Task scheduling method and device based on cloud edge collaboration

Abstract

The invention provides a task scheduling method and device based on cloud edge collaboration. The method comprises the steps of obtaining a to-be-allocated task from an edge end; acquiring a time queue of resource load values of the edge end and the cloud end; extracting resource load time sequence characteristics of the edge end and the cloud end; carrying out spatial distance measurement-based detailed feature interaction and resource load time sequence collaborative features based on resource load time sequence features; and according to the resource load time sequence cooperation characteristics, splitting the to-be-allocated task and scheduling the to-be-allocated task to an edge end and a cloud end respectively, so that the edge end and the cloud end perform task execution respectively. According to the method, the edge end resource load time sequence features and the cloud end resource load time sequence features are extracted, the resource load conditions of the cloud end and the edge end are considered, then detailed feature interaction based on spatial distance measurement is carried out, resource load time sequence cooperation features are obtained, flexible splitting and scheduling of tasks are achieved, and the task scheduling efficiency is improved. The overall operation stability of the system can be improved, and the task processing efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the field of task scheduling, and in particular to a task scheduling method and device based on cloud-edge collaboration. Background Art

[0002] With the access of new energy and the widespread application of distributed energy, adjustable loads and energy storage facilities in the power grid, the dispatching and control mode of the power system is changing from the traditional "source follows load" to the multi-level coordinated control mode of "source-grid-load-storage". This change requires the power grid to have higher real-time perception capabilities and precise control capabilities for distributed energy, loads and energy storage resources.

[0003] In order to meet this challenge, a two-level collaborative power dispatching system based on "cloud + edge" has emerged. This system provides data services by integrating network, storage and computing technologies on the edge side to achieve the purpose of improving system operation efficiency. However, when dealing with large-scale power system computing and analysis tasks, on the one hand, although edge devices can respond quickly, they may be unable to cope with complex tasks due to limited computing power and storage space; on the other hand, although cloud computing centers have powerful computing power, they may encounter delay problems when processing large amounts of real-time data. Therefore, existing systems often face the problem of unbalanced loads between edge networks and the cloud, which affects the stability of the overall operation of the system. How to effectively integrate the advantages of edge computing and cloud computing to form complementarity has become an urgent problem to be solved. Summary of the invention

[0004] The present invention provides a task scheduling method and device based on cloud-edge collaboration to solve the technical problem of unbalanced task distribution between the edge and the cloud, and improve the stability of the overall operation of the system.

[0005] In order to solve the above technical problems, an embodiment of the present invention provides a task scheduling method based on cloud-edge collaboration, including:

[0006] Acquire a task to be assigned from an edge end, where the task to be assigned is a power calculation task;

[0007] Obtaining a time queue of resource load values ​​at the edge and a time queue of resource load values ​​at the cloud;

[0008] Extracting edge resource load timing characteristics based on the time queue of the edge resource load value; extracting cloud resource load timing characteristics based on the time queue of the cloud resource load value;

[0009] Based on the time series characteristics of edge resource load and cloud resource load, detailed feature interaction based on spatial distance measurement is performed to obtain resource load time series coordination characteristics;

[0010] According to the resource load timing coordination characteristics, the task to be allocated is split to obtain a first power calculation subtask to be allocated and a second power calculation subtask to be allocated. The first power calculation subtask to be allocated and the second power calculation subtask to be allocated are respectively scheduled to the edge and the cloud, so that the edge and the cloud respectively execute the first power calculation subtask to be allocated and the second power calculation subtask to be allocated.

[0011] Correspondingly, the present invention application also provides a task scheduling device based on cloud-edge collaboration, including a task acquisition module, a queue acquisition module, a feature extraction module, a feature interaction module and a scheduling module; wherein,

[0012] The task acquisition module is used to acquire the task to be assigned from the edge end, and the task to be assigned is a power calculation task;

[0013] The queue acquisition module is used to acquire the time queue of the resource load value of the edge end and the time queue of the resource load value of the cloud end;

[0014] The feature extraction module is used to extract the time series features of the edge resource load based on the time series of the edge resource load value; and to extract the time series features of the cloud resource load based on the time series of the cloud resource load value;

[0015] The feature interaction module is used to perform detailed feature interaction based on spatial distance measurement based on the edge resource load time series characteristics and the cloud resource load time series characteristics to obtain resource load time series collaborative characteristics;

[0016] The scheduling module is used to split the task to be allocated according to the resource load timing coordination characteristics to obtain a first power calculation subtask to be allocated and a second power calculation subtask to be allocated, and schedule the first power calculation subtask to be allocated and the second power calculation subtask to be allocated to the edge end and the cloud end respectively, so that the edge end and the cloud end respectively execute the first power calculation subtask to be allocated and the second power calculation subtask to be allocated.

[0017] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0018] The present invention application provides a task scheduling method and device based on cloud-edge collaboration, which extracts the timing characteristics of edge resource load and cloud resource load, takes into account the resource load conditions of the cloud and edge, and then performs detailed feature interaction based on spatial distance measurement to obtain resource load timing collaboration characteristics, thereby realizing flexible task splitting and scheduling. Compared with the existing technology, it can realize load balancing of power computing tasks between edge and cloud, effectively integrate the advantages of edge and cloud, improve the stability of the overall system operation and the intelligence of task scheduling, and improve task processing efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 : A flow chart of an embodiment of a task scheduling method based on cloud-edge collaboration provided in the present invention application.

[0020] Figure 2 : A structural diagram of an embodiment of a task scheduling device based on cloud-edge collaboration provided in the present invention application.

[0021] Figure 3 : A structural block diagram of an electronic device for a task scheduling method based on cloud-edge collaboration provided in the present invention application.

[0022] Figure 4 : A schematic diagram of an application scenario of an embodiment of a task scheduling method based on cloud-edge collaboration provided in the present invention application. DETAILED DESCRIPTION

[0023] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0024] Embodiment one:

[0025] Please refer to Figure 1 , Figure 1 A task scheduling method based on cloud-edge collaboration provided by an embodiment of the present invention includes steps S101 to S105; wherein:

[0026] Step S101, obtaining a task to be assigned from an edge end, wherein the task to be assigned is a power calculation task.

[0027] Step S102, obtaining a time queue of resource load values ​​at the edge and a time queue of resource load values ​​at the cloud.

[0028] Step S103, extracting edge resource load timing characteristics based on the time queue of the edge resource load value; extracting cloud resource load timing characteristics based on the time queue of the cloud resource load value.

[0029] Step S104, based on the edge resource load timing characteristics and the cloud resource load timing characteristics, a detailed feature interaction based on spatial distance measurement is performed to obtain the resource load timing coordination characteristics.

[0030] Step S105, according to the resource load timing coordination characteristics, split the task to be allocated to obtain a first power calculation subtask to be allocated and a second power calculation subtask to be allocated, and schedule the first power calculation subtask to be allocated and the second power calculation subtask to be allocated to the edge and the cloud respectively, so that the edge and the cloud respectively execute the first power calculation subtask to be allocated and the second power calculation subtask to be allocated.

