Fault prediction methods and apparatuses, and device, storage medium and program product

Abstract

Provided in the present disclosure are fault prediction methods and apparatuses, and a device, a storage medium and a program product. A fault prediction method comprises: acquiring an initial sequence corresponding to a plurality of abnormal events of a target device, wherein the initial sequence comprises event information of each abnormal event; determining a plurality of target sub-sequences in the initial sequence, wherein the amount of event information comprised in each target sub-sequence is less than or equal to a first threshold value; and on the basis of the plurality of target sub-sequences, determining a fault prediction result for the target device. Thus, the accuracy of fault prediction for target devices is improved.

Description

[0001] This disclosure claims priority to Chinese patent application number 202410346716.0, filed with the Patent Office of the People's Republic of China on March 25, 2024, entitled "Fault Prediction Method, Apparatus, Device, Storage Medium, and Program Product," the entire contents of which are incorporated herein by reference. Technical Field: This disclosure relates to the field of computers, and more particularly to a fault prediction method, apparatus, device, storage medium, and program product. Background: During operation, a device may experience abnormal events, which may cause the device to malfunction. Therefore, based on abnormal events recorded by the device within a preset historical period, it is possible to predict whether the device is likely to malfunction in the future. In related art, an electronic device can obtain multiple abnormal events from a target device within multiple preset historical time periods, and determine statistical information for each abnormal event within the multiple preset historical time periods (i.e., the total number of occurrences, the maximum number of occurrences, and the minimum number of occurrences of each abnormal event within the multiple preset time periods). Based on the statistical information for each abnormal event, the device can then predict whether a fault will occur in the target device. However, in this approach, the statistical information for each abnormal event cannot accurately represent each abnormal event, resulting in low accuracy in fault prediction for the target device. SUMMARY OF THE INVENTION Various aspects of the present disclosure provide a fault prediction method, apparatus, device, storage medium, and program product to address the problem of low accuracy in fault prediction for a target device. In a first aspect, embodiments of the present disclosure provide a fault prediction method, comprising: obtaining an initial sequence corresponding to multiple abnormal events of a target device, the initial sequence including event information for each abnormal event; determining multiple target subsequences within the initial sequence, wherein the amount of event information included in the target subsequences is less than or equal to a first threshold; and determining a fault prediction result for the target device based on the multiple target subsequences. In a second aspect, embodiments of the present disclosure provide a fault prediction method, comprising: obtaining an initial sequence corresponding to multiple abnormal events of a target device, the initial sequence including event information for each abnormal event, where the target device is a cloud server; determining multiple target subsequences in the initial sequence, where the amount of event information included in the target subsequences is less than or equal to a first threshold; and determining a fault prediction result for the target device based on the multiple target subsequences.In a third aspect, embodiments of the present disclosure provide a fault prediction device, comprising: an acquisition module, a first determination module, and a second determination module. The acquisition module is configured to acquire an initial sequence corresponding to multiple abnormal events of a target device, the initial sequence including event information for each abnormal event; the first determination module is configured to determine multiple target subsequences within the initial sequence, the amount of event information included in the target subsequences being less than or equal to a first threshold; and the second determination module is configured to determine a fault prediction result for the target device based on the multiple target subsequences. In a fourth aspect, embodiments of the present disclosure provide a fault prediction device, comprising: an acquisition module, a first determination module, and a second determination module. The acquisition module is configured to acquire an initial sequence corresponding to multiple abnormal events of a target device, the initial sequence including event information for each abnormal event, wherein the target device is a cloud server; the first determination module is configured to determine multiple target subsequences within the initial sequence, the amount of event information included in the target subsequences being less than or equal to the first threshold; and the second determination module is configured to determine a fault prediction result for the target device based on the multiple target subsequences. In a fifth aspect, embodiments of the present disclosure provide an electronic device, comprising: a memory and a processor; the memory storing computer-executable instructions; and the processor executing the computer-executable instructions stored in the memory, causing the processor to perform the method described in either the first or second aspect. In a sixth aspect, embodiments of the present disclosure provide a computer-readable storage medium storing computer-executable instructions, which, when executed by the processor, implement the method described in either the first or second aspect. In a seventh aspect, embodiments of the present disclosure provide a computer program product, comprising a computer program, which, when executed by the processor, implements the method described in either the first or second aspect. Embodiments of the present disclosure provide a fault prediction method, apparatus, device, storage medium, and program product. A computing device can obtain initial sequences corresponding to multiple abnormal events of a target device, determine multiple target subsequences within the initial sequences, and then determine a fault prediction result for the target device based on the multiple target subsequences. Because the initial sequence includes event information for each abnormal event, it can more accurately represent each abnormal event compared to the statistical information of each abnormal event in the existing technology. Furthermore, multiple target subsequences can be determined from the initial sequence. When fault prediction is performed based on these multiple target subsequences, each target subsequence can be processed more accurately, thereby comprehensively improving the accuracy of fault prediction for the target device.BRIEF DESCRIPTION OF THE DRAWINGS The accompanying drawings described herein are intended to provide a further understanding of the present disclosure and constitute a part of the present disclosure. The exemplary embodiments of the present disclosure and their descriptions are intended to explain the present disclosure and do not constitute undue limitations thereon. In the accompanying drawings: FIG1 is a schematic diagram of a scenario provided by an exemplary embodiment of the present disclosure; FIG2 is a schematic flow diagram of a fault prediction method provided by an exemplary embodiment of the present disclosure; FIG3 is a schematic diagram of the structure of a target model provided by an exemplary embodiment of the present disclosure; FIG4 is a schematic flow diagram of another fault prediction method provided by an exemplary embodiment of the present disclosure; FIG5 is a schematic diagram of determining a target subsequence in an initial sequence provided by an exemplary embodiment of the present disclosure; FIG6 is a schematic diagram of the structure of a fault prediction apparatus provided by an exemplary embodiment of the present disclosure; FIG7 is a schematic diagram of the structure of a computing device provided by an exemplary embodiment of the present disclosure. DETAILED DESCRIPTION OF THE EMBODIMENTS It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, and displayed data, etc.) involved in the present disclosure are all information and data authorized by the user or fully authorized by all parties. The collection, use, and processing of the relevant data must comply with relevant laws, regulations, and standards, and corresponding operation portals are provided for the user to choose to authorize or reject. To further clarify the objectives, technical solutions, and advantages of this disclosure, the technical solutions of this disclosure will be described clearly and completely below in conjunction with specific embodiments of this disclosure and the corresponding drawings. Obviously, the described embodiments represent only a portion of the embodiments of this disclosure, and are not exhaustive. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of this disclosure without inventive effort are within the scope of protection of this disclosure. Figure 1 is a schematic diagram of a scenario provided by an exemplary embodiment of this disclosure. Referring to Figure 1 , a target device and a computing device may occur. Multiple abnormal events may occur on the target device. For example, the target device may experience abnormal event 1, abnormal event 2, ..., abnormal event z (where z is an integer greater than or equal to 1). The computing device may obtain multiple abnormal events from the target device and, based on these multiple abnormal events, perform fault prediction on the target device to determine a fault prediction result for the target device. In related technologies, a computing device may obtain multiple abnormal events of a target device within multiple preset historical time periods, and determine statistical information of each abnormal event within the multiple preset historical time periods (i.e., the total number of times each abnormal event occurred within the multiple preset time periods, as well as the maximum number of times, and the minimum number of times). Based on the statistical information of each abnormal event, it may then be possible to predict whether the target device will malfunction.However, in the above approach, the statistical information of each abnormal event cannot accurately represent each abnormal event, resulting in low accuracy in fault prediction for the target device. In an embodiment of the present disclosure, a computing device can obtain multiple abnormal events for a target device and determine an initial sequence based on the multiple abnormal events. The initial sequence can include event information for each abnormal event. The computing device can then determine multiple target subsequences from the initial sequence and process the multiple target subsequences to determine a fault prediction result for the target device. Because the initial sequence includes event information for each abnormal event, the initial sequence can more accurately represent each abnormal event compared to the statistical information of each abnormal event in the prior art. Furthermore, the computing device can determine multiple target subsequences from the initial sequence. When performing fault prediction based on the multiple target subsequences, each target subsequence can be processed more accurately, thereby comprehensively improving the accuracy of fault prediction for the target device. The technical solutions presented in this disclosure are described in detail below through specific embodiments. It should be noted that the following embodiments may exist independently or in combination with each other, and identical or similar content will not be repeated in different embodiments. Figure 2 is a flow chart of a fault prediction method provided by an exemplary embodiment of the present disclosure. Referring to FIG. 2 , the method may include:

