CN116339939A - Cloud service platform task scheduling method and system based on artificial intelligence - Google Patents

Abstract

The invention discloses a cloud service platform task scheduling method and system based on artificial intelligence, comprising the following steps: firstly, invoking a task classification model which is trained in advance, classifying cloud service tasks and obtaining task type information to be processed; determining undetermined task equipment information aiming at the task type information to be processed according to the task type information to be processed; then, determining the adaptation degree between the equipment information of the undetermined task and the cloud service task; finally, when the adaptation degree reaches a preset adaptation degree threshold value, the cloud service task is sent to task processing equipment corresponding to the information of the undetermined task equipment to be processed, so that the cloud service task can be quickly matched with the proper task processing equipment, the cloud service task and the task processing equipment can be guaranteed to have good adaptation degree, and the task scheduling efficiency of the cloud service platform is improved.

Description

Cloud service platform task scheduling method and system based on artificial intelligence

Technical Field

The invention relates to the field of artificial intelligence, in particular to a cloud service platform task scheduling method and system based on artificial intelligence.

Background

With the development of the internet, various cloud service platforms provide various services for users, and when a user needs to perform a certain operation, the user generally sends the needs to the cloud service platform in a form of initiating a cloud service task so that the cloud service platform responds. In the prior art, when the service is expanded or the number of users is increased, in order to relieve the pressure of task processing equipment, a load balancing strategy based on the performance of the equipment is generally adopted, but the same equipment is caused to process multiple tasks, so that the processing efficiency is low.

Disclosure of Invention

The invention aims to provide a cloud service platform task scheduling method and system based on artificial intelligence.

In a first aspect, an embodiment of the present invention provides a cloud service platform task scheduling method based on artificial intelligence, including:

responding to a task initiation request aiming at a cloud service platform, and analyzing the task initiation request to obtain a cloud service task corresponding to the task initiation request;

invoking a task classification model which is trained in advance to classify cloud service tasks to obtain task type information to be processed;

according to the task type information to be processed, determining the equipment information of the task to be determined aiming at the task type information to be processed;

Determining the adaptation degree between the equipment information of the undetermined task and the cloud service task;

and when the adaptation degree reaches a preset adaptation degree threshold, sending the cloud service task to task processing equipment corresponding to the information of the undetermined task equipment for processing.

In a second aspect, an embodiment of the present invention provides a server system, including a server, configured to perform the method of the first aspect.

Compared with the prior art, the invention has the beneficial effects that: by adopting the cloud service platform task scheduling method and system based on artificial intelligence, the cloud service tasks are classified by calling a task classification model which is trained in advance, so that task type information to be processed is obtained; determining undetermined task equipment information aiming at the task type information to be processed according to the task type information to be processed; then, determining the adaptation degree between the equipment information of the undetermined task and the cloud service task; finally, when the adaptation degree reaches a preset adaptation degree threshold value, the cloud service task is sent to task processing equipment corresponding to the information of the undetermined task equipment to be processed, so that the cloud service task can be quickly matched with the proper task processing equipment, the cloud service task and the task processing equipment can be guaranteed to have good adaptation degree, and the task scheduling efficiency of the cloud service platform is improved.

Drawings

In order to more clearly illustrate the technical solution of the embodiments of the present invention, the drawings that are required to be used in the embodiments will be briefly described. It is appreciated that the following drawings depict only certain embodiments of the invention and are therefore not to be considered limiting of its scope. Other relevant drawings may be made by those of ordinary skill in the art without undue burden from these drawings.

Fig. 1 is a schematic flow chart of steps of a task scheduling method of a cloud service platform based on artificial intelligence according to an embodiment of the present invention;

fig. 2 is a schematic block diagram of a server according to an embodiment of the present invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the embodiments of the present invention more clear, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

Referring to fig. 1 in combination, fig. 1 is a schematic step flow diagram of an artificial intelligence-based task scheduling method for a cloud service platform according to an embodiment of the present invention, and the detailed description of the artificial intelligence-based task scheduling method for the cloud service platform is provided below.

Step S101, responding to a task initiation request aiming at a cloud service platform, analyzing the task initiation request to obtain a cloud service task corresponding to the task initiation request.

And step S102, invoking a task classification model which is trained in advance, classifying the cloud service task, and obtaining the type information of the task to be processed.

Step S103, according to the task type information to be processed, determining the equipment information of the task to be determined aiming at the task type information to be processed.

Step S104, the adaptation degree between the equipment information of the undetermined task and the cloud service task is determined.

Step S105, when the adaptation degree reaches a preset adaptation degree threshold, the cloud service task is sent to task processing equipment corresponding to the undetermined task equipment information for processing.

In the embodiment of the invention, the cloud server platform can refer to a platform capable of providing online cloud service for users, including a shopping platform, a game platform, a live broadcast platform, a data storage platform and the like, is not limited herein, and because the number of users accessing the cloud service platform is large, cloud service tasks initiated by all users are processed on the same server device, so that excessive service load is easily caused, and corresponding task archiving records are not well counted. Therefore, a task classification model which is trained in advance can be called first to classify the cloud service task, and the type information of the task to be processed is obtained. In the embodiment of the invention, the task types to be processed can include, but are not limited to, task item categories, task priority categories, task emergency categories and the like. After the task type information to be processed of the cloud service task is determined, the information of the task equipment to be processed for processing the task type information to be processed can be further obtained, and under the condition that the adaptation degree between the information of the task equipment to be processed and the cloud service task is judged to reach the preset adaptation degree threshold value, namely, the task processing equipment corresponding to the information of the task equipment to be processed is reasonably processed by the cloud service task, the task processing equipment corresponding to the information of the task equipment to be processed is used as equipment for processing the corresponding cloud service task. By the design, the cloud service task can be quickly matched with proper task processing equipment, the task processing equipment and the cloud service task can be guaranteed to have higher adaptation degree, the task scheduling rate of the cloud service platform can be improved, and convenience is provided for follow-up operations such as archiving, sorting and feature extraction of the cloud service task.

Step 201, acquiring a plurality of training task sets and task classifications of preset training tasks in the plurality of training task sets, wherein the training task sets comprise two preset training tasks with close types.

In the embodiment of the invention, a plurality of training task sets can be acquired, and any training task set comprises two preset training tasks with similar types. Optionally, two tasks are randomly extracted from any task files contained in the task archiving database, and if the two tasks are of the same type, the two tasks are used as a training task set. The method and the device for acquiring the task type are not limited by the acquisition mode of the two task type approach, and are used for marking whether the two tasks are of the same type or not by an operator, if the two tasks are of the same type, the two tasks are acquired, and if the two tasks are of different types, the two tasks are acquired. Or, firstly, acquiring the task feature vector of each task according to the pixel information of each task, and then calculating the distance between the two tasks according to the task feature vectors of the two tasks. If the distance between the two tasks is smaller than the distance threshold, marking the two tasks as the same type of tasks, and if the distance between the two tasks is not smaller than the distance threshold, marking the two tasks as different types of tasks. The distance threshold may be obtained according to experience, historical experimental data, and the like, and the value of the distance threshold is not limited herein. It should be understood that, in an embodiment of the present invention, task features may include attribute features such as task name, semantic information of content, sponsor, executor, task priority, time of initiation, and so on.

For any training task set, task classification of each preset training task in the any training task set needs to be obtained. Any training task set should be classified by at least one task, and any task classification is used for determining a classification category to which any preset training task belongs. The embodiment of the invention does not limit the task attribution type content of any task classification.

