Rule generation method, fault prediction method and related device

Abstract

The invention provides a rule generation method, a fault prediction method and a related device, and relates to the technical field of data processing. The rule generation method can comprise the steps of obtaining a target sample; wherein the target sample comprises a positive sample and a negative sample, the positive sample refers to a sample with a target fault, and the negative sample refers to a sample without a fault; according to an output time sequence, searching at least one positive rule of which the occurrence frequency is greater than a frequency threshold value in the positive sample, and searching at least one negative rule of which the occurrence frequency is greater than the frequency threshold value in the negative sample; and obtaining a combined rule for fault prediction based on the positive rule and the negative rule. According to the technical scheme provided by the invention, the problems of limited coverage range and low accuracy of artificially defined fault prediction rules in the prior art can be solved.

Description

Technical Field

[0001] This application relates to the field of data processing technology, specifically to the field of fault prediction technology, and more specifically to a rule generation method, a fault prediction method, and related apparatus. Background Technology

[0002] In modern industry and various automated systems, diagnostic fault codes (DTCs) play a crucial role. By reading and analyzing DTCs, maintenance personnel can quickly locate the cause of faults and improve maintenance efficiency.

[0003] With technological advancements, diagnostic fault codes can not only be used to quickly pinpoint the cause of faults, but also to construct fault prediction rules. Currently, these rules are primarily defined manually by experts based on experience or historical data. However, manually defined rules have limited coverage, making it difficult to encompass all possible fault scenarios, especially complex or rare fault modes. Furthermore, the accuracy of these rules is highly dependent on the expert's skill level, which can easily lead to missed or false alarms during prediction, thus affecting the reliability of fault prediction. Summary of the Invention

[0004] Based on the defects and shortcomings of the prior art, this application proposes a rule generation method, a fault prediction method, and related apparatus, which can solve the problems of limited coverage and low accuracy of manually defined fault prediction rules in the prior art.

[0005] According to a first aspect of the embodiments of this application, a rule generation method is provided, the method comprising:

[0006] Obtain target samples; wherein each target sample includes at least one fault reminder message ordered in chronological order of output time; the target samples include positive samples and negative samples, wherein positive samples refer to samples in which the target fault has occurred, and negative samples refer to samples in which the fault has not occurred;

[0007] In accordance with the output time sequence, at least one positive rule is found in the positive samples whose frequency is greater than or equal to a frequency threshold, and at least one negative rule is found in the negative samples whose frequency is greater than or equal to a frequency threshold; wherein, both the positive rule and the negative rule consist of at least one fault reminder message ordered in chronological order of output time.

[0008] Based on the positive and negative rules, a combined rule for fault prediction is obtained; wherein, if the fault reminder information sequence output by the target device hits the positive rule in the combined rule but does not hit any negative rule in the combined rule, it indicates that the target fault may occur.

[0009] According to a second aspect of the embodiments of this application, a fault prediction method is provided, the method comprising:

[0010] Obtain a sequence of fault alert information output by the target device; wherein the sequence of fault alert information includes at least one fault alert message;

[0011] Determine whether the fault alert information sequence matches at least one of multiple combined rules; wherein, the multiple combined rules are used for fault prediction, each combined rule includes at least one positive rule and at least one negative rule, and both the positive rule and the negative rule include at least one fault alert information ordered by output time; if the fault alert information sequence output by the target device matches the positive rule in the combined rules and does not match any of the negative rules in the combined rules, it indicates that a fault corresponding to the combined rule may occur;

[0012] If the fault alert information sequence is determined to match at least one of multiple combined rules, fault prediction information is sent to the target object; wherein, the fault prediction information includes possible faults of the target device; the target object includes at least one of the following: the target device, the after-sales service terminal, and the user terminal associated with the target device.

[0013] According to a third aspect of the embodiments of this application, a rule generation apparatus is provided, the apparatus comprising:

[0014] The first acquisition module is used to acquire target samples; wherein each target sample includes at least one fault reminder message ordered in chronological order of output time; the target samples include positive samples and negative samples, wherein the positive samples are samples in which the target fault has occurred, and the negative samples are samples in which the fault has not occurred;

[0015] The mining module is used to search for at least one positive rule in the positive samples whose frequency is greater than or equal to a frequency threshold, and to search for at least one negative rule in the negative samples whose frequency is greater than or equal to a frequency threshold, according to the output time sequence; wherein, both the positive rule and the negative rule are composed of at least one fault reminder message ordered according to the output time sequence;

[0016] The first processing module is used to obtain a combined rule for fault prediction based on the positive rule and the negative rule; wherein, if the fault reminder information sequence output by the target device hits the positive rule in the combined rule but does not hit any negative rule in the combined rule, it indicates that the target fault may occur.

[0017] According to a fourth aspect of the embodiments of this application, a fault prediction apparatus is provided, the apparatus comprising:

[0018] The second acquisition module is used to acquire a sequence of fault reminder information output by the target device; wherein the sequence of fault reminder information includes at least one fault reminder information;

[0019] A matching module is used to determine whether the fault alert information sequence matches at least one of a plurality of combined rules; wherein, the plurality of combined rules are used for fault prediction, each combined rule includes at least one positive rule and at least one negative rule, and both the positive rule and the negative rule include at least one fault alert information ordered by output time; if the fault alert information sequence output by the target device matches the positive rule in the combined rules and does not match any of the negative rules in the combined rules, it indicates that a fault corresponding to the combined rule may occur;

[0020] The second processing module is used to send fault prediction information to a target object when it is determined that the fault reminder information sequence hits at least one of multiple combination rules; wherein, the fault prediction information includes possible faults of the target device; the target object includes at least one of the following: the target device, an after-sales service terminal, and a user terminal associated with the target device.

[0021] According to a fifth aspect of the embodiments of this application, an electronic device is provided, including: a memory and a processor;

[0022] The memory is connected to the processor and is used to store programs;

[0023] The processor is used to implement the rule generation method as described in the first aspect or the fault prediction method as described in the second aspect by running the program in the memory.

[0024] According to a sixth aspect of the embodiments of this application, a storage medium is provided, on which a computer program is stored, and when the computer program is run by a processor, it implements the rule generation method as described in the first aspect or the fault prediction method as described in the second aspect.

[0025] According to a seventh aspect of the embodiments of this application, a computer program product or a computer program is provided, the computer program product including the computer program, wherein when a processor executes the computer program, it implements the steps in the rule generation method as described in the first aspect or the fault prediction method as described in the second aspect.

[0026] Beneficial effects:

[0027] In the technical solution provided in this application, the target sample is used as input. Based on the chronological order of fault alert information output and the frequency of occurrence, the system automatically searches for frequently occurring positive rules in positive samples and frequently occurring negative rules in negative samples. Finally, based on the positive and negative rules, a combined rule for fault prediction is obtained. This method requires no expert involvement in rule construction, significantly reducing the knowledge modeling threshold and operational costs. It overcomes the drawback of existing systems where rule construction relies on manual intervention. Furthermore, the resulting combined rule is highly comprehensive, reducing omissions and improving rule coverage. In addition, the combined rule includes not only positive rules but also negative rules. When performing fault prediction, it not only hits positive rules but also excludes cases where negative rules are hit, reducing the false alarm rate. Furthermore, when performing rule mining, this application considers the occurrence logic of fault alert information in the time dimension. As a result, the mined rules also retain the output time features. When performing fault prediction, it is necessary not only to hit the fault alert information included in the rules, but also to match the rules in the time dimension. This can improve the accuracy of prediction and further reduce the false alarm rate. Attached Figure Description

[0028] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0029] Figure 1 A flowchart illustrating a rule generation method provided in an embodiment of this application;

[0030] Figure 2 A schematic diagram of a system structure provided in an embodiment of this application;

[0031] Figure 3 A flowchart illustrating a multi-objective optimization process provided in an embodiment of this application;

[0032] Figure 4 A flowchart illustrating a fault prediction method provided in an embodiment of this application;

[0033] Figure 5 A block diagram of a rule generation apparatus provided in an embodiment of this application;

[0034] Figure 6 A block diagram of a fault prediction device provided in an embodiment of this application;

[0035] Figure 7 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. Detailed Implementation

[0036] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0037] Application Overview

[0038] In modern industry and various automated systems, diagnostic fault codes play a crucial role. A diagnostic fault code is a specific code generated by an electronic control system to indicate a fault or abnormal condition that occurs during the operation of the system or equipment. When a sensor, actuator, or other critical component of the system malfunctions, the relevant electronic control unit (ECU) can generate a corresponding diagnostic fault code based on preset logic and algorithms, and store it in memory so that maintenance personnel can quickly and accurately locate and resolve the problem.

