A rail transit fault prediction method, device, equipment and storage medium

Abstract

The embodiments of the present application disclose a fault prediction method, device, equipment and storage medium for rail transit. The method obtains operation log data of rail transit, classifies and predicts the operation log data through a preset classification model, and obtains a corresponding prediction mode category; compares the prediction mode category with the preset target mode category to confirm the matching status of the prediction mode category and the preset target mode category; if the prediction mode category is the same as one of the preset target mode categories, the operation log data is judged to be normal; if the prediction mode category is different from the preset target mode category, the operation log data is judged to be abnormal, and the deviation rate is calculated according to the prediction mode category and the preset probability set, and the preset probability set includes all mode categories and corresponding probabilities; the probability of a fault in the operation process corresponding to the operation log data is determined according to the deviation rate, which can solve the problem of low accuracy in rail transit fault prediction and improve the prediction accuracy.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of artificial intelligence technology, and in particular to a method, device, equipment and storage medium for predicting faults in rail transit. Background Art

[0002] With the development of society, rail transit, as an important mode of transportation, has brought great convenience to people and played a very important role in promoting the economic development of society. However, when rail transit has a malfunction, if it cannot be discovered in time, it will not only affect the passenger experience, but also affect the safety of rail transit itself.

[0003] According to actual operation and maintenance experience, the cause of the failure is often hidden in the operation and maintenance log data. Traditional methods for predicting faults in log data include prediction based on keyword matching and prediction based on configuration parsing rules. Among them, the prediction method based on keyword matching is to match the given keywords to determine whether it is abnormal. This method is difficult to give corresponding keywords for different types of log data, which increases the difficulty of prediction. Even if the defined keywords appear, it does not necessarily mean that there is a fault problem in the log. This depends on the developer's habits, which increases the uncertainty of the prediction. The prediction method based on configuration parsing rules requires different parsing rules to be formulated according to different types of log data. The parsing work is very complicated and requires a lot of time for debugging.

[0004] Therefore, the existing method of predicting faults in logs cannot achieve fully automated fault prediction, and the accuracy of the obtained prediction results is also relatively low. Summary of the invention

[0005] The embodiments of the present application provide a rail transit fault prediction method, device, equipment and storage medium, which can solve the problem of low rail transit fault prediction accuracy, improve the degree of automation of fault prediction, and improve prediction accuracy.

[0006] In a first aspect, an embodiment of the present application provides a rail transit fault prediction method, comprising:

[0007] Acquire rail transit operation log data, and classify and predict the operation log data through a preset classification model to obtain a corresponding prediction mode category;

[0008] Comparing the predicted mode category with the preset target mode category to confirm the matching status between the predicted mode category and the preset target mode category;

[0009] If the predicted mode category is the same as one of the preset target mode categories, then judging that the operation log data is normal;

[0010] If the predicted mode category is different from the preset target mode category, the operation log data is judged to be abnormal, and the deviation rate is calculated according to the predicted mode category and a preset probability set, wherein the preset probability set includes all mode categories and corresponding probabilities;

[0011] The probability that a fault exists in the operation process corresponding to the operation log data is determined according to the deviation rate.

[0012] Further, the calculating the deviation rate according to the prediction mode category and the preset probability set includes:

[0013] Obtaining the prediction probability corresponding to the prediction mode category in the preset probability set;

[0014] The deviation rate is obtained according to the difference between the total probability and the predicted probability.

[0015] Furthermore, the operation log data is classified and predicted by a preset classification model to obtain a corresponding prediction mode category, including:

[0016] Inputting the operation log data into a preset classification model, and performing word segmentation processing on the operation log data in the preset classification model to obtain word sequence data;

[0017] Extracting a target template according to the word order data, the target template including at least one historical pattern category;

[0018] The historical pattern category in the target template is input into the preset LSTM model, and the corresponding prediction pattern category is output.

[0019] Furthermore, before the operation log data is classified and predicted by a preset classification model, the following steps are included:

[0020] Acquire training log data, perform word segmentation processing on the training log data, and obtain word sequence data;

[0021] extracting a template according to the word order data, the template comprising at least one historical pattern category;

[0022] Input the historical pattern category into the preset LSTM model and output the corresponding pattern category

[0023] Perform data processing on all obtained pattern categories to obtain the probability corresponding to each pattern category;

[0024] The preset probability set is obtained according to the pattern category and the corresponding probability.

[0025] Furthermore, the data processing of all obtained pattern categories to obtain the probability corresponding to each pattern category includes:

[0026] For all pattern categories output by the preset LSTM model, data processing is performed through softmax to obtain the probability corresponding to each pattern category.

[0027] Further, after obtaining the preset probability set according to the mode category and the corresponding probability, the method further comprises:

[0028] Sorting the pattern categories in the preset probability set from large to small according to probability;

[0029] A mode category corresponding to a preset sequence number is obtained as the preset target mode category.

[0030] Further, after calculating the deviation rate according to the prediction mode category and the preset probability set, the method further comprises:

[0031] Sort the corresponding deviation rates of all abnormal log data from large to small to obtain the sorting result;

[0032] The sorting results, the corresponding prediction mode categories and the corresponding operation log data identifiers are displayed in the corresponding display interface.

[0033] In a second aspect, an embodiment of the present application provides a rail transit fault prediction device, comprising:

[0034] A classification module is used to obtain rail transit operation log data, and classify and predict the operation log data through a preset classification model to obtain a corresponding prediction mode category;

[0035] An abnormality judgment module, used for comparing the predicted mode category with the preset target mode category, and confirming the matching status between the predicted mode category and the preset target mode category;

[0036] If the predicted mode category is the same as one of the preset target mode categories, then judging that the operation log data is normal;

[0037] If the predicted mode category is different from the preset target mode category, the operation log data is judged to be abnormal, and the deviation rate is calculated according to the predicted mode category and a preset probability set, wherein the preset probability set includes all mode categories and corresponding probabilities;

[0038] A prediction module is used to determine the probability of a fault existing in the operation process corresponding to the operation log data according to the deviation rate.

