Fault prediction method and apparatus, and device, storage medium and program product

Abstract

Provided in the present disclosure are a fault prediction method and apparatus, and a device, a storage medium and a program product. The method comprises: acquiring a plurality of pieces of initial performance information corresponding to a plurality of performance indicators of a target device, wherein the initial performance information comprises a plurality of pieces of performance data of the target device under the performance indicators within a historical time period; acquiring event information of a plurality of abnormal events of the target device, wherein occurrence moments of the plurality of abnormal events are within the historical time period; and on the basis of the plurality of pieces of initial performance information and the event information, determining a fault prediction result of the target device. Therefore, the accuracy of fault prediction for a target device is improved.

Description

[0001] The present disclosure claims priority to Chinese patent application No. 202410636949.4, filed on May 21, 2024, with the Chinese Patent Office, entitled "Fault prediction method, device, equipment, storage medium and program product", the entire content of which is incorporated herein by reference. TECHNICAL FIELD The present disclosure relates to the field of computers, and in particular, to a fault prediction method, device, equipment, storage medium and program product. BACKGROUND TECHNIQUE Equipment may fail during operation, thereby affecting the business running in the equipment, so in order to avoid affecting the business, the equipment can be fault predicted. During operation, each time an abnormal event occurs in the target equipment, an abnormal log can be generated accordingly. In related technologies, a plurality of abnormal logs of the target equipment can be obtained, the plurality of abnormal logs can be analyzed and processed and feature extraction processed to obtain a feature sequence, and then the target equipment can be fault predicted according to the feature sequence. However, in the above manner, the feature sequence obtained based on the abnormal log does not express comprehensive information, resulting in low accuracy of fault prediction of the target equipment. SUMMARY The present disclosure provides a fault prediction method, device, equipment, storage medium and program product to solve the problem of low accuracy of fault prediction of the target equipment. In a first aspect, the present disclosure provides a fault prediction method, comprising: obtaining a plurality of initial performance information corresponding to a plurality of performance indicators of a target equipment, the initial performance information comprising a plurality of performance data of the target equipment under the performance indicators within a historical period; obtaining event information of a plurality of abnormal events of the target equipment, the occurrence time of the plurality of abnormal events being within the historical period; determining a fault prediction result of the target equipment according to the plurality of initial performance information and the event information. In a possible implementation manner, the fault prediction result of the target equipment is determined according to the plurality of initial performance information and the event information, comprising: determining the number of events of the plurality of abnormal events; performing feature mapping processing on the plurality of initial performance information according to the number of events to obtain a plurality of target performance information, the number of performance features included in the target performance information being the same as the number of events; determining the fault prediction result according to the plurality of target performance information and the event information.In a possible implementation, for any one initial performance information, the initial performance information is subjected to feature mapping processing according to the number of events, to obtain target performance information corresponding to the initial performance information, including: determining a preset dimension; subjecting the initial performance information to feature extraction processing and mapping processing according to the number of events and the preset dimension, to obtain the target performance information. In a possible implementation, the fault prediction result is determined according to the plurality of target performance information and the event information, including: subjecting the plurality of target performance information to splicing processing, to obtain spliced performance information; processing the spliced performance information and the event information, to obtain the fault prediction result. In a possible implementation, the spliced performance information and the event information are processed to obtain the fault prediction result, including: subjecting the spliced performance information and the event information to N times of fusion processing, to obtain target performance features and target event features, where N is an integer greater than or equal to 1; and the fault prediction result is determined according to the target performance features and the target event features. In a possible implementation, the fault prediction result is determined according to the target performance features and the target event features, including: subjecting the target performance features and the target event features to splicing processing, to obtain target features; subjecting the target features to enhancement processing, to obtain target enhanced features; and the fault prediction result is determined according to the target enhanced features. In a possible implementation, the spliced performance information and the event information are subjected to N times of fusion processing, to obtain target performance features and target event features, including: subjecting the spliced performance information and the event information to fusion processing, to obtain first performance features and first event features; subjecting ith performance features and ith event features to fusion processing, to obtain i-th performance features and i-th event features; where i is 2, 3, …, N in turn, the N-th performance features are determined as the target performance features, and the N-th event features are determined as the target event features. In a possible implementation, the event information of a plurality of abnormal events of the target device is obtained, including: determining an event type and an occurrence time of each abnormal event; for any one abnormal event, determining a multi-dimensional event feature of the abnormal event according to the event type of the abnormal event; and generating the event information according to the occurrence time of each abnormal event and the multi-dimensional event feature of each abnormal event, where the multi-dimensional event feature of each abnormal event is included in the event information.In a possible implementation, the multi-dimensional event feature of the abnormal event is determined according to the event type of the abnormal event, including: obtaining preset feature information, the preset feature information including a plurality of event types and event features corresponding to each event type; and determining the multi-dimensional event feature of the abnormal event according to the event type of the abnormal event and the preset feature information. In a possible implementation, the event information is generated according to the occurrence time of each abnormal event and the multi-dimensional event feature of each abnormal event, including: combining and processing the multi-dimensional event features of the plurality of abnormal events in descending order of occurrence time to obtain a combined event feature; generating a time feature according to the occurrence time of each abnormal event; and fusing the combined event feature and the time feature to obtain the event information. In a possible implementation, the time feature is generated according to the occurrence time of each abnormal event, including: determining the latest occurrence time from the plurality of occurrence times of the plurality of abnormal events; determining a plurality of time differences between each occurrence time and the latest occurrence time; determining a feature value corresponding to each time difference, and combining and processing the feature values of the plurality of time differences to obtain the time feature. In a second aspect, an embodiment of the present disclosure provides a fault prediction method, including: obtaining a plurality of initial performance information corresponding to a plurality of performance indicators of a target device, the initial performance information including a plurality of performance data of the target device in a historical period under the performance indicators, and the target device being a cloud server; obtaining event information of a plurality of abnormal events of the target device, the occurrence times of the plurality of abnormal events being in the historical period; and determining a fault prediction result of the target device according to the plurality of initial performance information and the event information. In a possible implementation, the fault prediction result of the target device is determined according to the plurality of initial performance information and the event information, including: determining the number of events of the plurality of abnormal events; performing feature mapping processing on the plurality of initial performance information according to the number of events to obtain a plurality of target performance information, the number of performance features included in the target performance information being the same as the number of events; and determining the fault prediction result according to the plurality of target performance information and the event information.In a possible implementation, the event information of the plurality of abnormal events of the target device is acquired, including: determining an event type and an occurrence time of each abnormal event; for any abnormal event, determining a multi-dimensional event feature of the abnormal event according to the event type of the abnormal event; and generating the event information according to the occurrence time of each abnormal event and the multi-dimensional event feature of each abnormal event, wherein the multi-dimensional event feature of each abnormal event is included in the event information. In a third aspect, an embodiment of the present disclosure provides a fault prediction apparatus, including: a first acquisition module, a second acquisition module, and a determination module, wherein the first acquisition module is configured to acquire a plurality of initial performance information corresponding to a plurality of performance indicators of a target device, the initial performance information including a plurality of performance data of the target device in a historical period and under the performance indicators; the second acquisition module is configured to acquire event information of a plurality of abnormal events of the target device, the occurrence time of the plurality of abnormal events being located in the historical period; and the determination module is configured to determine a fault prediction result of the target device according to the plurality of initial performance information and the event information. In a possible implementation, the determination module is specifically configured to: determine a number of events of the plurality of abnormal events; perform feature mapping processing on the plurality of initial performance information according to the number of events to obtain a plurality of target performance information, the number of performance features included in the target performance information being the same as the number of events; and determine the fault prediction result according to the plurality of target performance information and the event information. In a possible implementation, for any initial performance information, the determination module is specifically configured to: determine a preset dimension; and perform feature extraction processing and mapping processing on the initial performance information according to the number of events and the preset dimension to obtain the target performance information. In a possible implementation, the determination module is specifically configured to: perform splicing processing on the plurality of target performance information to obtain spliced performance