Optionally, task classification of each preset training task in any training task set may be manually obtained, or task classification of each preset training task in any training task set may be obtained by using a classification model. The training task set can be utilized to train the initial model to obtain the classification model, or the classification model applied on line can be directly obtained. The training task set illustratively includes a plurality of tasks retrieved from a task archiving database. And obtaining the prediction result of each preset training task in any training task set by using the classification model. The prediction result of any one preset training task is the probability that any one preset training task belongs to each classification category, and if the probability that any one preset training task belongs to any classification category exceeds a classification threshold, the any classification category is used as a task classification of any one preset training task; if the probability that any one of the preset training tasks belongs to any of the classification categories does not exceed the classification threshold, the any classification category is not one task classification of any one of the preset training tasks. The classification threshold may be updated according to the actual situation, and optionally, the classification threshold is 0.5.

Step 202, task feature vectors of each preset training task in a plurality of training task sets are obtained according to an initial network structure.

In the embodiment of the invention, a plurality of training task sets are input into an initial network structure, and task feature vectors of each preset training task in the plurality of training task sets are acquired and output by the initial network structure. Optionally, each round of training, the plurality of training task sets are divided into a plurality of batches of input initial network structures. For example, for each round of training, the Z training task sets are respectively Z/Z batches, and each batch contains Z training task sets. That is, each time Z training task sets are input into the initial network structure, task feature vectors of each preset training task in the Z training task sets are acquired and output by the initial network structure, and the input is performed for Z/Z times. Wherein Z, Z and Z/Z are positive integers. In the embodiment of the invention, the initial network structure is required to be trained for multiple times to obtain the task classification model. The embodiment of the invention does not limit the structure, the size and the like of the initial network structure.

Step 203, obtaining a first classification vector of each preset training task in the plurality of training task sets according to the task feature vector of each preset training task in the plurality of training task sets, where the first classification vector of the preset training task is used for determining a task attribution type of the preset training task to be classified.

In the embodiment of the invention, the task feature vector of each preset training task in the plurality of training task sets can be input into the classification layer, and the classification layer acquires and outputs the first classification vector of each preset training task in the plurality of training task sets.

In the embodiment of the invention, a first classification vector of a preset training task is a matrix of 1 XZH, and any data in the matrix is a floating point number. The task attribution type of the preset training task to be classified (namely the prediction type of the preset training task) is determined through the first classification vector of the preset training task, and the task attribution type of the preset training task to be classified is at least one.

Step 204, obtaining second classification vectors of each preset training task in the training task sets according to task classifications of each preset training task in the training task sets, wherein the second classification vectors of the preset training tasks are used for determining task attribution types of marked preset training tasks.

In the implementation of the invention, for any one preset training task in a plurality of training task sets, a second classification vector of any one preset training task is obtained according to the task classification of the any one preset training task. The second classification vector of any preset training task is a matrix of 1×zh, and any data in the matrix is an integer (e.g. 0 or 1). The task attribution type of the marked preset training task (namely the task classification of the preset training task) is determined through the second classification vector of the preset training task, and the task attribution type of the marked preset training task is at least one.

Optionally, the task classification of the preset training task is at least one, and the obtaining the second classification vector of each preset training task in the plurality of training task sets according to the task classification of each preset training task in the plurality of training task sets includes: for any task classification of each preset training task in the training task sets, acquiring preset classification consistent with any task classification from the preset classifications, and acquiring task attribution type classification vectors of the preset classification consistent with any task classification as task attribution type classification vectors of any task classification; for any one preset training task in the plurality of training task sets, acquiring a second classification vector of any one preset training task according to a task attribution type classification vector of at least one task classification of any one preset training task.

In the embodiment of the invention, the task attribution type classification vector of each of a plurality of preset classifications can be obtained. And comparing whether each task class of any one preset training task in the plurality of training task sets is consistent with the plurality of preset classes or not to obtain a comparison result of each task class, and obtaining a second classification vector of any one preset training task through the comparison result of each task class.

Optionally, before acquiring the task attribution type classification vector of the preset classification consistent with the arbitrary task classification as the task attribution type classification vector of the arbitrary task classification, the method further includes: acquiring a classification aspect ratio group according to task classifications of preset training tasks in a plurality of training task sets, wherein the frequency of preset classifications corresponding to the transverse attribution of any data and preset classification corresponding to the longitudinal attribution of any data is determined by any data in the classification aspect ratio group, and one task classification is a preset classification; and acquiring task attribution type classification vectors of all preset classifications according to the preset orthogonal aspect ratio groups and the classification aspect ratio groups, wherein the order of the preset orthogonal aspect ratio groups exceeds the number of the preset classifications.

In the embodiment of the invention, the task classification of each preset training task in a plurality of training task sets can be subjected to statistical processing to obtain the classification aspect ratio array. The classifying aspect ratio array is a matrix of Z rows and Z columns, each transverse number column of the classifying aspect ratio array corresponds to each preset classification, each longitudinal number column of the classifying aspect ratio array corresponds to each preset classification, and Z is a positive integer. Wherein, the arbitrary preset classification is a task classification. That is, each preset training task in the plurality of training task sets is utilized to form a preset training task set, and the preset training task set totally comprises Z task classifications, and each task classification is a preset classification.

Optionally, any data in the classification aspect group may be a character or a positive integer, where the any data in the classification aspect group determines the number of times that the preset classification corresponding to the transverse attribution of any data is the same as the preset classification corresponding to the longitudinal attribution of any data. If any data in the classification aspect ratio array is a character, the preset classification corresponding to the transverse attribution of the any data is the same as the preset classification corresponding to the longitudinal attribution of the any data. If any data in the classification aspect ratio group is a positive integer, the preset classification corresponding to the transverse attribution of the any data is the same as the preset classification corresponding to the longitudinal attribution of the any data, and the same times of the classification are the positive integer.

Optionally, any data in the classification aspect group may be a probability value, where the probability value exceeds 0 and is less than or equal to 1, and at this time, the probability that the preset classification corresponding to the transverse attribution of any data is the same as the preset classification corresponding to the longitudinal attribution of any data is determined by any data in the classification aspect group.

In one possible implementation, the task classification of the preset training task is at least two, and the obtaining the classification aspect ratio array according to the task classification of each preset training task in the plurality of training task sets includes: acquiring an original aspect ratio array, wherein each transverse number row of the original aspect ratio array corresponds to each preset classification, and each longitudinal number row of the original aspect ratio array corresponds to each preset classification; for any one preset training task in a plurality of training task sets, setting a preset value at the intersection of a target transverse sequence and a target longitudinal sequence according to at least two task classifications of any one preset training task, wherein the target transverse sequence is a transverse sequence corresponding to a preset classification consistent with one task classification of any one preset training task in an original aspect ratio array, and the target longitudinal sequence is a longitudinal sequence corresponding to a preset classification consistent with other task classifications of any one preset training task except one task classification in the original aspect ratio array; and performing superposition operation on preset values at the intersections of each transverse array and each longitudinal array in the original aspect ratio array to obtain the classified aspect ratio array.

In the embodiment of the invention, an original aspect ratio array is obtained, and the original aspect ratio array is an array of Z rows and Z columns and can be displayed in a matrix form.

