A method and device for analyzing the root cause of communication equipment failure

Abstract

The present invention relates to the field of communication, and in particular to a method and device for analyzing the root cause of a communication device failure. The method includes: splitting the tree structure of a communication network into single chains based on a master control node; independently analyzing the probability of occurrence of the fault cause type of each single chain based on a recognition model, and using the ID of the faulty node and the corresponding fault cause type as the root cause of the fault of the single chain; summarizing the root causes of all single chains as the probability conclusion of the root cause of the fault of this fault; merging the feature data corresponding to all single chains associated with the root cause of the fault in this fault and the root cause of the fault of this fault as a fault training data, merging the feature data corresponding to all fault-free single chains as a fault-free training data, and iteratively training the recognition model. The present invention can analyze equipment failures in a tree-structured communication network, and realize the location of fault point sub-equipment and the identification of fault cause types.

Description

Technical Field

[0001] The present invention relates to the field of communications, and in particular to a method and device for analyzing root causes of communication equipment failures. Background Art

[0002] At present, Bayesian networks are widely used in the modeling of fault analysis systems with machine learning capabilities in the industrial field, such as transformer fault analysis in power supply networks. As a classifier model, Bayesian networks are mainly used for fault analysis of single-point equipment, that is, to identify the cause type of equipment failure under the premise of knowing the location of the fault point.

[0003] However, communication equipment and networks are not a single-point structure. When a fault occurs within a network that is not a single-point structure, operation and maintenance engineers usually cannot determine which level or node the fault is at. Therefore, the fault feature data collection will have a sub-device range. The fault analysis system needs to solve the following problems: 1. The root cause analysis model needs to distinguish between faulty nodes and normal nodes, that is, to locate the sub-device at the fault point. 2. The machine learning training model needs to distinguish and process the fault feature data of non-faulty sub-devices and faulty sub-devices.

[0004] In view of this, how to overcome the defects of the prior art and solve the problem of difficulty in analyzing the root cause of faults in a multi-layer network structure is a problem to be solved in this technical field. Summary of the invention

[0005] In view of the above defects or improvement requirements of the prior art, the present invention solves the problem of difficulty in analyzing the root cause of faults in a tree network structure.

[0006] The embodiment of the present invention adopts the following technical solution:

[0007] In the first aspect, the present invention provides a method for analyzing the root cause of a communication device failure, specifically: splitting the tree structure of the communication network into single chains based on the main control node, each node of the single chain corresponds to a layer of the communication network; independently analyzing the occurrence probability of each single chain failure cause type based on the recognition model, and taking the ID of the failure node and the corresponding failure cause type as the root cause of the failure of the single chain; summarizing and calculating the root causes of the failure of all single chains, and taking the summary result as the probability conclusion of the root cause of the failure of this failure; merging the feature data corresponding to all single chains associated with the root cause of the failure in this failure and the root cause of the failure of this failure as a failure training data, merging the feature data corresponding to all fault-free single chains as a fault-free training data, and adding them to the training data set of the recognition model to iteratively train the recognition model.

[0008] Preferably, the tree structure of the communication network is split into single chains based on the main control node, specifically including: representing the tree structure of the communication network as a tree structure, with the main control node as the root node of the tree structure, and each layer of the tree structure corresponding to a layer in the tree structure of the communication network; starting from the main control node, traversing each branch in the tree structure until reaching a leaf node, and all nodes on a traversal path from the root node to a leaf node are regarded as a single chain.

[0009] Preferably, when there are network connections in two directions on the main control node, the tree structure of the communication network is split into single chains based on the main control node, and also includes: splitting the upper tree structure and the lower tree structure of the main control node into partial single chains respectively; arranging the split partial single chains one-to-one into complete single chains with the main control node as the connection point, and copying the data of the upper node to the data of each single chain.

[0010] Preferably, before splitting the tree structure of the communication network into single chains based on the master control node, the method further includes: acquiring the tree structure range covered by the fault alarm information, and splitting the network nodes within the tree structure range into single chains.

[0011] Preferably, the probability of occurrence of the fault cause type corresponding to each node in each single chain is analyzed independently, specifically including: determining the node level corresponding to each fault cause type according to the network service characteristics of each sub-device; analyzing the fault feature data of each single chain based on the Bayesian network classifier, and outputting the relative probability of each fault cause type occurring at each node in each single chain.

