only reduces the amount of calculation, but also improves the accuracy of fault location, and can adapt to the complex distribution network operation environment.

[0042] 2. The present invention constructs a power outage area assessment model based on a dynamic fuzzy Bayesian network, incorporates the causal relationship between user power outage events and various areas of the distribution network into the analysis, integrates the distribution data on the user side and the line side, effectively solves the problem of power outage area assessment under conditions of incomplete information, and reduces dependence on data integrity.

[0043] 3. The present invention combines fuzzy theory with actual fault data to design membership functions and inference rules. It evaluates whether the probability obtained by reasoning needs to be corrected, takes the probability to be corrected as input for fuzzy processing, and then obtains the defuzzified value through defuzzification, thereby finally realizing the correction of the probability of the power outage area and improving the accuracy of the power outage area analysis.

[0044] 4. The method proposed in the present invention can also accurately judge the power outage area under the condition of incomplete fault information, and is superior to traditional methods.

[0045] 5. The present invention can not only identify the faulty feeder, but also conduct an in-depth analysis of the cause of the fault. Through the two-way analysis strategy, a more comprehensive fault analysis result is provided, which greatly improves the inspection and maintenance efficiency of the fault area by the operation and maintenance personnel, shortens the fault handling time, and ensures the stable operation of the distribution network. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 It is a flow chart of the present invention.

[0047] Figure 2 This is a schematic diagram of a simple distribution network topology.

[0048] Figure 3 This is a flow chart for analysis and judgment in the case of underreporting of fault information.

[0049] Figure 4 This is a flow chart for analyzing and judging the situation when fault information is falsely reported.

[0050] Figure 5 Schematic diagram of dynamic fuzzy Bayesian principle.

[0051] Figure 6 Schematic diagram of dynamic fuzzy Bayesian structure.

[0052] Figure 7 This is a typical structural diagram of medium and low voltage distribution networks.

[0053] Figure 8 Schematic diagram of the dynamic fuzzy Bayesian network structure.

[0054] Fig. 9 This is a schematic diagram of the actual dynamic fuzzy Bayesian network structure.

[0055] Fig.10 Schematic diagram of hierarchical fuzzy inference system.

[0056] Fig.11 It is a three-dimensional curve graph of fuzzy rules.

[0057] Fig.12 Schematic diagram of single power supply multi-T connection open-loop distribution network structure

[0058] Fig.13 This is a performance comparison chart between the method of the present invention and the traditional method.

[0059] Fig.14 It is a performance analysis diagram of the method of the present invention under the condition of incomplete information. DETAILED DESCRIPTION

[0060] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0061] like Figure 1 As shown, a bidirectional analysis method for distribution network faults combining correlation matrix and dynamic Bayesian includes the following steps:

[0062] Step 1: Build a correlation matrix through the actual topological structure of the distribution network, and construct a fault analysis matrix by combining the fault information matrix.

[0063] Step 1 needs to construct the correlation matrix according to the actual topological structure of the distribution network. Figure 2 For example, the specific steps of step 1 are:

[0064] (1-1) When a fault occurs on a feeder in the distribution network, the fault indicator connected to the head end of the fault feeder will monitor whether a fault current exceeding the set value flows through the fault indicator, and upload the monitored signal to the cloud platform. The fault indicator installed at the end of the fault feeder does not send a signal. The cloud platform generates a row vector matrix with the same number of columns as D′ as the fault information matrix G based on the uploaded information. D′ is the causal association matrix between the feeder section and the switch node;

[0065] (1-2) The inverse matrices of G and D′ are multiplied to obtain the fault analysis matrix Q as follows:

[0066]

[0067] In the formula, when the element in the pth row of Q is 1, it means that the pth feeder has a fault, and when the element in the pth row is 0, it means that there is no fault in the pth feeder; when the element in the qth row of G is 1, it means that a fault current exceeding the set value flows through the fault indicator on the qth node, and when the element in the qth row is 0, it means that no fault current exceeding the set value flows through the fault indicator on the qth node; the element d in the pth row and qth column of D′ is pq When it is 1, it means that the fault of the pth feeder affects the qth fault indicator, d pq When it is 0, it means that the pth feeder is not associated with the qth fault indicator.

[0068] Step 2: Establish fault judgment criteria for missed and false alarms of node fault information, so as to realize top-down fault feeder identification.