For any one of a plurality of training task sets, the original aspect ratio set is updated using at least two task classifications for the any one of the preset training tasks. Optionally, a preset value is set at the intersection of the target transverse number series and the target longitudinal number series according to at least two task classifications of any one preset training task, and the preset value may be 1. The target transverse number sequence is a transverse number sequence corresponding to a preset classification consistent with one task classification of any preset training task in the original aspect ratio array, and the target longitudinal number sequence is a longitudinal number sequence corresponding to a preset classification consistent with other task classifications of any preset training task except one task classification in the original aspect ratio array. In this way, the original aspect ratio array is updated with at least two task classifications for each of the preset training tasks in the plurality of training task sets. And performing superposition operation on preset values at the intersections of each transverse array and each longitudinal array in the original aspect ratio array to obtain the classified aspect ratio array. Optionally, in updating the original aspect ratio array using at least two task classifications of each preset training task, a superposition operation is performed on preset values at intersections of each lateral number array and each longitudinal number array in the original aspect ratio array.

For example, any training task set corresponds to task classification 1-3, the first transverse number row and the first longitudinal number row of the original aspect ratio array each correspond to a preset classification 1, the second transverse number row and the second longitudinal number row of the original aspect ratio array each correspond to a preset classification 2, the third transverse number row and the third longitudinal number row of the original aspect ratio array each correspond to a preset classification 3, and the preset classifications 1-3 are task classifications 1-3, respectively. Then for task class 1 of any one of the preset training tasks, the target transverse array is the first transverse array of the original longitudinal array, the target longitudinal array comprises the second longitudinal array and the third longitudinal array of the original longitudinal array, 1 is added at the intersection of the first transverse array and the second longitudinal array of the original longitudinal array (determining that the preset class 1 and the preset class 2 are the same once), and 1 is also added at the intersection of the first transverse array and the third longitudinal array of the original longitudinal array (determining that the preset class 1 and the preset class 3 are the same once). For task class 2 of any one preset training task, the target transverse array is the second transverse array of the original aspect array, the target longitudinal array comprises the first longitudinal array and the third longitudinal array of the original aspect array, 1 is added at the intersection of the second transverse array and the first longitudinal array of the original aspect array (the preset class 2 and the preset class 1 are determined to be the same once), and 1 is also added at the intersection of the second transverse array and the third longitudinal array of the original aspect array (the preset class 2 and the preset class 3 are determined to be the same once). For task class 3 of any one of the preset training tasks, the target transverse array is the third transverse array of the original aspect array, the target longitudinal array comprises the first longitudinal array and the second longitudinal array of the original aspect array, 1 is added at the intersection of the third transverse array and the first longitudinal array of the original aspect array (the preset class 3 and the preset class 1 are determined to be the same once in a class), and 1 is also added at the intersection of the third transverse array and the second longitudinal array of the original aspect array (the preset class 3 and the preset class 2 are determined to be the same once in a class).

Alternatively, the category number symbols may be updated according to at least two task classifications for each preset training task. If any training task set should be classified into task categories 1-3, the number of categories sign z1=z1+3 is updated, and it is determined that the total number of task categories is more than 3 after statistics on any preset training task.

In the embodiment of the invention, the original aspect ratio array is updated by utilizing at least two task classifications of each preset training task, and after the preset values at the junction of each transverse number array and each longitudinal number array in the original aspect ratio array are subjected to superposition operation, the updated original aspect ratio array is obtained. The updated original aspect ratio array can be used as a classification aspect ratio array, and at this time, any data in the classification aspect ratio array determines the same times of preset classification corresponding to the transverse attribution of any data and preset classification corresponding to the longitudinal attribution of any data. And dividing each data in the updated original aspect ratio array by the category number symbol to obtain a normalized original aspect ratio array, wherein the normalized original aspect ratio array is used as a classification aspect ratio array, and at the moment, any data in the classification aspect ratio array determines the probability that the preset classification corresponding to the transverse attribution of any data is the same as the preset classification corresponding to the longitudinal attribution of any data.

Under the condition that the classification aspect arrays are obtained, a Hadamard matrix can be initialized to be used as the classification aspect arrays, and task attribution type classification vectors of all preset classifications can be obtained by utilizing the preset orthogonal aspect arrays and the classification aspect arrays, wherein the task attribution type classification vectors of any preset classifications can be transverse number array data or longitudinal number array data of the preset orthogonal aspect arrays.

For any data of the preset orthogonal aspect group, if the data is 1, the position of the data is called a concerned bit, and if the data is-1, the position of the data is called an un-concerned bit.

In the embodiment of the invention, the order (i.e. the number of rows and the number of columns) of the initialized preset orthogonal aspect ratio array is recorded as n, and n needs to satisfy: and n > Z. Wherein, the task feature vector of any preset training task is a matrix of 1×zh (i.e. the length of the task feature vector of any preset training task is Zh), the classifying aspect array is a matrix of Z rows and Z columns (Z is the number of preset classifications), and n, zh and Z are all positive integers. The task attribution type classification vector of any preset classification is the transverse data or one column of data of the preset orthogonal aspect array, so that the order of the preset orthogonal aspect array exceeds the number of preset classifications. Because the task feature vector of the preset training task needs to determine the whole preset training task, and the task attribution type classification vector of any preset classification is a high-level abstract representation of the preset training task, the order of the preset orthogonal aspect ratio array should be smaller than or equal to the length Zh of the task feature vector of the preset training task. The task feature vector of the preset training task needs to contain the information specific to the preset training task besides the task feature vector used for classification, and the information specific to the preset training task is used for distinguishing different tasks of the same classification, so that the information specific to the preset training task is determined by using the attention bit.

In one possible implementation, obtaining the task attribution type classification vector of each preset classification according to the preset orthogonal aspect ratio array and the classification aspect ratio array includes: randomly acquiring a first preset classification from preset classifications of non-acquired classification vectors; any unprocessed transverse data is obtained from the preset orthogonal longitudinal and transverse array and used as a task attribution type classification vector of the first preset classification; acquiring task attribution type classification vectors of preset classifications corresponding to longitudinal attribution of all preset values in transverse number columns corresponding to the first preset classifications in the classification aspect groups according to the task attribution type classification vectors of the first preset classifications and the preset orthogonal aspect groups; if the preset classification of the non-acquired classification vector exists, repeating the operation from the preset classification of the non-acquired classification vector to the first preset classification arbitrarily acquired until the preset classification of the non-acquired classification vector disappears.

In the embodiment of the invention, one preset classification is arbitrarily acquired from preset classifications of the unoccupied classification vectors, and the arbitrarily acquired preset classification is used as a first preset classification. And optionally acquiring unprocessed transverse data except the first transverse sequence data from the preset orthogonal cross-bar array as a task attribution type classification vector of the first preset classification. The first transverse array data of the preset orthogonal array is all 1, that is, the positions of the first transverse array data of the preset orthogonal array are all concerned bits, not half concerned bits, and the other half is not concerned bits. The first lateral series data is located in all the concerned bits, so that the un-concerned bits which are not reserved for other preset classifications to determine are not beneficial to determining at least one class of the preset training task, and therefore, the lateral data except the first lateral series data are selected.

And then, according to the task attribution type classification vector of the first preset classification and the preset orthogonal longitudinal and transverse longitudinal number groups, acquiring the task attribution type classification vector of the preset classification corresponding to the longitudinal attribution of each preset value in the transverse number groups corresponding to the first preset classification in the classification longitudinal and transverse number groups. If the preset classification of the non-acquired categorization vector exists, a loop is executed from the step of randomly acquiring the first preset classification from the preset classification of the non-acquired categorization vector until the preset classification of the non-acquired categorization vector disappears.