[0012] Preferably, the root causes of failures of all single chains are summarized and calculated, specifically including: obtaining the occurrence probability value of the root cause of failures existing in each single chain, sorting the root causes of failures according to the occurrence probability value, and using the sorting result as the summary calculation result of the root causes of failures.

[0013] Preferably, obtaining the occurrence probability value of the root cause of the fault in each single chain further includes: when the same root cause of the fault exists in multiple single chains, taking the average of the probability values ​​of the root cause of the fault in each single chain as the occurrence probability value of the root cause of the fault.

[0014] Preferably, the characteristic data corresponding to all single chains associated with the root cause of the fault in a fault and the root cause of the fault are combined as a fault training data, specifically including: combining the probability value of the fault occurrence and the fault cause type as the prior probability affecting the fault cause type; regarding the no-fault situation as a fault cause type, and taking the average of the prior probabilities of all other fault cause types as the prior probability of the no-fault type, or taking the inverse of the number of fault cause types as the prior probability of the no-fault type.

[0015] Preferably, the characteristic data corresponding to all single chains in a fault and the root cause of the fault are used as training data, and also include: when the root cause of the fault is associated with multiple single chains, each fault characteristic value of the associated multiple single chains is merged into one training data of the recognition model in a mode manner.

[0016] On the other hand, the present invention provides a device for analyzing the root cause of a communication device failure, specifically comprising: comprising at least one processor and a memory, wherein the at least one processor and the memory are connected via a data bus, and the memory stores instructions that can be executed by the at least one processor, and after the instructions are executed by the processor, they are used to complete the method for analyzing the root cause of a communication device failure in the first aspect.

[0017] Compared with the prior art, the beneficial effect of the embodiment of the present invention is that the communication network is split into a single-chain structure, each single chain is analyzed independently, and the analysis results are summarized to obtain the analysis conclusion of the root cause of the fault and the training data for fault analysis. The method can analyze the equipment failure in the tree-structured communication network, realize the location of the fault point sub-device and the identification of the type of fault cause, and perform corresponding processing on the machine learning training model, and propose a matching machine learning method, which solves the problem of root cause analysis of tree-structured network failures that cannot be solved by the Bayesian network single-point analysis method, and has significant practical value. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments of the present invention. Obviously, the drawings described below are only some embodiments of the present invention, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0019] Figure 1 A flow chart of a method for analyzing the root cause of a communication device failure provided by an embodiment of the present invention;

[0020] Figure 2 A flow chart of another method for analyzing the root cause of a communication device failure provided by an embodiment of the present invention;

[0021] Figure 3 A schematic diagram of a network structure in a certain scenario of an embodiment of the present invention;

[0022] Figure 4 A flow chart of another method for analyzing the root cause of a communication device failure provided by an embodiment of the present invention;

[0023] Figure 5 A schematic diagram of a network structure in a certain scenario of an embodiment of the present invention;

[0024] Figure 6It is the node level of the fault cause in a certain scenario of an embodiment of the present invention;

[0025] Figure 7 It is multiple independent single-link fault feature data in a certain scenario of an embodiment of the present invention;

[0026] Figure 8 It is multiple independent single-link fault feature data in a certain scenario of an embodiment of the present invention;

[0027] Fig. 9 The ID of each single chain and the root cause analysis conclusion of its failure in a certain scenario of an embodiment of the present invention;

[0028] Fig.10 is the probability value of the root cause of the fault existing in each single chain in a certain scenario of an embodiment of the present invention;

[0029] Fig.11 The result after summarizing and sorting the failure probabilities of different single chains in a certain scenario of an embodiment of the present invention;

[0030] Fig.12 A schematic diagram of the data structure of training data used in a certain scenario of an embodiment of the present invention;

[0031] Fig.13 The fault feature data in a certain scenario of an embodiment of the present invention is decomposed into a single chain and training data with fault cause coding added;

[0032] Fig.14 A schematic diagram of the structure of a communication equipment fault root cause analysis device provided by an embodiment of the present invention;

[0033] The reference numerals are as follows:

[0034] 11: processor; 12: memory. DETAILED DESCRIPTION

[0035] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0036] The present invention is an architecture of a specific functional system, so the specific embodiments mainly illustrate the functional logical relationship between the various structural modules, and do not limit the specific software and hardware implementation methods.

