A method and device for generating interpretable language for automobile abnormality detection results

Abstract

The present invention provides a method for generating an interpretable language for automobile abnormality detection results, comprising: acquiring abnormality detection database data, abnormal event description data, abnormality detection cause explanation and vehicle body signal real-time data and analyzing and processing them, obtaining abnormal data description, result description, cause description, state description and abnormal ID and inputting them into a preset first model for training, and obtaining abnormal context information through the trained first model; inputting first question information for the abnormality into the trained first model to obtain first context information of the first question; inputting the first question information and the first context information into a second model, training the second model, and obtaining first explanation information of the first question through the trained second model; associating and storing the abnormality, the abnormality ID and the first explanation information; receiving the second question information; the second question information includes the abnormality ID; and querying and obtaining target explanation information of the abnormality ID according to the abnormality ID.

Description

Technical Field

[0001] The present invention relates to the field of data processing, and in particular to a method and device for generating interpretable language of automobile abnormality detection results. Background Art

[0002] With the iterative development of automotive Internet of Things technology, while the popularization of intelligent applications brings convenience to drivers, various means of theft, attack, and information theft against vehicles are emerging in an endless stream. Many automobile companies have built multi-scenario security and anomaly detection algorithms in the security field. Traditional anomaly detection algorithms mainly rely on machine learning or time series deep learning for anomaly detection. The interpretability of abnormal data mostly relies on the statistical knowledge between features in numerical values. However, the semantic link relationship between multiple fields is not utilized.

[0003] To date, there has been an endless stream of research on the interpretability of machine learning models. The closest implementation to the present invention may be to use a local explanatory model to disassemble the model for inference. This type of method selects some data samples from the training set, and the samples are usually related to specific decisions or prediction results. Then, a specific algorithm is used to build a local model on the selected data samples to observe the model reasoning process to explain the model. This method can achieve partial decoupling of input data in model reasoning, but it is impossible to incorporate human prior knowledge of data and data features into the reasoning, so that prior knowledge is reflected in model training but not in the model reasoning stage.

[0004] The existing technology mainly has the following disadvantages:

[0005] (a) The manual intervention in data is too analytical and cannot be automated or generalized within the field.

[0006] (b) The interpretability of the model reasoning stage does not incorporate prior knowledge and cannot incorporate hard rules into model reasoning.

[0007] (c) Explainability is more specialized and the cost for non-professionals to understand is high. Summary of the invention

[0008] The purpose of the embodiment of the present invention is to provide a method for generating an interpretable language for automobile abnormality detection results to solve the problems existing in the prior art.

[0009] In a first aspect, the present invention provides a method for generating an interpretable language for an abnormality detection result of an automobile, the method comprising:

[0010] Obtain anomaly detection database data, abnormal event description data, anomaly detection cause explanation and vehicle body signal real-time data;

[0011] Analyze and process the abnormality detection database data, abnormal event description data, abnormality detection cause explanation and vehicle body signal real-time data to obtain abnormal data description, result description, cause description, status description and abnormality ID;

[0012] Inputting the data description, result description, cause description, status description and abnormal ID of the exception into a preset first model for training, and obtaining the context information of the exception through the trained first model;

[0013] Inputting first question information for the exception into the trained first model to obtain first context information of the first question;

[0014] Inputting the first question information and the first context information into a second model, training the second model, and obtaining first explanation information of the first question through the trained second model;

[0015] Associating and storing the exception, the exception ID and the first explanation information;

[0016] receiving second problem information; the second problem information including an abnormality ID;

[0017] According to the exception ID, the target explanation information of the exception ID is obtained by querying.

[0018] In a possible implementation, the abnormal detection database data, abnormal event description data, abnormal detection cause explanation and vehicle body signal real-time data are analyzed and processed to obtain abnormal data description, result description, cause description, status description and abnormal ID, which specifically includes:

[0019] The data description corresponds to the abnormal events in the abnormal detection database data, and the status description corresponds to the real-time sensor data of the vehicle body; the result description and cause description are the result description and cause description of the abnormal events and the real-time sensor data of the vehicle body; the abnormal ID is added to the abnormal data description, result description, cause description and status description.

[0020] In a possible implementation, the method further includes:

[0021] Obtain real-time data of vehicle body signals;

[0022] Determine whether the real-time data of the vehicle body signal is abnormal. When the real-time data of the vehicle body signal is abnormal, store the real-time data of the vehicle body signal, the real-time data of the vehicle body signal at the previous moment, and the real-time data of the vehicle body signal at the next moment; the real-time data of the vehicle body signal includes: vehicle speed, rotation speed, fuel quantity, temperature and battery voltage.

[0023] In a possible implementation, the inputting of the abnormal data description, result description, cause description, state description and abnormal ID into a preset first model for training, and obtaining the abnormal context information through the trained first model specifically includes:

[0024] According to the scene information, adjust the proportion of abnormalities corresponding to the scene information;

[0025] According to the adjusted abnormal proportion, the abnormal data description, result description, cause description, state description and abnormal ID are input into the preset first model for training.