Optionally, according to the task attribution type classification vector and the preset orthogonal aspect ratio array of the first preset classification, the task attribution type classification vector of the preset classification corresponding to the longitudinal attribution of each preset value in the transverse number array corresponding to the first preset classification in the classification aspect ratio array is obtained, including: acquiring a second preset classification corresponding to the longitudinal attribution of the maximum preset value and not acquiring the classification vector according to each preset value in the transverse number row corresponding to the first preset classification in the classification aspect ratio group; acquiring the transverse data which has the maximum similarity with the task attribution type classification vector of the first category to be determined and is not processed from the preset orthogonal longitudinal and transverse array, and taking the transverse data as the task attribution type classification vector of the second preset category; if the transverse number columns corresponding to the first preset classification in the classification aspect ratio group have preset classifications which do not acquire the classification vectors and correspond to the preset value longitudinal attribution, repeating the operation from the second preset classification which acquire the maximum preset value and do not acquire the classification vectors according to each preset value in the transverse number columns corresponding to the first preset classification in the classification aspect ratio group until the transverse number columns corresponding to the first preset classification in the classification aspect ratio group do not have the preset classifications which do not acquire the classification vectors and correspond to the preset value longitudinal attribution.

In the embodiment of the invention, according to each preset value in the transverse number series corresponding to the first preset classification in the classification aspect ratio group, one preset classification corresponding to the longitudinal attribution of the maximum preset value and not acquiring the classification vector is acquired, and the preset classification is used as the second preset classification.

And acquiring the transverse data which has the maximum similarity with the task attribution type classification vector of the first preset category and is not processed from the preset orthogonal cross-bar array, and taking the transverse data as the task attribution type classification vector of the second preset category. Here, the maximum similarity means that the attention bit similarity is the maximum.

For example, the task attribution type classifying vector of the first waiting category is [1, -1, -1, -1], the attention bits of the task attribution type classifying vector of the first waiting category are 1 st, 3 rd and 4 th bits respectively, there are two unprocessed lateral data in the preset orthogonal aspect groups, [1, -1, -1] and [1, -1, -1] respectively. Since the attention bits of the task attribution type classification vector of the first class to be determined [1, -1, -1, -1] are 1 st, 3 rd, 4 th bits, respectively, and [1, -1, the bits of interest of-1 are 1 st, 2 nd, 4 th (one bit out of interest), respectively, [1, -1, -1] the bits of interest are 1 st, 2 nd 5 th (two bits out of interest), respectively. Thus, [1, -1, -1] is a task attribution type classification vector of the second preset classification.

If the transverse number columns corresponding to the first preset classification in the classification aspect array have preset classifications which do not acquire the classification vectors and correspond to the preset value longitudinal attribution, starting to perform circulation from the step of acquiring the second preset classification corresponding to the maximum preset value longitudinal attribution and not acquire the classification vectors according to each preset value in the transverse number columns corresponding to the first preset classification in the classification aspect array until the transverse number columns corresponding to the first preset classification in the classification aspect array do not have the preset classifications which do not acquire the classification vectors and correspond to the preset value longitudinal attribution.

By the method, task attribution type classification vectors of preset classifications corresponding to longitudinal attribution of all preset values in transverse number columns corresponding to the first preset classifications in the classification aspect ratio groups can be obtained.

Optionally, after the task attribution type classification vector of the preset classification corresponding to the longitudinal attribution of each preset value in the transverse number sequence corresponding to the first preset classification in the classification aspect ratio array is obtained, arranging each preset value in the transverse number sequence corresponding to the first preset classification according to the sequence from big to small, so as to obtain each preset value after arrangement corresponding to the first preset classification. And for any preset value after the arrangement corresponding to the first preset classification, the task attribution type classification vector of the preset classification corresponding to the transverse attribution of the any preset value is acquired, and at the moment, the preset classification corresponding to the transverse attribution of the any preset value can be used as a third preset classification.

In the embodiment of the invention, starting from a first preset value after arrangement corresponding to a first preset classification, sequentially executing 'taking a preset classification corresponding to transverse attribution of the preset value as a third preset classification', and acquiring a task attribution type classification vector of the preset classification corresponding to longitudinal attribution of each preset value in a transverse number sequence corresponding to the third preset classification in the classification aspect group according to a task attribution type classification vector of the third preset classification and a preset orthogonal aspect group 'until the last preset value after arrangement corresponding to the first preset classification'. And if the preset classification of the non-acquired classification vector exists, performing repeated operation from the first preset classification randomly acquired from the preset classification of the non-acquired classification vector until the preset classification of the non-acquired classification vector disappears.

The method for acquiring the task attribution type classification vector of the preset classification corresponding to the longitudinal attribution of each preset value in the transverse array corresponding to the third preset classification in the classification aspect array is similar to the method for acquiring the task attribution type classification vector of the preset classification corresponding to the longitudinal attribution of each preset value in the transverse array corresponding to the first preset classification in the classification aspect array according to the task attribution type classification vector of the first preset classification and the preset orthogonal aspect array, and is not repeated herein.

By the method, the task attribution type classification vector of each of the plurality of preset classifications can be obtained. Half of task attribution type classification vectors of any preset classification are concerned bits, and the other half are unfocused bits. Compared with single attention bit, the embodiment of the invention can better utilize all attention bits, avoid a large number of redundant attention bits and improve the characteristic determining capability. Even if the training task set is biased, the task classification model obtained through training of the training task set can still well extract task feature vectors, and the task determining capability of the model for tasks outside a preset training task domain is improved.

In the above manner, the transverse data (except the first transverse array data) of the preset orthogonal aspect array is used as a task attribution type classifying vector of a preset classification, and when the method is applied, one column of data (except the first longitudinal array data) of the hadamard matrix can be used as a task attribution type classifying vector of a preset classification, and the two are similar in implementation principle, so that the description is omitted.

Next, for any one preset training task in the plurality of training task sets, comparing whether each task class of the any one preset training task is consistent with the plurality of preset classes or not, and obtaining a comparison result of each task class.

For any task classification of the any preset training task, the preset classification consistent with the any task classification can be obtained from a plurality of preset classifications through the comparison result of the any task classification, and the task attribution type classification vector of the preset classification consistent with the any task classification is obtained as the task attribution type classification vector of the any task classification. In this way, the task attribution type classification vector of each task classification of the any one of the preset training tasks can be obtained.

If any one training task set should be classified by a task, the task attribution type classification vector of the task classification of any one preset training task is obtained as a second classification vector of any one preset training task. For example, the preset training task X corresponds to task class 1, and the task attribution type classification vector of task class 1 is [1, -1, -1, -1], then the second classification vector of preset training task X is [1, -1, -1, -1].

If the training task set is classified by at least two tasks, performing a combination operation on task attribution type classification vectors of each task classification of the preset training task to obtain second classification vectors of the preset training task, and in addition, obtaining weights of all concerned bits in the second classification vectors of the preset training task. For example, the preset training task Y corresponds to task classifications 1 and 2, and the task class 1 has a task attribution type classification vector of [1, -1, -1, -1] (the first, third, fourth bits are the bits of interest), the task class 2 has a task attribution type classification vector of [1, -1, -1] (the first, second, fourth bits are the bits of interest), then a combination operation is performed on the task attribution type classification vector of the task classification 1 and the task attribution type classification vector of the task classification 2 to obtain a second classification vector [1, -1, -1] (the first, second, third and fourth bits are concerned bits) of the preset training task Y. Meanwhile, the weights of the attention bits 1, 2, 3 and 4 are sequentially 2, 1 and 2, and the attention bit 1 is focused 2 times, the attention bit 2 is focused 1 time, the attention bit 3 is focused 1 time, and the attention bit 4 is focused 2 times. Therefore, for the preset training task Y, in the second classification vector of the preset training task Y, the attention bit 1 and the attention bit 4 are important for locating the task class 1, the task class 2 and the pending classes corresponding to the attention bit 1 and the attention bit 4 from the plurality of pending classes, and the attention bit 2 and the attention bit 3 are used as auxiliary information for quasi-acquisition of the specific class (quasi-acquisition of the task class 1 or the task class 2).