[0039] For example, in the automotive field, the engine control unit (ECU) can monitor various engine operating parameters in real time, such as coolant temperature, oil temperature, intake air pressure, and oxygen sensor voltage. When these parameters exceed the normal range, the ECU can generate corresponding diagnostic fault codes. By reading these fault codes, repair personnel can quickly understand the approximate location of the fault in the vehicle's engine system, thus enabling targeted repairs.

[0040] With technological advancements, diagnostic fault codes can not only be used to quickly pinpoint the cause of a fault, but also to construct fault prediction rules, where each rule consists of at least one diagnostic fault code. Currently, fault prediction rules are primarily defined manually by experts based on experience or historical data. By modeling the statistical co-occurrence relationship between diagnostic fault codes and actual faults, association rules with high support and high confidence can be identified from a large amount of historical maintenance data. For example, if DTC_105 and DTC_231 appear simultaneously, there is a high probability of a specific fault occurring. Therefore, DTC_105 and DTC_231 can be combined to form a fault prediction rule, with an AND relationship between them.

[0041] However, faced with complex and ever-changing system failure scenarios and constantly emerging new failure modes, rule selection becomes a bottleneck when there are too many rules. It is difficult for humans to exhaustively cover all possible failure combinations and relationships; that is, manually defined rules have limited coverage and cannot encompass all possible failure scenarios, especially complex or rare failure modes, easily leading to missed predictions. Furthermore, the accuracy of rules is greatly affected by the level of expert knowledge, easily leading to false positives and affecting the reliability of failure prediction. Therefore, there is an urgent need for a more intelligent and automated rule generation method to reduce human intervention and improve the coverage and accuracy of failure prediction.

[0042] To address this, this application provides a rule generation scheme. During rule generation, historical fault samples are used as input. Following the chronological order of fault alert output, and based on frequency of occurrence, the scheme automatically searches for frequently occurring positive rules in positive samples and frequently occurring negative rules in negative samples. Finally, based on the positive and negative rules, a combined rule for fault prediction is obtained. This application, by constructing an end-to-end data-driven rule generation process, eliminates reliance on expert experience, improves knowledge construction efficiency, and overcomes the drawback of manual rule construction in existing systems.

[0043] Furthermore, the resulting combined rules are highly comprehensive, reducing omissions and improving rule coverage. In addition, these combined rules include not only positive rules but also negative rules. When performing fault prediction, while hitting positive rules, cases where negative rules are hit are excluded, reducing the false alarm rate.

[0044] Furthermore, in the rule mining process of this application embodiment, the occurrence logic of fault reminder information in the time dimension is considered. As a result, the output time feature is also retained in the rules mined. When performing fault prediction, it is necessary not only to hit the fault reminder information included in the rules, but also to match the rules in the time dimension. This can improve the accuracy of prediction and reduce the false alarm rate.

[0045] Exemplary Methods

[0046] This application provides a rule generation method applied to a cloud server.

[0047] The method is described in detail below through some embodiments. The following embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0048] like Figure 1 As shown, the method may include steps 101 to 103, as described below:

[0049] Step 101: Obtain the target sample.

[0050] The target samples mentioned here refer to sequences of historical fault alerts output by different devices of the same type (such as different vehicles of the same model, different drones of the same model, etc.). There can be multiple target samples. For example... Figure 2 As shown, the ECU on the device side can collect fault alert information in real time, and upload it to the cloud after being encapsulated in the form of a timestamp sequence.

[0051] This sequence includes at least one fault alert message. If there are multiple fault alert messages, they are ordered according to their output time, with the first message appearing first and the last message appearing last. The output time mentioned here refers to the time when the corresponding device generates the fault alert message, i.e., the trigger time of the fault alert message.

[0052] The fault alert information mentioned here can be diagnostic fault codes, text prompts, or other types of fault alerts (such as audible alarms). Text prompts refer to text-based alerts that can appear as pop-ups on the central control screen or LCD instrument panel. Optionally, the text prompts mentioned here can specifically be classical Chinese prompts. Classical Chinese prompts refer to alarms or fault alerts presented in concise and refined text, such as "Engine overheating, do not continue driving," and may be accompanied by voice prompts. Their core purpose is to quickly convey key information through concise text.

[0053] When a sequence includes multiple fault alerts, it may include diagnostic fault codes and / or text prompts. For example, target sample A may be [DTC105, DTC231, ..., DTC105, text prompt 1], target sample B may be [DTC105, DTC231, DTC301, ..., DTC105], and target sample C may be [text prompt 1, text prompt 2, ..., text prompt n].

[0054] The target samples in this application embodiment include positive samples and negative samples. Positive samples refer to samples where the target fault has occurred, and negative samples refer to samples where the fault has not occurred. The target fault mentioned here can be any kind of fault that occurs in the equipment. For example, when the equipment is a vehicle, the target fault can be an engine fault, a transmission fault, or a brake failure fault, etc.

[0055] After obtaining historical fault alert information sequences from different devices in the cloud, fault tags can be added to these sequences based on whether a fault actually occurred. This labels indicate whether each sequence corresponds to a fault, thus forming target samples—that is, labeled sample data. For example, a positive sample could be [DTC105, DTC231, DTC105], indicating a fault occurred; a negative sample could be [DTC301, DTC231, DTC231], indicating no fault occurred. Figure 2 As shown, the cloud includes a sample library, which contains at least historical samples obtained from the device.

[0056] Whether a malfunction has occurred can be determined by maintenance personnel. Maintenance personnel can store information indicating whether a malfunction has occurred in the equipment, so the equipment can know whether it has malfunctioned.

[0057] Optionally, the device can periodically upload fault alert information sequences, such as weekly or monthly, depending on actual needs. Sometimes, the device outputting fault alert information does not necessarily indicate an actual malfunction. Therefore, if the device does not malfunction within a reporting cycle, the fault alert information sequence will be sent to the cloud only when the reporting time arrives. If a malfunction is confirmed within a reporting cycle, the fault alert information sequence from the start of that cycle to the confirmation of the malfunction will be sent to the cloud, and the timing will restart to collect fault alert information for the next cycle. Since maintenance personnel store information indicating whether a malfunction has occurred in the device, the device can know whether it has malfunctioned and thus determine whether to report a fault alert information sequence. Of course, when maintenance personnel determine that a vehicle malfunction has occurred, they can also manually operate the device to report the fault alert information sequence and simultaneously report the malfunction information.

[0058] Step 102: According to the output time sequence, find at least one positive rule in the positive samples whose frequency is greater than or equal to the frequency threshold, and find at least one negative rule in the negative samples whose frequency is greater than or equal to the frequency threshold.

[0059] As mentioned earlier, the rules used for fault prediction refer to what faults might occur when certain fault alerts appear. The higher the frequency of a certain fault occurring when certain fault alerts appear, the higher the accuracy of using such fault alerts to predict that fault. Therefore, after obtaining the target sample, rules for fault prediction can be found based on the frequency of occurrence of certain fault alerts in the target sample.

[0060] Furthermore, currently used frequent itemset mining methods primarily focus on the co-occurrence relationships of items, failing to reflect the temporal logic of fault alerts. The temporal order in which fault alerts are triggered may also be correlated with the occurrence of the fault. For example, in road test data of a hybrid vehicle, P0C73 (low inverter cooling pump speed) was recorded sequentially, followed by P0C78 (high inverter temperature alarm) approximately two minutes later, and then P0C79 (inverter over-temperature shutdown) two minutes after that, after which the vehicle entered power-limited mode. This temporal sequence clearly reflects the causal chain of "decreased cooling capacity → temperature rise → protection shutdown." If only the disordered co-occurrence of the three DTCs is statistically analyzed, it becomes difficult to distinguish between the root cause and the result, and early intervention at the P0C73 or P0C78 stage is impossible. Conversely, in a few samples, a reverse order occurs where "P0C78 precedes P0C73," yet the vehicle does not ultimately shut down, indicating that the different order reflects different risk levels. Therefore, the serialization rules that preserve the chronological order can more accurately capture the fault evolution pattern and achieve earlier and more reliable prediction. Thus, in this embodiment, the rules with high frequency of occurrence are searched in the target sample according to the output time order. Among the searched rules, the fault reminder information is sorted according to the output time priority order. The rules obtained can reflect the occurrence logic of the fault reminder information in the time dimension, which is conducive to further improving the accuracy of fault prediction.