[0037] In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other. The present invention will be described in detail below with reference to the accompanying drawings and embodiments.

[0038] Embodiment 1:

[0039] The existing root cause analysis of network equipment failures uses Bayesian networks as a classifier model, but this model is only applicable to single-point analysis. Based on this model, the method provided in this embodiment locates the root cause of a faulty device whose internal topology is a non-single-point structure, especially a tree structure, including the location of the sub-device where the fault point is located and the corresponding analysis of each fault type.

[0040] like Figure 1 As shown, the specific steps of the method for analyzing the root cause of a communication device failure provided by an embodiment of the present invention are as follows:

[0041] Step 101: Split the tree structure of the communication network into single chains based on the master control node, and each node of the single chain corresponds to a layer of the communication network.

[0042] In some actual scenarios of current network systems, network devices are connected layer by layer according to the service level, and finally connected to the only master control device in this area. Therefore, the topological structure of the communication network can be regarded as a tree structure with the master control node as the root node, the root node of the tree structure is the master control node, and each layer of the tree structure corresponds to a layer of the network structure. Taking the optical line terminal (OLT) device at the access network local end as an example, its internal structure can be understood as a tree structure in two directions with the master control disk as the root node: the tree structure in the downlink direction and the tree structure in the uplink direction, among which the tree structure in the downlink direction is the four levels of master control-line card-passive optical network (Passive Optical Network, abbreviated as PON) port-optical network unit (Optical Network Unit, abbreviated as ONU), and the tree structure in the uplink direction is the three levels of master control-uplink disk-uplink port, each of which contains multiple sub-devices. Correspondingly, its downstream direction can be regarded as a tree structure including four levels of OLT master control, line card, PON port, and ONU. Its upstream direction can be regarded as a tree structure including three levels of master control, upstream disk, and upstream port. Each sub-device is regarded as a sub-node of the upper-layer device to which it is connected.

[0043] For communication devices with an internal tree structure, the collected fault feature data is also in a tree structure, with fault data nodes corresponding to network device nodes one by one. In order to use the Bayesian network model as a fault identification model to analyze faults of sub-devices at different levels, it is necessary to first split the tree structure into multiple independent single chains with the master control node as the root, and then perform fault root cause analysis on each single chain, including identifying whether it is a fault-free single chain. The specific splitting method is described below.

[0044] Step 102: Based on the identification model, the occurrence probability of the fault cause type of each single chain is independently analyzed, and the ID of the faulty node and the corresponding fault cause type are used as the root cause of the fault of the single chain.

[0045] During the operation of network equipment, the same sub-device may have multiple possible causes of failure. For example, the PON port may have optical module damage and optical module mismatch, and the probability of different failures is different. After splitting the tree structure into single chains, the Bayesian network classifier model can be used to perform an overall analysis of the fault feature data of each single chain to calculate the relative probability of each type of fault cause in each single chain. In the method provided in this embodiment, in order to facilitate unified analysis of data, "no fault" is also regarded as a type of fault cause.

[0046] On the contrary, each fault cause type will only occur on one sub-device, and the fault cause type has a many-to-one correspondence with the sub-device tree node level where the fault is located. Therefore, the network level where the fault occurs can be uniquely determined based on the fault cause type. Further, after the tree structure is split into a single chain in step 101, each layer of nodes on the single chain corresponds to only one sub-device. Therefore, determining the fault cause type also locates the network level where the sub-device of the single chain fault is located.

[0047] In order to facilitate expression and calculation, in this embodiment, any sub-device node can be uniquely identified by a hierarchical ID. For example, the sub-device in the downstream direction of the OLT can be expressed by an ID in the form of Root-Slot-PON port-device number (Onu). Each type of fault cause corresponds to a sub-device node level, and the sub-device node ID forms a corresponding relationship with the type of fault cause. Therefore, if the single chain where the fault point is located and the type of fault cause can be analyzed, the sub-device node ID where the fault point is located can be determined.

[0048] Through the ID of the fault node and the actual fault cause type, the sub-device where the fault occurred and the fault cause type can be uniquely determined, and the root cause of the fault that caused the fault can be uniquely determined. In this embodiment, only when the sub-device node and the fault type are the same will they be considered as the same root cause of the fault, and different sub-device nodes or different fault cause types are not considered as the same root cause of the fault.