[0031] In this embodiment, the edge end may be an edge gateway or an edge device. For the above step S103, the time queue based on the resource load value of the edge end is used to extract the time series characteristics of the resource load of the edge end; the time queue based on the resource load value of the cloud end is used to extract the time series characteristics of the resource load of the cloud end, specifically:

[0032] Input the edge resource load timing characteristics and the cloud resource load timing characteristics into the 1D-CNN model to obtain a sequence of edge resource load local timing correlation feature vectors and a sequence of cloud resource load local timing correlation feature vectors;

[0033] Performing feature enhancement aggregation representation processing of cluster analysis on the sequence of local time series correlation feature vectors of edge resource load to obtain the time series feature of the edge resource load; performing feature enhancement aggregation representation processing of cluster analysis on the sequence of local time series correlation feature vectors of the cloud resource load to obtain the time series feature of the cloud resource load;

[0034] The feature enhancement aggregation representation processing of clustering analysis on the sequence of local time series correlation feature vectors of edge resource load to obtain the edge resource load time series feature is specifically:

[0035] A cluster analysis is performed on the sequence of local timing correlation feature vectors of the edge resource load to obtain a self-supervised clustering representation vector of the local timing correlation feature of the edge resource load; based on the self-supervised clustering representation vector of the local timing correlation feature of the edge resource load, feature clustering field modulation aggregation is performed on the sequence of local timing correlation feature vectors of the edge resource load to obtain a significant aggregation representation vector of the edge resource load timing as the edge resource load timing feature.

[0036] It can be understood that since the sequence of local time series correlation feature vectors of the edge resource load and the sequence of local time series correlation feature vectors of the cloud resource load respectively contain local time series feature information about the resource load values ​​of the edge and the cloud, and these local time series features have contextual correlation feature information in the full time domain, this correlation feature information is of great significance for subsequent task allocation. Therefore, this embodiment can further perform feature enhancement aggregation representation of cluster analysis on the sequence of local time series correlation feature vectors of the edge resource load and the sequence of local time series correlation feature vectors of the cloud resource load, so as to obtain the time series features of the edge resource load and the time series features of the cloud resource load respectively. Furthermore, the processing of feature enhancement aggregation representation of cluster analysis aims to use the simulated field analysis theory to perform explicit modeling based on the field for the clustering contribution of each local time series correlation feature vector of the edge resource load and each local time series correlation feature vector of the cloud resource load, so as to improve the accuracy of feature aggregation analysis, thereby improving the performance and generalization ability of the model.

[0037] Further, the self-supervised clustering representation vector based on the local time series correlation feature of the edge resource load performs feature clustering field modulation aggregation on the sequence of the local time series correlation feature vector of the edge resource load to obtain the edge resource load time series significant aggregation representation vector as the edge resource load time series feature, which is specifically:

[0038] Calculate the implicit clustering contribution factor of each edge resource load local time series correlation feature vector in the sequence of the edge resource load local time series correlation feature vector relative to the edge resource load local time series correlation feature self-supervised clustering representation vector to obtain a sequence of implicit clustering contribution factors of the edge resource load local time series correlation feature;

[0039] Vectorizing and arranging the sequence of implicit clustering contribution factors of the local time series correlation characteristics of the edge resource load to obtain a distribution vector of the clustering contribution field of the local time series correlation characteristics of the edge resource load;

[0040] Input the distribution vector of the cluster contribution field of the local temporal correlation feature of the edge resource load into the cluster contribution field explicit modeling module based on the self-attention mechanism to obtain the modulation weight vector of the cluster contribution field of the local temporal correlation feature of the edge resource load;

[0041] Taking each eigenvalue in the field modulation weight vector of the local timing correlation feature cluster contribution of the edge resource load as the weight, the weighted sum of the corresponding positions of the sequence of the local timing correlation feature vector of the edge resource load is calculated to obtain the edge resource load timing significant aggregation representation vector.

[0042] Among them, for the step of feature-enhanced aggregate representation of cluster analysis, first, an unsupervised learning method can be used to perform cluster analysis on the sequence of the local temporal correlation feature vectors of the edge resource load, and the local temporal context correlation feature patterns and structures of the edge resource load in the set can be automatically discovered to generate a self-supervised cluster representation vector of the local temporal correlation feature of the edge resource load. It can be understood that self-supervised clustering can reveal the intrinsic structure of each local temporal context correlation feature of the edge resource load, which helps the model to learn meaningful feature representations without explicit labels, and enhances the model's understanding of the distribution of local temporal context correlation features of the edge resource load. Furthermore, by calculating the implicit clustering contribution factor between the local temporal correlation feature vector of the edge resource load and the self-supervised cluster representation vector of the local temporal correlation feature of the edge resource load, it is used to reflect the degree of contribution of each local temporal correlation feature vector of the edge resource load to its cluster. Accordingly, the calculation of implicit clustering contribution factors helps to identify and quantify the role of each edge resource load local timing correlation feature vector in the clustering process, and provides a basis for subsequent feature selection and weight allocation, so that the model can pay more attention to features that contribute significantly to clustering. Next, the sequence of implicit clustering contribution factors of the local timing correlation characteristics of the edge resource load calculated is converted into a clustering contribution field distribution vector of the local timing correlation characteristics of the edge resource load, that is, the sequence of implicit clustering contribution factors of the local timing correlation characteristics of the edge resource load is integrated into the vector field by data integration to construct a clustering contribution field. Subsequently, the self-attention mechanism is used to explicitly model the clustering contribution field distribution vector of the local timing correlation characteristics of the edge resource load to generate a modulation weight vector of the clustering contribution field of the local timing correlation characteristics of the edge resource load. That is, the self-attention mechanism is used to explicitly model the clustering contribution of the implicit clustering contribution factors of the local temporal correlation features of each edge resource load in the cluster contribution field based on the field autocorrelation global analysis, so that the focus on the feature vectors with large clustering contributions can be strengthened, while suppressing unimportant features, improving the quality of feature representation and the accuracy of subsequent power calculation and task scheduling. Finally, the sequence of the local temporal correlation feature vectors of the edge resource load is weighted fused using the cluster contribution field modulation weight vector of the local temporal correlation feature of the edge resource load to generate a spatial domain edge resource load significant aggregation representation vector, which provides a comprehensive feature representation, which not only contains the original local temporal correlation feature information of the edge resource load, but also integrates the knowledge learned in the clustering process, so that the model can more accurately capture the key local temporal semantics of the edge resource load when processing complex tasks, and provides strong feature support for the application of deep learning models in a variety of complex tasks, thereby improving the performance and generalization ability of the model.

[0043] In some preferred implementations, in the sequence of calculating the local temporal correlation feature vectors of the edge resource load, the implicit clustering contribution factor of each local temporal correlation feature vector of the edge resource load relative to the self-supervised clustering representation vector of the local temporal correlation feature of the edge resource load is obtained to obtain a sequence of implicit clustering contribution factors of the local temporal correlation feature of the edge resource load, specifically:

[0044] Calculate the absolute value of the division of the corresponding positions between the local time series correlation feature vector of the edge resource load and the local time series correlation feature self-supervised clustering representation vector of the edge resource load to obtain the local time series self-supervised clustering interactive correlation feature vector of the edge resource load;

[0045] Calculate the logarithmic function value of the feature value of each position in the local temporal self-supervised clustering interaction correlation feature vector of the edge resource load with a constant of 2 as the base to obtain the local temporal self-supervised clustering interaction logarithmic representation vector of the edge resource load;

[0046] Taking each position feature value in the local time series self-supervised clustering interaction logarithm representation vector of the edge resource load as a weighting coefficient, calculating the weighted sum of each position feature value in the local time series correlation feature vector of the edge resource load, so as to obtain the edge resource load local time series correlation feature clustering contribution representation factor;

[0047] The exponential function value of the clustering contribution representation factor of the local temporal correlation feature of the edge resource load is calculated with the natural constant e as the base to obtain the implicit clustering contribution factor of the local temporal correlation feature of the edge resource load.