[0002] S201. Obtain an initial sequence corresponding to multiple abnormal events of a target device. The execution subject of the embodiment of the present disclosure may be a computing device, or a fault prediction device provided in the computing device. The fault prediction device may be implemented through software or through a combination of software and hardware. The fault prediction device may be a processor in the computing device. For ease of understanding, the following description will be made using the computing device as the execution subject. The target device may be a device for which fault prediction is required. For example, the target device may be a local server, a cloud server in a cloud computing system, etc. During operation, the target device may generate multiple abnormal events. For any abnormal event, the abnormal event may be represented by a phrase. For example, if abnormal event 1 is an event in which a hardware error occurs, abnormal event 1 may be represented as "dmesg_unrecover_mce". For example, if the target device is cloud server 1, cloud server 1 may experience multiple abnormal events. Assume that cloud server 1 has 8,000 abnormal events, namely abnormal event 1, abnormal event 2, ..., abnormal event 8,000. oFor any abnormal event, the computing device can determine the event information of the abnormal event in the abnormal event library. Optionally, the event information of the abnormal event can be a preset number corresponding to the abnormal event. Optionally, the computing device can have a preset abnormal event library. The abnormal event library can include multiple preset abnormal events and a preset number corresponding to each preset abnormal event. The multiple preset abnormal events are different. For example, the abnormal event library can be as shown in Table 1: Table 1 For any abnormal event, the computing device can query the abnormal event library based on the abnormal event. If the abnormal event is the same as the target preset abnormal event, the preset number corresponding to the target preset abnormal event can be determined as the preset number corresponding to the abnormal event, that is, the event information. The target preset abnormal event can be any one of multiple preset abnormal events. For example, if 8000 abnormal events occur on cloud server 1, namely abnormal event 1, abnormal event 2, ..., abnormal event 8000, if the abnormal event library is as shown in Table 1, then for abnormal event 1, the computing device can query the abnormal event library. If abnormal event 1 is the same as the preset abnormal event 1, the preset number corresponding to the preset abnormal event 1 is 001, then the event information of abnormal event 1 can be determined to be 001; assuming that the computing device can determine the event information of the 8000 abnormal events as shown in Table 2: Table 2 After the computing device determines the event information of each of the multiple abnormal events, it can generate an initial sequence corresponding to the multiple abnormal events based on the event information of the multiple abnormal events. The initial sequence can include the event information of each abnormal event. The amount of event information included in the initial sequence can be greater than a first threshold. The first threshold can be manually preset. For example, the first threshold can be 512. Optionally, the length of the initial sequence can be represented by the amount of event information included in the initial sequence. The initial sequence can be an extra-long sequence. For example, if the initial sequence includes 8000 event information, the sequence length of the initial sequence can be 8000. For example, if 8000 abnormal events occur in cloud server 1, and the event information of the 8000 abnormal events is shown in Table 2 above, then the initial sequence 1 can be generated as [095, -, 031, -, 006, -, 011, -, 049, 125, -, 032, -, 003, 001] based on the event information of the 8000 abnormal events. The initial sequence 1 includes the 8000 event information, and the sequence length of the initial sequence 1 can be 8000. o

[0003] S202. Determine multiple target subsequences in the initial sequence. The amount of event information included in the target subsequences may be less than or equal to a first threshold. The amount of event information included in the multiple target subsequences may be the same, that is, the sequence lengths of the multiple target subsequences may be the same. The computing device may sample the initial sequence to obtain multiple target subsequences. For example, if the computing device samples the initial sequence 1 with a sequence length of 8000, three target subsequences may be obtained, namely, target subsequence 1, target subsequence 2, and target subsequence 3. The sequence lengths of the three target subsequences may all be 200. o