[0061] Specifically, in this embodiment, positive rules with an occurrence frequency greater than or equal to a frequency threshold are searched in positive samples, and at least one negative rule with an occurrence frequency greater than or equal to a frequency threshold is searched in negative samples. Both positive and negative rules consist of at least one fault alert message ordered chronologically by output time; essentially, both positive and negative rules are sequences of fault alert messages. A positive rule indicates that its occurrence may lead to the target fault; a negative rule indicates that its occurrence should not lead to the target fault.

[0062] The frequency threshold mentioned here can be set according to actual needs, and this application embodiment does not impose specific limitations on it.

[0063] It should be noted that the positive and negative rules found in this step are not the final rules used for fault prediction.

[0064] Step 103: Based on the positive and negative rules, obtain the combined rules for fault prediction.

[0065] The combined rule described here is the final rule used for fault prediction. In this combined rule, the positive and negative rules are related by AND, and the positive and negative rules are related by OR. The negative rules are not related in the combined rule. Therefore, if the fault reminder information sequence output by the target device hits at least one positive rule in the combined rule and does not hit any negative rule in the combined rule, it indicates that there is a high probability that the target fault has occurred or will occur.

[0066] The resulting combination rules can be structurally expressed and stored for subsequent deployment and real-time inference. Each combination structure can be saved as a Boolean logic expression, that is, the positive and negative rules are organized into a logical expression structure (i.e., a Boolean logic expression). An example expression could be: This means that rules can be combined in interpretable forms such as "(positive rule 1 or positive rule 2) and not (negative rule 3 or negative rule 4)", thereby improving the expressive flexibility of the rule system. In addition, each combined rule can be accompanied by descriptions such as applicable scenarios, hit conditions, and rule source, so as to facilitate deployment in the cloud.

[0067] After deploying the combined rules in the cloud, the latest deployed combined rule set can be invoked to match the fault alert information sequence uploaded by the device, determine whether the predictive fault triggering conditions are met, and output the prediction result to the target object. The target object can be the target device, after-sales service terminal, user terminal, etc.

[0068] Currently in the industry, the relationship between fault alerts and faults still mainly relies on expert experience for extraction and organization, lacking a systematic, scalable, and automated rule mining mechanism. This manual approach results in high costs and long development cycles for rule systems, and makes it difficult to cope with the rapid expansion of new systems or new fault scenarios.

[0069] In this embodiment, historical fault samples (i.e., target samples) are used as input. Following the chronological order of fault alert output, and based on frequency of occurrence, the system automatically searches for frequently occurring positive rules in positive samples and frequently occurring negative rules in negative samples. Finally, based on the positive and negative rules, a combined rule for fault prediction is obtained. This method requires no expert involvement in rule construction, significantly reducing the knowledge modeling threshold and operational costs. It overcomes the drawback of existing systems where rule construction relies on manual intervention, and the method exhibits good generalization and transferability. Furthermore, the resulting combined rule is highly comprehensive, reducing omissions. Moreover, this combined rule includes not only positive rules but also negative rules, ensuring coverage while significantly reducing false alarm rates during fault prediction. Furthermore, in the rule mining process of this application, the occurrence logic of fault alert information in the time dimension is considered. Therefore, the mined rules also retain output time characteristics. When performing fault prediction, it is necessary not only to match the fault alert information included in the rules but also to ensure that the information matches the rules in the time dimension. This improves prediction accuracy and reduces false alarm rate. Finally, the combined rules are lightweight and configurable, making them easy to deploy in cloud systems.

[0070] In some optional embodiments, step 102: searching for at least one positive rule in the positive samples whose frequency is greater than or equal to a frequency threshold, and searching for at least one negative rule in the negative samples whose frequency is greater than or equal to a frequency threshold, according to the output time sequence, may include steps A1 and A2, as described below:

[0071] Step A1: Encode the fault alert information based on the number of times the fault alert information appears in the target sample, and obtain the encoding result.

[0072] In this embodiment of the application, each fault reminder information in the target sample can be frequency discretized and encoded. That is, based on the number of times the same fault reminder information appears in the sample, it is converted into different discrete terms to obtain the encoding result.

[0073] The encoding result of each fault reminder message includes: a first identifier for representing the fault reminder message and a second identifier for representing the Nth time, where N is an integer greater than or equal to 1.

[0074] For example, target sample A is [DTC105, DTC231, DTC105, ..., Classical Chinese prompt 1], and a target failure occurs. In this sample, DTC105 appears twice, so the first appearance of DTC105 can be encoded as DTC105_1, and the second appearance of DTC105 can be encoded as DTC105_2; DTC231 appears once, so it can be encoded as DTC231_1; Classical Chinese prompt 1 appears once, so it can be encoded as MSG_1. The final encoding result is [DTC105_1, DTC231_1, DTC105_2, ..., MSG_1], and a target failure occurs.

[0075] By performing frequency discretization encoding on fault alert information, the ability to detect repeated error reporting can be enhanced during rule mining.

[0076] Optionally, for fault alert messages that appear frequently, the second identifier can be "K+", where K is a value greater than or equal to 3, indicating that a fault alert message is consistently high-frequency. The specific identifier can be set according to actual needs. For example, if K=3, the second identifier can be "3+" for fault alert messages that appear more than 3 times. For example, if DTC105 appears 5 times in its sample, then the 4th and 5th DTC105 occurrences can both be encoded as DTC105_3+.

[0077] For the same fault alert, the first and second reports can often distinguish between "occasional" and "repeated" states. More frequent occurrences contribute relatively little to the actual judgment result, and more triggers do not change the processing logic. If each fault alert is marked with its actual occurrence number, it easily generates a large number of sparse, low-support sequence terms during rule mining, increasing the computational burden and diluting statistical significance. Using "K+" to represent persistence can significantly reduce the number of candidate rules and improve mining efficiency.

[0078] Step A2: Based on the encoding results of each target sample, in the order of output time, find at least one positive rule in the positive samples whose frequency is greater than or equal to the frequency threshold, and find at least one negative rule in the negative samples whose frequency is greater than or equal to the frequency threshold.

[0079] In existing technologies, unordered frequent itemset methods (such as Apriori and FP-Growth) are generally used for rule mining. However, these methods do not consider the temporal order of fault alerts or the frequency differences of the same fault alert. They assume that each fault alert appears only once in each sample and ignore the evolutionary characteristics of repeated reporting. Structurally, this method is difficult to capture multiple warning behaviors that may exist before a fault occurs.

[0080] In this embodiment, by performing frequency discretization encoding on fault alert information, identical fault alert information appearing sequentially in the same sample can be treated as different fault alert information. For example, for the encoding result [DTC105_1, DTC231_1, DTC105_2, ..., MSG_1] in the above example, when calculating the frequency of occurrence of fault alert information in all positive samples, the frequency of occurrence of DTC105_1 and DTC105_2 in all positive samples are calculated separately. Combined with mining according to the output time order, the repeated reporting behavior and triggering sequence of fault alert information can be simultaneously characterized at the pattern level. This is beneficial for identifying more causal and evolutionary fault triggering paths, thereby uncovering rules that are more conducive to improving prediction accuracy.

[0081] In some alternative embodiments, temporally ordered pattern mining can be performed based on frequent sequence pattern mining algorithms. Such algorithms can preserve the order in which items (i.e., fault alert information) appear, which is beneficial for fully preserving the temporal context and extracting the sequential triggering path before the fault occurs. Details are as follows.

[0082] Step 102: According to the output time sequence, find positive rules in the positive samples whose frequency is greater than or equal to the frequency threshold, and find negative rules in the negative samples whose frequency is greater than or equal to the frequency threshold, including steps B1 and B2, as described below:

[0083] Step B1: According to the output time sequence, determine the frequency of each fault reminder message in each sample in the corresponding type of sample, and determine the frequency of any 2 to M fault reminder messages that appear in each sample in the corresponding type of sample.