[0048] Furthermore, the edge resource load local time series correlation feature cluster contribution field distribution vector is input into the cluster contribution field explicit modeling module based on the self-attention mechanism to obtain the edge resource load local time series correlation feature cluster contribution field modulation weight vector, specifically:

[0049] Calculate the matrix multiplication between the edge resource load local time series correlation feature cluster contribution field distribution vector and the query weight matrix, key weight matrix and value weight matrix respectively to obtain the edge resource load cluster contribution field distribution query vector, the edge resource load cluster contribution field distribution key vector and the edge resource load cluster contribution field distribution value vector;

[0050] Calculate the multiplication between the edge resource load cluster distribution query vector and the transposed vector of the edge resource load cluster contribution field distribution key vector, use the multiplication result as the dividend, use the square root of the length of the edge resource load cluster contribution field distribution key vector as the divisor, perform departure processing according to the corresponding position, and obtain the edge resource load cluster contribution field semantic interaction representation matrix;

[0051] The semantic interaction representation matrix of the edge resource load cluster contribution field is normalized by the Softmax function, and the product between the distribution value vector of the edge resource load cluster contribution field and the normalization result is calculated to obtain the modulation weight vector of the edge resource load local temporal correlation feature cluster contribution field.

[0052] In some preferred implementations, the feature enhancement aggregation representation processing of cluster analysis can be performed on the sequence of local time series correlation feature vectors of the edge resource load according to the following feature aggregation formula to obtain the edge resource load time series feature, and the formula is specifically:

[0053] X={x 1 ,x 2 ,...,x k ,...,x n};

[0054]

[0055] v d ={d 1 ;d 2 ;...;d j ;...;d n};

[0056]

[0057] v q =W q v d ;

[0058] v k =W k v d ;

[0059] v v =W b v d ;

[0060]

[0061] Where X is the sequence of local time series correlation feature vectors of the edge resource load, x 1 ,x 2 ,xk ,x n are the first, second, kth and nth edge resource load local timing correlation feature vectors in the sequence of the edge resource load local timing correlation feature vectors, respectively, i is the i-th edge resource load local timing correlation feature vector in the sequence of edge resource load local timing correlation feature vectors, N is the number of feature vectors in the sequence of edge resource load local timing correlation feature vectors, x c is the self-supervised clustering representation vector of the local temporal correlation feature of the edge resource load, x j is the jth edge resource load local timing correlation feature vector in the sequence of edge resource load local timing correlation feature vectors, is the characteristic value of the kth position in the local time series correlation characteristic vector of the jth edge resource load, is the feature value of the kth position in the self-supervised clustering representation vector of the local temporal correlation feature of the edge resource load, log represents the logarithmic function value with base 2, exp represents the natural exponential function value, and d j is the implicit clustering contribution factor of the local temporal correlation feature of the edge resource load corresponding to the jth local temporal correlation feature vector of the edge resource load, v d The distribution vector of the clustering contribution field of the local time series correlation characteristics of the edge resource load obtained by quantizing the implicit clustering contribution factors of the local time series correlation characteristics of the edge resource load, W q , W k and W b are the query weight matrix, key weight matrix and value weight matrix respectively, v q 、v k and v v are the edge resource load cluster contribution field distribution query vector, the edge resource load cluster contribution field distribution key vector and the edge resource load cluster contribution field distribution value vector respectively, L is the length of the edge resource load cluster contribution field distribution key vector, Softmax(·) is the Softmax function, is vector multiplication, v w Contribute to the field modulation weight vector for clustering the local temporal correlation characteristics of edge resource loads, x f A significant aggregation representation vector of the edge resource load timing.

[0062] In some embodiments, a cluster analysis is performed on the sequence of local timing correlation feature vectors of the edge resource load to obtain a self-supervised clustering representation vector of the local timing correlation feature of the edge resource load, including: calculating the mean vector of the sequence of local timing correlation feature vectors of the edge resource load to obtain a self-supervised clustering representation vector of the local timing correlation feature of the edge resource load.

[0063] In some implementations, for the above step S104, based on the edge resource load timing characteristics and the cloud resource load timing characteristics, a detailed feature interaction based on spatial distance measurement is performed to obtain the resource load timing coordination characteristics, specifically:

[0064] Fine-grained feature decoupling is performed on the edge resource load time series significant aggregation representation vector according to a preset scale to obtain a set of edge resource load time series sub-component feature vectors; fine-grained feature decoupling is performed on the cloud resource load time series significant aggregation representation vector according to the preset scale to obtain a set of cloud resource load time series sub-component feature vectors; features can be compared at a lower level to capture more subtle feature differences. This means that the system not only takes into account the overall load level at the edge and cloud, but also notices small changes in the load characteristics, which is crucial for dynamically adjusting task scheduling.

[0065] The set of the edge resource load timing sub-component feature vectors and the set of the cloud resource load timing sub-component feature vectors are subjected to (optimal) feature sub-component pairing screening and interactive fusion processing to obtain the edge-cloud (optimal) resource load timing collaborative feature.

[0066] In this embodiment, the effectiveness of feature representation can be enhanced through a detailed feature interaction method based on spatial distance measurement, thereby improving the accuracy of task allocation and dynamic scheduling decisions, and better integrating and interacting the edge resource load timing characteristics and the cloud resource load timing characteristics. In practical applications, the resource load status of the edge and cloud continues to change over time, and these changes may be very subtle and easily ignored, resulting in reduced accuracy of task allocation and dynamic scheduling decisions. Based on this, in the technical solution of the present application, the edge resource load timing characteristics and the cloud resource load timing characteristics are further subjected to detailed feature interaction based on spatial distance measurement to obtain edge-cloud resource load timing collaborative characteristics.

[0067] Furthermore, the set of the edge resource load time series sub-component feature vectors and the set of the cloud resource load time series sub-component feature vectors are subjected to feature sub-component pairing screening and interactive fusion processing to obtain resource load time series collaborative features, specifically:

[0068] Calculate the hyperbolic space distance measurement factor between the time series sub-component feature vector of each edge resource load and the time series sub-component feature vector of each cloud resource load;

[0069] According to the hyperbolic space distance measurement factor, the (best) characteristic sub-component pairing of each edge resource load timing sub-component characteristic vector and each cloud resource load timing sub-component characteristic vector is screened to obtain an optimal {edge resource load timing sub-component characteristic vector, cloud resource load timing sub-component characteristic vector} pairing set;

[0070] Pair each optimal {edge resource load timing sub-component feature vector, cloud resource load timing sub-component feature vector} in the pairing set and input it into the feature multi-dimensional interaction module to obtain a set of (edge-cloud optimal) load timing sub-component feature pairing interaction fusion feature vectors;

[0071] The set of the load timing sub-component feature pairing interactive fusion feature vectors is cascaded to obtain the (edge-cloud resource) resource load timing collaborative feature.

