Fault prediction method and device, vehicle, equipment and storage medium

Abstract

The invention relates to a fault prediction method and device, a vehicle, equipment and a storage medium. The fault prediction method comprises the following steps: acquiring a plurality of time sequences and a predicted fault probability corresponding to each time sequence, and if the predicted fault probabilities corresponding to the plurality of time sequences meet a preset condition, acquiring a conditional probability corresponding to the time sequence for each time sequence, and determining the confidence coefficient of the predicted fault probability based on the predicted fault probability and the conditional probability corresponding to the plurality of time sequences, so that the method can calculate the conditional probability corresponding to each time sequence, and then calculate the confidence coefficient of the predicted fault probability based on the predicted fault probability and the conditional probability of each time sequence. When the confidence coefficient is calculated, the same relevance of the predicted fault probabilities of the two corresponding time sequences can be considered, so that the calculated confidence coefficient is more accurate, and the accuracy of the predicted fault probabilities can be intuitively and accurately expressed.

Description

Technical Field

[0001] The present disclosure relates to the field of data processing, and in particular to a fault prediction method, apparatus, vehicle, equipment, and storage medium. Background Art

[0002] With the development of smart cars, more and more sensors, actuators, and software functions are being integrated into cars. These new components will bring new risks to the safety of electronic and electrical systems. Therefore, it is necessary to predict faults in smart cars to ensure the personal safety of users.

[0003] In related technologies, system health management through artificially formulated rules can no longer adapt to complex software and hardware systems. For example, the rules stipulate that when the CPU (Central Processing Unit) occupancy rate is greater than 99%, there is a system resource shortage fault. However, the actual situation is that when the CPU occupancy rate reaches 100%, the system may still operate normally, and when the CPU occupancy rate reaches 98%, there may be a system resource shortage problem. From the driver's perspective, when a traditional car breaks down, the driver is in driving state and can be aware of the fault in time and eliminate it manually. However, for smart cars, the intelligent system replaces part of the driver's work, and the driver may not be able to realize the fault in time, which may lead to danger.

[0004] As can be seen, because smart cars have more complex software and hardware systems than traditional cars, traditional fault prediction can no longer meet the needs of smart cars. Therefore, how to achieve fault prediction for smart cars has become a technical problem that needs to be solved urgently. Summary of the Invention

[0005] In order to solve the above technical problems, the present disclosure provides a fault prediction method, apparatus, vehicle, equipment and storage medium.

[0006] In a first aspect, the present disclosure provides a fault prediction method applied to a vehicle, comprising:

[0007] Obtaining multiple time series and a predicted failure probability corresponding to each time series, wherein the time series is monitoring data obtained by sequentially monitoring multiple pre-set checkpoints for fault detection;

[0008] If the predicted failure probabilities corresponding to the multiple time series meet a preset condition, then for each time series, a conditional probability corresponding to the time series is obtained, wherein the conditional probability is: when the predicted failure probability of the preset time series corresponding to the time series is a preset failure value, the predicted failure probability corresponding to the time series is also the preset failure value;

[0009] The confidence level of the predicted failure probability is determined based on the predicted failure probabilities and the conditional probabilities corresponding to the multiple time series.

[0010] In an optional embodiment, the preset conditions include that the predicted failure probability corresponding to at least one of the time series is greater than or equal to a first preset probability threshold, and / or that the average value of the predicted failure probabilities corresponding to the multiple time series is greater than or equal to a second preset probability threshold.

[0011] In an optional implementation, determining the confidence level of the predicted failure probability based on the predicted failure probabilities and the conditional probabilities corresponding to the multiple time series includes:

[0012] Multiplying the conditional probabilities corresponding to the multiple time series to obtain a first product value;

[0013] For each of the time series, multiplying the corresponding predicted failure probability and the conditional probability to obtain a first sub-product value, subtracting the corresponding predicted failure probability from a preset value to obtain a first difference value, subtracting the corresponding conditional probability from the preset value to obtain a second difference value, multiplying the first difference value and the second difference value to obtain a second sub-product value, and adding the first sub-product value and the second sub-product value to obtain a first sum value;

[0014] Multiplying the first sum values ​​corresponding to the multiple time series to obtain a second product value;

[0015] A confidence level of the predicted failure probability is determined based on the first product value and the second product value.

[0016] In an optional embodiment, the method further includes:

[0017] If the confidence level of the predicted fault probability is greater than a preset threshold, the fault information corresponding to the predicted fault probability with the preset fault value is sent to the cloud server, wherein the fault information is used to instruct the language model deployed on the cloud server to output a response measure corresponding to the fault information, and the language model is trained based on the maintenance dataset;

[0018] The countermeasure is received from the cloud server, and the countermeasure is provided to the user.

[0019] In an optional implementation, the process of obtaining the predicted failure probability corresponding to the time series includes:

[0020] The time series is input into a corresponding fault prediction model to obtain the predicted fault probability output by the corresponding fault prediction model, wherein the fault prediction model is trained based on a fault dataset by a model trainer deployed on a cloud server.

[0021] In an optional embodiment, the checkpoint includes a software checkpoint and / or a hardware checkpoint.

[0022] Accordingly, the process of obtaining the time series includes:

[0023] Receiving monitoring data corresponding to the checkpoint reported by the software module corresponding to the software checkpoint;

[0024] And / or, periodically monitor the hardware checkpoints to obtain corresponding monitoring data.

[0025] In an optional embodiment, the method further includes:

[0026] If no fault is detected based on the time series, a normal label is marked on the time series, and if a fault is detected based on the time series, a fault information label is marked on the time series to obtain training data;

[0027] The training data is sent to a cloud server so that the cloud server updates the training data in a fault data set, and if the fault information corresponds to a new fault, updates the fault information and its corresponding countermeasures in a maintenance data set.

[0028] In a second aspect, the present disclosure further provides a fault prediction method applied to a cloud server, comprising:

[0029] Train the fault prediction model based on the fault dataset through the model trainer;

[0030] The trained fault prediction model is deployed on a vehicle to determine a predicted fault probability corresponding to the time series based on the time series;

[0031] Among them, the predicted failure probability is used to indicate the vehicle. If the predicted failure probabilities corresponding to multiple time series meet preset conditions, then for each of the time series, the conditional probability corresponding to the time series is obtained, and the conditional probability is: when the predicted failure probability of the preset time series corresponding to the time series is the preset failure value, the predicted failure probability corresponding to the time series is also the probability of the preset failure value, and based on the predicted failure probabilities and the conditional probabilities corresponding to the multiple time series, the confidence of the predicted failure probability is determined, wherein the time series is monitoring data obtained by sequentially monitoring multiple pre-set checkpoints for fault detection.

[0032] In an optional embodiment, the method further includes:

[0033] receiving training data sent by the vehicle and including the training data in the fault dataset to update the fault dataset, wherein the training data is obtained by the vehicle marking the time series with a normal label when no fault is detected based on the time series, and marking the time series with a fault label for representing fault information when a fault is detected based on the time series;

[0034] The model trainer is used to retrain the fault prediction model based on the updated fault data set, wherein the retrained fault prediction model is used to update the fault prediction model historically deployed on the vehicle.

[0035] In an optional embodiment, the method further includes:

[0036] receiving, through the language large model, fault information corresponding to the predicted fault probability having a value of the preset fault value, sent by the vehicle, and outputting a response measure corresponding to the fault information;

[0037] The countermeasure is sent to the vehicle so that the vehicle provides the countermeasure to the user.