[0084] Where M is the maximum value of fault alert information included in the target sample. For example, if one or some target samples include the most fault alert information among all target samples, then that number is M.

[0085] Here, the corresponding sample type refers to either positive or negative samples. Specifically, it involves determining the frequency of each fault alert message in all positive samples, and determining the frequency of any 2 to M consecutive fault alert messages appearing in all positive samples. When this step is performed based on the encoding result of step A1, the same fault alert message appearing consecutively in the same sample needs to be treated as different fault alert messages, meaning its frequency needs to be counted separately.

[0086] Among them, any 2 to M fault alert messages appearing sequentially can be adjacent or non-adjacent. For example, taking the sample [DTC105_1,DTC231_1,DTC105_2] as an example, it is necessary to determine the occurrence frequency of DTC105_1, DTC231_1, and DTC105_2 in all positive samples; determine the occurrence frequency of DTC105_1 and DTC231_1, DTC105_1 and DTC105_2 in all positive samples, DTC231_1 and DTC105_2 in all positive samples; and determine the occurrence frequency of DTC105_1, DTC231_1, and DTC105_2 in all positive samples.

[0087] Step B2: Based on the previously determined frequency of occurrence and frequency threshold, at least one positive rule and at least one negative rule are obtained.

[0088] After obtaining the frequency of fault alerts for various situations through step B1, positive and negative rules with high frequency can be filtered out based on the preset frequency threshold.

[0089] Optionally, the aforementioned frequent sequence pattern mining algorithm can specifically be the PrefixSpan (Prefix-projected Sequential pattern mining) algorithm.

[0090] The PrefixSpan algorithm will be used as an example to illustrate the mining process of positive and negative rules.

[0091] Assume the encoded and labeled target samples are S1 to S5, as described below:

[0092] S1: [DTC105_1,DTC231_1,DTC105_2], Fault A has occurred;

[0093] S2: [DTC105_1,DTC231_1,DTC231_2,DTC105_2], Fault A has occurred;

[0094] S3: [DTC105_1,DTC105_2,DTC231_1], Fault A has occurred;

[0095] S4: [DTC105_1,DTC301_1,DTC105_2], No fault occurred;

[0096] S5: [DTC301_1,DTC231_1,DTC231_2], no fault occurred.

[0097] Then we perform frequent pattern mining, assuming a frequency threshold of 2.

[0098] First, perform positive mining on S1, S2, and S3 to obtain positive rules.

[0099] First level: Single item

[0100] [DTC105_1] exists in S1, S2, and S3, and appears 3 times. It is retained.

[0101] [DTC231_1] exists in S1, S2, and S3, and its frequency is 3. It is retained.

[0102] [DTC105_2] exists in S, S2, and S3, and its frequency is 3. It is retained.

[0103] [DTC231_2] exists in S2 and appears 1 time, so it is discarded.

[0104] The second layer consists of two ordered items, which expand from the first layer to ensure the orderliness of events.

[0105] [DTC105_1→DTC231_1] appears in S1, S2, and S3, with a frequency of 3, so it is retained.

[0106] [DTC105_1→DTC105_2] appears in S1, S2, and S3, with a frequency of 3, so it is retained.

[0107] [DTC231_1→DTC105_2] appears in S1 and S2, with a frequency of 2, so it is retained.

[0108] [DTC231_1→DTC231_2] appears in S2 with a frequency of 1, so it is discarded.

[0109] [DTC231_2,DTC105_2] appears in S2 with a frequency of 1, so it is discarded.

[0110] The third level: ordered triplets, which are an extension of the second level.

[0111] [DTC105_1→DTC231_1→DTC105_2] appears in S1 and S2, with a frequency of 2, so it is retained.

[0112] [DTC105_1→DTC231_1→DTC231_2] appears in S2 with a frequency of 1, so it is discarded.

[0113] Fourth level: None, end.

[0114] This yields seven positive rules, as shown below:

[0115] [DTC105_1], Fault A has occurred;

[0116] [DTC231_1], Fault A has occurred;

[0117] [DTC105_2], Fault A has occurred;

[0118] [DTC105_1→DTC231_1], Fault A occurred;

[0119] [DTC105_1→DTC105_2], Fault A occurred;

[0120] [DTC231_1→DTC105_2], Fault A occurred;

[0121] [DTC105_1→DTC231_1→DTC105_2], Fault A occurred.

[0122] Then, negative mining is performed on S4 and S5:

[0123] First level: Single item

[0124] [DTC301_1] exists in S4 and S5, with a support level of 2, so it is retained.

[0125] Second layer: None, terminated.

[0126] This results in a negative rule: [DTC301_1], indicating no fault occurred.

[0127] In some optional embodiments, all positive rules and all negative rules obtained in step 102 can be directly combined into a final combined rule for fault prediction. For example, if all positive rules obtained in step 102 include positive rule 1 and positive rule 2, and all negative rules obtained in step 102 include negative rule 1 and negative rule 2, then the final combined rule can be: (positive rule 1 or positive rule 2) and not (negative rule 1 or negative rule 2). The combined rule obtained in this way has high coverage and can reduce false negatives when applied to fault prediction.

[0128] In some alternative embodiments, the positive and negative rules obtained in step 102 can also be filtered. For example, positive rules can be filtered based on confidence level and / or false alarm rate, such as filtering out positive rules with confidence level greater than or equal to a confidence threshold and / or positive rules with false alarm rate less than or equal to a first false alarm rate threshold; negative rules can be filtered based on support level and / or false alarm rate, such as filtering out negative rules with support level greater than or equal to a support threshold and / or negative rules with false alarm rate less than or equal to a second false alarm rate threshold. Then, based on the filtered positive and negative rules, the final combined rules for fault prediction are obtained. Since the positive rules in the combined rules have high confidence level and / or low false alarm rate, and the negative rules in the combined rules have high support level and / or low false alarm rate, the accuracy of fault prediction can be improved and the false alarm rate can be reduced when using such combined rules for fault prediction.

[0129] Optionally, in this embodiment of the application, a first positive rule can be selected from all positive rules based on the confidence level and false positive rate (FPR) of each positive rule. This first positive rule has a confidence level greater than or equal to a confidence threshold and a false positive rate less than or equal to a first false positive rate threshold. Here, confidence level refers to the ratio of the number of times a positive rule appears in positive samples to the number of times the positive rule appears in all samples; false positive rate refers to the ratio of the number of times the positive rule appears in all negative samples to the total number of negative samples.

[0130] In this embodiment of the application, a first negative rule can be selected from all negative rules based on the support and false alarm rate of each negative rule, provided that its support is greater than or equal to a support threshold and its false alarm rate is less than or equal to a second false alarm rate threshold. Here, support refers to the frequency of occurrence of a negative rule in all negative samples, and the false alarm rate is the ratio of the number of occurrences of a negative rule in all positive samples to the total number of positive samples; the second false alarm rate threshold is less than the first false alarm rate threshold.

[0131] Combining positive rules with high confidence and low false alarm rate with negative rules with high support and low false alarm rate can improve the accuracy of fault prediction and reduce the false alarm rate when used for fault prediction.

[0132] The confidence level of a certain positive rule can be expressed by formula (1), as follows:

[0133]

[0134] r in formula (1) + Let P represent a positive rule, and let P represent the rule r. + The sequence of fault alert messages.

[0135] The false alarm rate of a certain positive rule can be expressed by formula (2), as follows:

[0136]

[0137] r in formula (2) + Let P represent a positive rule, and let P represent the rule r. + The sequence of fault alert messages.

[0138] The false alarm rate of a certain negative rule can be expressed by formula (3), as follows:

[0139]

[0140] r in formula (3) - Let P represent a negative rule, and let P represent the rule r. - The sequence of fault alert messages.

[0141] The support of a certain negative rule can be represented by formula (4), as follows:

[0142]

[0143] r in formula (4) - Let P represent a negative rule, and let P represent the rule r. - The sequence of fault alert messages.

[0144] The support of a certain positive rule can be represented by formula (5), as follows:

[0145]

[0146] r in formula (5) + Let P represent a positive rule, and let P represent the rule r. + The sequence of fault alert messages.