[0072] In this embodiment, after feature decoupling, the hyperbolic space distance measurement factor between different sub-component feature vectors is calculated, and the use of hyperbolic space distance measurement rather than traditional Euclidean distance or other measurement methods can more effectively represent situations with hierarchical structures or high-dimensional sparse data. Since the hyperbolic space grows faster as it moves away from the origin, this measurement method can better maintain the hierarchical structure of the data and is more suitable for processing data distributions with exponential growth characteristics, which is crucial for capturing the fine-grained relationship between the cloud resource load timing characteristics and the edge resource load timing characteristics, especially when dealing with complex or nonlinear relationships, so as to more accurately reflect the actual edge and cloud load coordination and dynamic change relationship. Then, based on the hyperbolic space distance measurement factor, the edge resource load timing sub-component feature vector and the cloud resource load timing sub-component feature vector are optimally paired, with the aim of finding a pair of sub-component feature vectors with the smallest distance in the hyperbolic space, thereby achieving the optimal match. The key to this step is to ensure that each pair of cloud resource load timing sub-components and edge resource load timing sub-components matches the best representation of the similarity between the two feature vector sets, so that even very subtle feature differences will not be ignored, and the subsequent fusion process will retain the most valuable information as much as possible. After the best pairing, these pairs are input into the feature multi-dimensional interaction module. This module is responsible for interacting between the paired feature vectors to generate feature pairing interaction fusion feature vectors. The interaction process may include but is not limited to addition, multiplication or other forms of combination operations, and its goal is to create new feature representations that can better reflect the complex relationship between the cloud resource load timing and the edge resource load timing in the original data. This step enhances the expressive power of the features and enables the model to learn deeper feature associations. Finally, the edge-cloud resource load optimal sub-component feature pairing interaction fusion feature vectors generated by all interactions are cascaded, that is, merged into a single edge-cloud resource load fine-grained optimal alignment fusion feature vector according to a certain order or rule. The purpose of cascade fusion is to integrate the information after all edge-cloud resource load sub-component interactions to form a comprehensive and optimized feature representation. The feature vector processed in this way not only contains the information of the original feature vector, but also integrates the new information generated by fine-grained matching and multi-dimensional interaction, so that it can characterize the data more comprehensively and improve the accuracy and efficiency of task scheduling.

[0073] Furthermore, the hyperbolic space distance measurement factor between the time series sub-component feature vector of each edge resource load and the time series sub-component feature vector of each cloud resource load is calculated, specifically:

[0074] Extracting the edge resource load timing sub-component feature vector at a preset position from the set of edge resource load timing sub-component feature vectors;

[0075] Extracting a cloud resource load timing sub-component feature vector at a preset position from the set of cloud resource load timing sub-component feature vectors;

[0076] Calculate the difference vector between the edge resource load timing sub-component feature vector at the preset position and the cloud resource load timing sub-component feature vector at the preset position, and perform square processing based on a norm of the difference vector to obtain a norm representation of the semantic difference of the load timing sub-component;

[0077] Calculate the square of a norm of a characteristic vector of a time series subcomponent of the edge resource load at the preset position to obtain a first square result; perform subtraction processing on a first constant (which may be a constant 1) and the first square result according to corresponding positions to obtain a norm representation of a time series subcomponent of the edge resource load;

[0078] Calculate the square of the first norm of the cloud resource load time series sub-component feature vector at the preset position to obtain a second square result; perform subtraction processing on the second square result according to the corresponding position by a second constant (which may be a constant 1) to obtain a first norm representation of the cloud resource load time series sub-component;

[0079] Calculate the multiplication between the one-norm representation of the edge resource load timing subcomponent and the one-norm representation of the cloud resource load timing subcomponent, use the multiplication result as the divisor, use the one-norm semantic difference of the load timing subcomponent as the dividend, perform division processing, and obtain the one-norm semantic implicit association representation of the load timing;

[0080] The load timing one-norm semantic implicit association representation is multiplied by a third constant (which may be constant 2), the product is added to a fourth constant (which may be constant 1), and the result of the addition is processed by an inverse hyperbolic cosine function to obtain the hyperbolic space distance measurement factor.

[0081] In some implementations, detailed feature interaction based on spatial distance metric is performed on the edge resource load timing feature and the cloud resource load timing feature to obtain the edge-cloud resource load timing coordination feature, which can be implemented according to the following formula:

[0082] Decouple(v 1 )={v 11 ,v 12 ,...,v 1i ,...,v 1n};

[0083] Decouple(v2 )={v 21 ,v 22 ,...,v 2i ,...,v 2n};

[0084] k=arg min j .d P (v 1i ,v 2j ) / ;

[0085]

[0086] v f =[v p1 ;v p2 ;...v pn ];

[0087] Among them, v 1 is the edge resource load time series significant aggregation representation vector, v 2 is the significant aggregation representation vector of the cloud resource load time series, Decouple is the fine-grained feature decoupling operation, and v 11 ,v 12 ,v 1i ,v 1n are the first, second, i-th and n-th edge resource load timing sub-component feature vectors in the set of edge resource load timing sub-component feature vectors, respectively, v 21 ,v 22 ,v 2i ,v 2n are the first, second, i-th and n-th cloud resource load time series sub-component feature vectors in the set of cloud resource load time series sub-component feature vectors, ‖·‖ represents the one-norm of the vector, arccosh is the inverse hyperbolic cosine function, d P (v 1i ,v 2j ) is v 1i and v 2j The hyperbolic space distance metric between j .d P (v 1i ,v 2j ) / is to take d P (v 1i ,v 2j ) is the j parameter corresponding to the minimum value, k is the optimal pairing position value, v 2k is the kth cloud resource load timing sub-component feature vector in the set of cloud resource load timing sub-component feature vectors, ⊙ and are positional subtraction, positional dot multiplication, and positional addition, α, β, and γ are weighted hyperparameters, and v pi is the i-th edge-cloud resource load timing optimal sub-component feature pairing interactive fusion feature vector in the set of edge-cloud resource load timing optimal sub-component feature pairing interactive fusion feature vectors, [·;·;·] is the cascade operation of the vector, v f A representation vector for the edge-cloud resource load timing coordination alignment.

[0088] In some embodiments, according to the hyperbolic space distance measurement factor, each edge resource load timing sub-component feature vector and each cloud resource load timing sub-component feature vector are subjected to (optimal) feature sub-component pairing screening to obtain an optimal {edge resource load timing sub-component feature vector, cloud resource load timing sub-component feature vector} pairing set, specifically: selecting the position of the cloud resource load timing sub-component feature vector corresponding to the minimum value of multiple hyperbolic space distance measurement factors as the optimal pairing position value of the edge resource load timing sub-component feature vector at the preset position; pairing the {edge resource load timing sub-component feature vector, cloud resource load timing sub-component feature vector} corresponding to the optimal pairing position value to obtain the optimal {edge resource load timing sub-component feature vector, cloud resource load timing sub-component feature vector} pairing.

[0089] In some implementations, the inputting of each pairing in the pairing set into a feature multi-dimensional interaction module to obtain a set of load time series sub-component feature pairing interaction fusion feature vectors is specifically:

[0090] The edge resource load timing subcomponent feature vector and the cloud resource load timing subcomponent feature vector in the optimal {edge resource load timing subcomponent feature vector, cloud resource load timing subcomponent feature vector} pairing are respectively subjected to positional difference, positional dot multiplication and positional summation processing to obtain the edge-cloud resource load timing subcomponent differential feature vector, the edge-cloud resource load timing subcomponent dot product feature vector and the edge-cloud resource load timing subcomponent sum feature vector; the weighted sum of the corresponding positions between the edge-cloud resource load timing subcomponent differential feature vector, the edge-cloud resource load timing subcomponent dot product feature vector and the edge-cloud resource load timing subcomponent sum feature vector is calculated to obtain the edge-cloud resource load timing optimal subcomponent feature pairing interactive fusion feature vector.