[0039] Furthermore, the abnormality judgment module is further used to obtain the prediction probability corresponding to the prediction mode category in the preset probability set;

[0040] The deviation rate is obtained according to the difference between the total probability and the predicted probability.

[0041] Furthermore, the classification module is further used to input the operation log data into a preset classification model, and perform word segmentation processing on the operation log data in the preset classification model to obtain word sequence data;

[0042] Extracting a target template according to the word order data, the target template including at least one historical pattern category;

[0043] The historical pattern category in the target template is input into the preset LSTM model, and the corresponding prediction pattern category is output.

[0044] Furthermore, the device also includes a training module;

[0045] The training module is used to obtain training log data, perform word segmentation processing on the training log data, and obtain word sequence data;

[0046] extracting a template according to the word order data, the template comprising at least one historical pattern category;

[0047] Input the historical pattern category into the preset LSTM model and output the corresponding pattern category

[0048] Perform data processing on all obtained pattern categories to obtain the probability corresponding to each pattern category;

[0049] The preset probability set is obtained according to the pattern category and the corresponding probability.

[0050] Furthermore, the training module is also used to perform data processing on all pattern categories output by the preset LSTM model through softmax to obtain the probability corresponding to each pattern category.

[0051] Furthermore, the device also includes a preset target mode category module;

[0052] The preset target pattern category module is used to sort the pattern categories in the preset probability set from large to small according to probability;

[0053] A mode category corresponding to a preset sequence number is obtained as the preset target mode category.

[0054] Furthermore, the device also includes a display module;

[0055] The display module is used to sort the corresponding deviation rates of all abnormal log data from large to small to obtain a sorting result;

[0056] The sorting results, the corresponding prediction mode categories and the corresponding operation log data identifiers are displayed in the corresponding display interface.

[0057] In a third aspect, an embodiment of the present application provides a rail transit fault prediction device, including:

[0058] memory and one or more processors;

[0059] The memory is used to store one or more programs;

[0060] When the one or more programs are executed by the one or more processors, the one or more processors implement the rail transit fault prediction method as described in the first aspect.

[0061] In a fourth aspect, an embodiment of the present application provides a storage medium storing computer executable instructions, which, when executed by a computer processor, are used to execute the rail transit fault prediction method as described in the first aspect.

[0062] The embodiment of the present application obtains the corresponding prediction mode category by performing classification prediction processing of the preset classification model on the operation log data of the rail transit, compares the prediction mode category with the preset target mode category, confirms the matching status of the prediction mode category and the preset target mode category, if the prediction mode category is the same as one of the preset target mode categories, then the corresponding operation log data is judged to be normal, if the prediction mode category and the preset target mode category are not the same, then the corresponding operation log data is judged to be abnormal, and the deviation rate is calculated according to the prediction mode category and the preset probability set, and the probability of the existence of a fault in the operation process corresponding to the corresponding operation log data is determined according to the deviation rate. By adopting the above technical means, the operation log data of the rail transit can be classified and predicted by the preset classification model to obtain the corresponding prediction mode category, and the corresponding operation log data can be determined whether it is abnormal by the matching status of the prediction mode category and the preset target mode, so as to realize automatic fault prediction, reduce the complexity of the prediction work, and make the prediction work more convenient and quick. In addition, when the predicted operation log data is abnormal, the deviation rate is calculated by the prediction mode category and the preset probability set, and the probability of the existence of a fault in the corresponding operation process is determined by the deviation rate, so that the fault prediction is quantified, thereby improving the accuracy of the fault prediction result. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 It is a flow chart of a rail transit fault prediction method provided by an embodiment of the present application;

[0064] Figure 2 This is a first schematic diagram of the prediction process provided by the embodiment of the present application;

[0065] Figure 3 is a schematic diagram of the FT-tree algorithm provided in an embodiment of the present application;

[0066] Figure 4 It is a schematic diagram of the LSTM model provided in the embodiment of the present application;

[0067] Figure 5 is a second schematic diagram of the prediction process provided by an embodiment of the present application;

[0068] Figure 6 is a third schematic diagram of the prediction process provided by the embodiment of the present application;

[0069] Figure 7 It is a structural schematic diagram of a rail transit fault prediction device provided in an embodiment of the present application;

[0070] Figure 8 It is a structural schematic diagram of a rail transit fault prediction device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0071] In order to make the purpose, technical scheme and advantages of the present application clearer, the specific embodiments of the present application are further described in detail below in conjunction with the accompanying drawings. It is understood that the specific embodiments described herein are only used to explain the present application, rather than to limit the present application. It should also be noted that, for the convenience of description, only the part related to the present application but not all the contents are shown in the accompanying drawings. Before discussing the exemplary embodiments in more detail, it should be mentioned that some exemplary embodiments are described as processes or methods depicted as flow charts. Although the flow chart describes each operation (or step) as a sequential process, many of the operations therein can be implemented in parallel, concurrently or simultaneously. In addition, the order of each operation can be rearranged. The process can be terminated when its operation is completed, but it can also have additional steps not included in the accompanying drawings. The process can correspond to a method, a function, a procedure, a subroutine, a subprogram, etc.