[0069] Figure 3 and Figure 4 The following are the analysis flow charts for the cases of missed and false alarms of fault information. When the node fault information is missed, the fault information uploaded by the feeder terminal unit FTU at the node is empty. Figure 2 Taking the FTU reporting information G = [1*1 0 0 0] as an example, the following two analysis rules are constructed when processing missed reporting information. The process is as follows Figure 3 shown.

[0070] Rule 1: When the fault information of the node's front and rear adjacent nodes are the same, the missed information is set to the same fault information as the front and rear adjacent nodes; Figure 2 Take the case where the FTU reports missing information at node 2: G = [1*1 0 0 0] as an example. The missing fault information in G can be set to 1, and the judgment result is that the feeder section (3) is faulty, which is correct.

[0071] Rule 2: When the fault information of the adjacent nodes before and after a node is different, the node is removed as an invalid node and is not considered in D and G, and the judgment is re-performed. Figure 2 Take the case where the FTU reports missing information at node 3: G = [1 1*0 0 0]. After removing node 3 from the network, reconstruct G and D to obtain Q = [0 1 0 0 0]. At this time, the judgment result is that the feeder section (2) is faulty, and the judgment result is correct.

[0072] When the fault information is falsely reported, it is manifested as the fault information uploaded by the FTU at the node is reversed, that is, the fault information is actually 1 but reported as 0, or actually 0 but reported as 1. The following two judgment rules are constructed, and the process is as follows: Figure 4 As shown:

[0073] Rule 3: After uploading the false alarm fault information, the judgment result shows that multiple section faults occur on the same line. At this time, the error correction is performed based on the node adjacent information. If the fault information of a node is different from the fault information uploaded by the adjacent nodes, the fault information of the node is falsely reported. The false alarm information is modified to the same information as the adjacent nodes. If the value of an element in the judgment matrix is ​​-1, the value is set to 0. When the fault information is falsely reported, the value of the element in the judgment matrix may be -1. At this time, the meaning of -1 is the same as 0, so -1 can be set to 0. Figure 2 The FTU reports false positive information at node 5: For example, we get Q = [0 1 1 0 1 0], and the judgment result is that sections (1), (3), and (5) have faults. Since the probability of simultaneous faults in multiple sections of a line is low, the fault information matrix needs to be adjusted. Since the fault information of the two nodes adjacent to node 5 is 0, and because the judgment result is a multi-section fault, the fault information of node 5 is modified to 0.

[0074] Rule 4: After uploading the false alarm fault information, the judgment result is wrong, and the fault information needs to be collected again. Figure 2 The FTU reports false positive information at node 3: For example, we get Q = [0 1 0 0 00], and the result is that section (2) is faulty, and the FTU on the node needs to re-report the fault information.

[0075] Step 3: Build a dynamic Bayesian network based on the actual topological structure of the distribution network and the “station-line-transformer-user” relationship to infer the fault probability of each area.

[0076] Traditional Bayesian networks mainly rely on clear probability distributions to describe the causal relationship and conditional dependency between events. However, in actual distribution networks, due to the complex operating environment, uncertain equipment status, and the influence of external factors, the probability of an event often shows significant ambiguity and uncertainty. In order to solve this problem, the distribution fuzzy Bayesian network theory introduces fuzzy sets and membership functions to defuzzify uncertain information, thereby combining fuzzy theory with Bayesian networks, so that the network can enhance the reasoning ability of the probability of event occurrence under fuzzy and uncertain conditions. Its basic principles are as follows: Figure 5 shown.

[0077] Fuzzy theory can quantify fuzzy information, so that the system can still draw reasonable and effective inference results when facing fuzzy input. In addition, fuzzy theory can also convert the probability information calculated in the Bayesian network into membership, and combine the fuzzy reasoning system and the corresponding reasoning rules to analyze and judge the fault state. Figure 6 shown.