[0049] Step 103: Calculate the root causes of all single-chain failures, and use the result as the probability conclusion of the root cause of the failure.

[0050] Since the failures between different devices in the network system may affect each other, after analyzing each single chain in step 102, it is also necessary to conduct a comprehensive analysis of the root causes of the failures of all single chains, so as to achieve a comprehensive calculation of the failure probabilities of the same root cause of failure in multiple single chains, and finally output a comprehensive conclusion on the probabilities of each root cause of failure.

[0051] Step 104: Combine the feature data corresponding to all single chains associated with the root cause of the fault and the root cause of the fault as one piece of fault training data, combine the feature data corresponding to all fault-free single chains as one piece of fault-free training data, and add them to the training data set of the recognition model to iteratively train the recognition model.

[0052] In order to facilitate subsequent fault analysis and prediction, the fault data that has been summarized and analyzed can also be used as training data to train the Bayesian network model. In a fault, the fault data of all nodes corresponds to the network structure in a tree structure, and the corresponding training data to be used is also a tree structure. Therefore, it is also necessary to decompose the tree structure training data into multiple single-chain training data, and then reintegrate it into a fault training data or fault-free training data through the algorithm, and add it to the analysis model to achieve the machine learning capability of the analysis model.

[0053] Furthermore, for the training model, failure and non-failure are different scenarios, so the training data of fault-free single chains need to be distinguished from the training data of faulty single chains. The former is used to train and identify fault-free single chains, and the latter is used to train and identify the type of fault cause. Therefore, two types of training data are needed in the training model: 1. Summarize the feature data corresponding to all single chains associated with the root cause of the fault in this fault (passing through the faulty node), and combine it with the root cause of this fault as a fault training data; 2. Summarize the feature data corresponding to all fault-free single chains in this fault as a fault-free training data. After obtaining the two types of training data, add the two types of training data to the training data set of the recognition model respectively to iteratively train the recognition model.

[0054] After steps 101 to 104 provided in this embodiment, the Bayesian network model can be used to analyze the network with non-single-point structure, obtain the root cause analysis result of the fault, and use the analysis result as training data to further train the recognition model.

[0055] In the method provided in this embodiment, it is necessary to split the tree structure of the network into single chains. Figure 2 As shown, the splitting of single strands can be accomplished by the following steps.

[0056] Step 201: The tree structure of the communication network is represented by a tree structure, the master control node is used as the root node of the tree structure, and each layer of the tree structure corresponds to a layer in the tree structure of the communication network.

[0057] In this embodiment, each fault cause type corresponds to a corresponding level, so first, the nodes in the network need to be represented in a tree structure according to the tree structure of the communication network. In each tree-structured network area, there is a master control node. When splitting a single chain, the master control node is also used as the root node of the tree structure. At the same time, the sub-devices in the network also need to correspond to the level of the tree structure according to the actual network level.

[0058] Step 202: Starting from the master control node, the tree structure is traversed until a leaf node is reached. All nodes on a traversal path from the root node to a leaf node are used as a single chain, and the data of the upper node is copied to the data of each single chain.

[0059] When splitting, the master node is traversed to each leaf node, and the traversal path is regarded as a single chain. If any node in the traversal path fails, the communication of the entire single chain will be abnormal, which also corresponds to the collection path of the fault data.

[0060] After the split, the data of the upper node needs to be copied to the data of each single chain. In this embodiment, regardless of the upper or lower device, the upper node represents a layer closer to the root node relative to the node, rather than a higher layer in the upper direction.

[0061] Through step 201 - step 202 , the tree structure is split and each single chain for analysis is obtained.

[0062] Take OLT equipment as an example. Figure 3As shown, the internal downward connection direction is a tree structure of four levels, namely, master control-line card-PON port-ONU, in which the ID of each device node is expressed as: Root-Slot-Port-Onu, where Root represents the root node (master control), which is abbreviated as R in the following text for the convenience of description, Slot represents the slot number, Port represents the PON port number, and Onu represents the device number of the ONU. For example, R-11 represents the line card in the 11th slot (slot11), and R-11-3-5 represents the 5th ONU under the 3rd PON port (port3) under the line card in the 11th slot (slot11). After the tree structure in the figure is split according to the method of step 201-step 202, the single chain obtained is: R-line card 11-PON port 1-ONU1 (R-11-1-1), R-line card 11-PON port 1-ONU2 (R-11-1-2), R-line card 11-PON port 1-ONU3 (R-11-1-3), R-line card 11-PON port 2 (R-11-2), R-line card 12 (R-12).