[0038] In an optional embodiment, the method further includes:

[0039] Acquire a countermeasure corresponding to fault information corresponding to the new fault, and include the fault information and the corresponding countermeasure in a maintenance data set to update the maintenance data set;

[0040] Based on the updated maintenance dataset, the language model is retrained to update the language model.

[0041] In a third aspect, the present disclosure further provides a fault prediction device, which is applied to a vehicle and includes:

[0042] A first acquisition module is configured to acquire a plurality of time series and a predicted failure probability corresponding to each time series, wherein the time series is monitoring data obtained by sequentially monitoring a plurality of pre-set checkpoints for fault detection;

[0043] A second acquisition module is configured to acquire, for each time series, a conditional probability corresponding to the time series if the predicted failure probabilities corresponding to the multiple time series meet a preset condition, wherein the conditional probability is the probability that, when the predicted failure probability of a preset time series corresponding to the time series is a preset failure value, the predicted failure probability corresponding to the time series is also the preset failure value;

[0044] A determination module is used to determine the confidence of the predicted failure probability based on the predicted failure probability and the conditional probability corresponding to the multiple time series.

[0045] In a fourth aspect, the present disclosure further provides a fault prediction device, applied to a cloud server, comprising:

[0046] A first training module is used to train a fault prediction model based on a fault dataset through a model trainer;

[0047] The trained fault prediction model is deployed on a vehicle to determine a predicted fault probability corresponding to the time series based on the time series;

[0048] Among them, the predicted failure probability is used to indicate the vehicle. If the predicted failure probabilities corresponding to multiple time series meet preset conditions, then for each of the time series, the conditional probability corresponding to the time series is obtained, and the conditional probability is: when the predicted failure probability of the preset time series corresponding to the time series is the preset failure value, the predicted failure probability corresponding to the time series is also the probability of the preset failure value, and based on the predicted failure probabilities and the conditional probabilities corresponding to the multiple time series, the confidence of the predicted failure probability is determined, wherein the time series is monitoring data obtained by sequentially monitoring multiple pre-set checkpoints for fault detection.

[0049] In a fifth aspect, the present disclosure further provides a vehicle, comprising at least one of the following:

[0050] The fault prediction device according to the third aspect;

[0051] The fault prediction device according to the eighth aspect;

[0052] The computer-readable storage medium described in the ninth aspect.

[0053] In a sixth aspect, the present disclosure further provides a cloud server, comprising:

[0054] A fault data set and a model trainer, wherein the model trainer is used to train a fault prediction model based on the fault data set;

[0055] The trained fault prediction model is deployed on a vehicle to determine a predicted fault probability corresponding to the time series based on the time series;

[0056] Among them, the predicted failure probability is used to indicate the vehicle. If the predicted failure probabilities corresponding to multiple time series meet preset conditions, then for each of the time series, the conditional probability corresponding to the time series is obtained, and the conditional probability is: when the predicted failure probability of the preset time series corresponding to the time series is the preset failure value, the predicted failure probability corresponding to the time series is also the probability of the preset failure value, and based on the predicted failure probabilities and the conditional probabilities corresponding to the multiple time series, the confidence of the predicted failure probability is determined, wherein the time series is monitoring data obtained by sequentially monitoring multiple pre-set checkpoints for fault detection.

[0057] In an optional embodiment, the cloud server is further configured to receive training data sent by the vehicle and include the training data in the fault dataset to update the fault dataset, wherein the training data is obtained by the vehicle marking the time series with a normal label when no fault is detected based on the time series, and marking the time series with a fault label for representing fault information when a fault is detected based on the time series;

[0058] The model trainer is further used to retrain the fault prediction model based on the updated fault data set, wherein the retrained fault prediction model is used to update the fault prediction model historically deployed on the vehicle;

[0059] The cloud server further includes: a maintenance data set and a large language model, the large language model being configured to receive fault information corresponding to a predicted fault probability having a value of the preset fault value, sent by the vehicle, output a countermeasure corresponding to the fault information, and transmit the countermeasure to the vehicle so that the vehicle provides the countermeasure to the user;

[0060] The cloud server is further configured to obtain a response measure corresponding to fault information corresponding to a new fault, and record the fault information and the response measure corresponding to the fault in a maintenance data set to update the maintenance data set;

[0061] The large language model is also used to retrain the large language model based on an updated maintenance dataset to update the large language model.

[0062] In a seventh aspect, the present disclosure further provides a fault prediction system, comprising:

[0063] The vehicle of the fifth aspect and the cloud server of the sixth aspect.

[0064] In an eighth aspect, the present disclosure further provides a computing device, including:

[0065] processor;

[0066] a memory for storing executable instructions;

[0067] The processor is configured to read executable instructions from the memory and execute the executable instructions to implement the fault prediction method of the first aspect or the fault prediction method of the second aspect.

[0068] In a ninth aspect, the present disclosure further provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the processor implements the fault prediction method of the first aspect or the fault prediction method of the second aspect.

[0069] The technical solution provided by the embodiments of the present disclosure has the following advantages over the prior art:

[0070] The fault prediction method of the embodiment of the present disclosure can calculate the corresponding conditional probability for each time series, and then calculate the confidence of the predicted fault probability based on the predicted fault probability and conditional probability of each time series. In this way, when calculating the confidence, the same correlation between the predicted fault probabilities of the two corresponding time series can be taken into account, so that the calculated confidence is more accurate. In this way, the accuracy of the predicted fault probability can be expressed intuitively and accurately. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the present disclosure.

[0072] In order to more clearly illustrate the embodiments of the present disclosure or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0073] Figure 1 A schematic diagram of a flow chart of a fault prediction method provided in an embodiment of the present disclosure;

[0074] Figure 2 A flowchart of a reporting checkpoint provided in an embodiment of the present disclosure;

[0075] Figure 3 A flowchart of another fault prediction method provided by an embodiment of the present disclosure;

[0076] Figure 4 A schematic diagram of the structure of a fault prediction system provided by an embodiment of the present disclosure;

[0077] Figure 5A schematic diagram of the workflow of a fault prediction system provided by an embodiment of the present disclosure;

[0078] Figure 6 A schematic diagram of the structure of a fault prediction device provided in an embodiment of the present disclosure;

[0079] Figure 7 A schematic structural diagram of another fault prediction device provided in an embodiment of the present disclosure;

[0080] Figure 8 A schematic diagram of the structure of a fault prediction device provided in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0081] The following describes embodiments of the present disclosure in more detail with reference to the accompanying drawings. Although certain embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as limited to the embodiments described herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are for illustrative purposes only and are not intended to limit the scope of protection of the present disclosure.

[0082] It should be understood that the various steps described in the method embodiments of the present disclosure may be performed in different orders and / or in parallel. In addition, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present disclosure is not limited in this respect.

[0083] As used herein, the term "including" and its variations are open-ended, i.e., "including but not limited to." The term "based on" means "based, at least in part, on." The term "one embodiment" means "at least one embodiment," the term "another embodiment" means "at least one additional embodiment," and the term "some embodiments" means "at least some embodiments." Other terms are defined in the following description.

[0084] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units.

[0085] It should be noted that the modifications of "one" and "multiple" mentioned in the present disclosure are illustrative rather than restrictive, and those skilled in the art should understand that unless otherwise clearly indicated in the context, they should be understood as "one or more".

[0086] The names of the messages or information exchanged between multiple devices in the embodiments of the present disclosure are only used for illustrative purposes and are not used to limit the scope of these messages or information.