And 205, updating and iterating the initial network structure according to the task feature vector, the first classifying vector and the second classifying vector of each preset training task in the training task sets to obtain a task classification model.

In the embodiment of the invention, according to the task feature vectors of each preset training task in a plurality of training task sets, the first classification vectors of each preset training task in a plurality of training task sets and the second classification vectors of each preset training task in a plurality of training task sets, the cost parameters of the initial network structure are obtained. And carrying out updating iteration on the initial network structure according to the cost parameter of the initial network structure to obtain the initial network structure after updating iteration. If the training ending condition is met (if the training times are the target times or the cost parameters of the initial network structure are within the target range), the initial network structure after the updating iteration is used as a task classification model. If the training ending condition is not met, updating and iterating the updated initial network structure again according to the modes from step 201 to step 205 until the training ending condition is met, and obtaining the task classification model.

In one possible implementation manner, according to a task feature vector, a first classification vector and a second classification vector of each preset training task in a plurality of training task sets, updating and iterating an initial network structure to obtain a task classification model, including: for any training task set, acquiring a first price parameter of the any training task set according to at least one task feature vector of a preset training task in the any training task set; acquiring sample task pairs of any training task set, wherein the sample task pairs of any training task set comprise one preset training task in any training task set and a target training task set, and the target training task set is a training task set which is less than a preset vector distance threshold value except any training task set in a plurality of training task sets; acquiring second cost parameters of any training task set according to task feature vectors of each preset training task in sample task pairs of any training task set; and updating and iterating the initial network structure according to the first classifying vector and the second classifying vector of each preset training task in the training task sets, and the first price parameter and the second cost parameter of the training task sets to obtain a task classification model.

In the embodiment of the invention, for any training task set, the task feature vector of any one preset training task in the training task set or the task feature vectors of two preset training tasks in the training task set can be utilized to acquire the first price parameter of the training task set.

Optionally, according to a task feature vector of at least one preset training task in the arbitrary training task set, acquiring a first price parameter of the arbitrary training task set includes: performing standardized processing on task feature vectors of at least one preset training task in any training task set to obtain standard task feature vectors of at least one preset training task in any training task set; and acquiring a first price parameter according to the task feature vector and the standard task feature vector of at least one preset training task in any training task set.

In the embodiment of the invention, the task feature vector of at least one preset training task in any training task set is subjected to standardized processing so as to convert floating point data of the task feature vector into standard data, and the standard task feature vector of at least one preset training task in any training task set is obtained.

And then, acquiring a first price parameter according to the task feature vector of at least one preset training task in any training task set and the standard task feature vector of at least one preset training task in any training task set. For example, according to a task feature vector of a preset training task in any training task set and a standard task feature vector of the preset training task in any training task set, a first price parameter of any training task set is obtained.

Optionally, obtaining a sample task pair of any training task set includes: acquiring a first difference coefficient between one preset training task in any training task set and one preset training task in each other training task set, wherein each other training task set is a training task set except any training task set in a plurality of training task sets; acquiring a target training task set with a first difference coefficient smaller than a first contrast difference coefficient from other training task sets; and acquiring sample task pairs of any training task set according to one preset training task and any training task set in the target training task set.

In the embodiment of the invention, the sample task pair of each training task set can be obtained according to all the training task sets. Sample task pairs of each training task set can also be obtained according to all training task sets in a batch. The principle of acquiring sample task pairs of training task sets is similar to that of the sample task pairs of each training task set, and the sample task pairs of each training task set are acquired according to all training task sets in a batch are taken as an example for explanation.

For one batch, acquiring target characteristics of one preset training task in any training task set and target characteristics of one preset training task in each other training task set, wherein the target characteristics can be task characteristic vectors, depth characteristics and depth characteristics after dimension reduction processing. And acquiring a first difference coefficient between one preset training task in the random training task set and one preset training task in the random other training task set by utilizing the target characteristic of one preset training task in the random training task set and the target characteristic of one preset training task in the random other training task set. Because the training task set includes two preset training tasks of similar types, the "first difference coefficient between one preset training task in any training task set and one preset training task in any other training task set" is equivalent to the "first difference coefficient between any training task set and any other training task set". In this way, a first coefficient of difference between any training task set and each of the other training task sets, respectively, may be obtained.

Next, a target training task set with a first difference coefficient smaller than the first contrast difference coefficient is obtained from each other training task set, so that one preset training task in the target training task set is obtained. The first difference coefficients between any training task set and each other training task set can be sequenced in order from small to large, the Kth first difference coefficient is taken as a first contrast difference coefficient, and fixed data can also be taken as the first contrast difference coefficient.

After one preset training task in the target training task set is obtained, a sample task pair of the arbitrary training task set is obtained according to the arbitrary training task set and one preset training task in the target training task set. The training task set comprises a sample task pair of any training task set, wherein two preset training tasks of the any training task set are used as positive sample pairs, and one preset training task in the any training task set and one preset training task in the target training task set are used as negative sample pairs.

In the embodiment of the invention, the sample task pair of any training task set can be calibrated, for example, X is one preset training task in any training task set, Y is another preset training task in any training task set, and Z is one preset training task in a target training task set.

It should be noted that, the target training task set is a training task set that is smaller than the preset vector distance threshold except any training task set in the plurality of training task sets, that is, the target training task set is another training task set that is smaller than the preset vector distance threshold. Wherein, the other training task set is "smaller than the preset vector distance threshold value" means that the first difference coefficient between one preset training task in the other training task set and one preset training task in any training task set satisfies the first contrast difference coefficient. Accordingly, other training task sets "not less than the preset vector distance threshold" means that the first coefficient of difference does not satisfy less than the first contrast coefficient of difference, i.e., the first coefficient of difference exceeds or is equal to the first contrast coefficient of difference.

Optionally, obtaining a target training task set with a first difference coefficient smaller than the first contrast difference coefficient from each other training task set includes: and acquiring a target training task set with a first difference coefficient smaller than a first contrast difference coefficient from a candidate training task set, wherein the candidate training task set is other training task sets with the first difference coefficient not lower than a second contrast difference coefficient in other training task sets, and the first contrast difference coefficient exceeds the second contrast difference coefficient.

In the embodiment of the invention, after the first difference coefficient between any training task set and each other training task set is obtained, other training task sets with the first difference coefficient smaller than the second contrast difference coefficient can be removed from each other training task set to obtain each removed other training task set, and at the moment, each removed other training task set is the candidate training task set. The first difference coefficients between any training task set and each other training task set can be sequenced in order from small to large, the M (M is a positive integer) th or M th first difference coefficient is taken as a second comparison difference coefficient, and fixed data can also be taken as the second comparison difference coefficient. Next, a target training task set with a first difference coefficient smaller than the first contrast difference coefficient is obtained from the removed other training task sets.

For example, a batch contains z (positive integer) training task sets, and for one training task set, a first difference coefficient between one preset training task in the training task set and one preset training task in z-1 other training task sets can be obtained, i.e. a first difference coefficient between the training task set and z-1 other training task sets can be obtained. The z-1 first difference coefficients are sequenced from small to large, the first 3% training task sets are removed, namely, after the first (2*z-2) 0.03 training task sets are removed (if z=64, the first 4 training task sets are removed), one preset training task in the first 10 target training task sets is taken, and the preset training task sets are combined with the training task sets respectively to form 10 sample task pairs of the training task sets. Wherein 10 sample task pairs can be obtained for each training task set, and thus, for this batch, a total of 10 x z sample task pairs are built.