[0004] S203. Determine a fault prediction result for the target device based on the multiple target subsequences. In an optional embodiment, the fault prediction result for the target device can be determined based on the multiple target subsequences in the following manner: determine the sequence features corresponding to each target subsequence and the weight value of each target subsequence; determine the probability of the target device failing in a future time period based on the sequence features and weight value of each target subsequence; and determine the fault prediction result based on the failure probability. The sequence features of the target subsequences can be represented by a matrix. The weight value can be a value between 0 and 1. The future time period can be manually preset or can be a model parameter determined during training of the target model. For example, the future time period can be 48 hours after the current time when the fault prediction result is obtained. The computing device can be provided with a target model. The target model can be used to process the multiple target subsequences to determine the fault prediction result for the target device. The target model is explained below with reference to FIG3 . FIG3 is a schematic diagram of the structure of the target model provided in an exemplary embodiment of the present disclosure. Referring to FIG3 , the target model can include multiple embedding layers, multiple encoding layers, and a feature fusion layer. Optionally, the target model may further include a classification layer. For example, the target model may include three embedding layers, namely embedding layer 1, embedding layer 2, and embedding layer 3; and three encoding layers, namely encoding layer 1, encoding layer 2, and encoding layer 3. For any embedding layer, a D*M mapping matrix E may be preset in the embedding layer. M represents the number of preset abnormal events in the abnormal event library; D represents the number of event dimensions. The mapping matrix E may include multiple eigenvalues ​​of the M preset abnormal events in D event dimensions. For example, the mapping matrix E may be as follows: Wherein, y represents the feature value of the Mth preset abnormal event in the Dth event dimension. The computing device can determine a target input subsequence based on the target subsequence. The computing device can input the target input subsequence into an embedding layer. The embedding layer can perform mapping and transformation processing on the target input subsequence to obtain intermediate features corresponding to the target input subsequence, i.e., intermediate features corresponding to the target subsequence. After obtaining the intermediate features corresponding to the target subsequence, the intermediate features can be input into the corresponding encoding layer. The encoding layer can perform feature extraction processing on the intermediate features to obtain sequence features corresponding to each target subsequence. The feature fusion layer may include a fully connected layer and a fusion layer. The fully connected layer can be used to calculate a weight value for each target subsequence based on the sequence features corresponding to each target subsequence; the fusion layer can be used to perform weighted fusion based on the weight value of each target subsequence and the sequence features corresponding to each target subsequence to obtain a fault probability. The classification layer can be used to perform classification based on the fault probability to obtain a fault prediction result. The fault prediction result may include a normal prediction and a fault prediction. It should be noted that the target model can be generated based on a transformer model. For example, if there are three target subsequences, namely target subsequence 1, target subsequence 2, and target subsequence 3, the computing device can determine the target input subsequences corresponding to the three target subsequences, process the three target sub-input sequences through the target model, and obtain the sequence features and weight values ​​corresponding to the three target subsequences, as shown in Table 3: Table 3 If the future period is 48 hours, the computing device can determine the probability of failure of cloud server 1 in the next 48 hours based on the sequence features and weight values ​​corresponding to the three target subsequences in Table 3. Assume that the probability of failure of cloud server 1 in the next 48 hours is 0.8 o Optionally, determining the fault prediction result based on the fault probability may include the following two methods: Method 1: The fault prediction result includes the fault probability. In this method, the computing device may determine the fault probability as the fault prediction result. For example, if the computing device determines that the fault probability of cloud server 1 sending a fault within the next 48 hours is 0.8, then the fault prediction result of cloud server 1 may be determined to be 0.8. OMethod 2: The fault prediction result is either predicted normal or predicted fault. In this method, the computing device can determine a preset threshold and, based on the preset threshold and the fault probability, determine the fault prediction result. If the fault probability is greater than or equal to the preset threshold, the fault prediction result can be determined as predicted fault; if the fault probability is less than the preset threshold, the fault prediction result can be determined as predicted normal. For example, if the computing device determines that the probability of cloud server 1 sending a fault within the next 48 hours is 0.8, and the preset threshold is 0.6, then because the fault probability of 0.8 is greater than the preset threshold of 0.6, the fault prediction result can be determined as predicted fault. In this embodiment of the present disclosure, the computing device can obtain initial sequences corresponding to multiple abnormal events of the target device, determine multiple target subsequences within the initial sequence, and then determine the fault prediction result for the target device based on the multiple target subsequences. Because the initial sequence includes event information for each abnormal event, it can more accurately represent each abnormal event compared to the statistical information of each abnormal event in the prior art. Furthermore, the computing device can identify multiple target subsequences with sequence lengths less than a first threshold within the very long initial sequence. When performing fault prediction based on these multiple target subsequences, each target subsequence can be processed more accurately, thereby comprehensively improving the accuracy of fault prediction for the target device. The following describes the fault prediction method in detail based on the embodiment shown in FIG2 and in conjunction with FIG4 . FIG4 is a flow chart of another fault prediction method provided by an exemplary embodiment of the present disclosure. Referring to FIG4 , the method may include:

[0005] S401. Obtain multiple abnormal events. Optionally, the computing device may obtain multiple logs generated by the target device in real time, identify multiple abnormal logs within the multiple logs, and then determine multiple abnormal events corresponding to the multiple abnormal logs based on expert knowledge and regular expressions. Optionally, the computing device stores expert knowledge. The expert knowledge may include multiple predefined abnormal logs. For example, logs beginning with "Mee" may be predefined as abnormal logs. For any log, the computing device may determine whether the log meets the criteria of the predefined abnormal logs. If the log meets any of the predefined abnormal logs, the log is determined to be an abnormal log; if the log does not meet any of the predefined abnormal logs, the log is determined to be a normal log. For any abnormal log, the abnormal log may record abnormality description information and the time of occurrence. Optionally, the computing device may include multiple regular expressions. These multiple regular expressions may be preset. For each regular expression, the regular expression may correspond to a preset abnormal event. Different regular expressions correspond to different preset abnormal events. For example, regular expression 1 can be "Mee. *". Among them, "Mee" represents an error, " represents any character, and "*" represents any number of occurrences of a character. For example, if there are 100 regular expressions, then the 100 regular expressions may correspond to 100 preset abnormal events, and the 100 preset abnormal events are all different. For any abnormal log, the abnormal event corresponding to the abnormal log may be determined in the following manner: the computing device may determine the abnormal description information in the abnormal log, and determine the target regular expression that the abnormal description information meets from multiple regular expressions, and then determine the preset abnormal event corresponding to the target regular expression as the abnormal event corresponding to the abnormal log. For example, for abnormal log 1, if the computing device determines that the abnormal description information in the abnormal log 1 is "mce: [Hardware Error]: Machine check events logged", if the computing device determines that the abnormal description information meets regular expression 1, then the preset abnormal event 1 corresponding to regular expression 1 may be determined as abnormal event 1 corresponding to abnormal log 1. Assume that abnormal event 1 can be expressed as "dmesg_unrecover_mce".For example, if the computing device can obtain 8000 exception logs of cloud server 1, namely exception log 1, exception log 2, , exception log 8000, the computing device can determine the 8000 exception events corresponding to the 8000 exception logs, namely exception event 1, exception event 2, . . . , exception event 8000, in the above manner. o The 8,000 abnormal events may include the same abnormal event. For example, abnormal event 1 and abnormal event 5 may be the same abnormal event.