[0063] From the above example, it can be seen that after splitting, the tree structure in the actual network is split into a single chain structure.

[0064] In some other practical scenarios, the master device may have more than one network connection. In this case, the network connections in multiple directions need to be split separately and combined into a single link. Figure 4 As shown, the splitting of the multi-directional tree structure can be completed through the following steps.

[0065] Step 301: split the upper link tree structure and the lower link tree structure of the master control node into partial single chains according to the methods provided in steps 201 and 202.

[0066] For each tree structure on the master node, the splitting method is the same as above and will not be repeated.

[0067] Step 302: The split parts of the single chain are connected one-to-one using the master node as the connection point to form a complete single chain.

[0068] In the actual network structure, the upper tree structure and the lower tree structure are connected through the master control node. Regardless of the direction of the tree structure, the master control node is used as the root node. After each tree structure is split, different single chains need to be connected through the master control node to form a complete single chain from the top to the bottom. Since the information of each upper single chain will interact with all lower single chains after passing through the master control node, and the information of each lower single chain will interact with all upper single chains after passing through the master control node, it is necessary to arrange and combine all upper single chains and lower single chains one by one to find all combined single chains.

[0069] After steps 301 to 303, the multi-directional tree structure network can be split into single chains.

[0070] Still taking OLT equipment as an example, Figure 5 As shown in the figure, in addition to the four-layer tree structure of main control-line card-PON port-ONU in the downlink direction, the network hierarchy is: R-1 layer-2 layer-3 layer. In the uplink direction of the main control, there is another tree structure: uplink port-uplink disk-main control, and the network hierarchy is: (-2 layer)-(-1 layer)-R, where the minus sign indicates the uplink direction.

[0071] First, according to step 301, the upper link tree structure and the lower link tree structure are respectively decomposed into a plurality of upper link single chains and a plurality of lower link single chains. Figure 5 In the example, the uplink tree structure can be split into a single chain: uplink port 1-uplink disk 1-R (1-1-R), uplink port 1-uplink disk 2-R (1-2-R). The downlink tree structure can be split into a single chain: R-line card 11-PON port 1-ONU1 (R-1-1-1), R-line card 11-PON port 1-ONU2 (R-1-1-2).

[0072] Then, according to step 302, permutations and combinations are performed to form all single chains. The permutations and combinations form 4 single chains: uplink port 1-uplink disk 1-R-line card 11-PON port 1-ONU1 (1-1-R-1-1-1), uplink port 1-uplink disk 2-R-line card 11-PON port 1-ONU1 (1-2-R-1-1-1), uplink port 1-uplink disk 1-R-line card 11-PON port 1-ONU2 (1-1-R-1-1-2), uplink port 1-uplink disk 2-R-line card 11-PON port 1-ONU2 (1-2-R-1-1-2).

[0073] It can be seen from the above example that after splitting, the network nodes in the bidirectional tree structure network can also be split into a single chain structure.

[0074] When collecting fault data, the sub-device location of the fault point is usually uncertain, so a tree structure range is selected, or a tree structure range is determined based on the alarm information coverage of the network management system. In a fault, if a tree structure sub-device range is selected, the data of this tree structure is decomposed and converted into multiple independent single-chain data for processing. Obtain the tree structure range covered by the fault alarm information, split the network nodes within the tree structure range into single chains, and collect the fault data corresponding to each node in the tree structure.

[0075] like Figure 3 In the tree structure network shown, the corresponding Figure 6 As shown in the fault cause node level, the fault data of a certain fault can be decomposed and transformed into the following Figure 7The figure shows multiple independent single-link fault characteristic data. Each line in the figure is the ID of a single link and its various fault characteristic values. The first three lines of single-link data are three different ONUs (ONU1 to ONU3). They share the same PON port 1 (port1) and its upper layer line card 11 (port11) and the main control (R) data. The ID of PON port 1 is expressed as: R-11-1. The 4th to 10th lines of single links are Figure 3 The single-chain data of the other 7 different PON ports (R-11-2 to R-11-8) not shown in the figure, these 7 PON ports and the first 3 rows of single chains share the data of the same line card 11 (port11) and the master (R), and the ID of line card 11 is represented as: R-11. All the above single chains share the master-level data, that is, the data on the root node R. Further, as Figure 8 As shown, in order to distinguish the upward direction from the downward direction, if there is a fault cause node level of the upward node, the node level of the upward direction can be expressed as a negative value.