In the embodiment of the invention, other training task sets with the first difference coefficient smaller than the second contrast difference coefficient are removed for consideration of noise. Since for tasks of close type, the task classification model should extract task feature vectors of close type, and for tasks of not close type, the task classification model should extract task feature vectors of large variance. Other training task sets which are close to one training task set in possible types are removed by removing other training task sets with the first difference coefficient smaller than the second contrast difference coefficient, one preset training task in the other training task sets which are dissimilar to the one training task set but are close to the one training task set is utilized to construct a sample task pair of the one training task set with the one training task set, and therefore training of a task classification model by utilizing indistinguishable preset training tasks is achieved, and accuracy of the task classification model is improved.

Optionally, after the first difference coefficient between the arbitrary training task set and each other training task set is obtained, other training task sets with the first difference coefficient not lower than the second contrast difference coefficient can be screened from each other training task set, so as to obtain screened other training task sets, and at this time, the screened other training task sets are candidate training task sets. And then, acquiring a target training task set with the first difference coefficient smaller than the first contrast difference coefficient from the other screened training task sets. The implementation principle of each other training task set obtained through screening is similar to the implementation principle of each other training task set obtained through removal, and is not described herein again.

Alternatively, the magnitude of the iterative second contrast difference coefficient may be updated according to a relationship of whether or not the types between every two training task sets within a batch are close. For example, the number of types approaching between each two training task sets in a batch is counted, if the number is larger, the second contrast difference coefficient is larger, and if the number is smaller, the second contrast difference coefficient is smaller.

After the sample task pairs of any training task set are obtained, obtaining second cost parameters of any training task set according to task feature vectors of all preset training tasks in the sample task pairs of any training task set.

Optionally, obtaining the second cost parameter of the arbitrary training task set according to the task feature vector of each preset training task in the sample task pair of the arbitrary training task set includes: aiming at a sample task pair of any training task set, acquiring a second difference coefficient between one preset training task and another preset training task in the any training task set according to a task feature vector of the one preset training task and a task feature vector of the other preset training task in the any training task set, and acquiring a third difference coefficient between the one preset training task and the one preset training task in the target training task set in the any training task set according to the task feature vector of the one preset training task and the task feature vector of the one preset training task in the target training task set in the any training task set; and acquiring a second cost parameter of any training task set according to the second difference coefficient and the third difference coefficient.

And then, acquiring cost parameters of the initial network structure according to the first classifying vector and the second classifying vector of each preset training task in the training task sets and the first price parameters and the second cost parameters of the training task sets, so as to update and iterate the initial network structure by utilizing the cost parameters of the initial network structure, and obtaining a task classification model.

Optionally, updating and iterating the initial network structure according to a first classification vector and a second classification vector of each preset training task in the plurality of training task sets and a first price parameter and a second cost parameter of the plurality of training task sets to obtain a task classification model, including: acquiring cost parameters of the classifying vectors according to a first classifying vector and a second classifying vector of each preset training task in a plurality of training task sets; acquiring task feature vector cost parameters according to the first cost parameters and the second cost parameters of the plurality of training task sets; and updating and iterating the initial network structure according to the classified vector cost parameters and the task feature vector cost parameters to obtain a task classification model.

In the embodiment of the invention, for each preset training task in a training task set, task attribution type classification vector cost parameters of the preset training task are obtained according to a first classification vector of the preset training task and a second classification vector of the preset training task.

According to the method, task attribution type classification vector cost parameters of all preset training tasks in any training task set can be obtained, and task attribution type classification vector cost parameters of all preset training tasks in any training task set are obtained according to the task attribution type classification vector cost parameters of all preset training tasks in any training task set. And classifying the vector cost parameters according to the task attribution types corresponding to the training task sets, and acquiring the task attribution type classifying vector cost parameters corresponding to the training task sets.

The embodiment of the invention does not limit the weight coefficient of the second cost parameter and the weight coefficient of the first cost parameter.

According to the mode, the task feature vector cost parameters of each preset training task in the random training task set can be obtained, and the task feature vector cost parameters corresponding to the random training task set are obtained according to the task feature vector cost parameters of each preset training task in the random training task set. And acquiring task feature vector cost parameters corresponding to the training task sets according to the task feature vector cost parameters corresponding to the training task sets.

After the task attribution type classification vector cost parameters corresponding to the plurality of training task sets and the task feature vector cost parameters corresponding to the plurality of training task sets are obtained, the cost parameters of the initial network structure can be obtained according to the task attribution type classification vector cost parameters corresponding to the plurality of training task sets and the task feature vector cost parameters corresponding to the plurality of training task sets. And updating and iterating the initial network structure according to the cost parameters of the initial network structure to obtain a task classification model.

In order to clearly describe the solution provided by the embodiment of the present invention, the foregoing step S104 may be implemented by executing in the following detailed manner, taking the terminal as an execution body as an example.

Step S301, a terminal acquires task type information to be processed of a cloud service task and task equipment information to be determined aiming at the task type information to be processed.

The task type information to be processed of the cloud service task is the task type of the cloud service task which is currently identified and classified, and the task equipment information to be determined is the specific equipment for processing the current task type of the cloud service task to be determined.

Step S302, the terminal determines past processing information of the cloud service task according to the archiving associated information of the cloud service task, wherein the archiving associated information comprises past task type information of the cloud service task and past task equipment information aiming at the past task type information.

The past task type information of the cloud service task is task type information which is classified by the past history of the cloud service task, and the past task device information is device which is used for executing the corresponding task by the past history of the cloud service task. The past processing information of the cloud service task can reflect the past classification information and the past processing equipment of the cloud service task, and it should be understood that, with the self-updating of the neural network model and the content adjustment of the task, the type corresponding to the same cloud service task may be optimized accordingly.

Step S303, the terminal determines an adaptation parameter according to the past processing information of the cloud service task, the type information of the task to be processed and the equipment information of the task to be determined, wherein the adaptation parameter is used for representing the adaptation degree between the equipment information of the task to be determined and the cloud service task.

The higher the adaptation degree is, the more the task processing equipment is adapted to the cloud service task, and the lower the adaptation degree is, the task processing equipment is not adapted to the cloud service task.

According to the embodiment of the invention, the past processing information of the cloud service task is determined through the archiving associated information of the cloud service task, and the past processing information can reflect the historical task category and the historical processing equipment of the cloud service task. The method and the device for the cloud service task are used for predicting according to the past processing information of the cloud service task and the type information of the task to be processed, so that the adaptive parameters between the equipment information of the task to be determined and the cloud service task can be obtained, rules are not used in the prediction process, and the auxiliary decision cost can be reduced.

In the embodiment of the invention, the following scheme is also provided:

step S401, the terminal acquires task type information to be processed of the cloud service task and task equipment information to be determined aiming at the task type information to be processed.

The information of the type of the task to be processed of the cloud service task is related information of the cloud service task which is currently classified, for example, description attributes and the like of the cloud service task which is currently classified are included; the pending task device information is related information of the processing device used for processing the currently classified cloud service task, and includes information such as description attribute, usage and the like of the target processing device.

Step S402, the terminal acquires archiving related information of the cloud service task, wherein the archiving related information comprises past task type information of the cloud service task and past task equipment information aiming at the past task type information.

Step S403, the terminal determines the past processing information of the cloud service task according to the archiving associated information of the cloud service task.