[0026] In a possible implementation, before inputting the first question information and the first context information into the second model and training the second model, the method further includes:

[0027] Determining whether the amount of information of the context information meets the requirement;

[0028] When the amount of information of the context information does not meet the requirement, data enhancement is performed on the context information.

[0029] In a second aspect, the present invention provides a device for generating an interpretable language for an abnormality detection result of an automobile, characterized in that the device comprises:

[0030] An acquisition module, the acquisition module is used to acquire abnormality detection database data, abnormal event description data, abnormality detection cause explanation and vehicle body signal real-time data;

[0031] A processing module, the processing module is used to analyze and process the abnormality detection database data, abnormal event description data, abnormality detection cause explanation and vehicle body signal real-time data to obtain abnormal data description, result description, cause description, state description and abnormality ID;

[0032] A first training module, wherein the first training module is used to input the abnormal data description, result description, cause description, state description and abnormal ID into a preset first model for training, and obtain the abnormal context information through the trained first model;

[0033] A prediction module, wherein the prediction module is used to input first question information for anomalies into a trained first model to obtain first context information of the first question;

[0034] A second training module, the second training module is used to input the first question information and the first context information into a second model, train the second model, and obtain first explanation information of the first question through the trained second model;

[0035] A storage module, the storage module is used to associate and store the exception, the exception ID and the first explanation information;

[0036] A receiving module, the receiving module is used to receive second problem information; the second problem information includes an abnormality ID;

[0037] A query module is used to query and obtain target explanation information of the exception ID according to the exception ID.

[0038] In a possible implementation, the device further includes: a judgment module and an enhancement module;

[0039] The judging module is used to judge whether the amount of the context information meets the requirement;

[0040] The enhancement module is used to perform data enhancement on the context information when the amount of information of the context information does not meet the requirement.

[0041] In a third aspect, the present invention provides a computer server, comprising: a memory, a processor and a transceiver;

[0042] The processor is used to couple with the memory, read and execute instructions in the memory, so as to implement the method for generating an interpretable language for an abnormality detection result of an automobile according to any one of the first aspects;

[0043] The transceiver is coupled to the processor, and the processor controls the transceiver to send and receive messages.

[0044] In a fourth aspect, the present invention provides a chip system, comprising a processor, the processor being coupled to a memory, the memory storing program instructions, and the method for generating an interpretable language for automobile abnormality detection results as described in any one of the first aspects being implemented when the program instructions stored in the memory are executed by the processor.

[0045] In a fifth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, and the computer program is executed by a processor to implement the method for generating an interpretable language for automobile abnormality detection results as described in any one of claims 1 to 5.

[0046] By applying the method and device for generating interpretable language of automobile abnormality detection results provided by the present invention, the following technical effects are achieved:

[0047] (1) Explanation sources from more angles: This application is pre-trained with data description, result description, cause description, and state description. Through the semantic understanding and logical reasoning ability of the large language model, the anomaly detection results can be explained from a human perspective.

[0048] (2) As the scale of the first model gradually increases, an emergence effect will occur, and its ability to handle complex tasks will increase dramatically. Compared with the local explanatory model, it can process data with larger dimensions and has a stronger ability to discover hidden representations of data.

[0049] (3) The interpretability of the second model for anomaly detection can rely on pre-training text experience and instruction fine-tuning. Its knowledge, abnormal data judgment, and analysis capabilities can all be regulated during the training phase, without the need for regulation during the reasoning phase. It is also generalizable and can explain a variety of abnormal scenarios and abnormal results after training.

[0050] This application adds a large amount of anomaly detection database data, abnormal event description data, anomaly detection cause explanation data, and vehicle body signal data in the pre-training stage of the language model, and uses the language model to sample and analyze the abnormal data in prior knowledge, and analyze and describe the causes and possibilities. Its ability can provide rich case descriptions and knowledge reserves to explain the anomaly detection results, and it has a wide range of applications in vehicle protection safety and abnormal trigger reminders. It fully considers the actual needs and complexity of automobile anomaly detection, and has obvious advantages in the accuracy and reliability of anomaly detection, the generalization ability and controllability of the model, and the expansion of the scope of application. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 This is a flow chart of a method for generating an interpretable language for anomaly detection results of an automobile provided in the first embodiment of the present invention;

[0052] Figure 2 A second flow chart of a method for generating an interpretable language for anomaly detection results of an automobile provided in the first embodiment of the present invention;

[0053] Figure 3 This is a schematic diagram of the structure of a device for generating an interpretable language for anomaly detection results of an automobile provided in the second embodiment of the present invention;

[0054] Figure 4 The second structural diagram of the device for generating the interpretable language of the automobile abnormality detection result provided by the second embodiment of the present invention;

[0055] Figure 5 The third structural diagram of the device for generating the interpretable language of the automobile abnormality detection result provided in the second embodiment of the present invention;

[0056] Figure 6 A schematic diagram of the computer server structure provided in Embodiment 3 of the present invention;

[0057] Figure 7 A schematic diagram of the chip system structure provided in the fourth embodiment of the present invention;

[0058] Figure 8 This is a schematic diagram of the chip system structure provided in Example 5 of the present invention. DETAILED DESCRIPTION

[0059] The present application will be further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the relevant invention, rather than to limit the invention. It should also be noted that, for ease of description, only the parts related to the relevant invention are shown in the accompanying drawings.