[0076] After the fault data is split, the fault feature data of each single chain can be analyzed based on the Bayesian network classifier, and the relative probability of each fault cause type (including no fault) of each single chain can be output. First, according to the network service characteristics of each sub-device, the node level corresponding to each fault cause type is obtained. Based on the Bayesian network classifier, the fault feature data of each single chain is analyzed, and the relative probability of each fault cause type occurring in each node in each single chain is output. The analysis results are as follows: Fig. 9 As shown in the figure, each row is the ID of a single chain and its root cause analysis conclusion: the relative probability of various types of fault causes (including no fault). In this embodiment, each network layer has only one sub-device, and the fault cause type of each layer corresponds to the sub-device node level. Therefore, the ID of the single chain and the node level corresponding to the fault cause type can jointly determine the fault sub-device ID, thereby uniquely determining the root cause of the fault. Fig. 9 As shown, it is a summary of the analysis results of the cause of each single-chain data failure.

[0077] After obtaining the fault analysis data of each single link, it is also necessary to output the overall priority of the root cause of the fault. Fig.10 The occurrence probability value of the root cause of the fault in each single chain is shown, and each root cause of the fault is sorted according to the occurrence probability value, and the sorting result is used as the summary calculation result of the root cause of the fault. Fig.11 As shown in , this is the result of summarizing and sorting the failure probabilities of different single chains in a certain actual scenario. Furthermore, when the same root cause of failure exists in multiple single chains, the average of the probability values ​​of the root cause of failure in each single chain is taken as the probability value of the root cause of failure. Fig.11Among the items with the same device node ID and fault root cause type, the three items with the highest probability are: PON port optical module failure with device node ID 1-11, ONU failure with device node ID 1-11-1-2, and ONU failure with device node ID 1-11-1-3. Among them, the PON port optical module failure with device node ID 1-11 is included in 3 single chains, so the average of the corresponding 3 items of data is taken as the final value; the other two items are only included in one single chain, so the corresponding item values ​​are used directly. Fig.11 After summarizing, it can be seen that 1-11-1-2 and 1-11-1-2 both correspond to the PON port 1-11-1 of 1-11. Therefore, the optical module fault of the PON port 1-11 can be further located to 1-11-1.

[0078] In order to improve the accuracy of the Bayesian network model in fault analysis and prediction, it is also necessary to combine the probability value of the fault occurrence with the actual fault cause type as training data for the fault scenario to further train the model. The fault feature data obtained from a fault and its actual root cause (fault point sub-device location + fault cause type) together form a Fig.12 Similar to the analysis model, the fault feature data obtained from a fault is extracted and decomposed into single chains, and a fault cause type is added to each single chain. Fig.13 As shown in the figure, the training data is extracted from the fault feature data and then decomposed into a single chain and added with the fault cause code. The last column error is the fault cause type code, where 0 represents no fault. In order to ensure the validity of the training data, the fault cause type here needs to use the actual real fault cause type or the verified correct fault cause type.

[0079] When collecting fault data, the following reasons may cause the sub-device range (tree structure) of data collection to be selected very broadly or very precisely: 1) When the fault analysis system is driven by alarm information from other systems, different fault scenarios (different types of fault causes) result in different tree structure ranges covered by the sub-device ID pointed to by the alarm information. 2) The personal experience, ability or habits of the operation and maintenance personnel are different. As a result, it is highly likely that one piece of fault-free training data will be aggregated and generated, resulting in a high prior probability of the fault-free type in the Bayesian network model. Therefore, in this embodiment, the fault-free situation is regarded as a type of fault cause and is identified together with other fault cause types. However, in the analysis model, the calculation of its prior probability requires special processing.