[0006] S402. Determine the event information and occurrence time of each abnormal event. Since any abnormal log records the abnormality description and occurrence time, after determining the abnormal event corresponding to the abnormal log, the occurrence time recorded in the abnormal log can be determined as the occurrence time of the corresponding abnormal event. For example, if abnormal log 1 records an occurrence time of 2024 / 3 / 18 / 14:00:00, and abnormal log 1 corresponds to abnormal event 1, the occurrence time of abnormal event 1 can be determined to be 2024 / 3 / 18 / 14:00:00. Since the computing device may have a preset abnormal event library, which may include multiple preset abnormal events and a preset number corresponding to each preset abnormal event, the computing device can query the abnormal event library based on the abnormal event to determine the preset number corresponding to the abnormal event, and then determine the preset number corresponding to the abnormal event as the event information of the abnormal event. For example, if a computing device determines 8,000 abnormal events, namely abnormal event 1, abnormal event 2, ..., abnormal event 8000, and if the abnormal event database is as shown in Table 1, the computing device can query the abnormal event database to determine the event information of each of the 8,000 abnormal events, that is, the preset numbers. Assume that the event information of the 8,000 abnormal events can be determined as shown in Table 2.

[0007] S403. Sort the event information of multiple abnormal events in reverse order of occurrence time to obtain an initial sequence. Since each abnormal event has a corresponding occurrence time, the computing device can sort the event information of multiple abnormal events in reverse order of occurrence time, i.e., from latest to earliest occurrence time, to obtain the initial sequence. The occurrence time of each abnormal event corresponding to each event information in the initial sequence can be arranged in reverse order. For example, if there are 8000 abnormal events, namely abnormal event 1, abnormal event 2, ..., abnormal event 8000, and the occurrence time and event information corresponding to each of the 8000 abnormal events are as shown in Table 4: Table 4 Then, the event information of multiple abnormal events can be sorted in the order of occurrence time from latest to earliest, and the order of the event information of the 8000 abnormal events is: 095, ..., 031, ..., 006, ..., 011, ..., 049, , 125, , 032, , 003, 001. The electronic device can then determine that the initial sequence 1 is [095, -, 031, -, 006, -, 011, -, 049, 125, -, 032, 003, 001]. o Optionally, a start symbol [start] may be added before the initial sequence. The start symbol [start] may be used to indicate the beginning of the initial sequence. For example, initial sequence 1 may be as follows:

[0008] [start, 095, •••, 031, •••, 006, …, 01 1, …, 049, •,,, 125, •••, 032, •••, 003, 001] o

[0009] S404. Determine multiple initial subsequences in the initial sequence. In an optional embodiment, multiple initial subsequences can be determined in the initial sequence in the following manner: Determine multiple segmentation lengths; for any segmentation length L, determine the first L event information in the initial sequence as the initial subsequence corresponding to the segmentation length, where L can be an integer greater than or equal to 1. The segmentation length can be represented by the number of event information in the initial subsequence. The segmentation length can be less than or equal to the number of event information included in the initial sequence. For example, if the segmentation length is 200, it means that the initial subsequence corresponding to the segmentation length includes 200 event information. Optionally, multiple segmentation lengths can be represented by "L, , , , , , . . . " respectively. The segmentation lengths can be manually preset or automatically determined by the computing device based on the sequence length of the initial sequence. For example, if the sequence length of the initial sequence is 8000, the segmentation length L can be 2000, the segmentation length 1_2 can be 4000, and the segmentation length 1_3 can be 8000. oDetermining an initial subsequence within an initial sequence is described below with reference to FIG5 . FIG5 is a schematic diagram illustrating determining a target subsequence within an initial sequence according to an exemplary embodiment of the present disclosure. Referring to FIG5 , initial sequence 1 may be [095, -, 031, -, 006, -, 011, -, 049, 125, -, 032, -, 003, 001]. Initial sequence 1 includes 8,000 pieces of event information, and therefore comprises 8,000 bits, i.e., bits 0 to 7,999. If the computing device can determine three segmentation lengths, namely segmentation length 1_=2000, segmentation length 1_2=4000, and segmentation length 1_3=8000, then the computing device can determine the first 2000 event information in initial sequence 1 according to segmentation length Li, that is, the 2000 event information corresponding to bits 0 to 1999, and determine these 2000 event information as initial subsequence 1 corresponding to segmentation length L]. Initial subsequence 1 can be [095, …, 031, …, 006, …, 011, …, 049, …, 125]. For segmentation length L2, the computing device can determine the first 4000 event information in initial sequence 1, that is, the 4000 event information corresponding to bits 0 to 3999, and determine these 4000 event information as initial subsequence 2 corresponding to segmentation length 1_2. Then the initial subsequence 2 can be [095, …, 031, …, 006, …, 011, …, 049, …, 125, …, 032]; for the segmentation length 1_3, since the segmentation length L' and the sequence length of the initial sequence are both 8000, the computing device can determine the initial sequence 1 as the initial subsequence 3 corresponding to the segmentation length 3, then the initial subsequence 3 can be

[0010] [095, -, 031, -, 006, -, 011, -, 049, 125, -, 032, 003, 001] o Then the computing device can determine three initial subsequences in the initial sequence 1, namely initial subsequence 1, initial subsequence 2 and initial subsequence 3.