[0087] With the development of intelligent connected vehicles, sensors, actuators, and software are being integrated into vehicles. These new components introduce new risks to the safety of electronic and electrical systems. Automakers must fully address the safety requirements of these new scenarios and design appropriate safety mechanisms to prevent smart cars from endangering personal safety.

[0088] Compared to smart cars, traditional cars have simpler and easier-to-understand functionalities and electrical and electronic architectures. For example, they lack integrated autonomous driving, and their controllers handle a single function. Traditional cars can ensure personal safety through fault detection and passive safety measures, but this approach cannot meet the needs of smart cars.

[0089] Specifically, traditional methods of managing system health through manually formulated rules are no longer adaptable to complex software and hardware systems. For example, a rule may indicate that when the CPU (Central Processing Unit) utilization rate exceeds 99%, there is a system resource shortage problem. However, in reality, when the CPU utilization rate reaches 100%, the system may still operate normally, but when the CPU utilization rate reaches 98%, there may already be a system resource shortage problem.

[0090] Furthermore, from the driver's perspective, when a traditional car malfunctions, the driver is still driving and can promptly recognize the problem and manually correct it. However, in a smart car, the intelligent system replaces some of the driver's work, and the driver may not be able to recognize the problem in time, which can lead to dangerous situations.

[0091] Therefore, smart cars need more advanced and comprehensive security mechanisms to ensure personal safety in order to better cope with security challenges in smart cars.

[0092] The embodiment of the present disclosure provides a fault prediction method, which obtains multiple time series and the predicted fault probability corresponding to each time series, wherein the time series is monitoring data obtained by sequentially monitoring multiple pre-set checkpoints for fault detection. If the predicted fault probabilities corresponding to the multiple time series meet the preset conditions, then for each time series, the conditional probability that the predicted fault probability value of the time series is also the preset fault value when the predicted fault probability of the time series corresponding to the preset time series is the preset fault value is obtained, and the confidence of the predicted fault probability is determined based on the predicted fault probabilities and conditional probabilities corresponding to the multiple time series. It can be seen that the embodiment of the present disclosure can calculate the corresponding conditional probability for each time series, and then calculate the confidence of the predicted fault probability based on the predicted fault probability and conditional probability of each time series. In this way, when calculating the confidence, the correlation between the same predicted fault probabilities of the two corresponding time series can be taken into account, so that the calculated confidence is more accurate, so that the accuracy of the predicted fault probability can be expressed intuitively and accurately.

[0093] Based on this, the present disclosure provides a fault prediction method, referring to Figure 1 , which is a flow chart of a fault prediction method provided in an embodiment of the present disclosure.

[0094] In some embodiments of the present disclosure, Figure 1 The method shown can be applied to a vehicle, wherein the vehicle includes a prediction subsystem and a diagnosis subsystem, specifically, Figure 1 The method shown is applied to a prognostic subsystem in a vehicle.

[0095] like Figure 1 As shown, the fault prediction method may include the following steps.

[0096] S101: Obtain multiple time series and the predicted failure probability corresponding to each time series.

[0097] The time series refers to monitoring data obtained by sequentially monitoring a plurality of pre-set checkpoints for fault detection.

[0098] In the disclosed embodiment, for the entire vehicle, the multiple time series may be time series corresponding to different time periods, and for the subsystems in the vehicle, the multiple time series may be time series corresponding to different subsystems.

[0099] In the embodiment of the present disclosure, the predicted failure probability can be the probability of future failure corresponding to each time series. Specifically, while obtaining the predicted failure probability corresponding to each time series, the fault information corresponding to each time series can also be obtained, such as a fault code or a keyword used to describe the fault.

[0100] In an optional embodiment, the checkpoint includes a software checkpoint and / or a hardware checkpoint. Accordingly, the process of obtaining the time series includes: receiving monitoring data corresponding to the checkpoint reported by the software module corresponding to the software checkpoint, and / or periodically monitoring the hardware checkpoint to obtain the corresponding monitoring data.

[0101] In the embodiments of the present disclosure, the checkpoint can be a software checkpoint pre-implanted by the diagnostic subsystem in the software with security requirements, or a hardware checkpoint pre-implanted by the diagnostic subsystem in the hardware with security requirements, or a software checkpoint pre-implanted by the diagnostic subsystem in the software with security requirements and a hardware checkpoint pre-implanted in the hardware with security requirements.

[0102] For example, the CPU usage can be checked every 1 second. For processes involving safety, multiple checkpoints can be set. For example, when automatic driving is turned on, checkpoints should be set for each obstacle that needs to be processed during the perception process and for each steering operation and power operation step during the control process.

[0103] Exemplarily, checkpoints include software checkpoints and hardware checkpoints. Software checkpoints are checkpoints set for programs in a software system. For example, checkpoint 1 is set at the first line of code, and checkpoint 2 is set at the tenth line of code. When the program runs to the first line of code, it feeds back the monitoring data corresponding to checkpoint 1, and when the program runs to the tenth line of code, it feeds back the monitoring data corresponding to checkpoint 2. Hardware checkpoints are checkpoints set for hardware.

[0104] At the same time, the diagnostic subsystem configures the correct rules for this checkpoint, specifically, refer to Figure 2 , which is a flowchart of the reporting checkpoint provided in an embodiment of the present disclosure.

[0105] The software can report checkpoints to the diagnostic subsystem as needed, that is, the diagnostic subsystem receives monitoring data corresponding to the checkpoints reported by the software module corresponding to the software checkpoints. The diagnostic subsystem can also periodically monitor hardware checkpoints to obtain corresponding monitoring data.

[0106] When a checkpoint in the corresponding time series stops reporting (i.e., the checkpoint is not reported), reports for timeout, reports in the wrong order, or shows anomalies (i.e., inconsistent with the configured correct rules), a fault is detected. The diagnostic subsystem can handle the fault, including restoring the system and setting a fault code. In addition, the diagnostic subsystem will also input the checkpoint data (i.e., the time series) into the prediction subsystem at the same time.

[0107] Afterwards, if no fault is detected based on the time series, the time series is labeled as normal, and if a fault is detected based on the time series, the time series is labeled with fault information, and training data is obtained. The training data is sent to the cloud server so that the cloud server updates the training data in the fault data set, and if the fault information corresponds to a new fault, the fault information and its corresponding countermeasures are updated in the maintenance data set.

[0108] In the embodiment of the present disclosure, when an abnormality such as reporting cessation, reporting timeout, or reporting sequence error occurs at a checkpoint in the corresponding time sequence, it can be considered that a fault is detected based on the time sequence.

[0109] For ease of understanding, please refer to the following Tables 1-3. Assume that Table 1 is a data table obtained after fault detection based on time series 1, Table 2 is a data table obtained after fault detection based on time series 2, and Table 3 is a data table obtained after fault detection based on time series 3. Tables 1-3 are all data tables obtained after fault detection based on time series.

[0110] Checkpoint 1 Checkpoint 2 Checkpoint 3 Checkpoint 4 Checkpoint 5 Checkpoint 6 Checkpoint 7 Label 4 2 1 3 normal

[0111] Table 1

[0112] Checkpoint 1 Checkpoint 2 Checkpoint 3 Checkpoint 4 Checkpoint 5 Checkpoint 6 Checkpoint 7 Label 11 2 1 3 4 2 normal

[0113] Table 2

[0114] Checkpoint 1 Checkpoint 2 Checkpoint 3 Checkpoint 4 Checkpoint 5 Checkpoint 6 Checkpoint 7 Label 3 1 5 1 1 1 1 U161040

[0115] Table 3

[0116] As shown in Tables 1-3 above, no fault is detected in time series 1 in Table 1, and its corresponding label is a normal label. No fault is detected in time series 2 in Table 2, and its corresponding label is also a normal label. A fault is detected in time series 3 in Table 3, and its corresponding label is a label of fault information (such as fault code U161040 shown in Table 3).