In a possible implementation manner, a terminal determines a plurality of target task type nodes corresponding to a cloud service task in a cloud task graph structure according to past task type information of the cloud service task, wherein the cloud task graph structure is used for representing contact information among the plurality of task type nodes. And the terminal determines a plurality of target task processing equipment nodes corresponding to the cloud service task in a task equipment graph structure according to the past task equipment information of the cloud service task, wherein the task equipment graph structure is used for representing contact information among the plurality of task processing equipment nodes. And the terminal determines the past processing information of the cloud service task according to the plurality of target task type nodes and the plurality of target task processing equipment nodes. In some embodiments, the cloud task graph structure and the task device graph structure are stored in a cloud system, and each device in the cloud system can review the cloud task graph structure and the task device graph structure. The cloud task graph structure and the task equipment graph structure are both vertical domain knowledge graphs, and the cloud task graph structure is a graph specially set for a cloud service task. The task equipment graph structure comprises the processing equipment description attribute, the processing equipment hardware facilities, the network connection condition and other contact information among the task processing equipment nodes.

In this embodiment, the terminal may associate past task type information of the cloud service task with the cloud task graph structure, and determine a plurality of target task type nodes corresponding to the cloud service task. And the terminal correlates the past task equipment information of the cloud service task with the task equipment graph structure and determines a plurality of target task processing equipment nodes corresponding to the cloud service task. The terminal can reflect the transformation state of the cloud service task to a certain extent according to the determined past processing information of the cloud service task determined by the plurality of target task type nodes and the plurality of target task processing equipment nodes, and the follow-up terminal can determine the adaptation degree between the equipment information of the undetermined task and the cloud service task according to the past processing information of the cloud service task.

In a possible implementation manner, a terminal performs graph convolution processing on a cloud task graph structure to obtain feature vectors of a plurality of task type nodes in the cloud task graph structure. And the terminal executes feature extraction operation on the task equipment graph structure to obtain feature vectors of a plurality of task processing equipment nodes in the task equipment graph structure.

For example, the terminal performs a feature extraction operation on the cloud task graph structure by training the first graph rolling network to obtain feature vectors of a plurality of task type nodes in the cloud task graph structure. And the terminal executes feature extraction operation on the task equipment graph structure by training the second graph rolling network to obtain feature vectors of a plurality of task processing equipment nodes in the task equipment graph structure.

Step S404, the terminal determines an adaptation parameter according to the past processing information of the cloud service task, the type information of the task to be processed and the equipment information of the task to be determined, wherein the adaptation parameter is used for representing the adaptation degree between the equipment information of the task to be determined and the cloud service task.

The higher the adaptation degree is, the more suitable the processing equipment corresponding to the target processing equipment information is for the cloud service task, and the lower the adaptation degree is, the less suitable the processing equipment corresponding to the target processing equipment information is for the cloud service task.

In one possible implementation manner, a terminal obtains a first weight between past processing information of a cloud service task and task type information to be processed. The terminal acquires a second weight between the past processing information of the cloud service task and the equipment information of the task to be determined. And the terminal performs combination operation on the task type information to be processed and the undetermined task equipment information according to the first weight and the second weight to obtain a task equipment matching vector of the cloud service task. And the terminal determines the adaptation parameters according to the past processing information of the cloud service task and the task equipment matching vector of the cloud service task.

According to the embodiment of the invention, the past processing information of the cloud service task is determined through the archiving associated information of the cloud service task, and the past processing information can reflect the historical task category and the historical processing equipment of the cloud service task. The method and the device for the cloud service task are used for predicting according to the past processing information of the cloud service task and the type information of the task to be processed, so that the adaptive parameters between the equipment information of the task to be determined and the cloud service task can be obtained, rules are not used in the prediction process, and the auxiliary decision cost can be reduced. The terminal displays the adaptation parameters on the adaptation parameter page, a user can determine the adaptation degree between the equipment information of the task to be determined and the cloud service task by checking the adaptation parameters, the user can conveniently adjust the equipment information of the task to be determined in time when the adaptation degree is not high, and the efficiency of man-machine interaction is high.

The embodiment of the invention provides a server 100, wherein the server 100 comprises a processor and a nonvolatile memory storing computer instructions, and when the computer instructions are executed by the processor, the server 100 executes the cloud service platform task scheduling method based on artificial intelligence. As shown in fig. 2, fig. 2 is a block diagram of a server 100 according to an embodiment of the present invention. The server 100 includes a memory 111, a processor 112, and a communication unit 113. For data transmission or interaction, the memory 111, the processor 112 and the communication unit 113 are electrically connected to each other directly or indirectly.

The foregoing description, for purpose of explanation, has been presented with reference to particular embodiments. However, the illustrative discussions above are not intended to be exhaustive or to limit the disclosure to the precise forms disclosed. Many modifications and variations are possible in light of the above teaching. The embodiments were chosen and described in order to best explain the principles of the disclosure and its practical application, to thereby enable others skilled in the art to best utilize the disclosure and various embodiments with various modifications as are suited to the particular use contemplated. The foregoing description, for purpose of explanation, has been presented with reference to particular embodiments. However, the illustrative discussions above are not intended to be exhaustive or to limit the disclosure to the precise forms disclosed. Many modifications and variations are possible in light of the above teaching. The embodiments were chosen and described in order to best explain the principles of the disclosure and its practical application, to thereby enable others skilled in the art to best utilize the disclosure and various embodiments with various modifications as are suited to the particular use contemplated.

Claims (10)

1. The cloud service platform task scheduling method based on artificial intelligence is characterized by comprising the following steps of:

responding to a task initiation request aiming at a cloud service platform, and analyzing the task initiation request to obtain a cloud service task corresponding to the task initiation request;

invoking a task classification model which is trained in advance, classifying the cloud service task and obtaining task type information to be processed;

determining pending task equipment information aiming at the pending task type information according to the pending task type information;

determining the adaptation degree between the undetermined task equipment information and the cloud service task;

and when the adaptation degree reaches a preset adaptation degree threshold, sending the cloud service task to task processing equipment corresponding to the undetermined task equipment information for processing.

2. The method of claim 1, wherein the task classification model is obtained by:

acquiring a plurality of training task sets and task classifications of preset training tasks in the plurality of training task sets, wherein the training task sets comprise two preset training tasks with close types;

acquiring task feature vectors of each preset training task in the training task sets according to an initial network structure;

Acquiring first classification vectors of all preset training tasks in the training task sets according to task feature vectors of all preset training tasks in the training task sets, wherein the first classification vectors of the preset training tasks are used for determining task attribution types of the preset training tasks to be classified;

acquiring second classification vectors of all preset training tasks in the training task sets according to task classifications of all preset training tasks in the training task sets, wherein the second classification vectors of the preset training tasks are used for determining task attribution types of the marked preset training tasks;

for any training task set, acquiring a first price parameter of the any training task set according to a task feature vector of at least one preset training task in the any training task set;

acquiring sample task pairs of the arbitrary training task set, wherein the sample task pairs of the arbitrary training task set comprise one preset training task of the arbitrary training task set and a target training task set, and the target training task set is a training task set which is not the arbitrary training task set and is smaller than a preset vector distance threshold value in the plurality of training task sets;

Acquiring a second cost parameter of the arbitrary training task set according to task feature vectors of each preset training task in the sample task pair of the arbitrary training task set;

and updating and iterating the initial network structure according to the first classifying vector and the second classifying vector of each preset training task in the training task sets, and the first price parameter and the second cost parameter of the training task sets to obtain a task classification model.