[0011] S405: Sampling is performed on each initial subsequence to obtain multiple target subsequences. For any initial subsequence, in an optional embodiment, the target subsequence corresponding to the initial subsequence can be obtained by: determining a sampling step length corresponding to the initial subsequence based on the segmentation length corresponding to the initial subsequence; determining an offset value corresponding to the initial subsequence; and sampling the initial subsequence according to the offset value and the sampling step length to obtain the target subsequence corresponding to the initial subsequence. For any segmentation length, the segmentation length has a corresponding sampling step length. Optionally, the sampling step length can be represented by S. For example, segmentation length L1 can correspond to sampling step length S; segmentation length L2 can correspond to sampling step S2; ...; segmentation length Ln can correspond to sampling step S. n o The ratio of the segmentation length corresponding to the initial subsequence to the sampling step length can be a preset value. If the preset value is represented by K, that is: For example, if the segmentation length L is 2000, the sampling step S can be 10; if the segmentation length L is 4000, the sampling step S2 can be 20; if the segmentation length L is 8000, the sampling step S3 can be 40. The offset value can be used to determine the initial sampling position when performing sampling processing in the initial subsequence. For example, if the offset value is 2, the initial sampling position when performing sampling processing in the initial subsequence can be determined to be the second position, that is, sampling processing starts from the second position. Optionally, a random index strategy can be used to determine the offset value corresponding to the initial subsequence. The offset value can be an integer greater than or equal to 0 and less than the sampling step, that is, the offset value can be [0, S n - 1]. That is to say For example, if sampling step S2 is 20, then when sampling is performed in initial subsequence 2 according to sampling step S2, the offset value may be in the range of [0, 19], i.e., the initial sampling position may be bit 0, bit 1, bit 2, ..., or bit 19 in initial subsequence 2. If sampling step S' is 40, then when sampling is performed in initial subsequence 3 according to sampling step S', the offset value may be in the range of [0, 39], i.e., the initial sampling position may be bit 0, bit 1, bit 2, ..., or bit 39 in initial subsequence 3. For any initial subsequence, after the computing device determines the offset value and sampling step corresponding to the initial subsequence, it may perform sampling on the initial subsequence according to the offset value and sampling step to obtain a target subsequence corresponding to the initial subsequence. The target subsequence may include a preset number of pieces of event information, i.e., the sequence length of the target subsequence is a preset value K. Multiple target subsequences may have the same sequence length. Optionally, the target subsequence can be represented by q. For example, as shown in FIG5 , for initial subsequence 1, if the segmentation length L corresponding to initial subsequence 1 is 2000, the corresponding sampling step length & can be 10, and the corresponding offset value can be 0. Then, the computing device can sample the initial subsequence 1 from bit 0 in initial subsequence 1 with a sampling step length & of 10 (i.e., sampling once every 10 bits) to obtain target subsequence 1. Target subsequence 1 is q, = [031, 006, 011, 049, 125]. OTarget subsequence 1 can include 200 event information, meaning the sequence length of target subsequence 1 is 200, as shown in Figure 5 . For initial subsequence 2, if the segmentation length 1_2 corresponding to initial subsequence 2 is 4000, the corresponding sampling step S2 is 2 (i.e., sampling once every 20 bits). If the corresponding offset value is 0, the computing device can sample initial subsequence 2 starting from bit 0, using a sampling step S2 of 20, to obtain target subsequence 2. Target subsequence 2 can be q2 = [006, 049, 125, -, 032]. Target subsequence 2 can include 200 event information, meaning the sequence length of target subsequence 2 is 200, as shown in Figure 5 . For the initial subsequence 3, if the segmentation length 1_3 corresponding to the initial subsequence 3 is 8000, the corresponding sampling step S3 is 40 (i.e., sampling is performed once every 40 bits). If the corresponding offset value is 0, the initial subsequence 3 can be sampled from the 0th bit in the initial subsequence 3 according to the sampling step S3 of 40 to obtain the target subsequence 3. The target subsequence 3 can be q3 = [049, 125, -, 032, -, 001]. The target subsequence 3 can include 200 event information, that is, the sequence length of the target subsequence 3 is 200, as shown in Figure 5. As shown in Figure 5, the sequence lengths of the target subsequences 1, 2, and 3 are the same, all 200. o

[0012] S406: For any target subsequence, determine the intermediate features corresponding to the target subsequence. Optionally, for any target subsequence, a [cis] identifier can be added before the target subsequence to obtain the target input subsequence. The [cis] identifier can be used to indicate the beginning of the target input subsequence. For example, if target subsequence 1 is [031, 006, 011, 049, …, 125], [cis] can be added before target subsequence 1 to obtain target input subsequence 1 as [cis, 031, 006, 011, 049, …, 125]. Optionally, the target input subsequence can be processed using an embedding layer in the target model to obtain the intermediate features. Since the embedding layer is preset with a D*M mapping matrix E, the embedding layer can perform mapping and transformation processing on the target input subsequence according to the mapping matrix E to obtain the intermediate features corresponding to the target subsequence, i.e., the intermediate features corresponding to the target subsequence. Optionally, since the mapping matrix E includes the eigenvalues ​​of M preset abnormal events across D event dimensions, the computing device can determine multiple eigenvalues ​​of the K abnormal events across D event dimensions in the mapping matrix E based on the event information of the K abnormal events in the target input sequence. The computing device can also randomly initialize the eigenvalues ​​of "cis" across the D event dimensions to obtain intermediate features corresponding to the target input sequence. Here, M is greater than or equal to K. Optionally, the intermediate features can be represented by a matrix A of D*(K+1). For example, if the target subsequence 1 is [031, 006, 011, 049, …, 125] and includes 200 event information, the corresponding target input subsequence 1 can be [cis, 031, 006, 011, 049, …, 125]. If D is 512 and the target model is as shown in Figure 3, then the target input subsequence 1 can be mapped and transformed by the embedding layer 1 of the target model according to the mapping matrix E to obtain the intermediate feature 1. The intermediate feature 1 can be represented as a 512*201 matrix Ai. Assume that the matrix Ai can be as follows: 0.10 0.05 - 0.11

[0013] . _ 0.21 0.16 - 0.46 AL . :

[0014] _0.17 0.32 - 0.82. Matrix A consists of 512 rows and 201 columns. The eigenvalue in the first column is the randomly initialized eigenvalue of "cis" across the 512 event dimensions; the eigenvalue in the second column is the eigenvalue corresponding to "031" across the 512 event dimensions; ...; the eigenvalue in the 201st column is the eigenvalue corresponding to "125" across the 512 event dimensions. Similarly, embedding layer 2 can be used to perform mapping transformation processing on target input subsequence 2 corresponding to target subsequence 2 to obtain intermediate feature 2. Assume that intermediate feature 2 can be represented by matrix A2. Embedding layer 3 can be used to perform mapping transformation processing on target input subsequence 3 corresponding to target subsequence 3 to obtain intermediate feature 3. Assume that intermediate feature 3 can be represented by matrix A3.

[0015] S407. Based on the correlation information between the event features in the intermediate features, feature extraction processing is performed on the intermediate features to obtain sequence features corresponding to the target subsequence. Optionally, feature extraction processing can be performed on the intermediate features by the encoding layer in the target model to obtain sequence features. Since the intermediate features include a column of feature values ​​corresponding to "cis," and feature extraction processing is performed on the intermediate features by the encoding layer in the target model, correlations between abnormal events in the target subsequence can be calculated based on each feature value in the intermediate features. That is, the column of feature values ​​corresponding to "cis" has been correlated with multiple feature values ​​corresponding to multiple abnormal events in the intermediate features. Therefore, after feature extraction processing is performed on the intermediate features by the encoding layer, target features corresponding to the target subsequence can be obtained. The target features may include correlation feature values ​​between abnormal events and feature values ​​corresponding to "cis." The feature values ​​corresponding to "cis" may represent the overall features of the target subsequence. The computing device may determine the feature values ​​corresponding to "cis" in the target features and determine the feature values ​​corresponding to "cis" as the sequence features corresponding to the target subsequence. Optionally, the sequence feature can be represented by a D*1 column matrix qds. For example, if the intermediate feature 1 is as shown in the above matrix, and if the target model is as shown in FIG3 , feature extraction processing can be performed on the intermediate feature 1 by encoding layer 1 to obtain the sequence feature 1 corresponding to the target subsequence 1. The sequence feature 1 corresponding to the target subsequence 1 can then be represented by the column matrix qs as follows: Similarly, the intermediate feature 2 can be processed by the encoding layer 2 to obtain the sequence feature 2 corresponding to the target subsequence 2, and the sequence feature 2 can be represented as a column matrix \(q_{2s}\); the intermediate feature 3 can be processed by the encoding layer 3 to obtain the sequence feature 3 corresponding to the target subsequence 3, and the sequence feature 3 can be represented as a column matrix \(q_{3s}\).