[0147] In this embodiment of the application, positive rules with high confidence (e.g., confidence ≥ set threshold (e.g., 0.6)) and low false alarm rate (e.g., false alarm rate ≤ set threshold (e.g., 0.2)) can be selected using the above method, and negative rules with high support (e.g., support ≥ set threshold (e.g., 0.1) and strong exclusivity (false alarm rate ≤ lower value (e.g., 0.05)) in negative samples can be selected, so as to obtain a combination rule with high accuracy in predicting faults and low false alarm rate.

[0148] In this embodiment, an ordered association rule mining algorithm (such as PrefixSpan) is used to perform ordered pattern mining on the historical sequence of fault alert information. This method can completely preserve the temporal context and extract the sequential triggering path before the fault occurs. This mechanism can automatically extract high-quality, highly discriminative temporal pattern rules from a large amount of historical data, effectively capturing the fault evolution trend.

[0149] Furthermore, by discretizing the number of fault alert reports and employing an ordered association rule mining algorithm for time-ordered pattern mining, more dynamic sequence patterns can be identified. This allows for the early identification of multiple warning signals prior to a fault, improving prediction accuracy, and is particularly suitable for rare but high-risk fault types. Compared to existing technologies that only process Boolean DTC items, the embodiments of this application significantly enhance the temporal context awareness capability during fault prediction.

[0150] In some optional embodiments, for all positive rules and all negative rules obtained through step 102, a better combination rule can be obtained by combining the target optimization parameters, such as... Figure 2 As shown, the cloud can not only perform association rule mining, but also rule combination optimization, as detailed below.

[0151] Step 103: Based on the positive and negative rules, obtain the combined rules for fault prediction, which may include steps C1 to C3, as described below:

[0152] Step C1: Randomly combine the positive and negative rules to obtain at least one random combination rule.

[0153] Assume that the positive rules include rule1 and rule2, and the negative rules include rule3 and rule4. Random combination of positive and negative rules means that at least one positive rule can be combined with at least one negative rule, such as rule1 combined with rule3 (i.e.,...). ), rule1 and rule4 combined (i.e. ), rule2 and rule3 combined (i.e. ), rule2 and rule4 combined (i.e. ), the combination of rule1, rule2 and rule3 (i.e. (and all possible combinations)

[0154] Step C2: Determine the value of the target optimization parameter corresponding to each random combination rule.

[0155] The target optimization parameters mentioned here may include at least one of the following: minimizing the false alarm rate, maximizing the fault coverage, and minimizing the combinatorial complexity.

[0156] Minimizing the false positive rate refers to the ratio of the number of negative samples that hit the random combination rule to the total number of negative samples, which can be expressed by formula (6), as shown below:

[0157]

[0158] The maximum fault coverage rate refers to the ratio of the number of positive samples that hit the random combination rule to the total number of positive samples, which can be expressed by formula (7) as follows:

[0159]

[0160] Minimizing combinatorial complexity refers to the number of positive and negative rules included in the random combination rules. After-sales engineers or reviewers need to quickly understand and confirm the triggering conditions; too many rules will create lengthy and difficult-to-trace logical chains, reducing readability. Furthermore, too many rules often mean fragmented matching of historical samples, making it easy to get caught in random noise. Therefore, the fewer rules included in the combinatorial rules, the better. The simplified core rules have better generalization capabilities and perform more stably when facing new car models or new operating conditions.

[0161] Step C3: Determine the combination rules for fault prediction based on the values ​​of the target optimization parameters corresponding to each random combination rule.

[0162] In existing technologies, the combination logic between rules (such as OR, AND, NOT) often relies on manual experience for configuration, lacking better filtering methods. In this embodiment, random combination rules can be generated and evaluated in conjunction with target optimization parameters. Then, based on actual business needs, a random combination rule with a low false positive rate, a simpler random combination rule, or a random combination rule with high coverage can be selected as the final combination rule. Compared to manual filtering, this allows the combination rules to better meet business requirements. The final combination rule can be at least one.

[0163] Optionally, the target optimization parameters may include at least two of the following: minimizing the false alarm rate, maximizing the fault coverage, and minimizing the combinatorial complexity, thereby achieving multi-objective optimization of the combinatorial rules.

[0164] Optionally, embodiments of this application may employ multi-objective optimization algorithms (such as NSGA-II multi-objective genetic algorithm) to optimize combined rules, perform global search and performance evaluation on the rule combination, so that the final combined rules can significantly reduce the false alarm rate while ensuring coverage, overcoming the problem that rules cannot be optimized collaboratively in the prior art.

[0165] The following example uses the NSGA-II algorithm to illustrate the multi-objective optimization process.

[0166] Assume that the target optimization parameters may include: minimizing the false alarm rate, maximizing the fault coverage, and minimizing the combinatorial complexity. Also assume that the positive rules include rule1, rule2, and rule3, and the negative rules include rule4, rule5, and rule6.

[0167] Before optimization, we can first design the individual encoding. Each individual represents a combination rule structure, and each individual is encoded as a binary string of length m+n, where m represents the number of positive rules and n represents the number of negative rules; each bit indicates whether the corresponding rule is enabled, that is, whether the corresponding rule participates in the combination. The binary string is as follows:

[0168] x = [x1, x2, ..., x m ,x m+1 ,...,x m+n ]

[0169] Where, x i Indicates whether the i-th positive rule is enabled (0 indicates disabled, 1 indicates enabled), x m+j This indicates whether the j-th negative rule is enabled.

[0170] Let's take positive rules (rule1, rule2, rule3) and negative rules (rule4, rule5, rule6) as examples to continue the explanation.

[0171] These rules can be combined into a second-order Boolean rule structure as follows:

[0172]

[0173] Whether each rule participates in the combination is determined by a 6-bit binary string, which is the individual code. For example:

[0174] Individual 110011 indicates that rule1, rule2, rule5, and rule6 are enabled, while rule3 and rule4 are disabled. The corresponding combination structure is as follows:

[0175] like Figure 3 As shown, the multi-objective optimization process is described in detail below:

[0176] Step 301: Initial population generation, randomly generating multiple individuals with rule combinations.

[0177] In the example above, multiple 6-bit binary strings are randomly generated to form the initial population.

[0178] Step 302: Evaluate each individual and calculate the three indicators corresponding to each combination structure.

[0179] Each individual represents a combination of candidate rules, and its performance is evaluated by the following three optimization objectives: minimizing the false positive rate, maximizing the coverage, and minimizing the combination complexity.

[0180] In this algorithm, minimizing the combinatorial complexity can be expressed by formula (8), as shown below:

[0181]

[0182] In formula (8), m represents the number of positive rules, n represents the number of negative rules, xi is 0 or 1, and x i A value of 0 indicates that the i-th rule has not been activated. i A value of 1 indicates that the i-th rule has been activated.

[0183] Step 303: Perform non-dominated sorting, stratify the population, and find the Pareto optimal front.

[0184] This also means stratifying the population according to Pareto optimality of the three objectives. Pareto optimality means that it is impossible to improve the third objective without worsening the other two.

[0185] Step 304: Calculate the crowding level and retain the elites in each layer.

[0186] This is to ensure the diversity of solution sets.

[0187] Step 305: Perform genetic operations, using selection, crossover, and mutation to generate the next generation of individuals, forming a new population.

[0188] Step 306: Evaluate each individual in the new population based on the aforementioned three indicators.

[0189] Step 307: Repeat the iteration cycle until the maximum number of iterations or the convergence criterion is met.

[0190] The multi-objective optimization result is a solution set containing multiple candidate schemes. Each scheme represents a combination rule structure, and its corresponding false alarm rate, coverage and complexity reach the comprehensive optimality.

[0191] Ultimately, deployment can be selected from the optimal solution based on actual business preferences (such as a preference for low false positives, high coverage, or simplicity). That is, filtering rules are set according to actual business preferences (prioritizing the lowest false positive rate, or prioritizing the simplest, etc.), thereby selecting the final combined rule from the optimal solution for output.

[0192] For example, the final output is a set of Pareto optimal individuals, such as: the optimal individual is 101010, and the corresponding combination is...