3. The method of claim 2, wherein the task classification of the preset training task is at least one, and the obtaining the second classification vector of each preset training task in the plurality of training task sets according to the task classification of each preset training task in the plurality of training task sets comprises:

for any task classification of each preset training task in the training task sets, acquiring preset classifications consistent with the any task classification from a plurality of preset classifications, wherein the task classifications of the preset training tasks are at least two;

acquiring an original aspect ratio array, wherein each transverse number row of the original aspect ratio array corresponds to each preset classification, and each longitudinal number row of the original aspect ratio array corresponds to each preset classification;

For any one preset training task in the plurality of training task sets, setting a preset value at the intersection of a target transverse array and a target longitudinal array according to at least two task classifications of the any one preset training task, wherein the target transverse array is a transverse array corresponding to a preset classification consistent with one task classification of the any one preset training task in the original aspect array, and the target longitudinal array is a longitudinal array corresponding to a preset classification consistent with other task classifications of the any one preset training task except the one task classification in the original aspect array;

performing superposition operation on preset values at the intersections of each transverse array and each longitudinal array in the original aspect ratio array to obtain a classification aspect ratio array, wherein any data in the classification aspect ratio array determines the frequency of the preset classification corresponding to the transverse attribution of any data to be the same as the frequency of the preset classification corresponding to the longitudinal attribution of any data, and one task classification is one preset classification;

randomly acquiring a first preset classification from preset classifications of non-acquired classification vectors;

any unprocessed transverse data is obtained from a preset orthogonal longitudinal and transverse array and used as a task attribution type classification vector of the first preset classification;

According to each preset value in the transverse number series corresponding to the first preset classification in the classification aspect ratio group, acquiring a second preset classification corresponding to the longitudinal attribution of the maximum preset value and not acquiring the classification vector;

acquiring the transverse data which has the maximum similarity with the task attribution type classification vector of the first preset classification and is not processed from the preset orthogonal cross-axis array, and taking the transverse data as the task attribution type classification vector of the second preset classification;

if the transverse sequence corresponding to the first preset classification in the classification aspect array has the preset classification corresponding to the longitudinal attribution of the preset value, repeating the operation from each preset value in the transverse sequence corresponding to the first preset classification in the classification aspect array to acquire the second preset classification corresponding to the longitudinal attribution of the maximum preset value and not acquiring the classification vector until the transverse sequence corresponding to the first preset classification in the classification aspect array does not have the preset classification corresponding to the longitudinal attribution of the preset value;

if the preset classification of the unoccupied classification vector exists, optionally acquiring a first preset classification from the preset classification of the unoccupied classification vector to execute repeated operation until the preset classification of the unoccupied classification vector disappears, wherein the order of the preset orthogonal aspect number group exceeds the number of the preset classification;

Acquiring task attribution type classification vectors of preset classifications consistent with the arbitrary task classifications as task attribution type classification vectors of the arbitrary task classifications;

and for any one preset training task in the plurality of training task sets, acquiring a second classification vector of the any one preset training task according to a task attribution type classification vector of at least one task classification of the any one preset training task.

4. The method according to claim 2, wherein the obtaining the first price parameter of the arbitrary training task set according to the task feature vector of at least one preset training task in the arbitrary training task set includes:

performing standardized processing on task feature vectors of at least one preset training task in the random training task set to obtain standard task feature vectors of at least one preset training task in the random training task set;

and acquiring the first price parameter according to the task feature vector and the standard task feature vector of at least one preset training task in the random training task set.

5. The method of claim 2, wherein the obtaining the sample task pair for the arbitrary training task set comprises:

Acquiring a first difference coefficient between one preset training task in the arbitrary training task set and one preset training task in each other training task set, wherein each other training task set is a training task set except the arbitrary training task set in the plurality of training task sets;

acquiring a target training task set with a first difference coefficient smaller than a first contrast difference coefficient from the other training task sets;

and acquiring sample task pairs of the arbitrary training task set according to one preset training task in the target training task set and the arbitrary training task set.

6. The method of claim 5, wherein the obtaining a target training task set having a first difference coefficient that is less than a first contrast difference coefficient from the respective other training task sets comprises:

and acquiring a target training task set with a first difference coefficient smaller than the first contrast difference coefficient from a candidate training task set, wherein the candidate training task set is other training task sets with the first difference coefficient not lower than a second contrast difference coefficient in the other training task sets, and the first contrast difference coefficient exceeds the second contrast difference coefficient.

7. The method according to claim 2, wherein the obtaining the second cost parameter of the arbitrary training task set according to the task feature vector of each preset training task in the sample task pair of the arbitrary training task set includes:

aiming at a sample task pair of the arbitrary training task set, acquiring a second difference coefficient between one preset training task and another preset training task in the arbitrary training task set according to a task feature vector of the one preset training task and a task feature vector of the other preset training task in the arbitrary training task set, and acquiring a third difference coefficient between one preset training task and one preset training task in the arbitrary training task set according to a task feature vector of the one preset training task and a task feature vector of the one preset training task in the target training task set;

and acquiring a second cost parameter of the arbitrary training task set according to the second difference coefficient and the third difference coefficient.

8. The method according to claim 2, wherein updating the initial network structure to obtain the task classification model according to the first classification vector and the second classification vector of each preset training task in the plurality of training task sets and the first price parameter and the second cost parameter of the plurality of training task sets comprises:

Acquiring a classifying vector cost parameter according to a first classifying vector and a second classifying vector of each preset training task in the plurality of training task sets;

acquiring task feature vector cost parameters according to the first cost parameters and the second cost parameters of the training task sets;

and updating and iterating the initial network structure according to the classifying vector cost parameters and the task feature vector cost parameters to obtain a task classifying model.

9. The method of claim 1, wherein the determining the degree of adaptation between the pending task device information and the cloud service task comprises:

acquiring to-be-processed task type information of a cloud service task and to-be-determined task equipment information aiming at the to-be-processed task type information;

acquiring at least one first candidate task type parameter from past task type information of the cloud service task, wherein the first candidate task type parameter comprises a description attribute of a first task type node;

determining a description attribute vector of the first task type node according to the first candidate task type parameter;

acquiring a first comparison vector of the cloud service task according to the description attribute vector;

Comparing the first comparison vector with description attribute vectors of a plurality of task type nodes in a cloud task graph structure, determining a plurality of target task type nodes from the plurality of task type nodes, wherein the vector distance between the description attribute vector of the target task type node and the first comparison vector does not exceed a preset distance threshold, and the cloud task graph structure is used for representing contact information among the plurality of task type nodes;

acquiring at least one second candidate task type parameter from past task equipment information of the cloud service task, wherein the second candidate task type parameter comprises a description attribute of a second task type node;

determining a plurality of target task processing equipment nodes corresponding to the cloud service task in the task equipment graph structure according to the second candidate task type parameter, wherein the task equipment graph structure is used for representing contact information among the plurality of task processing equipment nodes;

determining past processing information of the cloud service task according to the target task type nodes and the target task processing equipment nodes;

acquiring a first weight between past processing information of the cloud service task and the task type information to be processed;

Acquiring a second weight between the past processing information of the cloud service task and the equipment information of the task to be determined;

performing a combination operation on the task type information to be processed and the task equipment information to be determined according to the first weight and the second weight to obtain a task equipment matching vector of the cloud service task;

and determining an adaptation parameter according to the past processing information of the cloud service task and a task equipment matching vector of the cloud service task, wherein the adaptation parameter is used for representing the adaptation degree between the undetermined task equipment information and the cloud service task.

10. A server system comprising a server for performing the method of any of claims 1-9.

Images