information; and perform processing on the spliced performance information and the event information to obtain the fault prediction result. In a possible implementation, the determination module is specifically configured to: perform N times of fusion processing on the spliced performance information and the event information to obtain a target performance feature and a target event feature, N being an integer greater than or equal to 1; and determine the fault prediction result according to the target performance feature and the target event feature.In a possible implementation, the determining module is specifically configured to: perform splicing processing on the target performance feature and the target event feature to obtain a target feature; perform enhancement processing on the target feature to obtain a target enhanced feature; and determine the fault prediction result according to the target enhanced feature. In a possible implementation, the determining module is specifically configured to: perform fusion processing on the spliced performance information and the event information to obtain an i-th performance feature and an i-th event feature; and perform fusion processing on the i-th performance feature and the i-th event feature to obtain an (i+1)-th performance feature and an (i+1)-th event feature, where i is 2, 3, …, N in turn, the N-th performance feature is determined as the target performance feature, and the N-th event feature is determined as the target event feature. In a possible implementation, the second obtaining module is specifically configured to: determine an event type and an occurrence time of each abnormal event; determine, for any one abnormal event, a multi-dimensional event feature of the abnormal event according to the event type of the abnormal event; and generate the event information according to the occurrence time of each abnormal event and the multi-dimensional event feature of each abnormal event, where the multi-dimensional event feature of each abnormal event is included in the event information. In a possible implementation, the second obtaining module is specifically configured to: obtain preset feature information, where the preset feature information includes a plurality of event types and event features corresponding to the event types; and determine the multi-dimensional event feature of the abnormal event according to the event type of the abnormal event and the preset feature information. In a possible implementation, the second obtaining module is specifically configured to: perform combination processing on the multi-dimensional event features of the plurality of abnormal events in a descending order of occurrence time to obtain a combined event feature; generate a time feature according to the occurrence time of each abnormal event; and perform fusion processing on the combined event feature and the time feature to obtain the event information. In a possible implementation, the second obtaining module is specifically configured to: determine a latest occurrence time from a plurality of occurrence times of the plurality of abnormal events; determine a plurality of time differences between each occurrence time and the latest occurrence time; determine a feature value corresponding to each time difference, and perform combination processing on the feature values of the plurality of time differences to obtain the time feature.In a fourth aspect, the embodiments of the present disclosure provide a fault prediction device, which comprises a first obtaining module, a second obtaining module and a determining module. The first obtaining module is configured to obtain a plurality of initial performance information corresponding to a plurality of performance indexes of a target device. The initial performance information comprises a plurality of performance data of the target device in a historical period under the performance indexes. The target device is a cloud server. The second obtaining module is configured to obtain event information of a plurality of abnormal events of the target device. The occurrence time of the plurality of abnormal events is located in the historical period. The determining module is configured to determine a fault prediction result of the target device according to the plurality of initial performance information and the event information. In a possible implementation, the determining module is specifically configured to: determine the number of events of the plurality of abnormal events; perform feature mapping processing on the plurality of initial performance information according to the number of events to obtain a plurality of target performance information. The number of performance features included in the target performance information is the same as the number of events. The determining module is configured to determine the fault prediction result according to the plurality of target performance information and the event information. In a possible implementation, the second obtaining module is specifically configured to: determine the event type and occurrence time of each abnormal event; for any one abnormal event, determine the multi-dimensional event feature of the abnormal event according to the event type of the abnormal event; and generate the event information according to the occurrence time of each abnormal event and the multi-dimensional event feature of each abnormal event. The multi-dimensional event feature of each abnormal event is included in the event information. In a fifth aspect, the embodiments of the present disclosure provide an electronic device, which comprises a memory and a processor. The memory stores computer execution instructions. The processor executes the computer execution instructions stored in the memory, so that the processor executes the method in any one of the first aspect or the second aspect. In a sixth aspect, the embodiments of the present disclosure provide a computer readable storage medium, which stores computer execution instructions. When the computer execution instructions are executed by a processor, the computer execution instructions are used to implement the method in any one of the first aspect or the second aspect. In a seventh aspect, the embodiments of the present disclosure provide a computer program product, which comprises a computer program. When the computer program is executed by a processor, the computer program implements the method in any one of the first aspect or the second aspect. The embodiments of the present disclosure provide a fault prediction method, device, equipment, storage medium and program product. An electronic device can obtain a plurality of initial performance information corresponding to a plurality of performance indexes of a target device, and obtain event information of a plurality of abnormal events of the target device. Then, the electronic device can determine a fault prediction result of the target device according to the plurality of initial performance information and the event information.Compared with the prior art in which only a feature sequence obtained based on an abnormal log is used for fault prediction, the information expressed is more comprehensive, and thus the accuracy of fault prediction on the target device is improved. The accompanying drawings, which are included to provide a further understanding of the present disclosure and constitute a part of the present disclosure, serve to explain the illustrative embodiments of the present disclosure and their description, and do not constitute an improper limitation on the present disclosure. In the drawings: FIG. 1 is a schematic diagram of a scenario provided by an illustrative embodiment of the present disclosure; FIG. 2 is a flowchart of a fault prediction method provided by an illustrative embodiment of the present disclosure; FIG. 3 is a flowchart of another fault prediction method provided by an illustrative embodiment of the present disclosure; FIG. 4 is a schematic diagram of a process of generating event information provided by an illustrative embodiment of the present disclosure; FIG. 5 is a structural schematic diagram of a target model provided by an illustrative embodiment of the present disclosure; FIG. 6 is a flowchart of still another fault prediction method provided by an illustrative embodiment of the present disclosure; FIG. 7 is a schematic diagram of a process of a fault prediction method provided by an illustrative embodiment of the present disclosure; FIG. 8 is a structural schematic diagram of a fault prediction apparatus provided by an illustrative embodiment of the present disclosure; and FIG. 9 is a structural schematic diagram of a computing device provided by an illustrative embodiment of the present disclosure. 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 for analysis, stored data, displayed data, etc.) involved in the present disclosure are all information and data authorized by the user or authorized by all parties, and the collection, use and processing of the related data need to comply with relevant laws, regulations and standards, and provide corresponding operation entrances for the user to select authorization or refusal. In order to make the purpose, technical scheme and advantages of the present disclosure clearer, the technical scheme of the present disclosure will be described clearly and completely by combining the specific embodiments of the present disclosure and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of the present disclosure, not all the embodiments. Based on the embodiments in the present disclosure, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present disclosure. FIG. 1 is a schematic diagram of a scenario provided by an illustrative embodiment of the present disclosure. Please refer to FIG. 1, which includes a target device and an electronic device. The target device can have multiple performance indicators, and each performance indicator has corresponding initial performance information.For example, the performance indicator 1 of the target device can be a central processing unit (CPU) utilization rate, and the initial performance information 1 can include a plurality of CPU utilization rates in a historical period. In different cases, the initial performance information of the target device can be different, and thus the initial performance information of the target device can reflect the running situation of the target device. The target device can have a plurality of abnormal events in the historical period. The abnormal events can cause the target device to malfunction. The electronic device can obtain a plurality of initial performance information and a plurality of abnormal events in the target device, and perform fault prediction on the target device according to the plurality of initial performance information and the plurality of abnormal events to determine a fault prediction result of the target device. In the running process, the target device can generate an abnormal log corresponding to each abnormal event. In the related art, a plurality of abnormal logs of the target device can be obtained, and the plurality of abnormal logs can be parsed and processed and feature extraction processing can be performed to obtain a feature sequence, and then the target device can be fault predicted according to the feature sequence. However, in the above manner, the feature sequence obtained based on the abnormal log does not express comprehensive information, which leads to low accuracy of fault prediction on the target device. In the embodiment of the present disclosure, the electronic device can not only obtain event information of a plurality of abnormal events of the target device, but also obtain a plurality of initial performance information corresponding to a plurality of performance indicators of the target device, and then the fault prediction result of the target device can be determined according to the plurality of initial performance information and the event information. Compared with the prior art of performing fault prediction only according to the feature sequence obtained based on the abnormal log, the information expressed is more comprehensive, and thus the accuracy of fault prediction on the target device is improved. In the following, the technical solutions shown in the present disclosure will be described in detail through specific embodiments. It should be noted that the following embodiments can exist independently, or can be combined with each other, and for the same or similar content, the description will not be repeated in different embodiments. FIG. 2 is a flowchart of a fault prediction method according to an example embodiment of the present disclosure. Please refer to FIG. 2, the method can include:

[0002] S201, obtain a plurality of initial performance information corresponding to a plurality of performance indicators of a target device. The execution subject of the embodiment of the present disclosure can be an electronic device, or a fault prediction device arranged in the electronic device. The fault prediction device can be realized by software, or realized by the combination of software and hardware. The fault prediction device can be a processor in the electronic device. For ease of understanding, in the following, the execution subject is taken as an example of the electronic device. The target device can be a device that needs to be fault predicted. The target device can have a plurality of performance indicators. Each performance indicator has corresponding initial performance information. For any one performance indicator, the initial performance information can include a plurality of performance data of the target device in a historical period under the performance indicator. The historical period can be a period corresponding to a preset time length before the current time. The preset time length can be artificially preset. For example, if the current time is 2024 / 4 / 25 14:00, and the preset time length is 3 days, the historical period can be 2024 / 4 / 22 14:00 to 2024 / 4 / 25 14:00. If the performance indicator is CPU utilization, the corresponding initial performance information can include a plurality of CPU utilization of the target device between 2024 / 4 / 22 14:00 and 2024 / 4 / 25 14:00. Optionally, for any one performance indicator, the performance indicator can have a corresponding collection period. For example, the collection period of performance indicator 1 can be 1h (hour); the collection period of performance indicator 2 can be 2h. Optionally, the electronic device or a data collection device can determine a plurality of performance indicators of the target device. For any one performance indicator, the electronic device or the data collection device can collect a plurality of performance data of the target device under the performance indicator according to the collection period corresponding to the performance indicator, and store the plurality of performance data in a preset storage space. The preset storage space can be in the electronic device, or in the data collection device, or in other storage devices. The electronic device can determine a plurality of performance indicators of the target device, and thus can obtain a plurality of initial performance information corresponding to the plurality of performance indicators of the target device in the historical period in the preset storage space.For example, if there are two performance indicators, CPU utilization and memory utilization, and the current time is 2024 / 4 / 25 14:00, and the historical period is 2024 / 4 / 22 14:00 to 2024 / 4 / 25 14:00, the electronic device can obtain initial performance information 1 corresponding to the CPU utilization of the target device in the preset storage space. The initial performance information 1 can include a plurality of CPU utilizations of the target device in the historical period. Assuming that the initial performance information 1 can be as shown in Table 1: As shown in Table 1, the initial performance information 1 can include 72 CPU utilizations. Similarly, the electronic device can obtain initial performance information 2 corresponding to the memory utilization of the target device in the preset storage space, and the initial performance information 2 can include a plurality of memory utilizations of the target device in the historical period. Assuming that the initial performance information 2 can be as shown in Table 2: As shown in Table 2, the initial performance information 2 can include 36 memory utilizations.

[0003] S202, obtain event information of a plurality of abnormal events of the target device. The target device can occur a plurality of abnormal events in the running process. For example, the abnormal event 1 can be an event of hardware error. Since the target device generates an abnormal log corresponding to each abnormal event, the electronic device can obtain a plurality of abnormal logs of the target device to determine a plurality of abnormal events according to the plurality of abnormal logs. For example, if the historical period is 2024 / 4 / 22 14:00 to 2024 / 4 / 25 14:00, and the target device has occurred 100 abnormal events in the historical period, which are abnormal event 1, abnormal event 2, …, abnormal event 100, the event information can include the characteristics of the 100 abnormal events. Assuming that the event information can be as shown in matrix Ai:

[0004] 0.47 0.97 0.35 ••• 0.81

[0005] . 0.97 0.87 0.85 ••• 0.57

[0006] Ai=. : :... :

[0007] .0.61 0.30 0.49 - 1.00.

[0008] S203, determine the failure prediction result of the target device according to the plurality of initial performance information and the event information. In an optional embodiment, the failure prediction result of the target device can be determined according to the plurality of initial performance information and the event information in the following manner: determine the event quantity of the plurality of abnormal events; perform feature mapping processing on the plurality of initial performance information according to the event quantity to obtain a plurality of target performance information; and determine the failure prediction result according to the plurality of target performance information and the event information. Since there are L abnormal events, the electronic device can determine that the event quantity of the plurality of abnormal events is Lo For example, if there are 100 abnormal events, the event quantity is 100; if there are 2 initial performance information, which are initial performance information 1 and initial performance information 2, if the initial performance information 1 includes 72 CPU utilization rates as shown in Table 1, and if the initial performance information 2 includes 36 memory utilization rates as shown in Table 2, the electronic device can perform feature mapping processing on the initial performance information 1 according to the event quantity 100 to obtain target performance information 1; can perform feature mapping processing on the initial performance information 2 according to the event quantity 100 to obtain target performance information 2, FIG. 3 is a flowchart of another failure prediction method provided by an exemplary embodiment of the present disclosure. Referring to FIG. 3, the method can include:

[0009] 5301, obtain a plurality of initial performance information corresponding to a plurality of performance indicators of a target device. It should be noted that the execution process of step S301 can refer to the execution process of step S201, which will not be described here.

[0010] 5302, determine the event type and occurrence time of each abnormal event. Since the target device generates an abnormal log corresponding to each abnormal event, the electronic device can obtain a plurality of abnormal logs of the target device to determine a plurality of abnormal events according to the plurality of abnormal logs. Optionally, a plurality of event types can be preset. Each event type can be represented by a type identifier. For example, if there are m event types, the type identifier of event type 1 can be 01, the type identifier of event type 2 can be 02, and the type identifier of event type m can be m. For any abnormal event, the abnormal event can have a corresponding event type and occurrence time. Optionally, the event types of a plurality of abnormal events can be represented by a sequence table. If there are L abnormal events, the event types of the L abnormal events can be represented as a 1*L matrix E:

[0011] E = [^1, e2, e3,..., e L ] Wherein, e% represents the type identification of the event type of the abnormal event L. For example, if there are 100 abnormal events, which are abnormal event 1, abnormal event 2, abnormal event 3, ……, abnormal event 100 respectively, the event type and occurrence time of the 100 abnormal events can be as shown in Table 3: The event characteristics of the event type 1; the elements in the second column can represent the event characteristics of the event type 2; ……; the elements in the mth column can represent the event characteristics of the event type m. Optionally, the electronic device can obtain preset feature information in a preset storage space. For example, if m is 30 and d is 512, the electronic device can obtain the preset feature information as shown in the 512*30 matrix &:

[0012] 0.72 0.16 0.85 ••• 0.73 0.48 0.66 0.56 ••• 0.63

[0013] Fi= : : :

[0014] _0.91 0.30 0.41 - 0.06. In matrix Fi, columns 1 to 30 represent the event features of event type 1, event type 2, ..., event type 30, respectively. Since the preset feature information includes the event features corresponding to each event type, the electronic device can determine the multidimensional event features of each abnormal event based on the event type of each abnormal event from the preset feature information. If the preset feature information includes m event types and the event features corresponding to each event type in d event dimensions, then the multidimensional event features of each abnormal event can include the event features in d event dimensions. Optionally, the multidimensional event features can be represented by a d*l matrix G. For example, if the preset feature information is as shown in matrix &; if there are 100 abnormal events, and the event types of these 100 abnormal events are represented by matrix EL[02,30,02,…,01], then the electronic device can determine that the event type of abnormal event 1 is event type 2. Based on event type 2, the event feature 2 corresponding to event type 2 can be determined as the multidimensional event feature 1 of abnormal event 1 in the preset feature FD, as shown in matrix G below; similarly, the electronic device can determine the multidimensional event feature 2 of abnormal event 3 as shown in matrix G2 below; the electronic device can determine the multidimensional event feature 100 of abnormal event 100 as shown in matrix G2 below; …; the electronic device can determine the multidimensional event feature 100 of abnormal event 100 as shown in matrix G2 below; …

[0015] 0.161 [ 0.73 1 [0.16 [ 0.72 - r _ 0.66 r _ 0.63 r _ 0.66 0.48

[0016] 5. , 6. ,

[0017] .0,30 J |_ 0.06 J k.30 |_ 0,91 .