[0080] The calculation of the prior probability of the fault-free type is different from that of other fault cause types. It is not based on quantitative statistics, but needs to be calculated based on its probability characteristics. Specifically, the following method can be used for calculation. (1) The probability value of the fault occurrence and the fault cause type are combined as the prior probability affecting the fault cause type. (2) The fault-free situation is regarded as a fault cause type, and the average of the prior probabilities of all other fault cause types is used as the prior probability of the fault-free type, or the inverse of the number of fault cause types is used as the prior probability of the fault-free type, and the calculation is: the prior probability of the fault-free type = 1 / the number of fault cause types, and the result is a constant value. Therefore, when calculating the posterior probabilities of various fault types, the fault-free type and all other fault types are always treated equally in terms of prior probability.

[0081] In actual use, the acquisition of fault-free training data can also be independent of the faulty scenario. That is, any number of single-chain data can be randomly collected on a normal sub-device and imported into the analysis model as fault-free training data, while its prior probability remains unchanged.

[0082] Furthermore, in actual scenarios, a root cause of a fault may be associated with multiple single chains. In order to prevent the different numbers of single chains formed by data collection for each fault from affecting the calculation of the prior probability of each fault cause type in the Bayesian network model, it is also necessary to merge each fault feature value of the multiple associated single chains into one training data for the identification model in the manner of finding the majority. For example: a main control fault forms N single chain training data. Since it is a root node fault, these N training data are all fault training data, but the fault cause type is all the same main control fault. In the method of this embodiment, this fault should not be counted as N main control faults, but should be counted as one main control fault, and the relevant fault data should be merged into one training data for use.

[0083] Furthermore, the fault feature values ​​in different single chains can be combined in a summation manner. For example, assuming that there are N single chain training data for a root node fault, a fault feature item has two values, 0 and 1. The value of the fault feature item in a single chain is 0, and the value in b single chains is 1. When summarizing, the prior probability of the fault feature item is calculated as 0: a / (a+b), 1: b / (a+b) (where a+b = <N)。

[0084] The communication equipment root cause analysis method provided in this embodiment breaks down the tree-structured fault feature data into multiple single-chain data, analyzes each of them independently, and then summarizes and calculates them, and finally locates the sub-device where the fault point is located and its fault type. It mainly solves the problems of summarizing and outputting the conclusions of multiple single-chain analysis, and merging and restoring multiple single-chain training data into one fault training data, and solves the special processing problem of training data without fault types. This method mainly breaks through the limitation that the Bayesian network classifier model can only perform single-point analysis on the root cause of the fault, and solves the problem of the root cause analysis model and training model of the communication equipment fault with an internal tree structure.

[0085] Embodiment 2:

[0086] Based on the communication device failure root cause analysis method provided in the above embodiment 1, the present invention also provides a communication device failure root cause analysis device that can be used to implement the above method, such as Fig.14 , which is a schematic diagram of the device architecture of an embodiment of the present invention. The communication device fault root cause analysis device of this embodiment includes one or more processors 11 and a memory 12. Fig.14 A processor 11 is taken as an example.

[0087] The processor 11 and the memory 12 may be connected via a bus or other means. Fig.14 The example of connecting through bus is taken in the following.

[0088] The memory 12 is a non-volatile computer-readable storage medium for a method for analyzing root causes of failures of communication equipment, and can be used to store non-volatile software programs, non-volatile computer executable programs, and modules, such as the method for analyzing root causes of failures of communication equipment in Example 1. The processor 11 executes various functional applications and data processing of the apparatus for analyzing root causes of failures of communication equipment by running the non-volatile software programs, instructions, and modules stored in the memory 12, that is, the method for analyzing root causes of failures of communication equipment in Example 1 is implemented.

[0089] The memory 12 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage devices. In some embodiments, the memory 12 may optionally include a memory remotely arranged relative to the processor 11, and these remote memories may be connected to the processor 11 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0090] The program instructions / modules are stored in the memory 12, and when executed by one or more processors 11, the method for analyzing the root cause of a communication device failure in the above-mentioned embodiment 1 is executed, for example, the above-described Figure 1 , Figure 2 and Figure 4 The steps shown.

[0091] A person skilled in the art may understand that all or part of the steps in the various methods of the embodiments may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, and the storage medium may include: a read-only memory (ROM), a random access memory (RAM), a disk or an optical disk, etc.