[0078] The typical medium and low voltage distribution network structure on a feeder is as follows: Figure 7 shown. Figure 7 It includes medium-voltage lines, medium-voltage branch line switches, low-voltage line side of the power supply in the substation, low-voltage user side, meter boxes, and distribution terminals such as DTU, FTU, and smart meters on the line. When a power outage occurs on the low-voltage user side under the substation due to a fault somewhere in the distribution network, the smart meter at the user's location will automatically collect the power outage information of the user. As a terminal device, the smart meter can not only monitor the user's power consumption status in real time, but also transmit data to the concentrator or data acquisition terminal (DTU). At the same time, other distribution terminal devices such as feeder terminal units (FTUs) and remote terminal units (RTUs) will also generate or collect event data related to this fault. However, due to the complexity of the distribution network operating environment, there may be problems such as underreporting and false alarms in the data collection process, making it difficult to accurately locate the fault point with a single data source. Therefore, it is necessary to integrate multiple data sources and conduct in-depth analysis combined with a large amount of historical data to improve the accuracy and reliability of fault diagnosis.

[0079] On this basis, construct Figure 8 The dynamic Bayesian network shown in FIG. 4 can realize probabilistic reasoning between events by combining various event data in the distribution network, and can infer the probability of occurrence of other related events when a certain event occurs.

[0080] The specific steps of step three are:

[0081] (3-1) Integrate the fault data on all sides of the distribution network (station-line-transformer-user), perform data cleaning and integration, build a complete data set containing fault information, and divide it into a training set and a test set.

[0082] (3-2) Taking the power outage events of users in the substation as the analysis basis, combined with the dependency and time series characteristics between fault data, a dynamic Bayesian network DBN is constructed; DBN includes six key nodes: the power outage ratio of users in the substation, preliminary judgment of event type, complaint telephone number of users in the substation, event information of low-voltage branch line terminal, substation terminal information and final judgment event, as shown in Table 1.

[0083] Table 1 Bayesian network node information table

[0084]

[0085] The dynamic Bayesian network can be further expressed as Fig. 9 shown. Fig. 9 In the network, the power outage events of users in the substation are used as the analysis basis and combined with the temporal relationship between the distribution terminal information and data, the status of other nodes is inferred, and a comprehensive bottom-up analysis of the fault area is achieved to solve the problem of locating the power outage area.

[0086] (3-3) The node parameters of DBN are learned through the maximum likelihood estimation method, and the probability of failure in different areas in the test set is calculated.

[0087] For the i-th node x in the power distribution Bayesian network i , x i There are m power distribution events in the network. Assume that all events related to node x have been obtained. i The power distribution event E′ of the relevant node = {e 1 ,e 2 ,···,e n}, then x i The kth power distribution event e k The conditional probability of occurrence is as follows:

[0088]

[0089] T represents the number of time slices, represents the power distribution event e in the tth time slice k , Par represents the probability of power distribution event, P B is the transition probability;

[0090] When a fault occurs in the distribution network and causes a power outage, a fault event chain is represented by F(r)={ti│ x i,1 ,x i,2 ,…x i,j …x i,n}, t i Represents x i Fault record, x i,n Represents x i The corresponding nth power distribution event, the meaning of this formula is that the fault event F(r) occurring on the line is due to many power distribution events x i,n The occurrence or non-occurrence of the fault event F(r) on the line at this time is inferred by the Bayesian formula. r ), calculated using the chain rule as shown below:

[0091] P(F r )=P(x 1 )P(x 2 ∣x 1 )...P(x i ∣x 1 x 2 ...x i-1 )

[0092] P(x 1 ) represents the node x 1 The probability of occurrence, P(x 2 ∣x 1 ) represents x 1 In case of x 2 Probability of occurrence;

[0093] For a network with T time slices, the power distribution events in the event chain include power outage events A for users in the substation area, preliminary judgment of power outage types B, complaint calls from users in the substation area C, low-voltage branch line terminal events D, substation terminal events E, and fault events F; where the fault events {F = l, l = 1, 2, 3, 4} represent user events, meter box events, low-voltage branch line events, and substation events, respectively. Within the time slice T, the posterior probability of line faults caused by these power distribution events is further expressed by the following formula:

[0094]

[0095] Where, P(F=l) T represents the probability of a fault event l occurring in the line in the Tth time slice, (·) 1,T Represents the observed values ​​of the set of elements contained in a certain characteristic factor on time slices 1 to T.

[0096] Step 4: Construct membership function and fuzzy reasoning rules to obtain a fuzzy reasoning system to adjust the failure probability.