[0018] S304, Based on the time of occurrence of each abnormal event, and each Event characteristics, generate event information. The time feature is obtained by combining the feature values of the time differences. The latest occurrence time refers to the occurrence time of the latest abnormal event in the multiple abnormal events, i.e., the current time, because the latest abnormal event occurs at the current time. Optionally, the unit of the time difference can be minutes (min). The time difference can represent the proximity of the occurrence time of each abnormal event to the current time. For example, if there are 100 abnormal events, namely abnormal event 1, abnormal event 2, abnormal event 3, …, abnormal event 100, the occurrence times of the 100 abnormal events can be as shown in Table 3 in the order from late to early, the latest occurrence time can be determined as the occurrence time of abnormal event 1, i.e., 2024 / 4 / 25 14:00, and the time difference between each occurrence time and the latest occurrence time can be determined to obtain multiple time differences. It is assumed that the multiple time differences can be as shown in Table 4. Optionally, the feature value corresponding to each time difference can be determined in the following manner: determining at least one time granularity; performing down-rounding processing on each time difference according to the at least one time granularity to obtain at least one rounding result of each time difference; performing mapping processing on the at least one rounding result of each time difference by a dynamic embedding function to obtain at least one initial feature value corresponding to each time difference; and performing summation processing on the at least one initial feature value corresponding to each time difference to obtain the feature value corresponding to each time difference. The time granularity can be at least one of the following: day, hour, and minute. For example, if there are 100 time differences as shown in Table 4, and there are 3 time granularities, namely day, hour, and minute, the 100 time differences can be respectively subjected to down-rounding processing according to the 3 time granularities to obtain 300 rounding results, which can be as shown in Table 5. The electronic device can perform mapping processing on the 3 rounding results of each time difference in Table 5 by a dynamic embedding function to obtain 3 initial feature values corresponding to each time difference. As shown in FIG. 4, the 3 initial feature values corresponding to time difference 1 can be feature value 1T, feature value 1-2, and feature value 1-3; the 3 initial feature values corresponding to time difference 2 can be feature value 2-1, feature value 2-2, and feature value 2-3; …; and the 3 initial feature values corresponding to time difference 100 can be feature value 100-1, feature value 100-2, and feature value 100-3. It is assumed that the 3 initial feature values corresponding to each time difference can be as shown in Table 6. As shown in FIG. 4, the electronic device can perform summation processing on at least one initial feature value corresponding to each time difference, and then obtain a feature value corresponding to each time difference. It is assumed that the feature value corresponding to each time difference can be as shown in Table 7. Optionally, after determining the feature value corresponding to each time difference, the electronic device can perform combination processing on the feature values of the plurality of time differences to obtain a time feature. The time feature can be represented by a matrix W of 1*L. For example, if there are 100 time differences, and the feature value corresponding to each time difference is as shown in Table 7, the electronic device can perform combination processing on the feature values corresponding to the 100 time differences to obtain a time feature, and the time feature can be represented as a matrix W1=[0.31, 0.24, 0.19,..., 0.09]. After the electronic device determines the combined event feature and the time feature, the electronic device can determine the multi-dimensional event feature of each abnormal event in the combined event feature, and determine the feature value of the time difference corresponding to each abnormal event in the time feature, and add the feature values of the time differences to the multi-dimensional event feature to perform fusion processing on the combined event feature and the time feature, and obtain event information. For example, as shown in FIG. 4, the electronic device can add the feature value corresponding to the time difference 1 to the multi-dimensional event feature 1, add the feature value corresponding to the time difference 2 to the multi-dimensional event feature 2,..., and add the feature value corresponding to the time difference 100 to the multi-dimensional event feature 100 to perform fusion processing on the combined event feature and the time feature, and obtain event information. If the combined event feature is represented as a matrix Y, and the time feature is represented as a matrix W, the electronic device can perform fusion processing on the combined event feature and the time feature to obtain event information, which is represented as a matrix I. By adding the time feature to the combined event feature to obtain the event information, the feature that the time intervals of the abnormal events are not equal can be better expressed, and the representation of the time information can be enhanced.

[0019] 5305、 Determine the number of events of the plurality of abnormal events. Since there are L abnormal events, the electronic device can determine that the number of events of the plurality of abnormal events is L. For example, if there are 100 abnormal events, the electronic device can determine that the number of events is 100.

[0020] 5306、 According to the number of events, the plurality of initial performance information is processed by feature mapping to obtain a plurality of target performance information. In an optional embodiment, for any one initial performance information, according to the number of events, the initial performance information is processed by feature mapping to obtain the target performance information corresponding to the initial performance information in the following manner: determining a preset dimension D; according to the number of events L and the preset dimension D, the initial performance information is processed by feature extraction and mapping to obtain the target performance information. Optionally, the electronic device can process the initial performance information by D times of feature extraction and mapping according to the preset dimension D to obtain D intermediate performance features, and then splice the D intermediate performance features to obtain the target performance information. For any one feature extraction and mapping, the electronic device can process the initial performance information by feature extraction through a convolution layer and perform marginal 0 to obtain an initial performance feature; the electronic device can process the initial performance feature by mapping through a linear layer to obtain an intermediate performance feature. Optionally, a matrix P of X*L can be preset in the linear layer. Optionally, the initial performance information can be represented by a matrix J of 1*X; the initial performance feature can be represented by a matrix J of 1*X; and the intermediate performance feature can be represented by a matrix J of 1*L. For example, if there are two initial performance information, initial performance information 1 as shown in Table 1 and initial performance information 2 as shown in Table 2, since the initial performance information 1 includes 72 CPU utilization rates, the initial performance information 1 can be represented as L=[0.82, 0.80, 0.95, 0.75,...,0.80,0.70]; since the initial performance information 2 includes 36 memory utilization rates, the initial performance information 2 can be represented as L=[0.85, 0.89, 0.70, 0.60,…, 0.86,0.88]. If the preset dimension D is 512, for the initial performance information 1, the electronic device can process the initial performance information 1 by 512 times of feature extraction and mapping to obtain 512 intermediate performance features corresponding to the initial performance information 1. For example, in the first feature extraction and mapping, the electronic device can process L=[0.82, 0.80,0.95, 0.75,...,0.80, 0.70] by feature extraction through a convolution layer and perform marginal 0 to obtain an initial performance feature 1, assuming that the initial performance feature 1 can be represented as a matrix J of 1*72=[0.52,0.49,0.61,0.75,…, 0,0].The linear layer can be preset with a 72*100 matrix P1, and the electronic device can perform mapping processing on the initial performance feature 1 through the matrix P1 in the linear layer to obtain an intermediate performance feature 1. It is assumed that the intermediate performance feature 1 can be represented as a 1*100 matrix = [0.05, 0.21, 0.13, 0.56,..., 0.30, 0.41]. It is assumed that among the 512 intermediate performance features corresponding to the initial performance information 1, the intermediate performance feature 1 can be represented as a 1*100 matrix = [0.05, 0.21, 0.13, 0.56,..., 0.30, 0.41], the intermediate performance feature 2 can be represented as a 1*100 matrix = [0.03, 0.01, 0.25, 0.13,..., 0.79, 0.36], and the intermediate performance feature 512 can be represented as a 1*100 matrix = [0.01, 0.19, 0.34, 0.76,..., 0.45, 0.20]. The 512 intermediate performance features can be spliced to obtain the target performance information 1. It is assumed that the target performance information 1 can be as shown in the matrix 0. Similarly, the electronic device can perform 512 times of feature extraction processing and mapping processing on the initial performance information 2 to obtain 512 intermediate performance features corresponding to the initial performance information 2, and then perform splicing processing on the 512 intermediate performance features to obtain the target performance information 2. It is assumed that the target performance information 2 can be as shown in the matrix B2. In the technical solution of the present disclosure, according to the number of events L and the preset dimension D, the initial performance information is subjected to feature extraction processing and mapping processing to obtain the target performance information, so that the plurality of target performance information and the event information are aligned in time, and the accuracy of fault prediction is improved.