[0072] The fault prediction method, device, equipment and storage medium of rail transit provided by the present application are intended to classify and predict the operation log data of rail transit through a preset classification model to obtain the corresponding prediction mode category during fault prediction, and determine whether the corresponding operation log data is abnormal through the matching state of the prediction mode category and the preset target mode, so as to realize automatic fault prediction and reduce the complexity of the prediction work. In addition, when the corresponding operation log data is predicted to be abnormal, the deviation rate is calculated by the prediction mode category and the preset probability set, and the probability of the corresponding operation process being faulty is determined by the deviation rate, so that the fault prediction is quantified to improve the accuracy of the fault prediction result. According to the operation and maintenance experience, the cause of the fault is often hidden in the log data, so the corresponding fault prediction can be made according to the log data. The fault prediction method includes two methods: prediction based on keyword matching and prediction based on configuration parsing rules. Among them, the prediction method based on keyword matching is to match the fixed keywords given, detect whether there are corresponding keywords in the log data, and thus determine whether the log data is abnormal. This method based on keyword matching prediction is difficult to give corresponding keywords when it comes to different types of log data, so that corresponding keywords need to be set for different types of log data, which increases the difficulty and workload of prediction. In addition, even if the defined keywords appear, it does not necessarily mean that there is a fault problem in the log data. This depends on the developer's habits, which increases the uncertainty of fault prediction. The prediction method based on configuration parsing rules requires specifying different parsing rules according to different types of log data. The parsing work is very complicated and requires a lot of time for debugging. Based on this, a fault prediction method for rail transit in an embodiment of the present application is provided to solve the problem of low prediction accuracy in the fault prediction process of existing rail transit.

[0073] Rail transit refers to a type of transportation vehicle or transportation system in which operating vehicles need to travel on specific tracks. The most typical rail transit is a railway system consisting of traditional trains and standard railways. With the diversified development of train and railway technology, rail transit has shown more and more types, not only covering long-distance land transportation, but also widely used in medium and short-distance urban public transportation. Common rail transit includes traditional railways (national railways, intercity railways and urban railways), subways, light rails and trams. New rail transit includes magnetic levitation rail systems, monorail systems (straddle rail systems and suspended rail systems) and automatic passenger rapid transit systems.

[0074] This embodiment is explained by taking the subway as an example. The integrated monitoring system plays a very important role in the operation of the subway. If the problems of the integrated monitoring system itself cannot be discovered in time, it will not only affect the passenger experience, but also affect the safety of the subway itself.

[0075] Figure 1 A flow chart of a rail transit fault prediction method provided in an embodiment of the present application is given. The rail transit fault prediction method provided in this embodiment can be executed by a rail transit fault prediction device, which can be implemented in software and / or hardware. The rail transit fault prediction device can be composed of two or more physical entities, or can be composed of one physical entity. Generally speaking, the rail transit fault prediction device can be a terminal device, such as a computer device, etc.

[0076] The following description is made by taking a computer device as the main body of the rail transit fault prediction method as an example. Figure 1 , the rail transit fault prediction method specifically includes:

[0077] S101, obtaining rail transit operation log data, and performing classification and prediction processing on the operation log data through a preset classification model to obtain a corresponding prediction mode category.

[0078] In the integrated monitoring system of rail transit, for log data with large volume and various formats, log anomaly prediction through machine learning, especially unsupervised real-time prediction and rapid fault location, is particularly important for the integrated monitoring of rail transit. The integrated monitoring system in rail transit is composed of various functional modules (Agents). The log data generated by each functional module affects each other. Therefore, the cause of the fault may also be caused by problems in other functional modules. Therefore, it is necessary to predict the anomaly of the operation log data of multiple modules, so as to determine which functional module is most likely to cause the problem, help quickly locate the problem, and reduce the pressure of operation and maintenance.

[0079] Figure 2 This is the first schematic diagram of the prediction process provided by the embodiment of the present application, referring to Figure 2Considering that the integrated monitoring system of rail transit is relatively large and complex, it is impossible to fully display the prediction process of all functional modules. Therefore, in this embodiment, one of the broadcast function modules and related modules is selected to perform a process description of fault prediction. The operation log data corresponding to the broadcast function module and related modules include paManager operation log data, paAgent operation log data and Iotgtw_pa operation log data. Obtain paManager operation log data, paAgent operation log data and Iotgtw_pa operation log data, and use the prediction model to predict and process these operation log data, and output the corresponding prediction results. If the prediction result is abnormal, the corresponding deviation rate is output at the same time, and the order is sorted according to the deviation rate. Generally speaking, the higher the deviation rate, the higher the probability of failure. The corresponding abnormal results and deviation rates are output to the display interface for the operation and maintenance staff to understand the specific prediction results, so that they can quickly locate and handle problems according to the prediction results.

[0080] Obtain the operation log data that needs to be predicted, input the operation log data into the preset classification model, summarize in the preset classification model, perform word segmentation on the operation log data, and obtain word sequence data. Extract the target template according to the word sequence data, and the target template includes at least one historical pattern category. Input the historical pattern category in the extracted target template into the preset LSTM model, and output the corresponding prediction pattern category. By extracting the log template according to the operation log data, and performing prediction processing through the LSTM model to obtain the corresponding prediction pattern category, fault prediction is performed through the pattern category, and fault prediction is performed based on the integrity of the log data to improve the accuracy of the prediction.

[0081] Exemplarily, the operation log data is segmented, and a complete operation log data is divided into a sequence of several words. Word segmentation methods include word segmentation methods based on string matching, word segmentation methods based on understanding, and word segmentation methods based on statistics. From the perspective of whether it is combined with the part-of-speech tagging process, word segmentation processing also includes an integrated method combining a simple word segmentation method and tagging. The format of the massive amount of operation log data in the process of rail transit operation is ever-changing and cannot be analyzed by tagging. Therefore, the word segmentation processing method provided in this embodiment is a simple word segmentation method.

[0082] For example, the operation log data is often very large, and tens of thousands of log data can be generated in one day. By analyzing the log pattern category, the log data can be compressed into hundreds or dozens of templates. The compressed templates serve the subsequent prediction pattern category. This embodiment provides an implementation method for template extraction using the FT-TREE algorithm. Figure 3is a schematic diagram of the FT-tree algorithm provided in the embodiment of the present application, refer to Figure 3 , assuming that the received operation log data is DM = (M1, M2, ..., M n ) is a collection of log messages, each M i Represents a log message. I=(a1,a2,…,a m ) is a combination of different words that appear in the log message. A word combination a i The higher the frequency of occurrence, the more it is a template. The second column of the following table shows the set of log messages DM = (M1, M2, ..., M n ), where each log is of type "SIF".