[0097] The specific steps of step 4 are:

[0098] (4-1) Based on the complete data set constructed in step (3-1), the fuzzy clustering algorithm is used to determine the interval of the input membership function and formulate fuzzy inference rules; according to the probability of occurrence of different fault events, the output membership function is divided into four levels: low (L), medium (M), high (H), and extremely high (E), and fuzzy weights are assigned 0.1, 0.3, 0.6, and 1, respectively. For example, a power outage event in a substation is usually accompanied by a large-scale power outage for users, so its corresponding risk level is extremely high. The constructed fuzzy rules are shown in Table 2. The fuzzy rules between different features in Table 2 are compared Fig.10 The fuzzy reasoning system in the first layer generates the rules, which outputs the risk level. The fuzzy reasoning system in other layers generates corresponding fuzzy rules based on the relationship between risk level and characteristics. The three-dimensional curve of fuzzy rules is shown in Fig.11 shown.

[0099] Table 2 Fuzzy reasoning rules

[0100]

[0101]

[0102] (4-2) Based on the probability of power outages occurring in each area inferred in step (3-3), an output membership function is constructed to characterize the degree of fault.

[0103] (4-3) Determine whether the power outage probability of each area obtained in step (3-3) needs to be corrected. If it needs to be corrected, the power outage probability is input into the fuzzy inference system as input, and a defuzzified value is obtained through a defuzzification process. The defuzzified value is used to correct the original maximum probability state to achieve adjustment and optimization of the power outage probability.

[0104] (4-4) The results obtained from the test set are compared with the actual recorded data to evaluate the performance of the model in power outage area assessment under conditions of complete information and incomplete information.

[0105] Step 5: Finally, the most likely fault area is inferred to achieve bottom-up fault area analysis.

[0106] Example

[0107] (1) Top-down fault feeder identification

[0108] Using Python programming language Fig.12 The single-source multi-T connection open-loop distribution network in the example is analyzed. The distribution system consists of 12 switch nodes and 12 feeder sections. 1 ~L 12For each feeder section, each switch node is equipped with an FTU to monitor in real time whether there is a fault current flowing. Considering the actual application scenario, the present invention sets six typical fault scenarios in three cases of normal fault information, missed alarm and false alarm to verify the accuracy and feasibility of the improved matrix algorithm. The test results are shown in Table 3.

[0109] Table 3 Fault feeder identification results

[0110]

[0111] From the results in the above table, it can be seen that the proposed matrix algorithm can correctly judge the faulty feeder section when the fault information is missed or falsely reported in the case of single-section fault and multi-section fault. 8 With L 10 Failures occur at the same time, and node S 4 In case of false alarm of fault information, analysis shall be performed.

[0112] When segment L 8 With L 10 When a failure occurs, node S 4 In the case of fault information false alarm, the fault information matrix is The constructed causal association matrix D′ is shown in the following formula.

[0113]

[0114] The fault judgment matrix Q can be obtained as Q = [0 0 1 0 0 0 0 1 0 1 0 0]. Multi-segment faults appear in the judgment results. According to the judgment criteria, the fault information matrix needs to be checked. It is found that the fault information uploaded by node 4 and its two adjacent nodes is different. Therefore, the fault information uploaded by node 4 is set to 0, and then the result is obtained again.

[0115] (2) Bottom-up fault area analysis

[0116] In order to verify the performance of the fault area analysis model proposed in the present invention under the condition of complete information, the training set and the test set are divided into a ratio of 8:2, and the same set of data is tested using the traditional Bayesian network, the dynamic Bayesian network and the method of the present invention respectively. The receiver operating characteristic (ROC) curve is used in the present invention to evaluate the analysis performance. By drawing the ROC curve, the classification performance of the analysis model under different thresholds can be intuitively displayed. In the comparison process, in addition to the ROC curve, three key indicators, standard error (SE), confidence interval (CI) and accuracy A%, are introduced to comprehensively evaluate the performance of the model. The results are shown in Tables 4 and Fig.13 shown.

[0117] Table 4 Comparison data of evaluation indicators

[0118]

[0119] Depend on Fig.13 As shown in Table 4, compared with the direct use of the Bayesian network method, the method of the present invention improves AU ROC, SE and CI by 10.63%, 35.96% and 7.95% respectively; compared with the dynamic Bayesian network method, the method of the present invention improves AUROC, SE and CI by 7.57%, 19.78% and 6.46% respectively. In terms of A%, the method of the present invention improves by 10.7% and 5.9% respectively compared with the Bayesian network and the dynamic Bayesian network.