[0117] In the embodiment of the present disclosure, after obtaining the training data, the training data is sent to the cloud server so that the cloud server updates the training data in the fault data set. Since the fault data set is used to train the fault prediction model, the fault prediction model is updated by continuously updating the fault data set, thereby increasing the fault prediction model's ability to recognize new faults.

[0118] In addition, if the fault information corresponds to a new fault, that is, the fault information and the corresponding countermeasures are not in the maintenance data set, the fault information and its corresponding countermeasures can also be updated in the maintenance data set so that the vehicle can have corresponding countermeasures when a new fault occurs.

[0119] After obtaining the time series based on the checkpoint, the predicted failure probability corresponding to the time series can be obtained based on the fault prediction model. In an optional embodiment, the process of obtaining the predicted failure probability corresponding to the time series includes: inputting the time series into the corresponding fault prediction model and obtaining the predicted failure probability output by the corresponding fault prediction model.

[0120] Among them, the fault prediction model is trained by the model trainer deployed on the cloud server based on the fault dataset.

[0121] In the disclosed embodiment, for the entire vehicle, there is only one fault prediction model, while for multiple subsystems, there are multiple corresponding fault prediction models.

[0122] In the embodiment of the present disclosure, the fault prediction model in the prediction subsystem may be pre-deployed by the prediction subsystem. In addition, the fault prediction model may be updated as the fault data set in the cloud service is updated.

[0123] Specifically, the fault dataset can contain multiple data tables, each of which has the data structure of [checkpoint vector, label]. The checkpoint vector contains all checkpoints recorded within the maximum warning time. For example, if a 1-second advance warning is required, the checkpoint vector contains all checkpoints recorded within 1 second. The label is the classification category, i.e., normal / fault number. The fault number can be a diagnostic trouble code defined for the vehicle, as shown in Tables 1-3 above.

[0124] In addition, a data table may be created for the entire vehicle system, each subsystem in the entire vehicle system, or a specific process. For example, a data table may be created only for the CPU load, as shown in Tables 4 to 6 below.

[0125] CPU usage 1 CPU usage 2 CPU usage 3 CPU usage 4 CPU usage 5 CPU usage 6 CPU usage 7 Label 1％ 1％ 1％ 1％ 1％ 1％ 1％ normal

[0126] Table 4

[0127] CPU usage 1 CPU usage 2 CPU usage 3 CPU usage 4 CPU usage 5 CPU usage 6 CPU usage 7 Label 1％ 10％ 10％ 40％ 40％ 80％ 80％ normal

[0128] Table 5

[0129] CPU usage 1 CPU usage 2 CPU usage 3 CPU usage 4 CPU usage 5 CPU usage 6 CPU usage 7 Label 100％ 100％ 100％ 100％ 100％ 100％ 100％ U161040

[0130] Table 6

[0131] It should be noted that the above table is only for illustrative purposes and should not be considered as limiting.

[0132] In the disclosed embodiment, the time series needs to be input into the corresponding fault prediction model. Assuming that data table 1 is the data in vehicle subsystem A, and fault prediction model A can be trained based on data table 1, then the time series in subsystem A obtained during the operation of the vehicle system needs to be input into the fault prediction model A, so as to obtain the predicted failure probability output by the fault prediction model A, that is, the predicted failure probability corresponding to the time series in subsystem A.

[0133] S102: If the predicted failure probabilities corresponding to the multiple time series meet a preset condition, then for each time series, obtain the conditional probability corresponding to the time series.

[0134] The conditional probability is: when the predicted failure probability of the preset time series corresponding to the time series is the preset failure value, the predicted failure probability corresponding to the time series is also the probability of the preset failure value.

[0135] In an optional embodiment, the preset conditions may include that the predicted failure probability corresponding to at least one time series is greater than or equal to a first preset probability threshold, and / or the average value of the predicted failure probabilities corresponding to multiple time series is greater than or equal to a second preset probability threshold.

[0136] The first preset probability threshold and the second preset probability threshold can be set based on demand. Specifically, the first preset probability threshold can be the same as the second preset probability threshold or different. For example, the first preset probability threshold and the second preset probability threshold can be 100%.

[0137] In the embodiment of the present disclosure, the preset time series can be obtained by looking up a table, and the table records the corresponding relationship between each time series and its corresponding preset time series. In the embodiment of the present disclosure, taking the case where the predicted failure probability corresponding to time series 1 among multiple time series meets the preset condition as an example, the preset time series corresponding to time series 1 is first obtained by looking up the table, and the predicted failure probability of the preset time series corresponding to time series 1 is taken as a condition to determine the conditional probability that the preset failure probability of time series 1 is also the preset failure value. Among them, the predicted failure probability corresponding to the time series is the preset failure value, which means that the module where the checkpoint corresponding to the time series is located is predicted to have a failure.

[0138] S103: Determine the confidence level of the predicted failure probability based on the predicted failure probability and conditional probabilities corresponding to the multiple time series.

[0139] In the embodiment of the present disclosure, when the system is running, multiple time series are input into the corresponding fault prediction models respectively. If any fault prediction model outputs a fault (that is, the preset fault probability meets the preset conditions), the confidence of the fault can be calculated according to the following method.

[0140] In an optional embodiment, the conditional probabilities corresponding to multiple time series are multiplied to obtain a first product value, for each time series, the corresponding predicted failure probability and the conditional probability are multiplied to obtain a first sub-product value, the corresponding predicted failure probability is subtracted from a preset value to obtain a first difference value, the corresponding conditional probability is subtracted from the preset value to obtain a second difference value, the first difference value and the second difference are multiplied to obtain a second sub-product value, the first sub-product value and the second sub-product value are added to obtain a first sum value, the first sum values ​​corresponding to multiple time series are multiplied to obtain a second product value, and the confidence of the predicted failure probability is determined based on the first product value and the second product value.

[0141] In the embodiment of the present disclosure, the preset value can be set based on demand, for example, the preset value can be 1.

[0142] To facilitate understanding of the above embodiment, a detailed introduction is given in conjunction with the following description:

[0143] First, using the fault dataset, the predicted failure probability corresponding to each time series is calculated.

[0144] Time Series 1 Time Series 2 … Time series n P(time series 1) P(time series 2) … P(time series n)

[0145] Table 7

[0146] When a failure is predicted to occur (i.e., the predicted failure probabilities corresponding to multiple time series meet preset conditions), the confidence level of the predicted failure probability is determined based on the predicted failure probabilities and conditional probabilities corresponding to the multiple time series.

[0147] Time Series 1 Time Series 2 Time Series 3 Time Series 4 … Time series n Confidence level Fault 1 0 0 0 0 p(fault|time series combination)

[0148] Table 8

[0149] Specifically, the calculation formula is as follows:

[0150]

[0151] Where C is the confidence level; n is the number of time series; i is the time series number; ai is the predicted failure probability of the i-th time series, which is 0 when normal and 1 when faulty; P(r|S i ) is the conditional probability that the predicted failure probability value of the time series Si is also the preset failure value when the predicted failure probability of the corresponding preset time series is the preset failure value (i.e. 1), that is, r is the condition for the time series Si to calculate the conditional probability.