[0083]

[0084] When extracting the template, first, the FT-TREE algorithm performs a first scan on the DM and obtains a list in descending order of the frequency of each occurrence, as shown in the table above. After that, the root point of the tree is created, which is marked by the information type, such as Figure 3 As shown in the first step, in this embodiment, the root node of the tree is "SIF". After that, the FT-TREE algorithm performs a second scan on DM. By processing M1, the first path / branch of the tree is obtained, and these words are sorted according to the order of the words in L. When processing M2, because the sorted word list of M2 shares a common prefix with the existing path / branch, the FT-TREE algorithm needs to create a new path / branch as a subtree of the node "to". Similarly, the remaining 6 system log messages are processed in the same way. Finally, the obtained FT-TREE is as follows Figure 3 Finally, prune the FT-TREE until it satisfies the constraint that the number of nodes is below the target number of nodes. Figure 3 As shown, each log message type should have only a small number of subtypes, and for each subtype, there should be many different system log messages matching it. Therefore, if a node in FT-TREE has too many child nodes, all of its child nodes are deleted from FT-TREE. As a result, the child node becomes a leaf node, and the nodes remaining in FT-TREE represent constants. Then, FT-TREE considers log data containing the same constant to belong to the same pattern category. The template definition of the final log data is shown in the following table:

[0085]

[0086] DM log data consists of multiple templates, each of which includes a pattern category (pattern number / pattern value) and a template constant part, and each log message can find the corresponding pattern category in the pattern table.

[0087] It should be noted that when the software is upgraded, the log data generated will be different. At this time, a new template needs to be generated to match these new sub-types of system log messages. At this time, this can be done by inserting new nodes into the original FT-TREE without having to rebuild a new FT-TREE.

[0088] Since the subtype of the detailed information field in the system log message is usually the longest combination of frequently occurring words, the template extraction method using the FT-TREE algorithm provided above is equivalent to identifying the longest combination of frequently occurring words from the system log message, supports incremental learning, and can better process streaming logs. Compared with most traditional template extraction methods (such as common clustering methods), FT-TREE can complete higher-precision extraction in linear time, and has obvious advantages when the amount of training sample data is large (tens of millions).

[0089] The implementation method of template extraction through the FT-TREE algorithm provided above obtains the template corresponding to each log data and the corresponding pattern category (pattern number / pattern value). Therefore, after obtaining the word sequence data after word segmentation processing, the method of template extraction through the FT-TREE algorithm provided above can obtain the target template corresponding to the operation log data to be predicted.

[0090] The implementation method of template extraction through the FT-TREE algorithm provided above can obtain the template corresponding to each log data and the corresponding pattern category (pattern number / pattern value), form a pattern table, and use the LSTM model to perform model training based on the obtained pattern table.

[0091] Exemplarily, an implementation method for model training using an LSTM model is provided. Figure 4 This is a schematic diagram of the LSTM model provided in the embodiment of the present application, refer to Figure 4 , RNN network is a network model used for time series data prediction, and LSTM model is a model that adds three gates to the standard RNN network: forget gate, input gate, and output gate, thereby avoiding the gradient vanishing problem of RNN and better storing feature values. Therefore, this embodiment selects LSTM model as the classification prediction model, which is as follows Figure 4 As shown, Figure 4It belongs to one layer of the network. The subsequent layers are the same and can be stacked in sequence. Using t-1 log messages of length h, (t-1) pattern values ​​(pattern categories) can be obtained, forming a pattern value stream that can be used as the input of the network, and the pattern value of the tth position is used as the output of the network, and the network is trained to obtain the LSTM model. For example: 2 log messages of length 3 (sta The), according to the 2 pattern values ​​of the 2 log messages of length 3, the corresponding third pattern value of length 3 can be obtained by the above FT-TREE algorithm, so that the 2 pattern values ​​of the 2 log messages of length 3 are used as input, and the corresponding third pattern value of length 3 obtained by the above FT-TREE algorithm is used as output to train the LSTM model. After determining the corresponding model parameters, the trained preset LSTM model can be obtained.

[0092] After obtaining the preset LSTM model, the target template corresponding to the predicted operation log data obtained in the above implementation mode is input into the preset LSTM model, and the corresponding prediction pattern category is output through the historical pattern category in the target template.

[0093] The LSTM model is a neural network. Compared with traditional machine learning methods, it has a stronger ability to learn features, strong robustness and fault tolerance, and can fully approximate any complex nonlinear relationship, thereby making the prediction more accurate. In this embodiment, the LSTM model predicts the possible situation of n based on the data before n-1, and has obvious time characteristics.

[0094] S102: Compare the prediction mode category with a preset target mode category to confirm a matching status between the prediction mode category and the preset target mode category.

[0095] In order to determine whether the log data corresponding to the corresponding pattern category is abnormal, a corresponding reference target pattern category is required for comparison and determination. By obtaining training log data, the training log data can be historical operation log data or simulated training log data. The training log data is segmented to obtain word order data. A template is extracted according to the word order data and the above-mentioned FT-TREE algorithm, and the template includes at least one historical pattern category. It should be noted that the template obtained here is a template corresponding to each log message in the training log data, and each template includes at least one historical pattern category. Each template is input into a preset LSTM model, and in the preset LSTM model, the corresponding pattern category is output according to the corresponding historical pattern category. Therefore, the pattern category corresponding to each log message in the training log data can be obtained to form a pattern table, and the pattern table includes all pattern categories. For all pattern categories output by the preset LSTM model, data processing is performed through softmax to obtain the probability corresponding to each pattern category. The preset probability set is obtained according to the pattern category and the corresponding probability. The pattern categories in the preset probability set are sorted from large to small according to probability, and the pattern category corresponding to the preset sorting number is obtained as the preset target pattern category.