[0016] 5408. Determine the weight value of each target subsequence. Optionally, the sequence features corresponding to each target subsequence can be processed by a fully connected layer in the feature fusion layer of the target model to obtain the weight value of each target subsequence. Optionally, if the weight value is represented by \(w\), the following formula (1) can be preset in the fully connected layer: \(w = Softmax(WQ)\) Formula (1) where, Softmax represents the activation function; \(W\) represents the weight matrix; if there are \(m\) target subsequences, then \(Q=(q_{1s},...,q_{ms})\). For example, if there are 3 target subsequences, where the target subsequence 1 corresponds to the sequence feature 1, the target subsequence 2 corresponds to the sequence feature 2, and the target subsequence 3 corresponds to the sequence feature 3, then the 3 sequence features can be processed by formula (1) respectively to obtain the weight value 1 of the target subsequence 1, the weight value 2 of the target subsequence 2, and the weight value 3 of the target subsequence 3, as shown in Table 3. 5409. Determine the failure probability of the target device in the future period according to the sequence features corresponding to each target subsequence and the weight value of each target subsequence. Among them, the value of \(i\) is 1, 2,..., \(m\), \(m\) is the number of target subsequences, \(m\) is an integer greater than or equal to 1, \(Rep\) represents the failure probability; \(\omega_i\) represents the weight value of the \(i\)-th target subsequence; \(q_{is}\) represents the sequence feature of the \(i\)-th target subsequence. Formula (2) means: after multiplying the weight values of \(m\) target subsequences by the corresponding sequence features respectively and then accumulating, the weighted sum \(Rep\) is obtained, and \(Rep\) is the failure probability. For example, if there are 3 target subsequences, the sequence feature 1 corresponding to the target subsequence 1 is \(q_{1s}\), the sequence feature 2 corresponding to the target subsequence 2 is \(q_{2s}\), and the sequence feature 3 corresponding to the target subsequence 3 is \(q_{3s}\), if the weight values of the 3 target subsequences are as shown in Table 3, then through formula (2), it can be calculated that:

[0017] \(Rep = 0.2*q_{1s}+0.7*q_{2s}+0.3*q_{3s}\) lsAssume that Rep = 0.8 can be calculated. If the future period is the next 48 hours, then it can be determined that the failure probability of cloud server 1 in the next 48 hours is 0.8 O