[0021] S307, The plurality of target performance information is spliced to obtain spliced performance information. Since each target performance information is obtained by transforming the initial performance information of the target device, the plurality of target performance information can represent the characteristics of the target device under a plurality of performance indicators, and thus the spliced performance information obtained by splicing the plurality of target performance information can represent the overall performance characteristics of the target device. If there are n target performance information, each target performance information can be represented by a feature matrix B of D*L, and the n target performance information can be spliced to obtain spliced performance information, which can be represented by a matrix Z of (n*D)*L. For example, if there are two target performance information, the target performance information 1 is shown in matrix B1, and the target performance information 2 is shown in matrix B2, the electronic device can splice the two target performance information to obtain spliced performance information, which can be represented as a matrix Z of 1024*100 as follows: The layer 1, the sub-encoding layer 2, the attention layer 1, and the attention layer 2o The attention layer can be fused by using a cross attention mechanism. In the attention layer 1 in the i-th encoding layer, the following formula (1) can be preset, and in the attention layer 2, the following formula (2) can be preset. Formula (1) and formula (2) can be used to calculate cross attention:

[0022] . T

[0023] CrossAtt(Q^ Ki) = Softmax(V^ TQ^ Ki) Formula 1

[0024] . T

[0025] CrossAtt(Qi, Ai) = Softmax(Qi, Ai) Formula (2) Wherein, Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and Qi, Ai, and QiWhen i is equal to N, the electronic device can perform fusion processing on the NT performance feature and the N-1 event feature through the Nth encoding layer to obtain an Nth performance feature and an Nth event feature, and can determine the Nth performance feature as a target performance feature and determine the Nth event feature as a target event feature. Optionally, the fault prediction result can be determined according to the target performance feature and the target event feature in the following manner: performing splicing processing on the target performance feature and the target event feature to obtain a target feature; performing enhancement processing on the target feature to obtain a target enhanced feature; and determining the fault prediction result according to the target enhanced feature. Optionally, the target performance feature can be represented by a matrix Oz; and the target event feature can be represented by a matrix OA. The electronic device can perform splicing processing on the matrix Oz and the matrix OA through a fusion layer in the target model to obtain the target feature, and the target feature can be represented by a matrix OP. Optionally, the electronic device can perform enhancement processing on the target feature through a first fully connected layer and an activation layer in the target model to obtain the target enhanced feature. The target enhanced feature can be represented by a one-dimensional matrix O, and the O can be calculated according to the following formula (3). F : o F = TanhCOp)]^ + b F Formula (3) wherein, Tanh(-) represents an activation function; represents a coefficient vector; and b represents a bias vector. Optionally, after obtaining the target enhanced feature, the electronic device can perform processing on the target enhanced feature through a second fully connected layer in the target model to obtain a classification vector, and the classification vector can be used to represent the fault prediction result. For example, if the spliced performance information is represented by a matrix Z and the event information is represented by a matrix I, the electronic device can perform processing on the spliced performance information and the event information through an encoder, a dense layer, a first normalization layer, a feedforward layer and a second normalization layer in the target model to obtain a target performance feature and a target event feature, and it is assumed that the target performance feature is a 1*1000 matrix. [0.32, 0.58, 0.09, 0.14, …, 0.63], and the target event feature is a 1*1000 matrix. [0.12, 0.69, 0.48, 0.07, …, 0.24], the electronic device can perform splicing processing on the target performance feature and the target event feature to obtain a target feature. It is assumed that the target feature can be represented by the following 2*1000 matrix OP:

[0026] _F0.32, 0.58, 0.09, 0.14,...,0.63

[0027] Up=The electronic device can further enhance the target feature by using formula (3), and obtain a target enhanced feature. Assuming that the target enhanced feature can be represented as F = Q22, 0.27, 0.05, 0.62,..., 0.91. The electronic device can process the target enhanced feature by using the second full connection layer in the target model, and obtain a classification vector (0.8, 0.2). If the future preset time length of the target model is 48 hours in the future, it can be determined that the failure probability of the target device in the next 48 hours is 0.8, and the normal probability is 0.2. Since the target model uses an encoder based on an attention mechanism, better fusion processing of the spliced performance information and event information is achieved, and for elements far apart, the influence of the forgetting phenomenon of the recurrent neural network is avoided, so that the accuracy of failure prediction is improved by processing the spliced performance information and event information by using the target model. It should be noted that during the training of the model, a plurality of sample data and a sample label corresponding to each sample data can be determined. After processing each sample data by using the model to obtain a failure prediction result corresponding to each sample data, the loss value (Loss) of the model can be calculated according to the failure prediction result corresponding to each sample data and the sample label, and the parameters are updated by using the gradient descent method through back propagation, so as to train the target model. In the embodiment of the present disclosure, the electronic device can obtain a plurality of initial performance information corresponding to a plurality of performance indicators of the target device, and determine the event type and occurrence time of each abnormal event. For any one abnormal event, the electronic device can determine the multi-dimensional event feature of the abnormal event according to the event type of the abnormal event. The electronic device can generate event information according to the occurrence time of each abnormal event and the multi-dimensional event feature of each abnormal event. The electronic device can determine the number of events of a plurality of abnormal events, and perform feature mapping processing on a plurality of initial performance information according to the number of events, to obtain a plurality of target performance information, and then perform splicing processing on the plurality of target performance information to obtain spliced performance information. The electronic device can process the spliced performance information and the event information to obtain a failure prediction result.Compared with the prior art in which fault prediction is performed only according to a feature sequence obtained based on an exception log, more comprehensive information is expressed in the present disclosure, because the electronic device can perform fault prediction based on event information of multiple exception events and multiple initial performance information. Moreover, time information that multiple exception events have different time intervals is better expressed by adding time features in the event information, so that the accuracy of fault prediction on the target device is comprehensively improved. If the target device in the embodiments of FIG. 2 or FIG. 3 is a cloud server, the present disclosure further provides a fault prediction method. FIG. 6 is a flowchart of another fault prediction method provided by an exemplary embodiment of the present disclosure. Referring to FIG. 6, the method comprises the following steps.

[0028] S601, Obtain a plurality of initial performance information corresponding to a plurality of performance indicators of a target device. The target device can be a cloud server that needs to be fault predicted. The target device can have a plurality of performance indicators. Each performance indicator has corresponding initial performance information. For any one performance indicator, the initial performance information can include a plurality of performance data of the target device in a historical period under the performance indicator. For example, if the target device is cloud server 1, the current time is 2024 / 4 / 25 14:00, and the preset time length is 3 days, the historical period can be 2024 / 4 / 22 14:00 to 2024 / 4 / 25 14:00. If the performance indicator is CPU utilization, the corresponding initial performance information can include a plurality of CPU utilization of cloud server 1 between 2024 / 4 / 22 14:00 and 2024 / 4 / 25 14:00. Optionally, the electronic device or the data collection device can determine a plurality of performance indicators of the target device. For any one performance indicator, the electronic device or the data collection device can collect a plurality of performance data of the target device under the performance indicator according to the collection period corresponding to the performance indicator, and store the plurality of performance data in a preset storage space. The preset storage space can be in the electronic device, in the data collection device, or in other storage devices. The electronic device can determine a plurality of performance indicators of the target device, and then obtain a plurality of initial performance information corresponding to a plurality of performance indicators of the target device in a historical period in the preset storage space. For example, if the target device is cloud server 1, if there are 2 performance indicators, CPU utilization and memory utilization, and the current time is 2024 / 4 / 25 14:00, the historical period is 2024 / 4 / 22 14:00 to 2024 / 4 / 25 14:00, the electronic device can obtain initial performance information 1 corresponding to the CPU utilization of cloud server 1 and initial performance information 2 corresponding to the memory utilization of cloud server 1 in the preset storage space. Assuming that the initial performance information 1 can be as shown in Table 1 above, and the initial performance information 2 can be as shown in Table 2 above.