[0096] Exemplarily, the FT-TREE algorithm provided in the above embodiment can obtain all the pattern values ​​(pattern categories) corresponding to the training log data and obtain a pattern table. Let K = (k1, k2, ..., kn) represent the pattern table, then each log message can be parsed to obtain a corresponding pattern value K through the constant part of the log template. Then, we construct a window W = {w t-h , w t-h+1 ,…,w t-1} represents the sequence of pattern values ​​before position t, then position t can output any value in set K. Therefore, log anomaly prediction based on pattern values ​​can be transformed into a multi-classification problem (n classification). Each value in K is regarded as a separate class. Through the historical pattern value sequence of position t, the conditional probability of each class in K appearing at position t is calculated, and p(y t =x i |w). The total probability of each pattern value in K is sorted from high to low to obtain a pattern value probability set table (preset probability set) (that is, the probability of a single pattern value appearing in this type of log data) for subsequent fault prediction.

[0097] In one embodiment, t-1 historical log messages of operation log data with a length of h are subjected to template extraction through the above-mentioned FT-TREE algorithm to obtain corresponding pattern values ​​(pattern categories), thereby forming a pattern value stream with a number of (t-1), and the obtained (t-1) pattern value streams are input into the trained preset LSTM model to obtain an n-classification, and then the softmax function is used to normalize the probability, and the softmax function is as follows:

[0098] o t =soffmax(Vh t +b v )

[0099] Among them, h t represents the output of the previous layer, the input of the next layer (here is the last layer), V represents the weight vector of this layer, b v Represents the bias vector.

[0100] Thus, the probability of occurrence of each mode value (mode category) can be obtained, which is calculated by the following formula:

[0101]

[0102] Among them, C represents the input history sequence (operation log data), x t represents the target mode value (target mode category), V′(x t ) represents the output word vector of the softmax layer, and V′ is the derivative of V. Each pattern value (pattern category) is regarded as the target pattern value (target pattern category) and calculated once, so that the probability value corresponding to each pattern value (pattern category) can be obtained, thereby obtaining the corresponding preset probability set. The pattern values ​​(pattern categories) in the preset probability set are sorted from large to small according to probability, and the pattern category corresponding to the preset sort number is obtained as the preset target pattern category.

[0103] Exemplarily, assume that there are pattern categories: pattern 1, pattern 2, pattern 3 and pattern 4. The corresponding preset probability set obtained by the above-mentioned probability calculation method is {pattern value 1: 40%, pattern value 2: 30%, pattern value 3: 20%. Pattern value 4: 10%}. Then arrange the probability values ​​in the preset probability set from large to small, and select the first b to form set B, which is the preset target pattern category set. Assuming that the first 2 pattern values ​​(pattern categories) are selected as the target pattern values ​​(pattern categories), the B set is {pattern value 1, pattern value 2}.

[0104] It should be noted that the specific value of the preset sort number (b) in the above embodiment is only for illustration, and the specific value of the preset sort number (b) corresponding to the preset target mode category can be set according to actual conditions and is not limited in this embodiment.

[0105] From the above implementation, the prediction mode category can be obtained, and the obtained prediction mode category is compared with the preset target mode category in set B to confirm the matching status of the prediction mode category and the preset target mode category. The probability is normalized by the softmax function so that the classification corresponding to each mode category is quantified, so that the subsequent judgment of whether there is a fault is more accurate.

[0106] S103: If the prediction mode category is the same as one of the preset target mode categories, it is determined that the operation log data is normal.

[0107] According to the above implementation, the prediction mode category can be obtained, and the obtained prediction mode category is compared with the preset target mode category in the B set to confirm the matching status of the prediction mode category and the preset target mode category. If the prediction mode category is the same as one of the preset target mode categories, it is judged that the corresponding operation log data is normal and there is no abnormality in the corresponding operation process.

[0108] Exemplarily, from the above embodiment, it can be obtained that the B set is {mode value 1, mode value 2}, and the corresponding preset target mode categories are mode 1 and mode 2. Assuming that the obtained prediction mode category is mode 2, the prediction mode category is the same as one of the preset target mode categories, then the corresponding predicted operation log data is judged to be normal.

[0109] The FT-TREE algorithm extracts templates, which makes it possible to train the LSTM neural network. Because the first premise of neural network training is that there must be input and output, and for fault prediction, the output is of course normal or abnormal, so the data required for training needs to have both normal and abnormal data, and the number of data for the two cannot differ too much, otherwise it will lead to overfitting (that is, biased towards normal or abnormal situations). However, for comprehensive monitoring systems, most of the log data obtained are normal, and there are few abnormalities, so this common training method is a bit unsuitable. Combined with the FT_tree algorithm, the number of pattern values ​​in the template extracted by the FT-tree is used as the output, and the pattern value stream is used as the input to train the LSTM model, thereby obtaining a prediction model (all trained with normal data), and then when using the LSTM model for prediction, the B set method is used to determine whether it is abnormal. In this way, the problem caused by the imbalance of normal and abnormal data can be avoided.

[0110] S104, if the predicted mode category is different from the preset target mode category, the operation log data is judged to be abnormal, and a deviation rate is calculated according to the predicted mode category and a preset probability set, wherein the preset probability set includes all mode categories and corresponding probabilities;

[0111] From the above implementation, the prediction mode category can be obtained, and the obtained prediction mode category is compared with the preset target mode category in the B set to confirm the matching status of the prediction mode category and the preset target mode category. If the prediction mode category and the preset target mode category are not the same, the corresponding operation log data is judged to be abnormal. Assuming that the prediction mode category is mode 4, based on the fact that mode 4 is different from the preset target mode category of the B set {mode value 1, mode value 2}, it is judged that the operation log data corresponding to mode 4 is abnormal, and the corresponding operation process is abnormal.

[0112] After predicting that there are abnormalities in the operation log data, based on the possibility that there are many abnormal operation log data, in order to prioritize the abnormal operation log data, a method is provided to characterize the probability of a fault by calculating the corresponding deviation rate based on the probability corresponding to each pattern category. The probability value corresponding to each pattern value (pattern category) can be obtained by the above-mentioned softmax function, thereby obtaining the corresponding preset probability set. Therefore, according to the preset probability set, the predicted probability value corresponding to the corresponding predicted pattern category can be obtained, and the deviation rate can be obtained according to the difference between the total probability value of the preset probability set and the predicted probability value. The higher the deviation rate, the higher the probability of the fault.