[0091] Furthermore, for the above step S105, according to the resource load timing coordination feature, the task to be allocated is split to obtain a first power calculation subtask to be allocated and a second power calculation subtask to be allocated, and the first power calculation subtask to be allocated and the second power calculation subtask to be allocated are respectively scheduled to the edge end and the cloud end, specifically:

[0092] Inputting the resource load time series coordination feature into a classifier-based task splitting module to obtain a first power calculation subtask to be allocated and a second power calculation subtask to be allocated;

[0093] The first power calculation subtask to be allocated and the second power calculation subtask to be allocated are allocated to the edge and the cloud, respectively.

[0094] Exemplarily, the first power calculation subtask to be allocated and the second power calculation subtask to be allocated may correspond to a first ratio and a second ratio, respectively, and the sum of the first ratio and the second ratio is one. In this way, this implementation can achieve more intelligent dynamic task scheduling to ensure load balancing and system stability between the edge network and the cloud in the cloud-edge system, and help optimize the task processing efficiency of the entire system.

[0095] In some implementations, the resource load time series coordination feature is input into a classifier-based task splitting module to obtain a first power calculation subtask to be allocated and a second power calculation subtask to be allocated, specifically:

[0096] Determine the maximum eigenvalue and the minimum eigenvalue of the edge-cloud resource load timing coordination alignment representation vector, and calculate the timing coordination alignment mean and the timing coordination alignment standard deviation of the feature set of the edge-cloud resource load timing coordination alignment representation vector;

[0097] Calculating a quotient of the timing collaborative alignment mean and the timing collaborative alignment standard deviation to obtain a timing collaborative alignment statistical normalization value;

[0098] Calculate the inverse of each eigenvalue of the edge-cloud resource load timing collaborative alignment representation vector, multiply it by the difference between the maximum eigenvalue of the timing collaborative alignment and the minimum eigenvalue of the timing collaborative alignment, and then subtract it from the timing collaborative alignment statistical normalization value to obtain a timing collaborative alignment distribution approximation vector;

[0099] Calculate an exponential function with a natural constant as the base and each eigenvalue of the time series collaborative alignment distribution approximation vector as the exponent to obtain a time series collaborative alignment distribution class approximation vector;

[0100] The timing collaborative alignment distribution class approximation vector is dot-added with the timing collaborative alignment statistical normalization value, and the base-two logarithm of the absolute value of each eigenvalue of the dot-added vector is calculated to obtain an optimized edge-cloud resource load timing collaborative alignment representation vector; and the optimized edge-cloud resource load timing collaborative alignment representation vector is input into a classifier-based task segmentation module to obtain a segmentation result, namely, the first power calculation subtask to be allocated and the second power calculation subtask to be allocated mentioned above.

[0101] The resource load timing coordination alignment representation vector, denoted as V, is optimally represented as:

[0102]

[0103]

[0104]

[0105] δ=v max -v min ;

[0106] η is the statistical normalized value of the time series collaborative alignment, δ is the difference between the maximum eigenvalue of the time series collaborative alignment and the minimum eigenvalue of the time series collaborative alignment, log is the logarithm with base 2, v max is the maximum eigenvalue of the timing coordination alignment representation vector of the edge-cloud resource load timing coordination alignment, v min is the minimum eigenvalue of the timing coordination alignment representation vector of the edge-cloud resource load timing coordination alignment, μ is the mean of the timing coordination alignment, σ is the standard deviation of the timing coordination alignment, V ⊙-1 is the reciprocal of each eigenvalue of the edge-cloud resource load timing coordination alignment representation vector, exp(·) is an exponential function calculated with a natural constant as the base and each eigenvalue of the vector as the exponent, V e is the time series coordinated alignment distribution class approximation vector, log is the logarithmic value with base 2, V ′ It is an optimized edge-cloud resource load timing coordination alignment representation vector. It is added by location point. is point-wise subtraction, and ⊙ is point-wise multiplication.

[0107] In implementing this preferred embodiment, considering that there are distribution differences between the time queue of the resource load value at the edge and the time queue of the resource load value in the time series source domain, after performing local time series feature extraction and local feature time series aggregation, the local fine-grained difference in features between the edge resource load time series significant aggregation representation vector and the cloud resource load time series significant aggregation representation vector will be amplified. In this way, when it is input into a feature interaction network based on optimal fine-grained feature matching, the obtained edge-cloud resource load time series collaborative alignment representation vector will also have a diversified set expression distribution due to the local fine-grained feature matching offset. Therefore, it is expected to improve the balance between the accuracy and completeness of the regression mapping when the edge-cloud resource load time series collaborative alignment representation vector is input into the task segmentation module based on the classifier for class regression, thereby improving the accuracy of the segmentation results.

[0108] Therefore, by performing random statistical normalization on the diverse feature set of the edge-cloud resource load timing collaborative alignment representation vector, the confidence space constructed based on the overall feature value of the edge-cloud resource load timing collaborative alignment representation vector is approximated by the standardized continuous probability density distribution relative to the response hypothesis test of each feature value of the edge-cloud resource load timing collaborative alignment representation vector, thereby establishing the target reachability of the diverse feature distribution of the edge-cloud resource load timing collaborative alignment representation vector to the unified regression target, so as to achieve the balanced executableness between mapping accuracy and mapping completeness in the class regression process based on the diverse feature distribution of the edge-cloud resource load timing collaborative alignment representation vector, and improve the accuracy of the segmentation result obtained by the task segmentation module based on the classifier of the edge-cloud resource load timing collaborative alignment representation vector input. In this way, it is possible to more accurately split a large task into multiple subtasks according to the collaborative association and dynamic changes between the resource load value of the edge and the resource load value of the cloud, thereby achieving more intelligent task dynamic scheduling, so as to ensure the load balancing and system stability of the edge network and the cloud in the cloud-edge system.

[0109] In summary, the above scheme is adopted to collect the resource load values ​​of the edge and the cloud in real time through real-time monitoring, and then introduce data processing and analysis algorithms based on artificial intelligence and deep learning in the back end to perform time series collaborative analysis and feature interaction on the resource load values ​​of the edge and the cloud, so as to learn and capture the time series collaborative changes and dynamic correlation interaction characteristics between the resource loads of the edge and the cloud, so as to obtain the proportion allocated to the edge and the cloud, and then split and allocate the total power computing tasks to be allocated based on the proportion. In this way, more intelligent dynamic scheduling of tasks can be achieved to ensure load balancing and system stability of the edge network and the cloud in the cloud-edge system, which helps to optimize the task processing efficiency of the entire system.

[0110] Please refer to Figure 2 The embodiment of the present invention further provides a task scheduling device 200 based on cloud-edge collaboration, including a task acquisition module 201, a queue acquisition module 202, a feature extraction module 203, a feature interaction module 204 and a scheduling module 205; wherein,

[0111] The task acquisition module 201 is used to acquire the task to be assigned from the edge end, and the task to be assigned is a power calculation task;

[0112] The queue acquisition module 202 is used to acquire the time queue of the resource load value of the edge end and the time queue of the resource load value of the cloud end;

[0113] The feature extraction module 203 is used to extract the time series features of the edge resource load based on the time series of the edge resource load value; and to extract the time series features of the cloud resource load based on the time series of the cloud resource load value;

[0114] The feature interaction module 204 is used to perform detailed feature interaction based on spatial distance measurement based on the edge resource load time series feature and the cloud resource load time series feature to obtain the resource load time series coordination feature;

[0115] The scheduling module 205 is used to split the task to be allocated according to the resource load timing coordination characteristics to obtain a first power calculation subtask to be allocated and a second power calculation subtask to be allocated, and schedule the first power calculation subtask to be allocated and the second power calculation subtask to be allocated to the edge and the cloud respectively, so that the edge and the cloud respectively execute the first power calculation subtask to be allocated and the second power calculation subtask to be allocated.