[0060] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0061] Embodiment 1

[0062] Embodiment 1 of the present invention provides a method for generating an interpretable language for an automobile abnormality detection result. The execution subject of the present application is a server, or a system or device with a computing and processing function, such as Figure 1 As shown, the method comprises the following steps:

[0063] Step 110, acquiring abnormality detection database data, abnormal event description data, abnormality detection cause explanation and vehicle body signal real-time data;

[0064] Among them, the anomaly detection database data refers to the abnormal events detected by the anomaly detection algorithm that have been labeled and sorted. It includes the operation data of the car in various abnormal scenarios, the type of anomaly, and the timestamp of the abnormality. The anomaly detection database data includes abnormal events in algorithm knowledge, programming knowledge, vehicle system knowledge, and network security knowledge. These abnormal events provide a historical reference, enabling the large language model to understand what circumstances are considered abnormal.

[0065] In a specific example, an anomaly detection database data may be:

[0066] Timestamp: 2013-11-13, 08:15:32;

[0067] Abnormal detection point: engine temperature sensor;

[0068] Detected abnormal values: Temperature reading: 105°C;

[0069] Normal temperature range: 85℃-95℃.

[0070] Abnormal event description data describes each event marked as abnormal, including the specific situation in which the abnormality occurred, the vehicle operating parameters related to the abnormality, and any possible external factors. Through these descriptions, the big language model can gain a deeper understanding of the background and context of the abnormal event.

[0071] In a specific example, the abnormal event description data may be:

[0072] Event description: Engine overheat alarm;

[0073] Duration: 5 minutes

[0074] Impact: Engine performance decreases and the vehicle automatically slows down.

[0075] Anomaly detection reason explanation is a detailed explanation of the cause of each abnormal event. Each anomaly detection result is associated with one or more reasons, explaining why the event is considered abnormal. These reason explanations help large language models generate more accurate and insightful analysis.

[0076] In a specific example, the anomaly detection reason explanation may be:

[0077] Possible causes: Failure of the cooling system, resulting in poor coolant circulation;

[0078] Secondary reasons: long-term high-speed driving and high external temperature;

[0079] Recommended Action: Immediately slow down and check the cooling system, stop the vehicle and contact service if necessary.

[0080] Real-time vehicle body signal data comes from various sensors and systems of the vehicle, and records real-time information when abnormal events occur, including but not limited to vehicle speed, rotation speed, fuel level, temperature, etc. This type of data provides a real-time context for the large language model, which helps the model understand the specific context of the abnormal event.

[0081] In a specific implementation, the real-time data of the vehicle body signal may be:

[0082] Real-time data of vehicle body signals (2023-11-13 08:15:32)

[0083] Engine speed: 3000rpm;

[0084] Speed: 120km / h;

[0085] Coolant temperature: 90°C;

[0086] Oil pressure: 3.5bar;

[0087] Fuel quantity: 50%;

[0088] External temperature: 30℃;

[0089] Battery voltage 12.6V.

[0090] Among them, before step 110, it can also include: acquiring real-time data of vehicle body signals; judging whether the real-time data of vehicle body signals is abnormal, and when the real-time data of vehicle body signals is abnormal, storing the real-time data of vehicle body signals, the real-time data of vehicle body signals at the previous moment, and the real-time data of vehicle body signals at the next moment; the real-time data of vehicle body signals includes: at least one of vehicle speed, rotation speed, fuel quantity, temperature, and battery voltage.

[0091] Specifically, the real-time data of the vehicle body signals of multiple vehicles, such as vehicle speed, rotation speed, fuel quantity, temperature and battery voltage, and the timestamps corresponding to the vehicle speed, rotation speed, fuel quantity, temperature and battery voltage, can be obtained in real time, and the vehicle speed, rotation speed, fuel quantity, temperature and battery voltage are compared with the vehicle speed threshold, rotation speed threshold, fuel quantity threshold, temperature threshold and battery voltage threshold in turn. If it is greater than the corresponding threshold, it can be judged that the real-time data of the vehicle body signal is abnormal. When the real-time data of the vehicle body signal is abnormal, the real-time data of the vehicle body signal corresponding to the previous moment and the next moment of the timestamp can be stored, for example, the temperature is 20.1°C at 9:01, 23°C at 9:02, and 20.1°C at 9:03. So that when the context prediction is performed through the first model later, the predicted context information can be compared with the stored context information.

[0092] Step 120, analyzing and processing the abnormality detection database data, abnormal event description data, abnormality detection cause explanation and vehicle body signal real-time data to obtain abnormal data description, result description, cause description, status description and abnormality ID;

[0093] Specifically, the data description records the abnormal events in the abnormal detection database data, the state description records the real-time sensor data of the vehicle body, and the result description and cause description are the result description and cause description of the abnormal events and the real-time sensor data of the vehicle body.