[0018] S410. Determine a fault prediction result based on the fault probability. For example, if the fault prediction result includes the fault probability, if the computing device determines that the fault probability of cloud server 1 sending a fault within the next 48 hours is 0.8, then the computing device determines that the fault prediction result of cloud server 1 is 0.8. oFor example, if the fault prediction result is predicted normal or predicted fault, and if the computing device determines that the probability of cloud server 1 sending a fault message within the next 48 hours is 0.8, and if the preset threshold is 0.6, then because the failure probability of 0.8 is greater than the preset threshold of 0.6, the fault prediction result can be determined to be a predicted fault. In the technical solution of the present disclosure, multiple target subsequences with sequence lengths less than a first threshold can be determined from an extremely long initial sequence. These target subsequences are then input into the target model for processing, significantly increasing the model information content and reducing the model computational complexity. It should be noted that the various processing steps (S401-S410) shown in the embodiment of FIG4 do not constitute a specific limitation on the fault prediction process. In other embodiments of the present disclosure, the fault prediction process may include more or fewer steps than the embodiment of FIG4. For example, the fault prediction process may include some of the steps in the embodiment of FIG4, or some steps in the embodiment of FIG4 may be replaced by steps with the same functionality, or some steps in the embodiment of FIG4 may be split into multiple steps. In an embodiment of the present disclosure, a computing device may obtain multiple abnormal events and determine the event information and occurrence time of each abnormal event. The computing device may sort the event information of the multiple abnormal events in reverse order of occurrence time to obtain an initial sequence, determine multiple initial subsequences within the initial sequence, and then sample each initial subsequence to obtain multiple target subsequences. The computing device may determine intermediate features corresponding to any target subsequence and, based on correlation information between event features within the intermediate features, perform feature extraction on the intermediate features to obtain sequence features corresponding to the target subsequence. The computing device may determine a weight value for each target subsequence and, based on the sequence features corresponding to each target subsequence and the weight value of each target subsequence, determine the probability of a target device failing in a future time period. A fault prediction result may then be determined based on the failure probability. Because the initial sequence includes event information for each abnormal event and is arranged in reverse chronological order, it contains information about the order of the abnormal events. Compared to the statistical information of each abnormal event in the prior art, the initial sequence can more accurately represent each abnormal event, which helps the target model extract richer features. Furthermore, multiple target subsequences with sequence lengths less than a first threshold can be identified within an extremely long initial sequence. When fault prediction is performed based on these multiple target subsequences, the computational complexity of the target model is reduced, and each target subsequence can be processed more accurately. Consequently, the accuracy of fault prediction for the target device is comprehensively improved.It should be noted that, through steps S401-S410, the initial model can also be trained to obtain a target model. The structure of the initial model can be shown in Figure 3. During the model training process, multiple initial sequences can be obtained through steps S401-S403, and these initial sequences can be divided into positive samples (i.e., faulty samples) and negative samples (i.e., normal samples). Each sample is labeled, and a labeling result for each sample is determined. The labeling result can be faulty or normal. Optionally, for any faulty sample, the duration between the occurrence of each abnormal event and the time of device downtime can also be determined. If this duration is greater than a preset impact duration, the abnormal event can be labeled as 0, indicating that the abnormal event is likely unrelated to the device downtime. If this duration is less than or equal to the preset impact duration, the abnormal event can be labeled as 1, indicating that the abnormal event is likely related to the device downtime. The preset impact duration can be manually preset. The preset duration indicates that abnormal events transmitted within this duration are likely to cause device downtime. For example, the preset impact duration can be 48 hours. For any sample, steps S404 and S405 can be performed (in step S405, multiple sampling processes can be performed according to a random indexing strategy) to obtain multiple target subsequences corresponding to the sample. Then, steps S406 to S410 can be performed on the multiple target subsequences using the initial model to obtain a fault prediction result corresponding to the sample. The computing device can compare the fault prediction result corresponding to each sample with the labeled result and use a gradient descent algorithm to backpropagate the loss gradient to update the model parameters of the initial model to obtain a target model. During model training, multiple target subsequences can be identified from an extremely long initial sequence and input into the initial model for processing, which can increase the amount of model information and reduce the computational complexity of the initial model. A random indexing strategy can also be used to determine the offset value, and the target subsequence can be sampled based on the offset value and the sampling step size, so that the representation of the target subsequence can be more generalized, allowing the initial model to be fully learned. In the initial model, the sequence features of multiple target subsequences can be fused with each other through a feature fusion layer, improving the initial model's ability to learn the overall features of the extremely long sequence and the overall performance of the initial model, thereby training a target model with better predictive performance.The target model in this disclosure has two major advantages over the Logarithmic Bidirectional Encoder Representation from Transformers (LogBERT) pre-trained model: (1) it does not require direct learning of the original log text, but instead learns through a pre-processed abnormal event library; (2) the segmentation method can support initial sequences of more than 8000 words in length, and the target model can process longer initial sequences to obtain a larger amount of information, ensuring computational efficiency while improving accuracy and recall. Compared with the extreme gradient boosting (XGBoost) model, the target model of the present disclosure also has two major advantages: (1) The BERT-based algorithm has fully demonstrated its feature extraction capabilities in fields such as natural language processing, and is more convenient and applicable than the manual feature extraction of XGBoost; (2) The feature information extracted by XGBoost is not rich. For example, it does not contain the sequence information of abnormal events, the information between abnormal events, etc., while the target model of the present disclosure can effectively integrate the sequence information of abnormal events and the information between abnormal events. In summary, through the technical solution of the present disclosure, the sequence length of the initial sequence can be more effectively expanded, so that the target model can incorporate more abnormal information to extract more representative feature expressions, thereby improving the performance of the trained classification layer, improving the accuracy and recall rate of fault prediction, and being able to find out the cloud servers that are about to crash in the cloud computing system more quickly and accurately for operation and maintenance, thereby reducing the crash rate or reducing the loss caused by the crash to users, and greatly increasing the stability and reliability of the cloud computing system. Figure 6 is a structural diagram of a fault prediction device provided by an embodiment of the present disclosure. Referring to Figure 6 , the fault prediction device 10 may include: an acquisition module 11, a first determination module 12, and a second determination module 13. The acquisition module 11 is configured to acquire an initial sequence corresponding to multiple abnormal events of a target device, wherein the initial sequence includes event information for each abnormal event; the first determination module 12 is configured to determine multiple target subsequences within the initial sequence, wherein the amount of event information included in each target subsequence is less than or equal to a first threshold; and the second determination module 13 is configured to determine a fault prediction result for the target device based on the multiple target subsequences. The fault prediction device provided in the embodiments of the present disclosure can implement the technical solutions shown in the aforementioned method embodiments. The implementation principles and beneficial effects are similar and are not further described here.In one possible implementation, the number of event information included in the initial sequence is greater than the first threshold; the first determination module 12 is specifically configured to: determine multiple initial subsequences in the initial sequence; and sample each initial subsequence to obtain the multiple target subsequences. In one possible implementation, the occurrence times of the abnormal events corresponding to the event information in the initial sequence are arranged in reverse order; the first determination module 12 is specifically configured to: determine multiple segmentation lengths, each of which is less than or equal to the number of event information included in the initial sequence; and for any segmentation length L, determine the first L event information in the initial sequence as the initial subsequence corresponding to the segmentation length, where L is an integer greater than or equal to 1. In one possible implementation, for any initial subsequence, the first determination module 12 is specifically configured to: determine a sampling step length corresponding to the initial subsequence based on a segmentation length corresponding to the initial subsequence, where the ratio of the segmentation length corresponding to the initial subsequence to the sampling step length is a preset value; determine an offset value corresponding to the initial subsequence, where the offset value is an integer greater than or equal to 0 and less than the sampling step length; and perform sampling processing on the initial subsequence according to the offset value and the sampling step length to obtain a target subsequence corresponding to the initial subsequence. In one possible implementation, the second determination module 13 is specifically configured to: determine a sequence feature corresponding to each target subsequence and a weight value for each target subsequence; determine a failure probability of the target device occurring in a future time period based on the sequence feature corresponding to each target subsequence and the weight value for each target subsequence; and determine the failure prediction result based on the failure probability. In one possible implementation, for any target subsequence, the second determination module 13 is specifically configured to: determine intermediate features corresponding to the target subsequence, where the intermediate features include event features of each abnormal event corresponding to the target subsequence, and the event features include feature values ​​of the abnormal event in multiple event dimensions; and perform feature extraction processing on the intermediate features based on association information between the event features in the intermediate features to obtain sequence features corresponding to the target subsequence.In one possible embodiment, the second determination module 13 is specifically configured to: add a start marker to the target subsequence to obtain a target input subsequence; process the target input subsequence using an embedding layer in the target model to obtain the intermediate features; in one possible embodiment, the second determination module 13 is specifically configured to: perform feature extraction processing on the intermediate features using an encoding layer in the target model to obtain the sequence features; in one possible embodiment, the second determination module 13 is specifically configured to: process the sequence features corresponding to each target subsequence and the weight value of each target subsequence using a feature fusion layer in the target model to obtain the fault probability. In one possible embodiment, the acquisition module 11 is specifically configured to: obtain the multiple abnormal events; determine the event information and occurrence time of each abnormal event; and sort the event information of the multiple abnormal events in reverse order of occurrence time to obtain the initial sequence. The fault prediction device provided in the embodiments of the present disclosure can implement the technical solutions shown in the above-mentioned method embodiments. The implementation principles and beneficial effects are similar and are not further described here. Figure 7 is a schematic structural diagram of a computing device provided in an exemplary embodiment of the present disclosure. Referring to Figure 7 , the computing device 20 may include a processor 21 and a memory 22. Exemplarily, the processor 21 and the memory 22 are interconnected via a bus 23. The memory 22 stores computer-executable instructions; the processor 21 executes the computer-executable instructions stored in the memory 22, causing the processor 21 to perform the method described in the above method embodiment. Accordingly, an embodiment of the present disclosure provides a computer-readable storage medium storing computer-executable instructions. When the computer-executable instructions are executed by the processor, they are used to implement the method described in the above method embodiment. Accordingly, an embodiment of the present disclosure may also provide a computer program product, including a computer program. When executed by the processor, the computer program can implement the method described in the above method embodiment. Those skilled in the art will appreciate that the embodiments of the present disclosure may be provided as methods, systems, or computer program products. Therefore, the present disclosure may take the form of entirely hardware embodiments, entirely software embodiments, or embodiments combining software and hardware. Furthermore, the present disclosure may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.The present disclosure is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present disclosure. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, such that execution of the instructions by the processor of the computer or other programmable data processing device produces means for implementing the functions specified in one or more processes in the flowcharts and / or one or more blocks in the block diagrams. These computer program instructions can also be stored in a computer-readable memory capable of directing the computer or other programmable data processing device to operate in a specific manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means that implement the functions specified in one or more processes in the flowcharts and / or one or more blocks in the block diagrams. These computer program instructions can also be loaded onto a computer or other programmable data processing device, causing the computer or other programmable device to execute a series of operational steps to produce a computer-implemented process. The instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one or more flow charts and / or one or more blocks in a block diagram. In a typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory. Memory may include non-permanent storage in a computer-readable medium, random access memory (RAM), and / or non-volatile memory, such as read-only memory (ROM) or flash memory. OMemory is an example of computer-readable media. Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can implement information storage using any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change RAM (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic disk storage or other magnetic storage devices, or any other non-transmitting medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves. It should also be noted that the terms "comprise," "include," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, product, or apparatus comprising a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, product, or apparatus. Without further limitation, an element specified by the phrase "comprising a..." does not preclude the presence of additional identical elements in the process, method, product, or apparatus comprising the recited element. The foregoing description is merely an example of the present disclosure and is not intended to limit the present disclosure. Various modifications and variations of the present disclosure will be apparent to those skilled in the art. Any modifications, equivalent substitutions, improvements, and the like made within the spirit and principles of the present disclosure are intended to be encompassed by the claims of the present disclosure.