[0092] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for analyzing the root cause of a communication device failure, characterized in that: Specifically include: The tree structure of the communication network is split into single chains based on the master node, and each node of the single chain corresponds to a layer of the communication network; Based on the recognition model, the occurrence probability of the fault cause type of each single chain is independently analyzed, and the ID of the faulty node and the corresponding fault cause type are used as the root cause of the fault of the single chain, where the occurrence probability of the fault cause type includes the prior probability of the non-fault type and the prior probability of the fault type; Summarize and calculate the root causes of all single-chain failures, and use the summary results as the probability conclusion of the root cause of the failure; The feature data corresponding to all single chains associated with the root cause of the fault and the root cause of the fault are combined as a piece of fault training data, and the feature data corresponding to all single chains without faults are combined as a piece of non-fault training data, and added to the training data set of the recognition model to iteratively train the recognition model; Among them, the feature data corresponding to all single chains associated with the root cause of the fault in this fault and the root cause of the fault of this fault are combined as a fault training data, and the feature data corresponding to all fault-free single chains are combined as a fault-free training data, specifically including: combining the probability value of the fault occurrence and the actual fault cause type as the prior probability affecting the fault cause type; regarding the fault-free situation as a fault cause type, and taking the average of the prior probabilities of all other fault cause types as the prior probability of the fault-free type, or taking the inverse of the number of fault cause types as the prior probability of the fault-free type.

2. The method for analyzing the root cause of a communication device failure according to claim 1, characterized in that: The tree structure of the communication network is split into single chains based on the master control node, specifically including: The tree structure of the communication network is represented by a tree structure, the master control node is used as the root node of the tree structure, and each layer of the tree structure corresponds to a layer in the tree structure of the communication network; Starting from the master node, each branch in the tree structure is traversed until a leaf node is reached. All nodes on a traversal path from the root node to a leaf node are treated as a single chain, and the data of the upper node is copied to the data of each single chain.

3. The method for analyzing the root cause of a communication device failure according to claim 2, characterized in that: When there are two-directional network connections on the master control node, the tree structure of the communication network is split into single chains based on the master control node, and further includes: Split the upper link tree structure and the lower link tree structure of the master control node into partial single chains respectively; The split parts of the single chain are connected with the master node as the connection point, and are arranged one by one to form a complete single chain.

4. The method for analyzing the root cause of a communication device failure according to claim 1, characterized in that: Before splitting the tree structure of the communication network into single chains based on the master control node, the method further includes: The tree structure range covered by the fault alarm information is obtained, the network nodes within the tree structure range are split into single chains, and the data of the upper nodes are copied to the data of each single chain.

5. The method for analyzing the root cause of a communication device failure according to claim 1, characterized in that: The independent analysis of the occurrence probability of the fault cause type corresponding to each node in each single chain specifically includes: Determine the node level corresponding to each fault cause type based on the network service characteristics of each sub-device; The fault feature data of each single chain is analyzed based on the Bayesian network classifier, and the relative probability of each fault cause type occurring at each node in each single chain is output.

6. The method for analyzing the root cause of a communication device failure according to claim 1, characterized in that: The root causes of all single chain failures are summarized and calculated, specifically including: Obtain the occurrence probability value of the root cause of the fault in each single chain, sort the root causes of the fault according to the occurrence probability value, and use the sorting result as the summary calculation result of the root cause of the fault.

7. The method for analyzing the root cause of a communication device failure according to claim 6, characterized in that: The step of obtaining the occurrence probability value of the root cause of the fault in each single chain further includes: When the same root cause of failure exists in multiple single chains, the average of the probability values ​​of the root cause of failure in each single chain is taken as the occurrence probability value of the root cause of failure.

8. The method for analyzing the root cause of a communication device failure according to claim 1, characterized in that: The method further includes combining the characteristic data corresponding to all single links associated with the root cause of the fault and the root cause of the fault as one piece of fault training data, and combining the characteristic data corresponding to all single links without fault as one piece of fault training data. When the root cause of the fault is associated with multiple single chains, each fault feature value of the associated multiple single chains is merged into one training data of the recognition model in a mode-based manner.

9. A communication equipment failure root cause analysis device, characterized in that: It includes at least one processor and a memory, the at least one processor and the memory are connected via a data bus, the memory stores instructions that can be executed by the at least one processor, and the instructions, after being executed by the processor, are used to complete the method for analyzing the root cause of a communication device failure as described in any one of claims 1 to 8.

Images