[0029] 5602、 obtain event information of a plurality of abnormal events of a target device. The target device can be in a running process, and a plurality of abnormal events can occur. For any one abnormal event, the abnormal event has a corresponding occurrence time. The occurrence times of the plurality of abnormal events can be located in a historical period. In an optional embodiment, the event information of the plurality of abnormal events of the target device is obtained: the event type and the occurrence time of each abnormal event are determined; for any one abnormal event, the multi-dimensional event feature of the abnormal event is determined according to the event type of the abnormal event; and the event information is generated according to the occurrence time of each abnormal event and the multi-dimensional event feature of each abnormal event, and the multi-dimensional event feature of each abnormal event is included in the event information. The event information can include the feature of each abnormal event in d event dimensions. Optionally, the event information can be represented as a matrix A of d*L. For example, if the target device is a cloud server 1, the cloud server 1 has 100 abnormal events, which are abnormal event 1, abnormal event 2, abnormal event 3, … …, abnormal event 100, respectively. The event type and the occurrence time of the 100 abnormal events can be as shown in Table 3. For the 100 abnormal events, the electronic device can determine the multi-dimensional event feature of each abnormal event according to the event type of each abnormal event. Assuming that the electronic device can determine the multi-dimensional event features of the 100 abnormal events as shown in the above-mentioned matrix G1, matrix G2, matrix G3, … …, matrix G100, respectively, the electronic device can generate the event information according to the occurrence time of each abnormal event shown in Table 3 and the 100 multi-dimensional event features, as shown in matrix I.