[0113] Exemplarily, assuming that the prediction mode category is mode 4, based on the fact that mode 4 is different from the preset target mode category of the B set of {mode value 1, mode value 2}, it is determined that the operation log data corresponding to mode 4 is abnormal, and the corresponding operation process is abnormal. The preset probability set is {mode value 1: 40%, mode value 2: 30%, mode value 3: 20%. Mode value 4: 10%}, then according to the preset probability set, the prediction probability value corresponding to the prediction mode category (mode 4) can be obtained as 10%, then according to the difference between the total probability (100%) and the prediction probability value (10%), the corresponding deviation rate is 90%.

[0114] S105. Determine, according to the deviation rate, the probability that a fault exists in the operation process corresponding to the operation log data.

[0115] The higher the deviation rate, the higher the probability of failure. The probability of failure in the operation process corresponding to the corresponding operation log data can be directly represented according to the numerical value of the deviation rate.

[0116] Figure 5 This is a second schematic diagram of the prediction process provided in the embodiment of the present application, referring to Figure 5, which is divided into a training process and a prediction process. The training process is to obtain historical log data, pre-process the historical log data, and obtain training log data. Perform template extraction on the training log data to obtain the corresponding input sequence data, which is the pattern value stream data. Input the input sequence data into the LSTM neural network model for training to obtain the corresponding trained LSTM neural network model. The training outputs all corresponding pattern values ​​(categories), and processes them through the softmax function to obtain a preset probability set table of pattern values. In the prediction process, real-time log data is obtained, and the real-time log data can be the actual operation log data of rail transit. The obtained real-time log data is pre-processed, and the pre-processed real-time log data is input into the trained LSTM neural network model, and the corresponding predicted pattern value (prediction pattern category) is output. The predicted probability value is calculated by the softmax function, and combined with the preset probability set table of pattern values, it is judged whether the predicted pattern value is normal.

[0117] Figure 6 This is a third schematic diagram of the prediction process provided in the embodiment of the present application, referring to Figure 6 , perform fault prediction on three log data, the three log data are paManager_info.log, paAgent_control.log and iotgtw_control.log, input the three log data into the classification prediction model and output the corresponding prediction mode category A, prediction mode category B and prediction mode category C respectively. Determine whether the obtained prediction mode category A, prediction mode category B and prediction mode category C are in the preset target mode category set B set. If the prediction mode category is in the B set, it means that the log data corresponding to the corresponding prediction mode category is normal. If the prediction mode category is not in the B set, it means that the log data corresponding to the corresponding prediction mode is abnormal, and then calculate the corresponding deviation rate. Assume that the deviation rate of paManager_info.log is calculated to be 90%, and the deviation rate of paAgent_control.log is 70%. Then the calculated deviation rate is sorted and displayed from high to low on the corresponding display interface according to the deviation rate, so as to prompt the operation and maintenance personnel to prioritize the maintenance of the higher deviation rate according to the deviation rate, so that the operation and maintenance personnel can quickly locate the optimal log data that may have problems, saving the time of fault prediction and improving the accuracy of fault prediction.

[0118] In the above, the log data generated by the integrated monitoring system is analyzed by combining the FT-TREE algorithm and the LSTM neural network algorithm, wherein the FT-TREE algorithm extracts log templates from log messages, and then uses the LSTM neural network algorithm for feature extraction, uses the prediction model to predict whether the next step of the historical data will be a fault, and displays the fault situation on the interface according to the deviation rate, so as to determine the module that is most likely to cause the problem, and help the operation and maintenance personnel to quickly locate the problem. The above technical means can better process streaming logs through the FT-TREE algorithm, extract templates with higher accuracy in linear time, and support incremental learning. As the number of logs increases, it does not need to be re-predicted, and only needs to use new log data for dynamic learning, which improves the flexibility of prediction and thus improves the work efficiency of fault prediction. In addition, the deviation rate of each log data is calculated using the results of the classification prediction model and the preset pattern probability table, and is dynamically displayed on the interface according to the deviation rate, so that the operation and maintenance personnel can start troubleshooting from the most likely log data, which speeds up the efficiency of locating fault problems.

[0119] In the above, the corresponding prediction mode category is obtained by performing classification prediction processing of the preset classification model on the operation log data of the rail transit, and the prediction mode category is compared with the preset target mode category to confirm the matching status of the prediction mode category and the preset target mode category. If the prediction mode category is the same as one of the preset target mode categories, the corresponding operation log data is judged to be normal. If the prediction mode category and the preset target mode category are not the same, the corresponding operation log data is judged to be abnormal, and the deviation rate is calculated according to the prediction mode category and the preset probability set, and the probability of the existence of a fault in the operation process corresponding to the corresponding operation log data is determined according to the deviation rate. By adopting the above technical means, the operation log data of the rail transit can be classified and predicted by the preset classification model to obtain the corresponding prediction mode category, and the corresponding operation log data can be determined whether it is abnormal by the matching status of the prediction mode category and the preset target mode, which can realize automatic fault prediction, reduce the complexity of the prediction work, and make the prediction work more convenient and quick. In addition, when the predicted operation log data is abnormal, the deviation rate is calculated by the prediction mode category and the preset probability set, and the probability of the existence of a fault in the corresponding operation process is determined by the deviation rate, so that the fault prediction is quantified, thereby improving the accuracy of the fault prediction result.

[0120] Based on the above embodiments, Figure 7 This is a schematic diagram of the structure of a rail transit fault prediction device provided in an embodiment of the present application. Figure 7 The rail transit fault prediction device provided in this embodiment specifically includes: a classification module 21, an abnormality judgment module 22 and a prediction module 23.