Claims

Claims 1. A fault prediction method, wherein: include: Acquire an initial sequence corresponding to multiple abnormal events of a target device, wherein the initial sequence includes event information of each abnormal event; Determine a plurality of target subsequences in the initial sequence, where the amount of event information included in the target subsequences is less than or equal to a first threshold; and determine a fault prediction result of the target device according to the plurality of target subsequences.

2. A fault prediction method, wherein: include: Acquire an initial sequence corresponding to multiple abnormal events of a target device, wherein the initial sequence includes event information of each abnormal event, and the target device is a cloud server; Determine a plurality of target subsequences in the initial sequence, where the amount of event information included in the target subsequences is less than or equal to a first threshold; and determine a fault prediction result of the target device according to the plurality of target subsequences.

3. The method according to claim 1 or 2, wherein: The amount of event information included in the initial sequence is greater than the first threshold; Determining a plurality of target subsequences in the initial sequence includes: determining a plurality of initial subsequences in the initial sequence; Sampling is performed on each initial subsequence to obtain the multiple target subsequences.

4. The method according to claim 3, wherein: The occurrence time of the abnormal events corresponding to each event information in the initial sequence is arranged in reverse order; Determining a plurality of initial subsequences in the initial sequence includes: determining a plurality of segmentation lengths, wherein the segmentation lengths are less than or equal to the number of event information included in the initial sequence; For any segmentation length L, the first L event information in the initial sequence is determined as the initial subsequence corresponding to the segmentation length L, where L is an integer greater than or equal to 1.

5. The method according to claim 3 or 4, wherein: For any initial subsequence, sampling processing is performed on the initial subsequence to obtain a target subsequence corresponding to the initial subsequence, including: determining a sampling step length corresponding to the initial subsequence based on a segmentation length corresponding to the initial subsequence, where a ratio of the segmentation length corresponding to the initial subsequence to the sampling step length is a preset value; determining an offset value corresponding to the initial subsequence, where the offset value is an integer greater than or equal to 0 and less than the sampling step length; and sampling processing is performed on the initial subsequence based on the offset value and the sampling step length to obtain a target subsequence corresponding to the initial subsequence.

6. The method according to any one of claims 1 to 5, wherein: Determining a fault prediction result for the target device based on the multiple target subsequences includes: determining a sequence feature corresponding to each target subsequence and a weight value for each target subsequence; determining a failure probability of the target device failing in a future time period based on the sequence feature corresponding to each target subsequence and the weight value for each target subsequence; and determining the fault prediction result based on the failure probability.

7. The method according to claim 6, wherein: For any target subsequence; determining a sequence feature corresponding to the target subsequence, including: determining an intermediate feature corresponding to the target subsequence, wherein the intermediate feature includes an event feature of each abnormal event corresponding to the target subsequence, and the event feature includes feature values ​​of the abnormal event in multiple event dimensions; According to the association information between the event features in the intermediate features, feature extraction processing is performed on the intermediate features to obtain sequence features corresponding to the target subsequence.

8. The method according to claim 7, wherein: Determining the intermediate features corresponding to the target subsequence includes: adding a start marker to the target subsequence to obtain a target input subsequence; processing the target input subsequence through an embedding layer in a target model to obtain the intermediate features; performing feature extraction processing on the intermediate features based on association information between event features in the intermediate features to obtain sequence features corresponding to the target subsequence, including: performing feature extraction processing on the intermediate features through an encoding layer in the target model to obtain the sequence features; determining a failure probability of the target device occurring in a future time period based on the sequence features corresponding to each target subsequence and a weight value of each target subsequence, including: processing the sequence features corresponding to each target subsequence and the weight value of each target subsequence through a feature fusion layer in the target model to obtain the failure probability.

9. The method according to any one of claims 1 to 8, wherein: Obtaining an initial sequence corresponding to multiple abnormal events of a target device includes: obtaining the multiple abnormal events; determining event information and an occurrence time of each abnormal event; and sorting the event information of the multiple abnormal events in reverse order of the occurrence time to obtain the initial sequence.

10. A fault prediction device, wherein: include: An acquisition module, a first determination module, and a second determination module, wherein the acquisition module is configured to acquire an initial sequence corresponding to multiple abnormal events of a target device, wherein the initial sequence includes event information of each abnormal event; the first determination module is configured to determine multiple target subsequences in the initial sequence, wherein the amount of event information included in the target subsequences is less than or equal to a first threshold; and the second determination module is configured to determine a fault prediction result for the target device based on the multiple target subsequences.

11. A computing device, wherein: include: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the computing device to perform the method according to any one of claims 1 to 9.

12. A computer-readable storage medium, wherein: The computer-readable storage medium stores computer-executable instructions, and when the processor executes the computer-executable instructions, the method according to any one of claims 1 to 9 is implemented.

13. A computer program product, comprising a computer program, wherein: When the computer program is executed by a processor, the method according to any one of claims 1 to 9 is implemented.

Images