[0094] In one example, result description and cause description are added to abnormal events and real-time sensor data of the vehicle body. At this time, the result description and cause description can be extracted, and the abnormal ID, data description, result description, cause description and status description are associated and stored.

[0095] Step 130, inputting the data description, result description, cause description, status description and abnormal ID of the exception into a preset first model for training, and obtaining the context information of the exception through the trained first model;

[0096] Specifically, this application requires training the model twice, and step 130 is the first training of the two trainings. In order to describe the process of the two trainings later, the first training can be regarded as the training of the first model, and the second training can be regarded as the training of the second model. The training of the first model this time is the text pre-training stage, which is the initial stage of large language model training. The task of the first model is to learn the input data and obtain contextual information. For example, the input of the first model is: The cat sit on table; the first prediction: The—>cat; the second prediction: The cat—>sit; the third prediction: The cat sit—>on; the fourth prediction: The cat sit on—>table.

[0097] In the present application, when the first model is trained, the input of the first model is data description, result description, cause description, and state description. The output of the first model is context information. Continuing with the previous example, the data description, result description, cause description and status description are input into the first model for training, and the abnormal context information can be obtained. The context information can be a piece of text, for example, the context information can be "The obtained ID12345 detected an abnormal engine temperature of 105 degrees Celsius at 2023-11-13 08:15:32. At this time, the vehicle engine speed is 3000rmp, the vehicle speed is 120km / h, the coolant temperature is 90 degrees Celsius, the oil pressure is 3.5bar, the fuel volume is 50%, the external temperature is 30 degrees Celsius, and the battery voltage is 12.6V. The abnormality caused the engine to overheat for five minutes, resulting in a decrease in engine performance and an automatic decrease in vehicle speed. After analysis, the engineer believes that the possible main reason is a failure in the vehicle's cooling system, resulting in poor coolant circulation. The secondary reason is long-term high-speed form and excessively high external temperature. It is recommended to slow down immediately and check the cooling system, stop the car and contact maintenance service if necessary."

[0098] During this training, the first model will compare the predicted words with the real words. When the difference in the comparison result is greater than a preset threshold, the weights and parameters of the first model will be updated until the difference between the predicted words and the real words is within a preset range. Thus, the first model trained in this application can predict abnormal context information based on the existing abnormal data description, result description, cause description, and state description.

[0099] Step 140, inputting the first question information for the anomaly into the trained first model to obtain first context information of the first question;

[0100] Specifically, after the first model training is completed, the first question information can be input into the trained first model to obtain the first context information. For example, the first question information is "What is the cause of the abnormal event numbered 12345?", and the first context information corresponding to ID12345 is retrieved in the database, that is, "the obtained ID12345 detected an abnormal engine temperature of 105 degrees Celsius at 2023-11-13 08:15:32. At this time, the vehicle engine speed is 3000rmp, the vehicle speed is 120km / h, the coolant temperature is 90 degrees Celsius, the oil pressure is 3.5bar, the fuel volume is 50%, the external temperature is 30 degrees Celsius, and the battery voltage is 12.6V. This abnormality caused the engine to continue to overheat for five minutes, resulting in a decrease in engine performance and an automatic decrease in vehicle speed. After analysis, the engineer believes that the main possible cause is a failure in the vehicle's cooling system, resulting in poor coolant circulation. The secondary reason is a long period of high speed and excessively high external temperature. It is recommended to slow down and check the cooling system immediately, and stop and contact maintenance service if necessary." The first context information is context information about data description, result description, cause description and status description.

[0101] Specifically, when the first model training is performed in step 140, the following steps may also be included:

[0102] According to the scene information, adjust the proportion of abnormalities corresponding to the scene information;

[0103] According to the adjusted abnormal proportion, the abnormal data description, result description, cause description, state description and abnormal ID are input into the preset first model for training.

[0104] The scene information may be the content of focus, for example, if the focus is on the engine, the scene information is engine abnormality, and if the focus is on the battery temperature, the scene information is battery temperature abnormality. At this time, the proportion of abnormal data under the scene information can be adjusted, for example, the proportion of engine abnormality data in the abnormal detection database data, abnormal event description data, abnormal detection cause explanation and vehicle body signal real-time data can be adjusted.

[0105] Step 150, inputting the first question and the first context information into the second model, training the second model, and obtaining first explanation information of the first question through the trained second model;

[0106] The large language model is further trained using the first context information and the first question information output by the first model, so as to obtain the explanation information of the first question. That is, the input of the second model is the second question information and the first context information, and the output is the explanation information.

[0107] The analysis information may be a reply information and suggestion information for the abnormality, and the explanation information may include the value of normal data, the value of abnormal data, the cause of the abnormality and suggestions. For example, the explanation information may be "Based on the abnormal data, it is found that the battery voltage is abnormal, only 11.5 volts, which is lower than the normal working voltage of 12 volts. The possible reasons are battery aging or problems with the battery charging system. It is recommended to check the battery and its charging system."