[0030] 5603、 The failure prediction result of the target device is determined according to the plurality of initial performance information and the event information. In an optional embodiment, the failure prediction result of the target device can be determined according to the plurality of initial performance information and the event information in the following manner: the number of events of the plurality of abnormal events is determined; the plurality of initial performance information is subjected to feature mapping processing according to the number of events, to obtain a plurality of target performance information, the number of performance features included in the target performance information being the same as the number of events; and the failure prediction result O is determined according to the plurality of target performance information and the event information. Optionally, the failure prediction result can include the failure probability of the target device and the normal probability of the target device. Optionally, the failure prediction result can be represented by a classification vector (C1, C2), where C1 can be used to represent the normal probability of the target device, and C2 can be used to represent the failure probability of the target device. For example, if there are 100 abnormal events, the number of events is 100; if there are 2 initial performance information, which are initial performance information 1 as shown in Table 1 and initial performance information 2 as shown in Table 2, since the initial performance information 1 includes 72 CPU utilization rates, the initial performance information 1 can be represented as L=[0.82, 0.80, 0.95, 0.75, …, 0.80, 0.70]; and since the initial performance information 2 includes 36 memory utilization rates, the initial performance information 2 can be represented as &=[0.85, 0.89, 0.70, 0.60, …, 0.86, 0.88]. The electronic device can perform feature mapping processing on the initial performance information 1 to obtain target performance information 1, which can be as shown in matrix B1; and perform feature mapping processing on the initial performance information 2 to obtain target performance information 2, which can be as shown in matrix B2. If the event information is as shown in the above matrix I, the electronic device can process the target performance information 1, the target performance information 2, and the event information to determine the failure prediction result of the target device. Assuming that the failure prediction result of the target device is (0.8, 0.2), the failure probability of the target device is 0.8, and the normal probability is 0.2. In the embodiments of the present disclosure, the electronic device can obtain a plurality of initial performance information corresponding to a plurality of performance indicators of the target device, and obtain event information of a plurality of abnormal events of the target device, and then determine the failure prediction result of the target device according to the plurality of initial performance information and the event information.Compared with the prior art in which only a feature sequence obtained based on an exception log is used for fault prediction, the information expressed is more comprehensive, and thus the accuracy of fault prediction on the target device is improved. Next, based on any of the above embodiments, the fault prediction method is further described in combination with FIG. 7. FIG. 7 is a process schematic diagram of a fault prediction method provided in an exemplary embodiment of the present disclosure. Referring to FIG. 7, the electronic device can obtain a plurality of abnormal events of the target device. Since each abnormal event has a corresponding event type and occurrence time, the electronic device can determine a plurality of event types of the plurality of abnormal events and a plurality of occurrence times of the plurality of abnormal events. The preset feature information can be preset in the dynamic embedding layer, and the preset feature information includes a plurality of event types and event features corresponding to each event type. The electronic device can process the event type of each abnormal event through the dynamic embedding layer to obtain a multi-dimensional event feature of each abnormal event. For example, if there are 100 abnormal events, the multi-dimensional event features of the 100 abnormal events can be obtained, and the multi-dimensional event feature of each abnormal event can be as shown in matrix G1, matrix G2, matrix G, …, matrix Gm. After determining the multi-dimensional event feature of each abnormal event, the multi-dimensional event features of the plurality of abnormal events can be combined and processed in the order from the latest occurrence time to the earliest occurrence time to obtain a combined event feature. For example, the combined event feature can be as shown in matrix G. The electronic device can also process the plurality of occurrence times of the plurality of abnormal events through the relative time embedding layer to obtain a time feature. The relative time embedding layer can be preset with a dynamic embedding function. Specifically, the electronic device can determine the latest occurrence time from the plurality of occurrence times of the plurality of abnormal events, and determine a time difference between each occurrence time and the latest occurrence time to obtain a plurality of time differences. The electronic device can determine at least one time granularity, and perform down-round processing on each time difference according to the at least one time granularity to obtain at least one integer result of each time difference, and then the relative time embedding layer can be used to map and process the at least one integer result of each time difference to obtain at least one initial feature value corresponding to each time difference. The electronic device can perform summation processing on the at least one initial feature value corresponding to each time difference to obtain a feature value corresponding to each time difference, and then the feature values of the plurality of time differences can be combined and processed to obtain the time feature. Since the time feature is determined according to the initial feature value corresponding to each time difference at different time granularities, the time feature strengthens the feature expression of different time differences.For example, if there are 100 abnormal events, the occurrence time of the 100 abnormal events is as shown in Table 3, the electronic device can determine 100 time differences, and then determine the feature values corresponding to the 100 time differences as shown in Table 7. The electronic device can perform combination processing on the feature values corresponding to the 100 time differences to obtain a time feature, such as matrix %. After the electronic device determines the combined event feature and the time feature, the electronic device can perform fusion processing on the combined event feature and the time feature to obtain event information, such as matrix I. The electronic device can obtain a plurality of initial performance information of the target device under a plurality of performance indicators. For example, the electronic device can obtain initial performance information 1 of the target device under CPU utilization, and initial performance information 2 under memory utilization. The electronic device can determine the event quantity L of the plurality of abnormal events, and perform feature mapping processing on each initial performance information through a convolution layer and a linear layer to obtain corresponding target performance information. Specifically, the electronic device can perform D times of feature extraction processing and mapping processing on the initial performance information through the convolution layer and the linear layer according to the event quantity L and the preset dimension D, to obtain D intermediate performance features, and then perform splicing processing on the D intermediate performance features to obtain the target performance information. Optionally, a matrix Po of X*L can be preset in the linear layer. For example, if there are 100 abnormal events, the event quantity is 100o If there are 2 initial performance information, which are initial performance information 1 as shown in Table 1 and initial performance information 2 as shown in Table 2. According to Table 1, the initial performance information 1 can be represented as L=[0.82, 0.80, 0.95, 0.75, …, 0.80, 0.70]; according to Table 2, the initial performance information 2 can be represented as £=[0.85, 0.89, 0.70, 0.60, …, 0.86, 0.88]; if the preset dimension D is 512, the electronic device can perform 512 times of feature extraction processing and mapping processing on the initial performance information 1 to obtain 512 intermediate performance features, and then can perform splicing processing on the 512 intermediate performance features to obtain target performance information 1, assuming that the target performance information 1 can be as shown in matrix B1; the electronic device can perform 512 times of feature extraction processing and mapping processing on the initial performance information 2 to obtain 512 intermediate performance features, and then can perform splicing processing on the 512 intermediate performance features to obtain target performance information 2, assuming that the target performance information 2 can be as shown in matrix B2. The electronic device can perform splicing processing on a plurality of target performance information to obtain spliced performance information.For example, if the target performance information 1 is as shown in matrix B1 and the target performance information 2 is as shown in matrix B2, the electronic device can perform splicing processing on the two pieces of target performance information to obtain spliced performance information, which can be matrix Z. After the electronic device determines the spliced performance information and the event information, the electronic device can input the spliced performance information and the event information into the target model, and process the spliced performance information and the event information through the target model to obtain a fault prediction result. It should be noted that the process of processing the spliced performance information and the event information through the target model to obtain the fault prediction result can be referred to FIG. 5, and will not be described here in detail. In the technical solution of the present disclosure, compared with the fault prediction model based on system abnormal logs, the target model can better extract and represent the features of the sample. Through the target model, the information fusion and enhancement are performed on the two different forms of data, i.e., the spliced performance information based on performance and the event information based on abnormal events, so that the target model can more accurately evaluate the overall state of the target device according to the spliced performance information and the event information, and thus make a more accurate fault prediction result. Compared with the model based on system abnormal logs (HitAnomaly), the advantages of the target model are as follows: in fault prediction, the event information based on abnormal events (abnormal logs) and the spliced performance information based on performance are added, and the cross-attention mechanism is used to fuse the information of different modalities, so that the state of the system can be more comprehensively evaluated, and the accuracy of fault prediction is improved. Compared with the Long Short-Term Memory (LSTM) model, the advantages of the target model are as follows: (1) the dynamic embedding method is used to vectorize multiple abnormal events, and the vectorization is automatically updated during the training process, so that the distance and similarity between different abnormal events can be better represented, and the loss caused by the feature extraction engineering on the abnormal event log information can be reduced. (2) the encoder based on the attention mechanism is used, so that the elements far apart will not be affected by the forgetting phenomenon of the recurrent neural network, and at the same time, the training time of the model can be reduced, and the features can be better extracted. In the embodiment of the present disclosure, the electronic device can obtain multiple abnormal events of a target device, and determine multiple event types and multiple occurrence times of the multiple abnormal events. The electronic device can determine the multi-dimensional event features of each abnormal event according to the event type of each abnormal event, and further determine the combined event features. The electronic device can generate the time features according to the occurrence times of each abnormal event. The electronic device can determine the event information according to the combined event features and the time features.The electronic device can obtain a plurality of initial performance information corresponding to a plurality of performance indicators of a target device, and perform feature mapping processing on the plurality of initial performance information to obtain a plurality of target performance information, and then perform splicing processing on the plurality of target performance information to obtain spliced performance information. The electronic device can process the spliced performance information and event information through a target model to obtain a fault prediction result. Since the electronic device can perform fault prediction based on event information of a plurality of abnormal events and a plurality of initial performance information, compared with the prior art that only performs fault prediction based on a feature sequence obtained based on abnormal logs, more comprehensive information is expressed; and the time information of the unequal time intervals of the plurality of abnormal events is better expressed by adding the time feature in the event information, so the accuracy of fault prediction on the target device is improved comprehensively. FIG. 8 is a structural schematic diagram of a fault prediction device provided by an example embodiment of the present disclosure. Please refer to FIG. 8, the fault prediction device 10 can include a first acquisition module 11, a second acquisition module 12, and a determination module 13, wherein the first acquisition module 11 is configured to obtain a plurality of initial performance information corresponding to a plurality of performance indicators of a target device, the initial performance information including a plurality of performance data of the target device in a historical period and under the performance indicators; the second acquisition module 12 is configured to obtain event information of a plurality of abnormal events of the target device, the occurrence time of the plurality of abnormal events being located in the historical period; and the determination module 13 is configured to determine a fault prediction result of the target device according to the plurality of initial performance information and the event information. The fault prediction device provided by the embodiments of the present disclosure can execute the technical solutions shown in the method embodiments, and the implementation principles and beneficial effects are similar, which will not be repeated here. In a possible implementation manner, the determination module 13 is specifically configured to: determine the number of events of the plurality of abnormal events; perform feature mapping processing on the plurality of initial performance information according to the number of events to obtain a plurality of target performance information, the number of performance features included in the target performance information being the same as the number of events; and determine the fault prediction result according to the plurality of target performance information and the event information. In a possible implementation manner, for any one initial performance information, the determination module 13 is specifically configured to: determine a preset dimension; and perform feature extraction processing and mapping processing on the initial performance information according to the number of events and the preset dimension to obtain the target performance information.In a possible implementation, the determining module 13 is specifically configured to: perform splicing processing on the plurality of target performance information to obtain spliced performance information; and perform processing on the spliced performance information and the event information to obtain the fault prediction result. In a possible implementation, the determining module 13 is specifically configured to: perform N times of fusion processing on the spliced performance information and the event information to obtain a target performance feature and a target event feature, where N is an integer greater than or equal to 1; and determine the fault prediction result according to the target performance feature and the target event feature. In a possible implementation, the determining module 13 is specifically configured to: perform splicing processing on the target performance feature and the target event feature to obtain a target feature; perform enhancement processing on the target feature to obtain a target enhanced feature; and determine the fault prediction result according to the target enhanced feature. In a possible implementation, the determining module 13 is specifically configured to: perform fusion processing on the spliced performance information and the event information to obtain a first performance feature and a first event feature; and perform fusion processing on an ith performance feature and an ith event feature to obtain an i-th performance feature and an i-th event feature, where i is 2, 3, …, N in turn, the (N-1)th performance feature is determined as the target performance feature, and the (N-1)th event feature is determined as the target event feature. In a possible implementation, the second obtaining module 12 is specifically configured to: determine an event type and an occurrence time of each abnormal event; determine, for any one abnormal event, a multi-dimensional event feature of the abnormal event according to the event type of the abnormal event; and generate the event information according to the occurrence time of each abnormal event and the multi-dimensional event feature of each abnormal event, where the multi-dimensional event feature of each abnormal event is included in the event information. In a possible implementation, the second obtaining module 12 is specifically configured to: obtain preset feature information, where the preset feature information includes a plurality of event types and event features corresponding to the event types; and determine the multi-dimensional event feature of the abnormal event according to the event type of the abnormal event and the preset feature information. In a possible implementation, the second obtaining module 12 is specifically configured to: perform combination processing on the multi-dimensional event features of the plurality of abnormal events in a sequence from late to early according to the occurrence time to obtain a combined event feature; generate a time feature according to the occurrence time of each abnormal event; and perform fusion processing on the combined event feature and the time feature to obtain the event information.In a possible implementation, the second obtaining module 12 is specifically configured to: determine a latest occurrence time from the multiple occurrence times of the multiple abnormal events; determine a time difference between each occurrence time and the latest occurrence time to obtain multiple time differences; determine a feature value corresponding to each time difference, and combine the feature values of the multiple time differences to obtain the time feature. The fault prediction apparatus provided by the embodiments of the present disclosure can execute the technical solutions shown in the above method embodiments, and the implementation principles and beneficial effects are similar, which will not be repeated here. FIG. 9 is a structural schematic diagram of an electronic device provided by an exemplary embodiment of the present disclosure. Please refer to FIG. 9, the electronic device 20 can include a processor 21 and a memory 22. Exemplarily, the processor 21, the memory 22, and each part are connected with each other through a bus 23. The memory 22 stores computer execution instructions; the processor 21 executes the computer execution instructions stored in the memory 22, so that the processor 21 executes the method shown in the above method embodiments. Accordingly, the embodiments of the present disclosure provide a computer readable storage medium, the computer readable storage medium stores computer execution instructions, and the computer execution instructions are used to execute the method provided by the above method embodiments when executed by the processor. Accordingly, the embodiments of the present disclosure can also provide a computer program product, including a computer program, which can execute the method shown in the above method embodiments when executed by the processor. Those skilled in the art should understand that the embodiments of the present disclosure can be provided as a method, a system, or a computer program product. Therefore, the present disclosure can be implemented in the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present disclosure can be implemented in 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 code. The present disclosure is described with reference to flowcharts and / or block diagrams of the method, device (system), and computer program product according to the embodiments of the present disclosure. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of the flows 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, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flow diagram block or blocks or in the block diagram block or blocks. These computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer readable memory produce an article of manufacture including instructions which implement the function specified in the flow diagram block or blocks or in the block diagram block or blocks. The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flow diagram block or blocks or in the block diagram block or blocks. In one typical configuration, the computer device includes one or more processors (CPU's), input / output interfaces, network interfaces, and memory. The memory can include non-persistent memory and / or persistent memory, which can be embodied in computer-readable media, random access memory (RAM), and / or read-only memory (ROM), flash memory, electrical medium such as wires, electrical connections, electrical pathways, optical medium, and / or other fixed (hard-coded) or floppy storage medium. oMemory is an example of a computer-readable medium. Computer-readable media includes permanent and non-permanent, removable and non-removable media implemented by any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (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 technology, compact disc read-only memory (CD-ROM), digital versatile discs (DVDs) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices or any other non-transmission medium that can be used to store information accessible to a computing device. According to the definition herein, computer-readable media does not include transitory media, such as modulated data signals and carrier waves. It is also noted that the terms "comprising", "including", or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can also include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without limitation, an element preceded by "comprises a" does not, without more constraints, foreclose the existence of additional identical elements in the process, method, article, or apparatus. The embodiments described above are only examples of the disclosure and are not intended to limit the disclosure. The disclosure can have various changes and modifications for those skilled in the art. Any modification, equivalent replacement, improvement, etc. within the spirit and principle of the disclosure should be included in the scope of the claims of the disclosure.