[0116] As a preferred solution, the feature extraction module 203 extracts the time series features of the edge resource load based on the time series of the edge resource load value; and extracts the time series features of the cloud resource load based on the time series of the cloud resource load value, including:

[0117] The feature extraction module 203 inputs the edge resource load timing features and the cloud resource load timing features into the 1D-CNN model to obtain a sequence of edge resource load local timing correlation feature vectors and a sequence of cloud resource load local timing correlation feature vectors;

[0118] Performing feature enhancement aggregation representation processing of cluster analysis on the sequence of local time series correlation feature vectors of edge resource load to obtain the time series feature of the edge resource load; performing feature enhancement aggregation representation processing of cluster analysis on the sequence of local time series correlation feature vectors of the cloud resource load to obtain the time series feature of the cloud resource load;

[0119] The feature enhancement aggregation representation processing of clustering analysis on the sequence of local time series correlation feature vectors of edge resource load to obtain the edge resource load time series feature is specifically:

[0120] A cluster analysis is performed on the sequence of local timing correlation feature vectors of the edge resource load to obtain a self-supervised clustering representation vector of the local timing correlation feature of the edge resource load; based on the self-supervised clustering representation vector of the local timing correlation feature of the edge resource load, feature clustering field modulation aggregation is performed on the sequence of local timing correlation feature vectors of the edge resource load to obtain a significant aggregation representation vector of the edge resource load timing as the edge resource load timing feature.

[0121] As a preferred solution, the feature extraction module 203 performs feature clustering field modulation aggregation on the sequence of the local time series correlation feature vectors of the edge resource load based on the self-supervised clustering representation vector of the local time series correlation feature of the edge resource load, so as to obtain the edge resource load time series significant aggregation representation vector as the edge resource load time series feature, which is specifically:

[0122] The feature extraction module 203 calculates the implicit clustering contribution factor of each edge resource load local time series correlation feature vector in the sequence of the edge resource load local time series correlation feature vector relative to the edge resource load local time series correlation feature self-supervised clustering representation vector to obtain a sequence of implicit clustering contribution factors of the edge resource load local time series correlation feature;

[0123] Vectorizing and arranging the sequence of implicit clustering contribution factors of the local time series correlation characteristics of the edge resource load to obtain a distribution vector of the clustering contribution field of the local time series correlation characteristics of the edge resource load;

[0124] Input the distribution vector of the cluster contribution field of the local temporal correlation feature of the edge resource load into the cluster contribution field explicit modeling module based on the self-attention mechanism to obtain the modulation weight vector of the cluster contribution field of the local temporal correlation feature of the edge resource load;

[0125] Taking each eigenvalue in the field modulation weight vector of the local timing correlation feature cluster contribution of the edge resource load as the weight, the weighted sum of the corresponding positions of the sequence of the local timing correlation feature vector of the edge resource load is calculated to obtain the edge resource load timing significant aggregation representation vector.

[0126] As a preferred solution, the feature extraction module 203 calculates the implicit clustering contribution factor of each edge resource load local time series correlation feature vector relative to the edge resource load local time series correlation feature self-supervised clustering representation vector in the sequence of the edge resource load local time series correlation feature vectors to obtain a sequence of implicit clustering contribution factors of the edge resource load local time series correlation feature, specifically:

[0127] The feature extraction module 203 calculates the absolute value of the division of the edge resource load local time series correlation feature vector and the edge resource load local time series correlation feature self-supervised clustering representation vector at the corresponding position to obtain the edge resource load local time series self-supervised clustering interactive correlation feature vector;

[0128] Calculate the logarithmic function value of the feature value of each position in the local temporal self-supervised clustering interaction correlation feature vector of the edge resource load with base 2 to obtain the local temporal self-supervised clustering interaction logarithmic representation vector of the edge resource load;

[0129] Taking each position feature value in the local time series self-supervised clustering interaction logarithm representation vector of the edge resource load as a weighting coefficient, calculating the weighted sum of each position feature value in the local time series correlation feature vector of the edge resource load, so as to obtain the edge resource load local time series correlation feature clustering contribution representation factor;

[0130] The exponential function value of the clustering contribution representation factor of the local temporal correlation feature of the edge resource load is calculated with the natural constant e as the base to obtain the implicit clustering contribution factor of the local temporal correlation feature of the edge resource load.

[0131] As a preferred solution, the feature extraction module 203 inputs the edge resource load local temporal correlation feature cluster contribution field distribution vector into the cluster contribution field explicit modeling module based on the self-attention mechanism to obtain the edge resource load local temporal correlation feature cluster contribution field modulation weight vector, specifically:

[0132] The feature extraction module 203 calculates the matrix multiplication between the edge resource load local time series correlation feature cluster contribution field distribution vector and the query weight matrix, the key weight matrix and the value weight matrix respectively, so as to obtain the edge resource load cluster contribution field distribution query vector, the edge resource load cluster contribution field distribution key vector and the edge resource load cluster contribution field distribution value vector;

[0133] Calculate the multiplication between the edge resource load cluster distribution query vector and the transposed vector of the edge resource load cluster contribution field distribution key vector, use the multiplication result as the dividend, use the square root of the length of the edge resource load cluster contribution field distribution key vector as the divisor, perform departure processing according to the corresponding position, and obtain the edge resource load cluster contribution field semantic interaction representation matrix;

[0134] The semantic interaction representation matrix of the edge resource load cluster contribution field is normalized by the Softmax function, and the product between the distribution value vector of the edge resource load cluster contribution field and the normalization result is calculated to obtain the modulation weight vector of the edge resource load local temporal correlation feature cluster contribution field.

[0135] As a preferred solution, the feature interaction module 204 performs detailed feature interaction based on spatial distance measurement based on the edge resource load timing characteristics and the cloud resource load timing characteristics to obtain resource load timing coordination characteristics, specifically:

[0136] The feature interaction module 204 performs fine-grained feature decoupling on the edge resource load time series significant aggregation representation vector according to a preset scale to obtain a set of edge resource load time series sub-component feature vectors; performs fine-grained feature decoupling on the cloud resource load time series significant aggregation representation vector according to the preset scale to obtain a set of cloud resource load time series sub-component feature vectors;

[0137] The set of the edge resource load timing sub-component feature vectors and the set of the cloud resource load timing sub-component feature vectors are subjected to feature sub-component pairing screening and interactive fusion processing to obtain resource load timing coordination features.

[0138] As a preferred solution, the feature interaction module 204 performs feature sub-component pairing screening and interactive fusion processing on the set of the edge resource load timing sub-component feature vectors and the set of the cloud resource load timing sub-component feature vectors to obtain the resource load timing collaborative feature, specifically:

[0139] The feature interaction module 204 calculates the hyperbolic space distance measurement factor between each edge resource load time series sub-component feature vector and each cloud resource load time series sub-component feature vector;

[0140] According to the hyperbolic space distance measurement factor, the characteristic sub-component pairing and screening of each edge resource load time series sub-component characteristic vector and each cloud resource load time series sub-component characteristic vector are performed to obtain a pairing set;

[0141] Input each pair in the pairing set into a feature multi-dimensional interaction module to obtain a set of load time series sub-component feature pairing interaction fusion feature vectors;

[0142] The set of the load time series sub-component feature pairing interactive fusion feature vectors is cascaded to obtain the resource load time series coordination feature.