[0152] Correspondingly, in the above formula Corresponding to the first product value, a i P(r|S i) corresponds to the first sub-product value, 1-a i Corresponding to the first difference, 1-P(r|S i ) corresponds to the second difference, (1-a i )(1-P(r|S i )) corresponds to the second sub-product value, a i P(r|S i )+(1-a i )(1-P(r|S i )) corresponds to the first sum value, Corresponding to the second product value, C is the confidence level of the predicted failure probability determined based on the first product value and the second product value.

[0153] In an optional implementation, if the confidence level of the predicted fault probability is greater than a preset threshold, the fault information corresponding to the predicted fault probability having the preset fault value is sent to the cloud server, the countermeasures sent by the cloud server are received, and the countermeasures are provided to the user.

[0154] Among them, the fault information is used to instruct the language model deployed on the cloud server to output the response measures corresponding to the fault information. The language model is trained based on the maintenance dataset.

[0155] In the embodiment of the present disclosure, the fault information may be a fault code or a keyword for description.

[0156] In an embodiment of the present disclosure, if the confidence level of the predicted fault probability is greater than a preset threshold, it indicates that the result of the fault prediction is credible, and the fault information corresponding to the predicted fault probability with a preset fault value is sent to the cloud server, thereby instructing the large language model deployed on the cloud server to output countermeasures corresponding to the fault information. The human-computer interface receives the countermeasures sent by the cloud server and provides the countermeasures to the user through an output module connected to the human-computer interface. Specifically, the countermeasures can be displayed in the form of text on the display screen or prompted by voice.

[0157] In the fault prediction method provided by the embodiment of the present disclosure, multiple time series and the predicted fault probability corresponding to each time series are obtained, wherein the time series is monitoring data obtained by sequentially monitoring multiple pre-set checkpoints for fault detection. If the predicted fault probabilities corresponding to the multiple time series meet the preset conditions, then for each time series, the conditional probability corresponding to the time series is obtained, wherein the conditional probability is: when the predicted fault probability of the preset time series corresponding to the time series is the preset fault value, the predicted fault probability corresponding to the time series is also the probability of the preset fault value. Based on the predicted fault probabilities and conditional probabilities corresponding to the multiple time series, the confidence of the predicted fault probability is determined. It can be seen that the present disclosure can calculate the corresponding conditional probability for each time series, and then calculate the confidence of the predicted fault probability based on the predicted failure probability and conditional probability of each time series. In this way, when calculating the confidence, the same correlation between the predicted fault probabilities of the two corresponding time series can be taken into account, so that the calculated confidence is more accurate. In this way, the accuracy of the predicted fault probability can be expressed intuitively and accurately.

[0158] The present disclosure also provides a fault prediction method for a cloud server. Figure 3 A flowchart of another fault prediction method provided in an embodiment of the present disclosure.

[0159] Specifically include:

[0160] S301: Train a fault prediction model based on a fault dataset using a model trainer.

[0161] The trained fault prediction model is deployed on the vehicle to determine the predicted fault probability corresponding to the time series based on the time series.

[0162] Among them, the predicted failure probability is used to indicate the vehicle. If the predicted failure probabilities corresponding to multiple time series meet the preset conditions, then for each time series, the conditional probability corresponding to the time series is obtained. The conditional probability is: when the predicted failure probability of the preset time series corresponding to the time series is the preset failure value, the predicted failure probability corresponding to the time series is also the probability of the preset failure value, and based on the predicted failure probabilities and conditional probabilities corresponding to the multiple time series, the confidence of the predicted failure probability is determined, wherein the time series is monitoring data obtained by sequentially monitoring multiple pre-set checkpoints for fault detection.

[0163] Specifically, the method of acquiring the fault data set is described in the above embodiment.

[0164] In the embodiment of the present disclosure, after obtaining the fault data set, the cloud server uses the fault data set as input to train the fault prediction model through a model trainer deployed on the cloud server.

[0165] After the fault prediction model is trained, the trained fault prediction model is deployed on the vehicle to determine the predicted fault probability corresponding to the time series based on the time series.

[0166] Specifically, how to determine the predicted failure probability corresponding to the time series based on the time series and the failure prediction model is described in the above embodiment.

[0167] In the embodiment of the present disclosure, the predicted failure probability and conditional probability corresponding to each time series are used to determine the confidence of the predicted failure probability. Specifically, the process of determining the confidence of the predicted failure probability is described in the above embodiment.

[0168] In an optional implementation, training data sent by the vehicle is received and included in the fault dataset to update the fault dataset, and the fault prediction model is retrained based on the updated fault dataset through a model trainer.

[0169] The training data is obtained by marking a normal label on the time series when no fault is detected based on the time series, and marking a fault label for representing the fault information on the time series when a fault is detected based on the time series.

[0170] Among them, the retrained fault prediction model is used to update the fault prediction model historically deployed on the vehicle.

[0171] In the disclosed embodiment, the training data sent by the vehicle is included in the fault data set for updating the fault data set. Specifically, the training data is obtained by the vehicle marking a normal label for the time series when no fault is detected based on the time series, and marking the time series with a fault label for characterizing the fault information when a fault is detected based on the time series, such as when an abnormality such as stop reporting, report timeout, or report sequence error occurs at a checkpoint in the corresponding time series.

[0172] The training data can then be passed in through the model trainer to retrain the fault prediction model to improve the accuracy of the fault prediction model. The retrained fault prediction model is used to update the fault prediction model deployed on the vehicle. That is, by using the updated fault prediction model, the fault prediction function on the vehicle can be improved and updated.

[0173] In the disclosed embodiment, by updating the fault prediction model on the vehicle, the accuracy and reliability of fault prediction can be improved. Through continuous iteration and updating, the vehicle system can continuously improve the fault prediction capability and improve the reliability and safety of the vehicle.

[0174] In an optional embodiment, the cloud server may also include a large language model. Through the large language model, it can receive fault information corresponding to the predicted fault probability with a preset fault value sent by the vehicle, and output countermeasures corresponding to the fault information, and send the countermeasures to the vehicle so that the vehicle can provide the countermeasures to the user.

[0175] In the disclosed embodiment, the large language model can be obtained by training the maintenance data set.

[0176] Specifically, the data structure of the maintenance dataset can be [input, language label], where the input can be fault information, such as a fault code or a keyword used to describe the fault, and the language label can be the handling suggestions for the fault proposed by safety personnel, maintenance personnel, or technical experts, as shown in Table 9 below:

[0177]

[0178] Table 9

[0179] As shown in Table 9 above, the input is a fault code, and the language label indicates a predicted basic software failure. This failure may cause the large screen to freeze or go dark. Please remain calm, quickly take control of the vehicle, and disable all intelligent driving features using Method 1. After parking in a safe area, try restarting the vehicle using Method 2. If the problem persists, contact a professional for repair.

[0180] It should be noted that Table 9 is only an example and should not be considered as limiting.

[0181] Furthermore, in order to improve the accuracy of the response measures corresponding to the fault information output by the language model, the speech model can also be retrained based on the updated maintenance data set. In an optional implementation, the response measures corresponding to the fault information corresponding to the new fault are obtained, and the fault information and its corresponding response measures are included in the maintenance data set to update the maintenance data set. Based on the updated maintenance data set, the language model is retrained to update the language model.

[0182] In the disclosed embodiment, fine-tuning the trained language model (i.e., retraining the language model) based on the updated maintenance dataset is a transfer learning technique. The fine-tuning method varies depending on the model used, but should include the following steps:

[0183] First, prepare the maintenance data set, load the language model, set the language model parameters, load the maintenance data set, set the language model training parameters, and train the language model.