[0108] Continuing with the previous example, the explanation information could be "The abnormal event ID12345 was at 08:15:32 on November 13, 2023. The vehicle detected that the temperature reading of the engine temperature sensor was 105°C, which exceeded the normal temperature range of 85°C-95°C. This caused the engine overheating alarm to be triggered, which lasted for 5 minutes. Due to this overheating condition, the engine performance decreased and the vehicle automatically slowed down for safety reasons."

[0109] Therefore, after training, the large language model can achieve the purpose of input data description, result description, cause description, status description, and output explanatory information.

[0110] In an optional implementation, the present application can also establish a vectorized knowledge base, which converts unstructured data such as text, images, and sounds into a structured numerical form, usually an array consisting of a series of numbers. In the present invention, the vectorized knowledge base is mainly aimed at the results of automobile abnormality detection. For example, if an abnormal event of engine overheating is detected, this event may be described as "the engine temperature exceeds the safety threshold." In the vectorized knowledge base, such an event is converted into a vector, such as [1,3,8,...,0,150]. With such a vectorized representation, search and matching can be performed in the knowledge base. When a new abnormal event is detected, the event is vectorized and compared with the vector in the knowledge base. If a very similar vector is found, it can be inferred that this is a known abnormal event. At this time, the corresponding explanation and suggestion are extracted as reference information for the second model.

[0111] In an optional implementation, for the context information output by the first model, if the amount of the context information does not meet the requirement, the following steps may be performed:

[0112] Determine whether the amount of context information meets the requirements; when the amount of context information does not meet the requirements, perform data enhancement on the context information.

[0113] Among them, a character threshold can be set. For example, if the first context information has fewer than a certain number of characters, it can be regarded as the amount of context information not meeting the requirement. When the amount of information does not meet the requirement, it can be expanded through data enhancement. In this application, data enhancement can be performed based on semantic replacement, context generation, etc.

[0114] Step 160, associating and storing the exception, the exception ID and the first explanation information;

[0115] Specifically, the trained first model in the present application can infer the context information of the exception. The trained second model can obtain explanation information for the context information and the first question information, and then the exception and the exception ID can be stored in the exception event repository.

[0116] In one example, the exception, the exception ID, and the first explanation information may be associated and stored in the exception event repository to expand the exception event repository. The expanded exception event repository may also be used as a database called by subsequent LLM reasoning services. Figure 2 Provide specific instructions.

[0117] Step 170, receiving second problem information; the second problem information includes an abnormality ID;

[0118] Specifically, the second question information may be the user's question information, which includes the abnormality ID, such as "What is the abnormality of ID12345?". The user's question information may be queried in the abnormality detection database to directly obtain the target explanation information about the second question information.

[0119] Step 180, according to the exception ID, query and obtain the target explanation information of the exception ID.

[0120] Specifically, according to the exception ID, the target explanation information corresponding to the exception ID is queried through the trained large language model.

[0121] Furthermore, in the present application, when the first model outputs the predicted context information, it can output the normal context information and the abnormal context information at the same time, so as to intuitively obtain the difference between the normal context information and the abnormal context information. The normal context information and the abnormal context information can be displayed in the form of text or in the form of graphics. For example, when the temperature is abnormal, the normal temperature and the abnormal temperature are displayed in the form of graphics in one or two pictures.

[0122] Figure 2 As a specific embodiment of the present application, Figure 2 As shown, this application includes the following steps:

[0123] Step 201: Real-time reporting of automobile IoT data.

[0124] Specifically, the automotive IoT will report anomaly detection database data, abnormal event description data, and explanations of the causes of anomaly detection.

[0125] Step 202: Kafka real-time data stream.

[0126] Among them, the automotive Internet of Things reports real-time data of vehicle body signals.

[0127] Step 203: time series database.

[0128] Specifically, the automobile IoT data is stored in the database according to the timestamp.

[0129] Step 204, anomaly detection algorithm.

[0130] Specifically, the anomaly detection algorithm mainly performs anomaly detection, and the anomaly detection algorithm can be specifically run in the first model.

[0131] Step 205: abnormal event is triggered.

[0132] Step 206, abnormality reporting.

[0133] Step 207: abnormal event storage.

[0134] Specifically, when an abnormal event is detected, the abnormality may be stored. For steps 205 - 207 , refer to the description of step 160 .

[0135] Step 208, the User inputs a question.

[0136] Specifically, this step is consistent with step 170, and at this time, the input second question information is received.

[0137] Step 209: LLM reasoning service.

[0138] Specifically, the reasoning service runs in the second model, and the second model can infer the corresponding explanation information.

[0139] Step 210, User gets an answer.

[0140] Specifically, the interpretation information is played back.