[0143] As a preferred solution, the feature interaction module 204 calculates the hyperbolic space distance measurement factor between each edge resource load time series sub-component feature vector and each cloud resource load time series sub-component feature vector, specifically:

[0144] The feature interaction module 204 extracts the edge resource load timing sub-component feature vector at a preset position from the set of edge resource load timing sub-component feature vectors;

[0145] Extracting a cloud resource load timing sub-component feature vector at a preset position from the set of cloud resource load timing sub-component feature vectors;

[0146] Calculate the difference vector between the edge resource load timing sub-component feature vector at the preset position and the cloud resource load timing sub-component feature vector at the preset position, and perform square processing based on a norm of the difference vector to obtain a norm representation of the semantic difference of the load timing sub-component;

[0147] Calculate the square of a norm of a characteristic vector of a time series subcomponent of the edge resource load at the preset position to obtain a first square result; perform subtraction processing on a first constant and the first square result according to corresponding positions to obtain a norm representation of a time series subcomponent of the edge resource load;

[0148] Calculate the square of a norm of a characteristic vector of a cloud resource load timing subcomponent at the preset position to obtain a second square result; perform subtraction processing on a second constant and the second square result according to corresponding positions to obtain a norm representation of a cloud resource load timing subcomponent;

[0149] Calculate the multiplication between the one-norm representation of the edge resource load timing subcomponent and the one-norm representation of the cloud resource load timing subcomponent, use the multiplication result as the divisor, use the one-norm semantic difference of the load timing subcomponent as the dividend, perform division processing, and obtain the one-norm semantic implicit association representation of the load timing;

[0150] The load time series one-norm semantic implicit association representation is multiplied by a third constant, the product is added to a fourth constant, and the result of the addition is processed by an inverse hyperbolic cosine function to obtain the hyperbolic space distance measurement factor.

[0151] As a preferred solution, the scheduling module 205 splits the task to be allocated according to the resource load timing coordination feature to obtain a first power calculation subtask to be allocated and a second power calculation subtask to be allocated, and schedules the first power calculation subtask to be allocated and the second power calculation subtask to be allocated to the edge and the cloud, respectively, as follows:

[0152] The scheduling module 205 inputs the resource load time series coordination feature into the classifier-based task splitting module to obtain a first power calculation subtask to be allocated and a second power calculation subtask to be allocated;

[0153] The first power calculation subtask to be allocated and the second power calculation subtask to be allocated are allocated to the edge and the cloud, respectively.

[0154] In addition, you can refer to Figure 3 , which shows a schematic diagram of the structure of an electronic device 600 suitable for implementing the embodiment of the present application. The terminal device in the embodiment of the present application may include but is not limited to mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 3 The electronic device shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0155] like Figure 3 As shown, the electronic device 600 may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 601, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 602 or a program loaded from a storage device 608 into a random access memory (RAM) 603. In the RAM 603, various programs and data required for the operation of the electronic device 600 are also stored. The processing device 601, the ROM 602, and the RAM 603 are connected to each other via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0156] Typically, the following devices may be connected to the I / O interface 605: an input device 606 including, for example, a touch screen, a touch pad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; an output device 607 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 608 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 609. The communication device 609 may allow the electronic device 600 to communicate with other devices wirelessly or by wire to exchange data. Although Figure 3The electronic device 600 is shown with various devices, but it should be understood that it is not required to implement or possess all the devices shown. More or fewer devices may be implemented or possessed instead.

[0157] In particular, according to an embodiment of the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present application includes a computer program product, which includes a computer program carried on a non-transitory computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication device 609, or installed from the storage device 608, or installed from the ROM 602. When the computer program is executed by the processing device 601, the above-mentioned functions defined in the method of the embodiment of the present application are executed.

[0158] It should be noted that the computer-readable medium mentioned above in the present application may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, device or device. In the present application, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries a computer-readable program code. This propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The computer readable signal medium may also be any computer readable medium other than a computer readable storage medium, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer readable medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

[0159] In some embodiments, the client and the server may communicate using any currently known or future developed network protocol such as HTTP (HyperText Transfer Protocol), and may be interconnected with any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network ("LAN"), a wide area network ("WAN"), an internet (e.g., the Internet), and a peer-to-peer network (e.g., an ad hoc peer-to-peer network), as well as any currently known or future developed network.

[0160] The computer-readable medium may be included in the electronic device, or may exist independently without being incorporated into the electronic device.

[0161] Computer program code for performing the operations of the present application may be written in one or more programming languages ​​or a combination thereof, including, but not limited to, object-oriented programming languages, such as Java, Smalltalk, C++, and conventional procedural programming languages, such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0162] The functions described above herein may be performed at least in part by one or more hardware logic components. For example, without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), and the like.

[0163] In the context of the present application, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0164] Figure 4 FIG. 1 is an application scenario diagram of a task scheduling method based on cloud-edge collaboration according to an exemplary embodiment. Figure 4 As shown, in this application scenario, first, the edge end (the edge gateway shown in the figure) obtains the power calculation task to be allocated (for example, Figure 4 C1 shown in FIG); obtaining a time queue of resource load values ​​of the edge end (for example, Figure 4 C2 shown in FIG); obtaining a time queue of resource load values ​​in the cloud (for example, Figure 4 Then, the obtained power calculation task to be allocated, the time queue of the resource load value of the edge end, and the time queue of the resource load value of the cloud end are input to a server deployed with a task scheduling based on cloud-edge collaboration (for example, Figure 4 In S) shown in the figure, the server is able to split the power calculation task to be allocated based on cloud-edge collaborative task scheduling to obtain a first power calculation subtask to be allocated and a second power calculation subtask to be allocated, and schedule the first power calculation subtask to be allocated and the second power calculation subtask to be allocated to the edge and the cloud respectively, so that the edge and the cloud respectively execute the first power calculation subtask to be allocated and the second power calculation subtask to be allocated.

[0165] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0166] The present invention application provides a task scheduling method and device based on cloud-edge collaboration. By extracting the timing characteristics of edge resource load and cloud resource load, the resource load conditions of the cloud and edge are taken into account, and then detailed feature interaction based on spatial distance measurement is performed to obtain resource load timing collaboration characteristics, thereby realizing flexible task splitting and scheduling. Compared with the existing technology, it can realize load balancing of power computing tasks between edge and cloud, effectively integrate the advantages of edge and cloud, improve the stability of the overall system operation and the intelligence of task scheduling, and improve task processing efficiency.

[0167] The specific embodiments described above further illustrate the purpose, technical solutions and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. It is particularly pointed out that for those skilled in the art, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the scope of protection of the present invention.

Claims

1. A task scheduling method based on cloud-edge collaboration, characterized in that: include: Acquire a task to be assigned from an edge end, where the task to be assigned is a power calculation task; Obtaining a time queue of resource load values ​​at the edge and a time queue of resource load values ​​at the cloud; Extracting edge resource load timing characteristics based on the time queue of the edge resource load value; Extracting cloud resource load timing characteristics based on the time queue of the cloud resource load value; Based on the time series characteristics of edge resource load and cloud resource load, detailed feature interaction based on spatial distance measurement is performed to obtain resource load time series coordination characteristics; According to the resource load timing coordination characteristics, the task to be allocated is split to obtain a first power calculation subtask to be allocated and a second power calculation subtask to be allocated. The first power calculation subtask to be allocated and the second power calculation subtask to be allocated are respectively scheduled to the edge and the cloud, so that the edge and the cloud respectively execute the first power calculation subtask to be allocated and the second power calculation subtask to be allocated.