[0193] Currently, most systems still rely on static rule bases or simple single-rule matching logic, lacking a structured optimization mechanism for overall prediction performance, resulting in low prediction accuracy and high false positive rates. To overcome the problems of single rules and inability to be structured and optimized in existing systems, this application constructs a positive prediction rule base and a negative exclusion rule base, mining judgment patterns from both faulty and non-faulty samples, and designs a Boolean logic expression-type rule combination structure, combining rules in interpretable forms such as "(positive rule 1 or positive rule 2) and non-negative rule 3" to improve the expressive flexibility of the rule system. Furthermore, a multi-objective evolutionary optimization algorithm is introduced, using at least two of the false positive rate, coverage, and complexity as objective functions to search for the optimal solution set in the rule combination space, achieving comprehensive performance optimization of rule selection, such as enhanced predictive ability and logical simplification optimization in the combination structure.

[0194] It is understandable that after selecting the first positive rule based on confidence and / or false alarm rate, and selecting the first negative rule based on support and / or false alarm rate, the combined rules for fault prediction can be obtained by combining the target optimization parameters, thereby reducing the amount of data processing for multi-objective optimization and improving the efficiency of multi-objective optimization.

[0195] In some embodiments, the present application can realize the mining of combination rules through a model, that is, inputting the target sample into a pre-trained model, having the model analyze the target sample, and outputting the final combination rules.

[0196] In addition to the above rule generation method, this application embodiment also provides a fault prediction method applied to a cloud server.

[0197] The method is described in detail below through some embodiments. The following embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0198] like Figure 4 As shown, the method may include steps 401 to 403, as described below:

[0199] Step 401: Obtain the sequence of fault alert information output by the target device.

[0200] The fault reminder information sequence includes at least one fault reminder information, which can be a diagnostic fault code, a text prompt, or an audible alarm, etc.

[0201] The target device can send fault alert information to the cloud immediately after generating it. The cloud can trigger a prediction operation upon receiving one or more fault alert messages, or it can trigger the prediction operation only when the received fault alert message sequence meets the target conditions. The target conditions mentioned here can be set according to actual needs. For example, the target conditions may include: the fault alert message sequence includes fault alerts related to major faults; or the frequency of occurrence of a certain fault alert message in the fault alert message sequence is greater than a threshold.

[0202] Step 402: Determine whether the fault alert message sequence matches at least one of the multiple combination rules.

[0203] The combined rules described here are obtained through the aforementioned rule generation method and are used for fault prediction. Each combined rule includes at least one positive rule and at least one negative rule, and both the positive and negative rules include at least one fault alert message ordered by output time.

[0204] Step 403: If the fault alert information sequence matches at least one of multiple combined rules, send fault prediction information to the target object.

[0205] If the fault alert information sequence output by the target device matches one or more positive rules in the combined rules, but does not match any negative rules in the combined rules, it indicates a high probability that the target fault corresponding to that combined rule has occurred or will occur. Otherwise, the probability of the target fault occurring is considered low.

[0206] If the fault alert message sequence matches at least one of the multiple combined rules, then fault prediction information (i.e., early warning information) can be sent to the target object. If the fault alert message sequence does not match any of the multiple combined rules, then no action is taken.

[0207] The fault prediction information may include potential faults of the target device. The target object may include at least one of the following: the target device, the after-sales service terminal, or the user terminal associated with the target device (for example, when the target device is a vehicle, the user terminal may be the vehicle owner's terminal associated with the vehicle).

[0208] Sending fault prediction information to the target device allows users of the target device to understand the device's status and take timely measures.

[0209] Sending fault prediction information to the user terminal associated with the target device allows the device user to understand the device status and take timely measures.

[0210] Sending fault prediction information to the after-sales service center can trigger service reminders or work order processes, enabling the after-sales service center to promptly provide inspection and repair services for the target equipment. Figure 2 As shown, the cloud sends the prediction results to the service ticket platform, which automatically creates service tickets to dispatch after-sales resources and enter the after-sales processing stage, realizing a closed loop of linkage from early warning to intervention.

[0211] The combined rules used for fault prediction in this embodiment include not only positive rules but also negative rules. When performing fault prediction, while hitting positive rules, it also excludes cases where negative rules are hit, thus reducing the false alarm rate. Furthermore, the combined rules retain the output time characteristics of fault alert information. When performing fault prediction, it is necessary not only to hit the fault alert information included in the rules but also to match the rules in the time dimension. This improves prediction accuracy and reduces the false alarm rate. Other advantages of the combined rules for fault prediction in this embodiment can be found in the foregoing embodiments and will not be repeated here.

[0212] Furthermore, to support proactive remote diagnostics, this application embodiment can also construct a rule-based inference engine and an after-sales service triggering interface, realizing an automated closed loop from data monitoring to after-sales service response and improving the intelligence level of the after-sales system. Specifically, after deploying the mined combination rules as configuration items to the cloud system, the cloud can receive fault reminder information sequences uploaded by vehicles in real time and determine online whether the fault reminder information sequence matches a combination rule for a certain type of fault. Once matched, it can automatically trigger warning signals and after-sales service processes (such as notifications, appointments, work orders, etc.). This realizes a closed-loop system of predictive judgment and cloud-based proactive service triggering, enhancing the remote intelligent service capabilities of equipment and providing feasible and scalable support for building an intelligent after-sales platform.

[0213] It should be noted that when the combination rules are obtained by mining the sequence of fault reminder information through frequency discretization encoding, the fault reminder information also needs to be frequency discretized during fault prediction.

[0214] In some optional embodiments, the fault prediction information in step 103 may also include: a fault reminder information sequence, the combination rules that the fault reminder information sequence hits, and the corresponding trigger path, etc.

[0215] In existing technologies, some fault prediction solutions employ artificial intelligence models such as machine learning and even deep learning. The output of these models is a risk score, or probability value, while after-sales service processes require explicit logic: "a work order is triggered when a certain condition is met." If the output fault probability is 73%, the system must pre-set a threshold, such as 70%, before dispatching a work order. If the threshold is set too low, there will be too many false alarms, consuming after-sales resources; if it's set too high, serious faults will be missed, creating safety hazards. Furthermore, the threshold itself lacks interpretable evidence; engineers cannot know which fault alerts or sensor features drive the score, nor can they determine whether the difference between "70% vs. 68%" is statistically significant. As equipment types or software versions change, model distribution drifts, requiring repeated manual parameter adjustments to the threshold, which is an uninterpretable parameter and difficult to maintain.

[0216] This application addresses the issue of uninterpretable existing methods by reasoning each prediction result based on structured logical combination rules. Upon each rule hit, the triggered rule details and matching input sequence are output, forming a clear and visualized reasoning chain that can serve as a basis for manual review, service intervention, and rule revision. This mechanism supports the traceability of prediction results, solving the problems of black-box models being difficult to interpret and reuse. It facilitates future rule version management and knowledge system iteration, and can be applied to safety-sensitive vehicle control scenarios.

[0217] like Figure 2 As shown, the cloud can not only perform rule matching, but also rule interpretation. The rule interpretation mentioned here refers to providing information such as the details of the triggered rule, the matched input sequence, and the corresponding trigger path.

[0218] In some alternative embodiments, when the target object includes an after-sales service provider, after step 103: sending fault prediction information to the target object, the method may include steps D1 and D2, as described below:

[0219] Step D1: Receive service result information sent by the after-sales service terminal.

[0220] The service result information includes confirmation information indicating whether the target device has malfunctioned.

[0221] Step D2: Based on the sequence of confirmation information and fault reminder information, generate new sample data and save it.

[0222] The new sample data includes fault alert information sequences and labels indicating whether a fault has occurred. The new sample data is used to update the combination rules.

[0223] After completing after-sales service for the target device, the after-sales service provider can upload confirmation information indicating whether a fault has indeed occurred. Based on this confirmation information, a tag can be added to the fault alert information sequence described in step 401. This tag indicates whether the fault alert information sequence corresponds to a fault. The tagged fault alert information sequence can then be saved as new sample data to update the sample library for subsequent updates and optimizations of the combined rules. For example, if the cloud detects an increase in the false alarm rate or false negative rate of certain combined rules, it can automatically trigger the rule update process, enabling the self-evolution of the rule system. When optimizing combined rules, optimization can be performed on the combined rules corresponding to a specific fault or on the entire combined rule library, depending on actual needs.

[0224] In some optional embodiments, the combined rules obtained through the rule generation method can also be deployed to the device, where the device performs fault prediction on the fault alert information sequence it generates. The rule base is maintained by the cloud, and the latest rules are distributed to the device after each rule base update. Alternatively, the combined rules can be deployed to edge devices for online inference at edge nodes. Here, edge devices refer to devices between the device and the cloud. Specific configurations can be tailored to actual needs.