[0121] The classification module 21 is used to obtain the operation log data of the rail transit, and classify and predict the operation log data through a preset classification model to obtain a corresponding prediction mode category;

[0122] An abnormality judgment module 22, used for comparing the predicted mode category with the preset target mode category, and confirming the matching status between the predicted mode category and the preset target mode category;

[0123] If the predicted mode category is the same as one of the preset target mode categories, then judging that the operation log data is normal;

[0124] If the predicted mode category is different from the preset target mode category, the operation log data is judged to be abnormal, and the deviation rate is calculated according to the predicted mode category and a preset probability set, wherein the preset probability set includes all mode categories and corresponding probabilities;

[0125] The prediction module 23 is used to determine the probability of a fault existing in the operation process corresponding to the operation log data according to the deviation rate.

[0126] Furthermore, the abnormality judgment module 22 is also used to obtain the prediction probability corresponding to the prediction mode category in the preset probability set;

[0127] The deviation rate is obtained according to the difference between the total probability and the predicted probability.

[0128] Furthermore, the classification module 21 is further used to input the operation log data into a preset classification model, and perform word segmentation processing on the operation log data in the preset classification model to obtain word sequence data;

[0129] Extracting a target template according to the word order data, the target template including at least one historical pattern category;

[0130] The historical pattern category in the target template is input into the preset LSTM model, and the corresponding prediction pattern category is output.

[0131] Furthermore, the device also includes a training module;

[0132] The training module is used to obtain training log data, perform word segmentation processing on the training log data, and obtain word sequence data;

[0133] extracting a template according to the word order data, the template comprising at least one historical pattern category;

[0134] Input the historical pattern category into the preset LSTM model and output the corresponding pattern category

[0135] Perform data processing on all obtained pattern categories to obtain the probability corresponding to each pattern category;

[0136] The preset probability set is obtained according to the pattern category and the corresponding probability.

[0137] Furthermore, the training module is also used to perform data processing on all pattern categories output by the preset LSTM model through softmax to obtain the probability corresponding to each pattern category.

[0138] Furthermore, the device also includes a preset target mode category module;

[0139] The preset target pattern category module is used to sort the pattern categories in the preset probability set from large to small according to probability;

[0140] A mode category corresponding to a preset sequence number is obtained as the preset target mode category.

[0141] Furthermore, the device also includes a display module;

[0142] The display module is used to sort the corresponding deviation rates of all abnormal log data from large to small to obtain a sorting result;

[0143] The sorting results, the corresponding prediction mode categories and the corresponding operation log data identifiers are displayed in the corresponding display interface.

[0144] In the above, the corresponding prediction mode category is obtained by performing classification prediction processing of the preset classification model on the operation log data of the rail transit, and the prediction mode category is compared with the preset target mode category to confirm the matching status of the prediction mode category and the preset target mode category. If the prediction mode category is the same as one of the preset target mode categories, the corresponding operation log data is judged to be normal. If the prediction mode category and the preset target mode category are not the same, the corresponding operation log data is judged to be abnormal, and the deviation rate is calculated according to the prediction mode category and the preset probability set, and the probability of the existence of a fault in the operation process corresponding to the corresponding operation log data is determined according to the deviation rate. By adopting the above technical means, the operation log data of the rail transit can be classified and predicted by the preset classification model to obtain the corresponding prediction mode category, and the corresponding operation log data can be determined whether it is abnormal by the matching status of the prediction mode category and the preset target mode, which can realize automatic fault prediction, reduce the complexity of the prediction work, and make the prediction work more convenient and quick. In addition, when the predicted operation log data is abnormal, the deviation rate is calculated by the prediction mode category and the preset probability set, and the probability of the existence of a fault in the corresponding operation process is determined by the deviation rate, so that the fault prediction is quantified, thereby improving the accuracy of the fault prediction result.

[0145] The rail transit fault prediction device provided in the embodiment of the present application can be used to execute the rail transit fault prediction method provided in the above embodiment, and has corresponding functions and beneficial effects.

[0146] The present application embodiment provides a rail transit fault prediction device, referring to Figure 8 The rail transit fault prediction device includes: a processor 31, a memory 32, a communication module 33, an input device 34 and an output device 35. The number of processors in the rail transit fault prediction device can be one or more, and the number of memories in the rail transit fault prediction device can be one or more. The processor, memory, communication module, input device and output device of the rail transit fault prediction device can be connected through a bus or other methods.

[0147] The memory 32, as a computer-readable storage medium, can be used to store software programs, computer executable programs and modules, such as the program instructions / modules corresponding to the fault prediction method of rail transit described in any embodiment of the present application (for example, the classification module, abnormal judgment module and prediction module in the fault prediction device of rail transit). The memory may mainly include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application required for at least one function; the storage data area may store data created according to the use of the device, etc. In addition, the memory 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 device. In some instances, the memory may further include a memory remotely arranged relative to the processor, and these remote memories may be connected to the device 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 a combination thereof.

[0148] The communication module 33 is used for data transmission.

[0149] The processor 31 executes various functional applications and data processing of the device by running the software programs, instructions and modules stored in the memory, that is, realizes the above-mentioned rail transit fault prediction method.

[0150] The input device 34 may be used to receive input digital or character information and generate key signal input related to user settings and function control of the device. The output device 35 may include a display device such as a display screen.

[0151] The rail transit fault prediction device provided above can be used to execute the rail transit fault prediction method provided in the above embodiment, and has corresponding functions and beneficial effects.

[0152] An embodiment of the present application also provides a storage medium storing computer executable instructions, which are used to execute a rail transit fault prediction method when executed by a computer processor. The rail transit fault prediction method includes: obtaining operation log data of the rail transit, and classifying and predicting the operation log data through a preset classification model to obtain a corresponding prediction mode category; comparing the prediction mode category with a preset target mode category to confirm the matching status of the prediction mode category and the preset target mode category; if the prediction mode category is the same as one of the preset target mode categories, the operation log data is judged to be normal; if the prediction mode category is different from the preset target mode category, the operation log data is judged to be abnormal, and the deviation rate is calculated according to the prediction mode category and a preset probability set, and the preset probability set includes all mode categories and corresponding probabilities; according to the deviation rate, the probability of a fault in the operation process corresponding to the operation log data is determined.