[0120] In order to verify the performance of the fault area judgment model proposed in this invention under the condition of incomplete information, a group test was designed. First, the dynamic Bayesian network model was trained using the complete training set data. Then, some terminal data was randomly deleted from the complete test set data to generate a data set with incomplete information, so as to simulate the data missing situation in actual applications, thereby evaluating the judgment ability of the model under the condition of incomplete information. The experimental design is divided into four groups: complete information, 10% missing information, 20% missing information and 30% missing information. The results are shown in Tables 5 and Fig.14 shown.

[0121] Table 5 Comparison data of evaluation indicators

[0122]

[0123] Depend on Fig.14 As shown in Table 5, when the degree of information loss is 10% and 20% respectively, the accuracy of the method of the present invention reaches 83.59% and 80.24% respectively.

[0124] In summary, the method proposed in the present invention, based on the traditional matrix algorithm, combines the causal relationship between the faulty feeder and the node, constructs a causal correlation matrix, improves the fault judgment criteria, and effectively solves the problem of feeder identification in the case of missed and false alarms of fault information. This method not only reduces the amount of calculation, but also improves the accuracy of fault location, and can adapt to the complex distribution network operation environment. By combining fuzzy theory and actual fault data, the present invention designs a membership function and inference rules, evaluates whether the probability obtained by reasoning needs to be corrected, takes the probability to be corrected as input for fuzzification, and then obtains the defuzzified value through defuzzification, and finally realizes the correction of the probability of the power outage area, which improves the accuracy of the fault area judgment.

Claims

1. A two-way analysis method for distribution network faults combining correlation matrix and dynamic Bayes, characterized in that: The following steps are involved: Step 1: Construct a correlation matrix based on the actual topological structure of the distribution network, and construct a fault analysis matrix based on the fault information matrix; Step 2: Establish fault judgment criteria for missed and false alarms of node fault information, so as to realize top-down fault feeder identification; Step 3: Build a dynamic Bayesian network based on the actual topological structure of the distribution network and the "station-line-transformer-user" relationship to infer the fault probability of each area; Step 4: Construct membership function and fuzzy reasoning rules to obtain a fuzzy reasoning system to adjust the fault probability; Step 5: Finally, the most likely fault area is inferred to achieve bottom-up fault area analysis.

2. The method for bidirectional analysis of distribution network faults combining correlation matrix and dynamic Bayesian according to claim 1 is characterized in that: The specific steps of step 1 are: (1-1) When a fault occurs on a feeder in the distribution network, the fault indicator connected to the head end of the fault feeder will monitor whether a fault current exceeding the set value flows through the fault indicator, and upload the monitored signal to the cloud platform. The fault indicator installed at the end of the fault feeder does not send a signal. The cloud platform generates a row vector matrix with the same number of columns as D′ as the fault information matrix G based on the uploaded information. D′ is the causal association matrix between the feeder section and the switch node; (1-2) The inverse matrices of G and D′ are multiplied to obtain the fault analysis matrix Q as follows: In the formula, when the element in the pth row of Q is 1, it means that the pth feeder has a fault, and when the element in the pth row is 0, it means that there is no fault in the pth feeder; when the element in the qth row of G is 1, it means that a fault current exceeding the set value flows through the fault indicator on the qth node, and when the element in the qth row is 0, it means that no fault current exceeding the set value flows through the fault indicator on the qth node; The element d in row p and column q in D′ pq When it is 1, it means that the fault of the pth feeder affects the qth fault indicator, d pq When it is 0, it means that the pth feeder is not associated with the qth fault indicator.

3. The method for bidirectional analysis of distribution network faults combining correlation matrix and dynamic Bayesian according to claim 1 is characterized in that: In the step 2, when the node fault information is missed, it is manifested that the fault information uploaded by the feeder terminal unit FTU at the node is empty. When processing the missed information, the following two judgment rules are constructed: Rule 1: When the fault information of the node's front and rear adjacent nodes are the same, the missed information is set to the same fault information as the front and rear adjacent nodes; Rule 2: When the fault information of the adjacent nodes before and after a node is different, the node is removed as an invalid node and is not considered in D and G, and a new assessment is performed.