[0184] In the disclosed embodiment, the maintenance data set can be updated by obtaining fault information corresponding to a new fault and its corresponding countermeasures and including them in the maintenance data set. This allows the maintenance data set to contain more fault information and corresponding solutions, so that more accurate and effective countermeasures can be provided in the event of future vehicle failures.

[0185] Afterwards, the language model can be retrained based on the updated maintenance data set to update the language model. Through retraining, the language model can learn the correlation between new fault information and response measures, thereby improving the response capability of fault identification.

[0186] It can be seen that the embodiment of the present disclosure can improve the accuracy and reliability of the language big model by updating the language big model, so that it can better understand and generate response measures applicable to various fault situations, and help provide more accurate and practical maintenance suggestions so that the vehicle can quickly solve the problem when a fault occurs.

[0187] The fault prediction method provided by the embodiment of the present disclosure can train a fault prediction model based on a fault data set through a model trainer deployed in a cloud service. The trained fault model can be deployed on a vehicle to determine the predicted fault probability of the time series based on a time series, wherein the predicted fault probability is used to indicate that if the predicted fault probabilities corresponding to multiple time series meet a preset condition, then for each time series, the conditional probability corresponding to the time series is obtained. The conditional probability is: when the predicted fault probability of a preset time series corresponding to the time series is a preset fault value, the predicted fault probability corresponding to the time series is also the probability of the preset fault value, and based on multiple The predicted failure probability and conditional probability corresponding to each time series are used to determine the confidence of the predicted failure probability, wherein the time series is monitoring data obtained by sequentially monitoring multiple pre-set checkpoints for fault detection. It can be seen that the embodiment of the present disclosure can calculate the corresponding conditional probability for each time series, and then calculate the confidence of the predicted failure probability based on the predicted failure probability and conditional probability of each time series. In this way, when calculating the confidence, the same correlation between the predicted failure probabilities of the two corresponding time series can be taken into account, so that the calculated confidence is more accurate. In this way, the accuracy of the predicted failure probability can be expressed intuitively and accurately.

[0188] The embodiments of the present disclosure also provide a vehicle, specifically, the vehicle includes a prediction subsystem and a diagnosis subsystem.

[0189] The prediction subsystem is used to obtain multiple time series and the predicted failure probability corresponding to each time series, wherein the time series is monitoring data obtained by sequentially monitoring multiple pre-set checkpoints for fault detection, and if the predicted failure probabilities corresponding to the multiple time series meet the preset conditions, then for each time series, the conditional probability corresponding to the time series is obtained, wherein the conditional probability is: when the predicted failure probability of the preset time series corresponding to the time series is the preset failure value, the predicted failure probability corresponding to the time series is also the probability of the preset failure value, and based on the predicted failure probabilities and conditional probabilities corresponding to the multiple time series, the confidence of the predicted failure probability is determined.

[0190] The prediction subsystem is further configured to send fault information corresponding to the predicted fault probability having a preset fault value to the cloud server if the confidence level of the predicted fault probability is greater than a preset threshold, wherein the fault information is used to instruct a large language model deployed on the cloud server to output a response measure corresponding to the fault information, the large language model being trained based on the maintenance dataset;

[0191] Human-machine interface, used to receive countermeasures sent by the cloud server;

[0192] Output module, used to provide response measures to users;

[0193] The prediction subsystem is further configured to input the time series into a corresponding fault prediction model and obtain the predicted failure probability output by the corresponding fault prediction model, wherein the fault prediction model is trained based on the fault dataset by a model trainer deployed on the cloud server;

[0194] The diagnostic subsystem is configured to receive monitoring data corresponding to a checkpoint reported by a software module corresponding to the software checkpoint; and / or periodically monitor hardware checkpoints to obtain corresponding monitoring data.

[0195] The diagnostic subsystem is further configured to label the time series as normal if no fault is detected based on the time series, and label the time series as fault information if a fault is detected based on the time series, thereby obtaining training data; and to send the training data to the cloud server so that the cloud server updates the training data in the fault data set, and if the fault information corresponds to a new fault, updates the fault information and its corresponding countermeasures in the maintenance data set.

[0196] It should be noted that the various processes and effects in the embodiment of the present disclosure are the same as those in the method embodiment of S101 to S103, and are not described in detail here.

[0197] The present disclosure also provides a vehicle, comprising at least one of the following:

[0198] The fault prediction device according to any one of the above embodiments;

[0199] The fault prediction device according to any one of the above embodiments;

[0200] The computer-readable storage medium according to any of the above embodiments.

[0201] The embodiment of the present disclosure also provides a cloud server. Specifically, the cloud server includes a fault data set, a maintenance data set, a fault model trainer, and a large speech model.

[0202] The model trainer is used to train the fault prediction model based on the fault dataset;

[0203] The trained fault prediction model is deployed on the vehicle to determine the predicted fault probability corresponding to the time series based on the time series;

[0204] Among them, the predicted failure probability is used to indicate the vehicle. If the predicted failure probabilities corresponding to multiple time series meet the preset conditions, then for each time series, the conditional probability corresponding to the time series is obtained. The conditional probability is: when the predicted failure probability of the preset time series corresponding to the time series is the preset failure value, the predicted failure probability corresponding to the time series is also the probability of the preset failure value, and based on the predicted failure probabilities and conditional probabilities corresponding to the multiple time series, the confidence of the predicted failure probability is determined, wherein the time series is monitoring data obtained by sequentially monitoring multiple pre-set checkpoints for fault detection.

[0205] The cloud server is further configured to receive training data sent by the vehicle and include the training data in the fault dataset to update the fault dataset. The training data is obtained by the vehicle assigning a normal label to the time series when no fault is detected based on the time series, and assigning a fault label to the time series to represent the fault information when a fault is detected based on the time series.

[0206] The model trainer is further used to retrain the fault prediction model based on the updated fault dataset, wherein the retrained fault prediction model is used to update the fault prediction model historically deployed on the vehicle;

[0207] The cloud server also includes: a maintenance data set and a large language model, the large language model being configured to receive fault information sent by the vehicle corresponding to a predicted fault probability having a value of a preset fault value, output a countermeasure corresponding to the fault information, and transmit the countermeasure to the vehicle so that the vehicle provides the countermeasure to the user;

[0208] The cloud server is further used to obtain the corresponding countermeasures of the fault information corresponding to the new fault, and include the fault information and the corresponding countermeasures in the maintenance data set to update the maintenance data set;

[0209] The language model is also used to retrain the language model based on an updated maintenance dataset to update the language model.

[0210] It should be noted that the various processes and effects in the embodiment of the present disclosure are the same as those in the method embodiment of S301 and are not described in detail here.

[0211] The present disclosure also provides a fault prediction system. Figure 4 , which is a structural diagram of a fault prediction system provided in an embodiment of the present disclosure.

[0212] The fault prediction system provided in the embodiments of the present disclosure has the function of predicting possible faults in the vehicle's electronic and electrical systems and promptly notifying the user of the corresponding methods for the faults to improve the controllability of the entire vehicle (controllability is an important indicator for evaluating system safety in the ISO26262 and ISO21448 standards).

[0213] like Figure 4 As shown, the fault prediction system 400 includes a vehicle 401 and a cloud service 402 .

[0214] The vehicle 401 includes a prediction subsystem and a diagnosis subsystem. The prediction subsystem is responsible for fault prediction, and the diagnosis subsystem is responsible for load fault monitoring and sample collection.