[0153] Storage medium - any of various types of memory devices or storage devices. The term "storage medium" is intended to include: installation media, such as CD-ROM, floppy disk or tape device; computer system memory or random access memory, such as DRAM, DDR RAM, SRAM, EDO RAM, Rambus RAM, etc.; non-volatile memory, such as flash memory, magnetic media (such as hard disk or optical storage); registers or other similar types of memory elements, etc. Storage media may also include other types of memory or combinations thereof. In addition, the storage medium may be located in the first computer system in which the program is executed, or may be located in a different second computer system, which is connected to the first computer system via a network (such as the Internet). The second computer system can provide program instructions to the first computer for execution. The term "storage medium" may include two or more storage media residing in different locations (for example, in different computer systems connected by a network). The storage medium may store program instructions (for example, embodied as a computer program) that can be executed by one or more processors.

[0154] Of course, the storage medium storing computer executable instructions provided in the embodiment of the present application, whose computer executable instructions are not limited to the rail transit fault prediction method described above, can also execute related operations in the rail transit fault prediction method provided in any embodiment of the present application.

[0155] The rail transit fault prediction device, storage medium and rail transit fault prediction equipment provided in the above embodiments can execute the rail transit fault prediction method provided in any embodiment of the present application. For technical details not described in detail in the above embodiments, please refer to the rail transit fault prediction method provided in any embodiment of the present application.

[0156] The above are only preferred embodiments of the present application and the technical principles used. The present application is not limited to the specific embodiments described herein, and various obvious changes, readjustments and substitutions that can be made by those skilled in the art will not deviate from the scope of protection of the present application. Therefore, although the present application is described in more detail through the above embodiments, the present application is not limited to the above embodiments, and may include more other equivalent embodiments without departing from the concept of the present application, and the scope of the present application is determined by the scope of the claims.

Claims

1. A rail transit fault prediction method, characterized in that: include: Acquire rail transit operation log data, and classify and predict the operation log data through a preset classification model to obtain a corresponding prediction mode category; Comparing the predicted mode category with the preset target mode category to confirm the matching status between the predicted mode category and the preset target mode category; If the predicted mode category is the same as one of the preset target mode categories, then judging that the operation log data is normal; If the predicted mode category is different from the preset target mode category, the operation log data is judged to be abnormal, and the deviation rate is calculated according to the predicted mode category and a preset probability set, wherein the preset probability set includes all mode categories and corresponding probabilities; Determining, according to the deviation rate, a probability that a fault exists in the operation process corresponding to the operation log data; The calculating the deviation rate according to the prediction mode category and the preset probability set includes: Obtaining the prediction probability corresponding to the prediction mode category in the preset probability set; The deviation rate is obtained according to the difference between the total probability and the predicted probability.

2. The method according to claim 1, characterized in that The operation log data is subjected to classification prediction processing through a preset classification model to obtain a corresponding prediction mode category, including: Inputting the operation log data into a preset classification model, and performing word segmentation processing on the operation log data in the preset classification model to obtain word sequence data; Extracting a target template according to the word order data, the target template including at least one historical pattern category; The historical pattern category in the target template is input into the preset LSTM model, and the corresponding prediction pattern category is output.

3. The method according to claim 1, characterized in that: Before the operation log data is classified and predicted by a preset classification model, the following steps are included: Acquire training log data, perform word segmentation processing on the training log data, and obtain word sequence data; extracting a template according to the word order data, the template comprising at least one historical pattern category; Input the historical pattern category into a preset LSTM model and output the corresponding pattern category; Perform data processing on all obtained pattern categories to obtain the probability corresponding to each pattern category; The preset probability set is obtained according to the pattern category and the corresponding probability.

4. The method according to claim 3, characterized in that The data processing of all obtained pattern categories to obtain the probability corresponding to each pattern category includes: For all pattern categories output by the preset LSTM model, data processing is performed through softmax to obtain the probability corresponding to each pattern category.

5. The method according to claim 3, characterized in that: After obtaining the preset probability set according to the mode category and the corresponding probability, the method includes: Sorting the pattern categories in the preset probability set from large to small according to probability; A mode category corresponding to a preset sequence number is obtained as the preset target mode category.

6. The method according to claim 1, characterized in that After calculating the deviation rate according to the prediction mode category and the preset probability set, the method includes: Sort the corresponding deviation rates of all abnormal log data from large to small to obtain the sorting result; The sorting results, the corresponding prediction mode categories and the corresponding operation log data identifiers are displayed in the corresponding display interface.

7. A rail transit fault prediction device, characterized in that: include: A classification module is used to obtain rail transit operation log data, and classify and predict the operation log data through a preset classification model to obtain a corresponding prediction mode category; An abnormality judgment module, used for comparing the predicted mode category with the preset target mode category, and confirming the matching status between the predicted mode category and the preset target mode category; If the predicted mode category is the same as one of the preset target mode categories, then judging that the operation log data is normal; If the predicted mode category is different from the preset target mode category, the operation log data is judged to be abnormal, and the deviation rate is calculated according to the predicted mode category and a preset probability set, wherein the preset probability set includes all mode categories and corresponding probabilities; A prediction module, used to determine the probability of a fault existing in the operation process corresponding to the operation log data according to the deviation rate; The abnormality judgment module is further used to obtain the prediction probability corresponding to the prediction mode category in the preset probability set; The deviation rate is obtained according to the difference between the total probability and the predicted probability.

8. A rail transit fault prediction device, characterized in that: include: memory and one or more processors; The memory is used to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 6.

9. A storage medium storing computer executable instructions, characterized in that: When the computer executable instructions are executed by a processor, they are used to perform the method according to any one of claims 1 to 6.

Images