[0141] By applying the method for generating an interpretable language for automobile anomaly detection results provided by the present invention and through fine-grained attack link modeling, it is possible to more accurately describe the correlation between attack behaviors and vulnerabilities in the automobile system, reduce dependence on expert experience and manual analysis, improve the efficiency of the tracing process, and automatically locate the attack link, thereby improving the accuracy of attack link positioning.

[0142] Combination Figure 2 To give a specific example, suppose a car equipped with IoT devices suddenly stalls while driving, and its onboard sensors collect the following data: the accelerator pedal position is 30%, the fuel tank oil level is 10%, the battery voltage is 11.5 volts, and the engine temperature is 90°C. This data information is sent to the data center in real time.

[0143] (a) The Kafka system in the data center receives the car’s data and quickly distributes it to the anomaly detection algorithm for reasoning.

[0144] (b) After analyzing this data, the anomaly detection algorithm identifies it as an "abnormal condition" because the battery voltage should be greater than 12 V under normal circumstances. The algorithm stores this set of data in the abnormal event repository and assigns a unique number to this event: EX12345.

[0145] (c) The car owner or car manufacturer discovers this exception number and asks: "What is the cause of the exception event number EX12345?". The exception event repository receives the query instruction, finds the corresponding exception data according to the number EX12345, and transmits the data to the LLM reasoning service. After analyzing the data, the LLM reasoning service responds: "Based on the exception data, it is found that the battery voltage is abnormal, only 11.5 volts, which is lower than the normal operating voltage of 12 volts. The possible reasons are battery aging or problems with the battery charging system. It is recommended to check the battery and its charging system." This example illustrates the entire process from the detection of anomalies by car sensors, to the processing in the data center, and then to the LLM reasoning service providing an explainable description of the anomaly.

[0146] By applying the method and device for generating interpretable language of automobile abnormality detection results provided by the present invention, the following technical effects are achieved:

[0147] (1) Explanation sources from more angles: This application is pre-trained with data description, result description, cause description, and state description. Through the semantic understanding and logical reasoning ability of the large language model, the anomaly detection results can be explained from a human perspective.

[0148] (2) As the scale of the first model gradually increases, an emergence effect will occur, and its ability to handle complex tasks will increase dramatically. Compared with the local explanatory model, it can process data with larger dimensions and has a stronger ability to discover hidden representations of data.

[0149] (3) The interpretability of the second model for anomaly detection can rely on pre-training text experience and instruction fine-tuning. Its knowledge, abnormal data judgment, and analysis capabilities can all be regulated during the training phase, without the need for regulation during the reasoning phase. It is also generalizable and can explain a variety of abnormal scenarios and abnormal results after training.

[0150] This application adds a large amount of anomaly detection database data, abnormal event description data, anomaly detection cause explanation data, and vehicle body signal data in the pre-training stage of the language model, and uses the language model to sample and analyze the abnormal data in prior knowledge, and analyze and describe the causes and possibilities. Its ability can provide rich case descriptions and knowledge reserves to explain the anomaly detection results, and it has a wide range of applications in vehicle protection safety and abnormal trigger reminders. It fully considers the actual needs and complexity of automobile anomaly detection, and has obvious advantages in the accuracy and reliability of anomaly detection, the generalization ability and controllability of the model, and the expansion of the scope of application.

[0151] Embodiment 2

[0152] Embodiment 2 of the present invention provides a device for generating an interpretable language for vehicle abnormality detection results, such as Figure 3 As shown, the device includes: an acquisition module 310, a processing module 320, a first training module 330, a prediction module 340, a second training module 350, a storage module 360, a receiving module 370 and a query module 380.

[0153] The acquisition module 310 is used to acquire abnormality detection database data, abnormal event description data, abnormality detection cause explanation and vehicle body signal real-time data;

[0154] The processing module 320 is used to analyze and process the abnormality detection database data, abnormal event description data, abnormality detection cause explanation and vehicle body signal real-time data to obtain abnormal data description, result description, cause description, state description and abnormality ID;

[0155] The first training module 330 is used to input the abnormal data description, result description, cause description, state description and abnormal ID into a preset first model for training, and obtain the abnormal context information through the trained first model;

[0156] The prediction module 340 is used to input the first question information for the anomaly into the trained first model to obtain the first context information of the first question;

[0157] The second training module 350 is used to input the first question information and the first context information into the second model, train the second model, and obtain the first explanation information of the first question through the trained second model;

[0158] The storage module 360 ​​is used to associate and store the exception, the exception ID and the first explanation information;

[0159] The receiving module 370 is used to receive the second problem information; the second problem information includes an abnormality ID;

[0160] The query module 380 is used to query and obtain target explanation information of the exception ID according to the exception ID.

[0161] Further, the processing module 320 analyzes and processes the abnormality detection database data, the abnormal event description data, the abnormality detection cause explanation and the vehicle body signal real-time data to obtain the abnormal data description, result description, cause description, state description and abnormality ID, including:

[0162] According to the description type, the abnormal detection database data, abnormal event description data, abnormal detection cause explanation and vehicle body signal real-time data are extracted and processed; the description type includes data class, result class, cause class and status class;

[0163] According to the data class, result class, cause class and status class, data description, result description, cause description and status description are extracted from the abnormal detection database data, abnormal event description data, abnormal detection cause explanation and vehicle body signal real-time data;

[0164] Add an exception ID to the exception data description, result description, cause description, and status description.