[0225] Exemplary device

[0226] Accordingly, this application also provides a rule generation device for use in the cloud.

[0227] like Figure 5 As shown, the device may include:

[0228] The first acquisition module 501 is used to acquire the target sample.

[0229] Each target sample includes at least one fault alert message ordered in chronological order of output time; the target samples include positive samples and negative samples, where positive samples are those that have experienced the target fault and negative samples are those that have not experienced the fault.

[0230] The mining module 502 is used to find at least one positive rule in the positive samples whose frequency is greater than or equal to a frequency threshold, according to the output time sequence, and to find at least one negative rule in the negative samples whose frequency is greater than or equal to a frequency threshold.

[0231] Both the positive and negative rules consist of at least one fault reminder message ordered in chronological order of output time.

[0232] The first processing module 503 is used to obtain a combined rule for fault prediction based on the positive rule and the negative rule.

[0233] Specifically, if the fault alert information sequence output by the target device matches a positive rule in the combined rules but does not match any negative rule in the combined rules, it indicates that the target fault may have occurred.

[0234] In some alternative embodiments, the mining module 502 may be specifically used for:

[0235] Based on the number of times the fault reminder information in each target sample appears in that target sample, the fault reminder information is encoded to obtain the encoding result; wherein, the encoding result of each fault reminder information includes: a first identifier for representing the fault reminder information and a second identifier for representing the Nth occurrence, where N is an integer greater than or equal to 1;

[0236] Based on the encoding results of each target sample, in the order of output time, at least one positive rule is found in the positive samples whose frequency is greater than or equal to a frequency threshold, and at least one negative rule is found in the negative samples whose frequency is greater than or equal to a frequency threshold.

[0237] In some alternative embodiments, the mining module 502 may be specifically used for:

[0238] According to the output time sequence, determine the frequency of each fault reminder information in the corresponding type of sample in each sample, and determine the frequency of any 2 to M fault reminder information appearing in the corresponding type of sample in each sample; where M is the maximum value of fault reminder information included in the target sample, and the corresponding type of sample is a positive sample or a negative sample.

[0239] Based on the occurrence frequency and the frequency threshold, the at least one positive rule and the at least one negative rule are obtained.

[0240] In some alternative embodiments, the apparatus may further include:

[0241] The first filtering module is used to filter out a first positive rule from all the positive rules based on the confidence level and false alarm rate of each positive rule, and the first positive rule has a confidence level greater than or equal to a confidence level threshold and a false alarm rate less than or equal to a first false alarm rate threshold.

[0242] The second filtering module is used to filter out a first negative rule from all the negative rules based on the support and false alarm rate of each negative rule, where the support is greater than or equal to the support threshold and the false alarm rate is less than or equal to the second false alarm rate threshold.

[0243] The first processing module 503 can be specifically used for:

[0244] Based on the selected first positive rule and the first negative rule, a combined rule for fault prediction is obtained.

[0245] In some alternative embodiments, the first processing module 503 may be specifically used for:

[0246] The positive rule and the negative rule are randomly combined to obtain at least one random combination rule;

[0247] The numerical values ​​of the target optimization parameters corresponding to each random combination rule are determined; wherein, the target optimization parameters include at least one of the following: minimizing the false positive rate, maximizing the fault coverage rate, and minimizing the combinatorial complexity; the minimizing the false positive rate refers to the ratio of the number of negative samples that hit the random combination rule to the total number of negative samples; the maximizing the fault coverage rate refers to the ratio of the number of positive samples that hit the random combination rule to the total number of positive samples; the minimizing the combinatorial complexity refers to the number of positive rules and negative rules included in the random combination rule;

[0248] Based on the value of the target optimization parameter corresponding to each random combination rule, the combination rule used for fault prediction is determined.

[0249] The rule generation apparatus provided in this embodiment belongs to the same concept as the rule generation method provided in the above embodiments of this application. It can execute the rule generation method provided in any of the above embodiments of this application and has the corresponding functional modules and beneficial effects of the execution method. Technical details not described in detail in this embodiment can be found in the specific processing content of the rule generation method provided in the above embodiments of this application, and will not be repeated here.

[0250] Accordingly, this application also provides a fault prediction device applied in the cloud.

[0251] like Figure 6 As shown, the device may include:

[0252] The second acquisition module 601 is used to acquire the fault reminder information sequence output by the target device.

[0253] The fault alert information sequence includes at least one fault alert information.

[0254] The matching module 602 is used to determine whether the fault reminder information sequence matches at least one of multiple combination rules.

[0255] The multiple combined rules are used for fault prediction. Each combined rule includes at least one positive rule and at least one negative rule. Both the positive rule and the negative rule include at least one fault reminder message ordered by output time. If the fault reminder message sequence output by the target device matches the positive rule in the combined rules but does not match any negative rule in the combined rules, it indicates that a fault corresponding to the combined rule may occur.

[0256] The second processing module 603 is used to send fault prediction information to the target object when it is determined that the fault reminder information sequence hits at least one of multiple combination rules.

[0257] The fault prediction information includes possible faults of the target device; the target object includes at least one of the following: the target device, the after-sales service terminal, and the user terminal associated with the target device.

[0258] In some optional embodiments, the fault prediction information further includes: the fault reminder information sequence and the combination rules for the fault reminder information sequence.

[0259] In some alternative embodiments, the method further includes:

[0260] The receiving module is used to receive service result information sent by the after-sales service terminal.

[0261] The service result information includes confirmation information indicating whether the target device has malfunctioned.

[0262] The sample generation module is used to generate new sample data and save it based on the sequence of the confirmation information and the fault reminder information.

[0263] The new sample data includes the fault alert information sequence and a label indicating whether a fault has occurred. The new sample data is used to update the combination rules.

[0264] The fault prediction device provided in this embodiment belongs to the same concept as the fault prediction method provided in the above embodiments of this application. It can execute the fault prediction method provided in any of the above embodiments of this application and has the corresponding functional modules and beneficial effects of the execution method. Technical details not described in detail in this embodiment can be found in the specific processing content of the fault prediction method provided in the above embodiments of this application, and will not be repeated here.

[0265] Exemplary electronic devices

[0266] The present application also provides an electronic device, such as Figure 7As shown, the electronic device includes a memory 700 and a processor 710.

[0267] The memory 700 is connected to the processor 710 and is used to store programs.

[0268] The processor 710 is used to implement the rule generation method in the above embodiments by running the program stored in the memory 700.

[0269] Specifically, the aforementioned electronic device may also include: a communication interface 720, an input device 730, an output device 740, and a bus 750.

[0270] The processor 710, memory 700, communication interface 720, input device 730, and output device 740 are interconnected via a bus. Among them:

[0271] Bus 750 may include a pathway for transmitting information between various components of a computer system.

[0272] The processor 710 can be a general-purpose processor, such as a general-purpose central processing unit (CPU), a microprocessor, etc., or an application-specific integrated circuit (ASIC), or one or more integrated circuits used to control the execution of the program of the present invention. It can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), an off-the-shelf programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0273] The processor 710 may include a main processor, as well as a baseband chip, modem, etc.

[0274] The memory 700 stores a program that executes the technical solution of this invention, and may also store an operating system and other key business functions. Specifically, the program may include program code, which includes computer operation instructions. More specifically, the memory 700 may include read-only memory (ROM), other types of static storage devices capable of storing static information and instructions, random access memory (RAM), other types of dynamic storage devices capable of storing information and instructions, disk storage, flash memory, etc.

[0275] Input device 730 may include a device for receiving user input data and information, such as a keyboard, mouse, camera, scanner, light pen, voice input device, touch screen, pedometer, or gravity sensor.

[0276] Output device 740 may include devices that allow information to be output to a user, such as a display screen, printer, speaker, etc.

[0277] The communication interface 720 may include a device that uses any transceiver to communicate with other devices or communication networks, such as Ethernet, Radio Access Network (RAN), Wireless Local Area Network (WLAN), etc.

[0278] The processor 710 executes the program stored in the memory 700 and calls other devices, which can be used to implement the various steps in the rule generation method or fault prediction method provided in the above embodiments of this application.