4. The method for bidirectional analysis of distribution network faults combining correlation matrix and dynamic Bayesian according to claim 3 is characterized in that: In step 2, when the fault information is falsely reported, it is manifested as the fault information uploaded by the FTU at the node is flipped, that is, the fault information is actually 1 but reported as 0, or actually 0 but reported as 1, and the following two judgment rules are constructed: Rule 3: After uploading the false alarm fault information, the analysis results show that multiple section faults occur on the same line. At this time, the error correction is compared based on the node adjacent information; If the fault information of a node is different from the fault information uploaded by the adjacent nodes, the fault information of the node is falsely reported. The falsely reported information is modified to be the same as the adjacent nodes. If the value of an element in the judgment matrix is ​​-1, the value is set to 0. Rule 4: After uploading the false alarm fault information, the analysis result is wrong, and the fault information needs to be collected again.

5. The method for bidirectional analysis of distribution network faults combining correlation matrix and dynamic Bayesian according to claim 1 is characterized in that: The specific steps of step three are: (3-1) Integrate the fault data on all sides of the distribution network (station-line-transformer-user), clean and integrate the data, build a complete data set containing fault information, and divide it into a training set and a test set; (3-2) Taking the power outage events of users in the substation as the analysis basis, combined with the dependency and time series characteristics between fault data, a dynamic Bayesian network DBN is constructed; the DBN includes six key nodes: the power outage ratio of users in the substation, preliminary judgment of event type, complaint telephone number of users in the substation, event information of low-voltage branch line terminal, substation terminal information and final judgment event; (3-3) The node parameters of DBN are learned through the maximum likelihood estimation method, and the probability of failure in different areas in the test set is calculated.

6. The method for bidirectional analysis of distribution network faults combining correlation matrix and dynamic Bayesian according to claim 5 is characterized in that: In the step (3-3), for the i-th node x in the power distribution Bayesian network i , x i There are m power distribution events in the network. Assume that all events related to node x have been obtained. i The power distribution event E′ of the relevant nodes is n }, then x i The kth power distribution event e k The conditional probability of occurrence is as follows: T represents the number of time slices, represents the power distribution event e in the tth time slice k , Par represents the probability of power distribution event, P B is the transition probability; When a fault occurs in the distribution network and causes a power outage, a fault event chain is represented by F(r)={t i│ x i,1 ,x i,2, …x i,j …x i,n }, t i Represents x i Fault record, x i,n Represents x i The corresponding nth power distribution event, the meaning of this formula is that the fault event F(r) occurring on the line is due to many power distribution events x i,n The occurrence or non-occurrence of the fault event F(r) on the line at this time is inferred by the Bayesian formula. r ), calculated using the chain rule as shown below: P(F r )=P(x1)P(x2∣x1)...P(x i ∣x1x2...x i-1 ) P(x1) represents the probability of node x1 occurring, and P(x2|x1) represents the probability of x2 occurring when x1 occurs; For a network with T time slices, the power distribution events in the event chain include power outage events A for users in the substation area, preliminary judgment of power outage types B, complaint calls from users in the substation area C, low-voltage branch line terminal events D, substation terminal events E, and fault events F; where the fault events {F = l, l = 1, 2, 3, 4} represent user events, meter box events, low-voltage branch line events, and substation events, respectively. Within the time slice T, the posterior probability of line faults caused by these power distribution events is further expressed by the following formula: Where, P(F=l) T represents the probability of a fault event l occurring in the line in the Tth time slice, (·) 1,T Represents the observed values ​​of the set of elements contained in a certain characteristic factor on time slices 1 to T.

7. The method for bidirectional analysis of distribution network faults combining correlation matrix and dynamic Bayesian according to claim 1 is characterized in that: The specific steps of step 4 are: (4-1) Based on the complete data set constructed in step (3-1), the interval of the input membership function is determined by using a fuzzy clustering algorithm, and fuzzy inference rules are formulated; according to the probability of occurrence of different fault events, the output membership function is divided into four levels: low, medium, high, and very high, and fuzzy weights are assigned to each level; (4-2) constructing an output membership function to characterize the degree of fault according to the probability of power outage in each area inferred in step (3-3); (4-3) Determine whether the power outage probability of each area obtained in step (3-3) needs to be corrected. If it needs to be corrected, the power outage probability is used as input to the fuzzy inference system, and a defuzzified value is obtained through a defuzzification process. The defuzzified value is used to correct the original maximum probability state to achieve adjustment and optimization of the power outage probability; (4-4) The results obtained from the test set are compared with the actual recorded data to evaluate the performance of the model in power outage area assessment under conditions of complete information and incomplete information.

Images