[0165] Further, such as Figure 4 As shown, the device also includes a judgment module 410.

[0166] The acquisition module 310 is also used to acquire real-time data of vehicle body signals;

[0167] The judgment module 410 judges whether the real-time data of the vehicle body signal is abnormal;

[0168] When the real-time data of the vehicle body signal is abnormal, the storage module 360 ​​stores the real-time data of the vehicle body signal, the real-time data of the vehicle body signal at the previous moment, and the real-time data of the vehicle body signal at the next moment; the real-time data of the vehicle body signal includes: vehicle speed, rotation speed, fuel level, and temperature.

[0169] Furthermore, the first training module 330 inputs the abnormal data description, result description, cause description, state description and abnormal ID into a preset first model for training, and the abnormal context information obtained by the trained first model specifically includes:

[0170] According to the scene information, adjust the proportion of abnormalities corresponding to the scene information;

[0171] According to the adjusted abnormal proportion, the abnormal data description, result description, cause description, state description and abnormal ID are input into the preset first model for training.

[0172] Further, combined with Figure 3 and Figure 4 In an optional implementation, Figure 5 As shown, the device also includes: an enhancement module 510.

[0173] The judging module 410 is used to judge whether the amount of the context information meets the requirement;

[0174] The enhancement module 510 is used to perform data enhancement on the context information when the amount of the context information does not meet the requirement.

[0175] The device provided in the second embodiment of the present invention can execute the method steps in the above-mentioned method embodiment 1, and its implementation principle and technical effects are similar, which will not be repeated here.

[0176] It should be noted that it should be understood that the division of the various modules of the above device is only a division of logical functions. In actual implementation, they can be fully or partially integrated into one physical entity, or they can be physically separated. And these modules can all be implemented in the form of software called by processing elements; they can also be all implemented in the form of hardware; some modules can also be implemented in the form of software called by processing elements, and some modules can be implemented in the form of hardware. For example, the determination module can be a separately established processing element, or it can be integrated in a chip of the above device. In addition, it can also be stored in the memory of the above device in the form of program code, and called and executed by a processing element of the above device. The implementation of other modules is similar. In addition, these modules can be fully or partially integrated together, or they can be implemented independently. The processing element described here can be an integrated circuit with signal processing capabilities. In the implementation process, each step of the above method or each module above can be completed by an integrated logic circuit of hardware in the processor element or instructions in the form of software.

[0177] For example, the above modules may be one or more integrated circuits configured to implement the above methods, such as one or more application specific integrated circuits (ASIC), or one or more microprocessors (DSP), or one or more field programmable gate arrays (FPGA). For another example, when a module above is implemented in the form of a processing element scheduling program code, the processing element may be a general-purpose processor, such as a central processing unit (CPU) or other processor that can call program code. For another example, these modules may be integrated together and implemented in the form of a system-on-a-chip (SOC).

[0178] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The above computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The above-mentioned computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the above-mentioned computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center through wired (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (such as infrared, wireless, Bluetooth, microwave, etc.) methods. The above-mentioned computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that includes one or more available media integrated. The above-mentioned available medium can be a magnetic medium (such as a floppy disk, a hard disk, a tape), an optical medium (such as a DVD), or a semiconductor medium (such as a solid state disk (SSD)), etc.

[0179] Embodiment 3

[0180] Embodiment 3 of the present invention provides a computer server, such as Figure 6 As shown, it includes: a memory, a processor and a transceiver;

[0181] The processor is used to couple with the memory, read and execute instructions in the memory, so as to implement any one of the methods for generating an interpretable language for an abnormality detection result of an automobile provided in the first embodiment above;

[0182] The transceiver is coupled to the processor, and the processor controls the transceiver to send and receive messages.

[0183] Embodiment 4

[0184] Embodiment 4 of the present invention provides a chip system, such as Figure 7 As shown, it includes a processor, a coupling between the processor and a memory, the memory stores program instructions, and when the program instructions stored in the memory are executed by the processor, a method for generating an interpretable language for automobile abnormality detection results as provided in Example 1 is implemented.

[0185] Embodiment 5

[0186] Embodiment 5 of the present invention provides a computer-readable storage medium, such as Figure 8 As shown, it includes a program or instruction. When the program or instruction is executed on a computer, it implements a method for generating an interpretable language for an abnormal vehicle detection result as provided in the first embodiment.