[0215] The cloud server 402 includes a fault data set, a maintenance data set, a fault model trainer, and a speech big model. The fault data set fault model trainer is used to train the fault prediction module, and the maintenance data set speech big model is used to provide users with suggestions for handling faults.

[0216] It should be noted that Figure 4 The fault prediction system shown is similar to Figure 1 、 Figure 3 The fault prediction method in the illustrated method embodiment has the same effect and will not be described in detail here.

[0217] Below, the fault prediction system provided by the embodiment of the present disclosure will be described in detail based on a specific example.

[0218] Figure 5 A schematic diagram of the workflow of a fault prediction system provided in an embodiment of the present disclosure.

[0219] like Figure 5 As shown, the fault prediction system workflow may specifically include the following steps 1 to 9.

[0220] Step 1: When the diagnostic subsystem identifies a fault, it uploads historical data (i.e., multiple time series) and fault codes (i.e., fault information) to the fault dataset on the cloud.

[0221] Step 2: The fault data set provides samples for training the fault prediction model to the fault model trainer.

[0222] Step 3: Safety personnel, maintenance personnel, technical experts, etc. formulate natural language suggestions based on the fault information in the fault dataset and enter them into the maintenance dataset.

[0223] Step 4: The language model is enhanced and fine-tuned based on the maintenance dataset.

[0224] Step 5: The fault model trainer retrains the fault prediction model based on the newly added data and updates it to the prediction subsystem on the automotive E / E system side.

[0225] Step 6: When the vehicle system is running, the diagnostic subsystem monitors the vehicle and records the monitored data.

[0226] Step 7: The diagnosis subsystem inputs the monitored checkpoints (ie, time series) into the fault prediction model.

[0227] Step 8: When the fault prediction model predicts a fault, it generates a corresponding fault prompt word (i.e., fault information) and uploads it to the language model.

[0228] Step 9: The language model outputs the response to the fault information and returns it to the human-machine interface (such as the cockpit screen). If the fault occurs in the future, the user has obtained the fault response method based on the human-machine interface and can operate the vehicle accordingly to avoid danger.

[0229] It can be seen that the fault prediction system of the embodiment of the present disclosure, which uses a data-driven, model-based and knowledge-based approach to perform health management, has the following advantages:

[0230] By adding a way to identify faults, dangers can be predicted in advance, making the system safer and adaptable to load systems that are difficult to analyze in the future. By updating the fault prediction model, it is possible to quickly adapt to increasingly complex software and hardware logic. The accumulation of fault data sets and maintenance data sets can also achieve better fault prediction results, making the fault detection system more "smart" with use. It can also capture hidden connections in faults to facilitate design improvements, provide users with fault handling advice, and improve the controllability of the entire vehicle. In addition, the use of NPU (Neural Processing Unit) for prediction saves CPU resources.

[0231] Based on the above method embodiment, the present disclosure also provides a fault prediction device, referring to Figure 6 , is a structural diagram of a fault prediction device provided in an embodiment of the present disclosure, which is applied to a vehicle.

[0232] like Figure 6 As shown, the fault prediction device may include:

[0233] A first acquisition module 601 is configured to acquire multiple time series and a predicted failure probability corresponding to each time series, wherein the time series is monitoring data obtained by sequentially monitoring multiple pre-set checkpoints for fault detection;

[0234] A second acquisition module 602 is configured to acquire, for each time series, a conditional probability corresponding to the time series if the predicted failure probabilities corresponding to the multiple time series meet a preset condition, wherein the conditional probability is the probability that, when the predicted failure probability of a preset time series corresponding to the time series is a preset failure value, the predicted failure probability corresponding to the time series is also the preset failure value;

[0235] The determination module 603 is configured to determine the confidence level of the predicted failure probability based on the predicted failure probabilities and the conditional probabilities corresponding to the multiple time series.

[0236] In an optional embodiment, the preset conditions include that the predicted failure probability corresponding to at least one of the time series is greater than or equal to a first preset probability threshold, and / or that the average value of the predicted failure probabilities corresponding to the multiple time series is greater than or equal to a second preset probability threshold.

[0237] In an optional implementation, the determining module 603 includes:

[0238] A first processing submodule, configured to multiply the conditional probabilities corresponding to the multiple time series to obtain a first product value;

[0239] a second processing submodule, configured to, for each of the time series, multiply the corresponding predicted failure probability and the conditional probability to obtain a first sub-product value, subtract the corresponding predicted failure probability from a preset value to obtain a first difference value, subtract the corresponding conditional probability from the preset value to obtain a second difference value, multiply the first difference value and the second difference value to obtain a second sub-product value, and sum the first sub-product value and the second sub-product value to obtain a first sum value;

[0240] a third processing submodule, configured to multiply the first sum values ​​corresponding to the multiple time series to obtain a second product value;

[0241] A determination submodule is configured to determine a confidence level of the predicted failure probability based on the first product value and the second product value.

[0242] In an optional embodiment, the method further includes:

[0243] A first sending module is configured to send fault information corresponding to the predicted fault probability having a preset fault value to a cloud server if the confidence level of the predicted fault probability is greater than a preset threshold, wherein the fault information is used to instruct a large language model deployed on the cloud server to output a response measure corresponding to the fault information, the large language model being trained based on a maintenance dataset;

[0244] The first receiving module is used to receive the countermeasure sent by the cloud server and provide the countermeasure to the user.

[0245] An optional implementation includes:

[0246] A third acquisition module is used to obtain the predicted failure probability corresponding to the time series;

[0247] The third acquisition module is specifically used to input the time series into the corresponding fault prediction model to obtain the predicted failure probability output by the corresponding fault prediction model, wherein the fault prediction model is trained by a model trainer deployed on the cloud server based on the fault data set.

[0248] In an optional embodiment, the checkpoint includes a software checkpoint and / or a hardware checkpoint.

[0249] Accordingly, the fourth acquisition module is used to obtain the time series process;

[0250] The fourth acquisition module includes:

[0251] A receiving submodule, configured to receive monitoring data corresponding to the checkpoint reported by the software module corresponding to the software checkpoint;

[0252] And / or, a monitoring submodule, configured to periodically monitor the hardware checkpoints to obtain corresponding monitoring data.

[0253] In an optional embodiment, the method further includes:

[0254] a labeling module, configured to label the time series as normal if no fault is detected based on the time series, and label the time series as fault information if a fault is detected based on the time series, thereby obtaining training data;

[0255] The second sending module is used to send the training data to the cloud server, so that the cloud server updates the training data in the fault data set, and if the fault information corresponds to a new fault, updates the fault information and its corresponding countermeasures in the maintenance data set.

[0256] It should be noted that Figure 6 The fault prediction device 600 shown can perform Figure 1 The various steps in the method embodiment shown are implemented Figure 1 The various processes and effects in the illustrated method embodiment are not described in detail here.

[0257] The present disclosure also provides a fault prediction device, referring to Figure 7 , which is a structural diagram of another fault prediction device provided in an embodiment of the present disclosure, applied to a cloud server.

[0258] like Figure 7 As shown, the fault prediction device may include:

[0259] A first training module 701 is configured to train a fault prediction model based on a fault dataset using a model trainer;

[0260] The trained fault prediction model is deployed on a vehicle to determine a predicted fault probability corresponding to the time series based on the time series;

[0261] Among them, the predicted failure probability is used to indicate the vehicle. If the predicted failure probabilities corresponding to multiple time series meet preset conditions, then for each of the time series, the conditional probability corresponding to the time series is obtained, and the conditional probability is: when the predicted failure probability of the preset time series corresponding to the time series is the preset failure value, the predicted failure probability corresponding to the time series is also the probability of the preset failure value, and based on the predicted failure probabilities and the conditional probabilities corresponding to the multiple time series, the confidence of the predicted failure probability is determined, wherein the time series is monitoring data obtained by sequentially monitoring multiple pre-set checkpoints for fault detection.