[0279] Exemplary computer program products and storage media

[0280] In addition to the methods and devices described above, embodiments of this application may also be computer program products, which include computer program instructions that, when executed by a processor, cause the processor to perform steps in the rule generation method or the fault prediction method described in the embodiments of this application.

[0281] The aforementioned computer program product can be implemented through hardware, software, or a combination thereof. In one optional embodiment, the computer program product is specifically embodied in a computer storage medium; in another optional embodiment, the computer program product is specifically embodied in a software product, such as a software development kit (SDK), etc.

[0282] The computer program product can be written in any combination of one or more programming languages ​​to perform the operations of the embodiments of this application. The programming languages ​​include object-oriented programming languages ​​such as Java and C++, as well as conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the user's computing device, partially on the user's computing device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0283] Furthermore, embodiments of this application may also be storage media storing a computer program, which is executed by a processor from the steps in the rule generation method or the fault prediction method described in the embodiments of this application.

[0284] In addition, embodiments of this application may also be chips, which include processors and data interfaces. The processor reads instructions stored in the memory through the data interface to execute the steps in the rule generation method or the fault prediction method described in the embodiments of this application.

[0285] For the foregoing method embodiments, in order to simplify the description, they are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, because according to this application, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0286] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For apparatus embodiments, since they are basically similar to method embodiments, the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0287] The steps in the methods of the various embodiments of this application can be adjusted, merged, or deleted in order according to actual needs, and the technical features described in each embodiment can be replaced or combined.

[0288] The modules and sub-modules in the devices and terminals in the various embodiments of this application can be merged, divided, and deleted according to actual needs.

[0289] It should be understood that the disclosed terminals, devices, and methods can be implemented in other ways, given the several embodiments provided in this application. For example, the terminal embodiments described above are merely illustrative. For instance, the division of modules or sub-modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple sub-modules or modules may be combined or integrated into another module, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or modules, and may be electrical, mechanical, or other forms.

[0290] The modules or submodules described as separate components may or may not be physically separate. The components that constitute a module or submodule may or may not be physical modules or submodules; that is, they may be located in one place or distributed across multiple network modules or submodules. Some or all of the modules or submodules can be selected to achieve the purpose of this embodiment's solution, depending on actual needs.

[0291] Furthermore, the functional modules or sub-modules in the various embodiments of this application can be integrated into one processing module, or each module or sub-module can exist physically separately, or two or more modules or sub-modules can be integrated into one module. The integrated modules or sub-modules described above can be implemented in hardware or in the form of software functional modules or sub-modules.

[0292] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0293] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software unit executed by a processor, or a combination of both. The software unit can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0294] Finally, 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, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

Claims

1. A rule generation method, characterized in that, The method includes: Obtain target samples; wherein each target sample includes at least one fault reminder message ordered in chronological order of output time; the target samples include positive samples and negative samples, wherein positive samples refer to samples in which the target fault has occurred, and negative samples refer to samples in which the fault has not occurred; In accordance with the output time sequence, at least one positive rule is found in the positive samples whose frequency is greater than or equal to a frequency threshold, and at least one negative rule is found in the negative samples whose frequency is greater than or equal to a frequency threshold; wherein, both the positive rule and the negative rule consist of at least one fault reminder message ordered in chronological order of output time. Based on the positive and negative rules, a combined rule for fault prediction is obtained; wherein, if the fault reminder information sequence output by the target device hits the positive rule in the combined rule but does not hit any negative rule in the combined rule, it indicates that the target fault may occur.

2. The rule generation method according to claim 1, characterized in that, The step of searching for at least one positive rule in the positive samples with a frequency greater than or equal to a frequency threshold, and searching for at least one negative rule in the negative samples with a frequency greater than or equal to a frequency threshold, according to the output time sequence, includes: Based on the number of times the fault reminder information in each target sample appears in that target sample, the fault reminder information is encoded to obtain the encoding result; wherein, the encoding result of each fault reminder information includes: a first identifier for representing the fault reminder information and a second identifier for representing the Nth occurrence, where N is an integer greater than or equal to 1; Based on the encoding results of each target sample, in the order of output time, at least one positive rule is found in the positive samples whose frequency is greater than or equal to a frequency threshold, and at least one negative rule is found in the negative samples whose frequency is greater than or equal to a frequency threshold.

3. The rule generation method according to claim 1 or 2, characterized in that, The step of searching for positive rules in the positive samples with a frequency greater than or equal to a frequency threshold, and searching for negative rules in the negative samples with a frequency greater than or equal to a frequency threshold, according to the output time sequence, includes: According to the output time sequence, determine the frequency of each fault reminder information in the corresponding type of sample in each sample, and determine the frequency of any 2 to M fault reminder information appearing in the corresponding type of sample in each sample; where M is the maximum value of fault reminder information included in the target sample, and the corresponding type of sample is a positive sample or a negative sample. Based on the occurrence frequency and the frequency threshold, the at least one positive rule and the at least one negative rule are obtained.

4. The rule generation method according to claim 1, characterized in that, After obtaining at least one of the positive rules and at least one of the negative rules, the method further includes: Based on the confidence level and false alarm rate of each positive rule, a first positive rule is selected from all the positive rules, with a confidence level greater than or equal to the confidence level threshold and a false alarm rate less than or equal to the first false alarm rate threshold. Based on the support and false alarm rate of each negative rule, a first negative rule is selected from all the negative rules, with a support greater than or equal to the support threshold and a false alarm rate less than or equal to the second false alarm rate threshold. The combined rules for fault prediction, derived based on the positive and negative rules, include: Based on the selected first positive rule and the first negative rule, a combined rule for fault prediction is obtained.

5. The rule generation method according to claim 1 or 4, characterized in that, The combined rules for fault prediction, derived based on the positive and negative rules, include: The positive rule and the negative rule are randomly combined to obtain at least one random combination rule; The numerical values ​​of the target optimization parameters corresponding to each random combination rule are determined; wherein, the target optimization parameters include at least one of the following: minimizing the false positive rate, maximizing the fault coverage rate, and minimizing the combinatorial complexity; the minimizing the false positive rate refers to the ratio of the number of negative samples that hit the random combination rule to the total number of negative samples; the maximizing the fault coverage rate refers to the ratio of the number of positive samples that hit the random combination rule to the total number of positive samples; the minimizing the combinatorial complexity refers to the number of positive rules and negative rules included in the random combination rule; Based on the value of the target optimization parameter corresponding to each random combination rule, the combination rule used for fault prediction is determined.

6. A fault prediction method, characterized in that, include: Obtain a sequence of fault alert information output by the target device; wherein the sequence of fault alert information includes at least one fault alert message; Determine whether the fault alert information sequence matches at least one of multiple combined rules; wherein, the multiple combined rules are used for fault prediction, each combined rule includes at least one positive rule and at least one negative rule, and both the positive rule and the negative rule include at least one fault alert information ordered by output time; if the fault alert information sequence output by the target device matches the positive rule in the combined rules and does not match any of the negative rules in the combined rules, it indicates that a fault corresponding to the combined rule may occur; If the fault alert information sequence is determined to match at least one of multiple combined rules, fault prediction information is sent to the target object; wherein, the fault prediction information includes possible faults of the target device; the target object includes at least one of the following: the target device, the after-sales service terminal, and the user terminal associated with the target device.

7. The fault prediction method according to claim 6, characterized in that, The fault prediction information also includes: the fault reminder information sequence and the combination rules that the fault reminder information sequence hits.

8. The fault prediction method according to claim 6 or 7, characterized in that, When the target object includes an after-sales service provider, after sending the fault prediction information to the target object, the method further includes: Receive service result information sent by the after-sales service terminal; wherein, the service result information includes confirmation information indicating whether the target device has malfunctioned; Based on the confirmation information and the fault alert information sequence, new sample data is generated and saved; wherein, the new sample data includes the fault alert information sequence and a label indicating whether a fault has occurred, and the new sample data is used to update the combination rules.

9. An electronic device, characterized in that, include: memory and processor; The memory is connected to the processor and is used to store programs; The processor is used to implement the rule generation method as described in claims 1 to 5 or the fault prediction method as described in any one of claims 6 to 8 by running the program in the memory.

10. A computer program product, characterized in that, The computer program product stores a computer program, which, when executed by a processor, implements the rule generation method as described in claims 1 to 5 or the fault generation method as described in any one of claims 6 to 8.

Images