[0187] The professionals should further realize that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to the function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0188] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be implemented using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0189] The above specific implementation methods further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above are only specific implementation methods of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for generating an interpretable language for automobile anomaly detection results, characterized in that: The method comprises: Obtain anomaly detection database data, abnormal event description data, anomaly detection cause explanation and vehicle body signal real-time data; Analyze and process the abnormality detection database data, abnormal event description data, abnormality detection cause explanation and vehicle body signal real-time data to obtain abnormal data description, result description, cause description, status description and abnormality ID; The data description, result description, cause description, status description and the abnormal ID of the exception are input into a preset first model for training, and the context information of the exception is obtained through the trained first model; the context information is the context information about the data description, result description, cause description and status description, and the context information is a piece of text; the input of the first model is the data description, result description, cause description and status description; the output of the first model is the context information; Inputting first question information for the exception into the trained first model to obtain first context information of the first question; Input the first question information and the first context information into a second model, train the second model, and obtain first explanation information of the first question through the trained second model; the first explanation information is explanation information for the first question, including reply information and suggestion information for the exception; the input of the second model is the second question information and the first context information, and the output is the first explanation information; Associating and storing the exception, the exception ID and the first explanation information; receiving second problem information; the second problem information including an abnormality ID; According to the exception ID, the target explanation information of the exception ID is obtained by querying.

2. The method according to claim 1, characterized in that The analysis and processing of the abnormality detection database data, abnormal event description data, abnormality detection cause explanation and vehicle body signal real-time data to obtain abnormal data description, result description, cause description, status description and abnormality ID specifically includes: The data description corresponds to the abnormal events in the abnormal detection database data, and the status description corresponds to the real-time sensor data of the vehicle body; the result description and cause description are the result description and cause description of the abnormal events and the real-time sensor data of the vehicle body; the abnormal ID is added to the abnormal data description, result description, cause description and status description.

3. The method according to claim 1, characterized in that: The method also includes: Obtain real-time data of vehicle body signals; Determine whether the real-time data of the vehicle body signal is abnormal. When the real-time data of the vehicle body signal is abnormal, store the real-time data of the vehicle body signal, the real-time data of the vehicle body signal at the previous moment, and the real-time data of the vehicle body signal at the next moment; the real-time data of the vehicle body signal includes: vehicle speed, rotation speed, fuel quantity, temperature and battery voltage.

4. The method according to claim 1, characterized in that: The abnormal data description, result description, cause description, state description and abnormal ID are input into a preset first model for training, and the abnormal context information obtained by the trained first model specifically includes: According to the scene information, adjust the proportion of abnormalities corresponding to the scene information; According to the adjusted abnormal proportion, the abnormal data description, result description, cause description, state description and abnormal ID are input into the preset first model for training.

5. The method according to claim 1, characterized in that Before inputting the first question information and the first context information into the second model and training the second model, the method further includes: Determining whether the amount of information of the context information meets the requirement; When the amount of information of the context information does not meet the requirement, data enhancement is performed on the context information.

6. A device for generating an interpretable language for automobile anomaly detection results, characterized in that: The device comprises: An acquisition module, the acquisition module is used to acquire abnormality detection database data, abnormal event description data, abnormality detection cause explanation and vehicle body signal real-time data; A processing module, the processing module is used to analyze and process the abnormality detection database data, abnormal event description data, abnormality detection cause explanation and vehicle body signal real-time data to obtain abnormal data description, result description, cause description, state description and abnormality ID; A first training module, the first training module is used to input the abnormal data description, result description, cause description, state description and abnormal ID into a preset first model for training, and obtain the abnormal context information through the trained first model; the context information is the context information about the data description, result description, cause description and state description, and the context information is a piece of text; the input of the first model is the data description, result description, cause description and state description; the output of the first model is the context information; A prediction module, wherein the prediction module is used to input first question information for anomalies into a trained first model to obtain first context information of the first question; a second training module, the second training module is used to input the first question information and the first context information into a second model, train the second model, and obtain first explanation information of the first question through the trained second model; the first explanation information is explanation information for the first question, including reply information and suggestion information for the exception; the input of the second model is the second question information and the first context information, and the output is the first explanation information; A storage module, the storage module is used to associate and store the exception, the exception ID and the first explanation information; A receiving module, the receiving module is used to receive second problem information; the second problem information includes an abnormality ID; A query module is used to query and obtain target explanation information of the exception ID according to the exception ID.

7. The device according to claim 6, characterized in that The device also includes: a judgment module and an enhancement module; The judging module is used to judge whether the amount of the context information meets the requirement; The enhancement module is used to perform data enhancement on the context information when the amount of information of the context information does not meet the requirement.

8. A computer server, characterized in that: include: memory, processors, and transceivers; The processor is used to couple with the memory, read and execute instructions in the memory, so as to implement the method for generating an interpretable language of an abnormality detection result of an automobile according to any one of claims 1 to 5; The transceiver is coupled to the processor, and the processor controls the transceiver to send and receive messages.

9. A chip system, characterized in that: The invention comprises a processor, the processor is coupled to a memory, the memory stores program instructions, and when the program instructions stored in the memory are executed by the processor, the method for generating an interpretable language for an abnormality detection result of an automobile according to any one of claims 1 to 5 is implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and the computer program is executed by a processor to implement the method for generating an interpretable language for automobile abnormality detection results as described in any one of claims 1 to 5.

Images