[0262] In an optional embodiment, the method further includes:

[0263] a second receiving module, configured to receive training data sent by the vehicle and include the training data in the fault dataset to update the fault dataset, wherein the training data is obtained by the vehicle marking the time series with a normal label when no fault is detected based on the time series, and marking the time series with a fault label for representing fault information when a fault is detected based on the time series;

[0264] The second training module is used to retrain the fault prediction model based on the updated fault data set through the model trainer, wherein the retrained fault prediction model is used to update the fault prediction model historically deployed on the vehicle.

[0265] In an optional embodiment, the method further includes:

[0266] a third receiving module, configured to receive, through the language large model, fault information corresponding to the predicted fault probability having a value equal to the preset fault value, sent by the vehicle, and output a countermeasure corresponding to the fault information;

[0267] The third sending module is used to send the countermeasure to the vehicle, so that the vehicle provides the countermeasure to the user.

[0268] In an optional embodiment, the method further includes:

[0269] a fifth acquisition module, configured to acquire a countermeasure corresponding to fault information corresponding to the new fault, and include the fault information and the corresponding countermeasure in a maintenance data set to update the maintenance data set;

[0270] The third training module is used to retrain the language model based on the updated maintenance dataset to update the language model.

[0271] It should be noted that Figure 7 The fault prediction device 700 shown can perform Figure 3 The various steps in the method embodiment shown are implemented Figure 3 The various processes and effects in the illustrated method embodiment are not described in detail here.

[0272] In addition, the present disclosure also provides a fault prediction device, see Figure 8 As shown, this may include:

[0273] Processor 801, memory 802, input device 803 and output device 804. The number of processors 801 in the fault prediction device can be one or more. Figure 8 In some embodiments of the present disclosure, the processor 801, the memory 802, the input device 803 and the output device 804 may be connected via a bus or other means, wherein: Figure 8 The bus connection is taken as an example.

[0274] The memory 802 can be used to store software programs and modules. The processor 801 executes the software programs and modules stored in the memory 802 to implement various functional applications and data processing of the fault prediction device. The memory 802 may primarily include a program storage area and a data storage area, wherein the program storage area may store an operating system, at least one application required for a function, and the like. In addition, the memory 802 may include high-speed random access memory and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device. The input device 803 may be used to receive input digital or character information and generate signal input related to user settings and function control of the fault prediction device.

[0275] Specifically in this embodiment, the processor 801 will load the executable files corresponding to the processes of one or more applications into the memory 802 according to the following instructions, and the processor 801 will run the applications stored in the memory 802, thereby realizing the various functions of the above-mentioned fault prediction device.

[0276] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or device comprising the element.

[0277] The foregoing description is intended only to provide specific embodiments of the present disclosure, intended to enable those skilled in the art to understand and implement the present disclosure. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present disclosure. Therefore, the present disclosure is not intended to be limited to the embodiments described herein, but rather to be construed in the broadest manner consistent with the principles and novel features disclosed herein.

Claims

1. A fault prediction method, characterized in that: include: Obtaining multiple time series and a predicted failure probability corresponding to each time series, wherein the time series is monitoring data obtained by sequentially monitoring multiple pre-set checkpoints for fault detection; If the predicted failure probabilities corresponding to the multiple time series meet a preset condition, then for each time series, a conditional probability corresponding to the time series is obtained, wherein the conditional probability is: when the predicted failure probability of the preset time series corresponding to the time series is a preset failure value, the predicted failure probability corresponding to the time series is also the preset failure value; The confidence level of the predicted failure probability is determined based on the predicted failure probabilities and the conditional probabilities corresponding to the multiple time series.

2. The method according to claim 1, characterized in that The preset conditions include that the predicted failure probability corresponding to at least one of the time series is greater than or equal to a first preset probability threshold, and / or that the average value of the predicted failure probabilities corresponding to the multiple time series is greater than or equal to a second preset probability threshold.

3. The method according to claim 1, characterized in that The determining, based on the predicted failure probabilities and the conditional probabilities corresponding to the multiple time series, a confidence level of the predicted failure probability, includes: Multiplying the conditional probabilities corresponding to the multiple time series to obtain a first product value; For each of the time series, multiplying the corresponding predicted failure probability and the conditional probability to obtain a first sub-product value, subtracting the corresponding predicted failure probability from a preset value to obtain a first difference value, subtracting the corresponding conditional probability from the preset value to obtain a second difference value, multiplying the first difference value and the second difference value to obtain a second sub-product value, and adding the first sub-product value and the second sub-product value to obtain a first sum value; Multiplying the first sum values ​​corresponding to the multiple time series to obtain a second product value; A confidence level of the predicted failure probability is determined based on the first product value and the second product value.

4. The method according to claim 1, wherein Also includes: If the confidence level of the predicted fault probability is greater than a preset threshold, the fault information corresponding to the predicted fault probability with the preset fault value is sent to the cloud server, wherein the fault information is used to instruct the language model deployed on the cloud server to output a response measure corresponding to the fault information, and the language model is trained based on the maintenance dataset; The countermeasure is received from the cloud server, and the countermeasure is provided to the user.

5. The method according to claim 1, wherein The process of obtaining the predicted failure probability corresponding to the time series includes: The time series is input into a corresponding fault prediction model to obtain the predicted fault probability output by the corresponding fault prediction model, wherein the fault prediction model is trained based on a fault dataset by a model trainer deployed on a cloud server.

6. The method according to claim 1, characterized in that The checkpoints include software checkpoints and / or hardware checkpoints. Accordingly, the process of obtaining the time series includes: Receiving monitoring data corresponding to the checkpoint reported by the software module corresponding to the software checkpoint; And / or, periodically monitor the hardware checkpoints to obtain corresponding monitoring data.

7. The method according to claim 6, characterized in that Also includes: If no fault is detected based on the time series, a normal label is marked on the time series, and if a fault is detected based on the time series, a fault information label is marked on the time series to obtain training data; The training data is sent to a cloud server so that the cloud server updates the training data in a fault data set, and if the fault information corresponds to a new fault, updates the fault information and its corresponding countermeasures in a maintenance data set.

8. A fault prediction device, characterized in that: The device comprises: A first acquisition module is configured to acquire a plurality of time series and a predicted failure probability corresponding to each time series, wherein the time series is monitoring data obtained by sequentially monitoring a plurality of pre-set checkpoints for fault detection; A second acquisition module is configured to acquire, for each time series, a conditional probability corresponding to the time series if the predicted failure probabilities corresponding to the multiple time series meet a preset condition, wherein the conditional probability is the probability that, when the predicted failure probability of a preset time series corresponding to the time series is a preset failure value, the predicted failure probability corresponding to the time series is also the preset failure value; A determination module is used to determine the confidence of the predicted failure probability based on the predicted failure probability and the conditional probability corresponding to the multiple time series.

9. A fault prediction device, characterized in that: The system comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that The storage medium stores a computer program, and when the computer program is executed by a processor, the processor implements the method according to any one of claims 1 to 7.

11. A vehicle, characterized in that: Include at least one of the following: The fault prediction device according to claim 8; The fault prediction device according to claim 9; The computer-readable storage medium of claim 10.

Images