Claims

CLAIM 1. A failure prediction method in which, The method comprises the following steps: obtaining a plurality of initial performance information corresponding to a plurality of performance indexes of a target device, wherein the initial performance information comprises a plurality of performance data of the target device under the performance indexes in a historical period; obtaining event information of a plurality of abnormal events of the target device, wherein the occurrence time of the plurality of abnormal events is in the historical period; and determining a fault prediction result of the target device according to the plurality of initial performance information and the event information.

2. The method of claim 1, wherein, According to the plurality of initial performance information and the event information, the fault prediction result of the target device is determined, comprising: determining the number of events of the plurality of abnormal events; performing feature mapping processing on the plurality of initial performance information according to the number of events to obtain a plurality of target performance information, wherein the number of performance features included in the target performance information is the same as the number of events; and determining the fault prediction result according to the plurality of target performance information and the event information.

3. The method of claim 2, wherein, For any one initial performance information, the initial performance information is processed by feature mapping according to the number of events to obtain target performance information corresponding to the initial performance information, comprising: determining a preset dimension; and performing feature extraction processing and mapping processing on the initial performance information according to the number of events and the preset dimension to obtain the target performance information.

4. The method according to claim 2 or 3, wherein, According to the plurality of target performance information and the event information, the fault prediction result is determined, comprising: performing splicing processing on the plurality of target performance information to obtain spliced performance information; and processing the spliced performance information and the event information to obtain the fault prediction result.

5. The method according to claim 4, wherein, The spliced performance information and the event information are processed to obtain the fault prediction result, comprising: performing N times of fusion processing on the spliced performance information and the event information to obtain target performance features and target event features, wherein N is an integer greater than or equal to 1; and determining the fault prediction result according to the target performance features and the target event features.

6. The method according to claim 5, wherein, According to the target performance features and the target event features, the fault prediction result is determined, comprising: performing splicing processing on the target performance features and the target event features to obtain target features; performing enhancement processing on the target features to obtain target enhanced features; and determining the fault prediction result according to the target enhanced features.

7. The method according to claim 5 or 6, wherein, The spliced performance information and the event information are fused N times to obtain target performance characteristics and target event characteristics, including: the spliced performance information and the event information are fused to obtain the first performance characteristics and the first event characteristics; the ith performance characteristics and the ith event characteristics are fused to obtain the i performance characteristics and the i event characteristics; wherein, the i is 2, 3, …, N in turn, and the N performance characteristics are determined as the target performance characteristics, and the N event characteristics are determined as the target event characteristics.

8. The method according to any one of claims 1 to 7, wherein, Obtain event information of a plurality of abnormal events of the target device, including: determining the event type and the occurrence time of each abnormal event; for any one abnormal event, determining the multi-dimensional event characteristics of the abnormal event according to the event type of the abnormal event; generating the event information according to the occurrence time of each abnormal event and the multi-dimensional event characteristics of each abnormal event, wherein the multi-dimensional event characteristics of each abnormal event are included in the event information. 24 According to the event type of the abnormal event, the multi-dimensional event characteristics of the abnormal event are determined, including: obtaining preset feature information, wherein the preset feature information includes a plurality of event types and event characteristics corresponding to each event type; according to the event type of the abnormal event and the preset feature information, the multi-dimensional event characteristics of the abnormal event are determined.

9. The method according to claim 8, wherein, According to the occurrence time of each abnormal event and the multi-dimensional event characteristics of each abnormal event, the event information is generated, including: combining the multi-dimensional event characteristics of the plurality of abnormal events in the order from late to early according to the occurrence time to obtain combined event characteristics; generating time characteristics according to the occurrence time of each abnormal event; and fusing the combined event characteristics and the time characteristics to obtain the event information.

10. The method according to claim 8 or 9, wherein, According to the occurrence time of each abnormal event, the time characteristics are generated, including: determining the latest occurrence time among the plurality of occurrence times of the plurality of abnormal events; determining the time difference between each occurrence time and the latest occurrence time to obtain a plurality of time differences; determining the feature value corresponding to each time difference, and combining the feature values of the plurality of time differences to obtain the time characteristics.

11. The method according to claim 10, wherein, including:

12. A failure prediction method in which, Obtain a plurality of initial performance information corresponding to a plurality of performance indicators of a target device, the initial performance information including a plurality of performance data of the target device under the performance indicators in a historical period, and the target device being a cloud server; Obtain event information of a plurality of abnormal events of the target device, the occurrence time of the plurality of abnormal events being located in the historical period; and determine a fault prediction result of the target device according to the plurality of initial performance information and the event information. ​ 13. The method according to claim 12, wherein, According to the plurality of initial performance information and the event information, a failure prediction result of the target device is determined, including: determining an event quantity of the plurality of abnormal events; according to the event quantity, performing feature mapping processing on the plurality of initial performance information to obtain a plurality of target performance information, the target performance information including the same number of performance features as the event quantity; and according to the plurality of target performance information and the event information, determining the failure prediction result.

14. The method according to claim 12 or 13, wherein, The event information of the plurality of abnormal events of the target device is obtained, including: determining an event type and an occurrence time of each abnormal event; and for any one abnormal event, determining a multi-dimensional event feature of the abnormal event according to the event type of the abnormal event. According to the occurrence time of each abnormal event and the multi-dimensional event feature of each abnormal event, the event information is generated, and the multi-dimensional event feature of each abnormal event is included in the event information.

15. An electronic device, wherein, Comprise: At least one processor; And the memory that communication connection with the at least one processor;Wherein, the memory has the instruction that can be executed by the at least one processor, the instruction is executed by the at least one processor, to make the electronic equipment execute the method of claim 1-11, or any one of claim 12-14.

16. A computer readable storage medium, wherein, The computer readable storage medium stores computer execution instructions, when the processor executes the computer execution instructions, realize as claim 1-11, or any one of claim 12-14 described method.

17. A computer program product comprising a computer program, wherein, The computer program is executed by the processor to realize the method as claimed in any one of claims 1-11, or claims 12-14. ​

Images