2. The task scheduling method based on cloud-edge collaboration according to claim 1, characterized in that: The time queue based on the resource load value of the edge end is used to extract the time series characteristics of the edge end resource load; Based on the time queue of the cloud resource load value, extracting the cloud resource load time series characteristics, including: Input the edge resource load timing characteristics and the cloud resource load timing characteristics into the 1D-CNN model to obtain a sequence of edge resource load local timing correlation feature vectors and a sequence of cloud resource load local timing correlation feature vectors; Performing feature enhancement aggregation representation processing of cluster analysis on the sequence of local time series correlation feature vectors of edge resource load to obtain the time series feature of the edge resource load; performing feature enhancement aggregation representation processing of cluster analysis on the sequence of local time series correlation feature vectors of the cloud resource load to obtain the time series feature of the cloud resource load; The feature enhancement aggregation representation processing of clustering analysis on the sequence of local time series correlation feature vectors of edge resource load to obtain the edge resource load time series feature is specifically: A cluster analysis is performed on the sequence of local timing correlation feature vectors of the edge resource load to obtain a self-supervised clustering representation vector of the local timing correlation feature of the edge resource load; based on the self-supervised clustering representation vector of the local timing correlation feature of the edge resource load, feature clustering field modulation aggregation is performed on the sequence of local timing correlation feature vectors of the edge resource load to obtain a significant aggregation representation vector of the edge resource load timing as the edge resource load timing feature.

3. The task scheduling method based on cloud-edge collaboration as claimed in claim 2, characterized in that: The self-supervised clustering representation vector based on the local time series correlation feature of the edge resource load is used to perform feature clustering field modulation aggregation on the sequence of the local time series correlation feature vector of the edge resource load to obtain the edge resource load time series significant aggregation representation vector as the edge resource load time series feature, which is specifically: Calculate the implicit clustering contribution factor of each edge resource load local time series correlation feature vector in the sequence of the edge resource load local time series correlation feature vector relative to the edge resource load local time series correlation feature self-supervised clustering representation vector to obtain a sequence of implicit clustering contribution factors of the edge resource load local time series correlation feature; Vectorizing and arranging the sequence of implicit clustering contribution factors of the local time series correlation characteristics of the edge resource load to obtain a distribution vector of the clustering contribution field of the local time series correlation characteristics of the edge resource load; Input the distribution vector of the cluster contribution field of the local temporal correlation feature of the edge resource load into the cluster contribution field explicit modeling module based on the self-attention mechanism to obtain the modulation weight vector of the cluster contribution field of the local temporal correlation feature of the edge resource load; Taking each eigenvalue in the field modulation weight vector of the local timing correlation feature cluster contribution of the edge resource load as the weight, the weighted sum of the corresponding positions of the sequence of the local timing correlation feature vector of the edge resource load is calculated to obtain the edge resource load timing significant aggregation representation vector.

4. The task scheduling method based on cloud-edge collaboration as claimed in claim 3, characterized in that: In the sequence of calculating the local temporal correlation feature vectors of the edge resource load, the implicit clustering contribution factor of each local temporal correlation feature vector of the edge resource load relative to the self-supervised clustering representation vector of the local temporal correlation feature of the edge resource load is obtained to obtain a sequence of implicit clustering contribution factors of the local temporal correlation feature of the edge resource load, specifically: Calculate the absolute value of the division of the corresponding positions between the local time series correlation feature vector of the edge resource load and the local time series correlation feature self-supervised clustering representation vector of the edge resource load to obtain the local time series self-supervised clustering interactive correlation feature vector of the edge resource load; Calculate the logarithmic function value of the feature value of each position in the local temporal self-supervised clustering interaction correlation feature vector of the edge resource load with base 2 to obtain the local temporal self-supervised clustering interaction logarithmic representation vector of the edge resource load; Taking each position feature value in the local time series self-supervised clustering interaction logarithm representation vector of the edge resource load as a weighting coefficient, calculating the weighted sum of each position feature value in the local time series correlation feature vector of the edge resource load, so as to obtain the edge resource load local time series correlation feature clustering contribution representation factor; The exponential function value of the clustering contribution representation factor of the local temporal correlation feature of the edge resource load is calculated with the natural constant e as the base to obtain the implicit clustering contribution factor of the local temporal correlation feature of the edge resource load.

5. The task scheduling method based on cloud-edge collaboration as claimed in claim 4, characterized in that: The edge resource load local time series correlation feature cluster contribution field distribution vector is input into the cluster contribution field explicit modeling module based on the self-attention mechanism to obtain the edge resource load local time series correlation feature cluster contribution field modulation weight vector, specifically: Calculate the matrix multiplication between the edge resource load local time series correlation feature cluster contribution field distribution vector and the query weight matrix, key weight matrix and value weight matrix respectively to obtain the edge resource load cluster contribution field distribution query vector, the edge resource load cluster contribution field distribution key vector and the edge resource load cluster contribution field distribution value vector; Calculate the multiplication between the edge resource load cluster distribution query vector and the transposed vector of the edge resource load cluster contribution field distribution key vector, use the multiplication result as the dividend, use the square root of the length of the edge resource load cluster contribution field distribution key vector as the divisor, perform departure processing according to the corresponding position, and obtain the edge resource load cluster contribution field semantic interaction representation matrix; The semantic interaction representation matrix of the edge resource load cluster contribution field is normalized by the Softmax function, and the product between the distribution value vector of the edge resource load cluster contribution field and the normalization result is calculated to obtain the modulation weight vector of the edge resource load local temporal correlation feature cluster contribution field.

6. The task scheduling method based on cloud-edge collaboration as claimed in claim 5, characterized in that: Based on the time series characteristics of the edge resource load and the cloud resource load, a detailed feature interaction based on the spatial distance measurement is performed to obtain the resource load time series coordination characteristics, which are specifically: Perform fine-grained feature decoupling on the edge resource load time series significant aggregation representation vector according to a preset scale to obtain a set of edge resource load time series sub-component feature vectors; perform fine-grained feature decoupling on the cloud resource load time series significant aggregation representation vector according to the preset scale to obtain a set of cloud resource load time series sub-component feature vectors; The set of the edge resource load timing sub-component feature vectors and the set of the cloud resource load timing sub-component feature vectors are subjected to feature sub-component pairing screening and interactive fusion processing to obtain resource load timing coordination features.

7. The task scheduling method based on cloud-edge collaboration as claimed in claim 6, characterized in that: The set of the edge resource load time series sub-component feature vectors and the set of the cloud resource load time series sub-component feature vectors are subjected to feature sub-component pairing screening and interactive fusion processing to obtain resource load time series collaborative features, specifically: Calculate the hyperbolic space distance measurement factor between the time series sub-component feature vector of each edge resource load and the time series sub-component feature vector of each cloud resource load; According to the hyperbolic space distance measurement factor, the characteristic sub-component pairing and screening of each edge resource load time series sub-component characteristic vector and each cloud resource load time series sub-component characteristic vector are performed to obtain a pairing set; Input each pair in the pairing set into a feature multi-dimensional interaction module to obtain a set of load time series sub-component feature pairing interaction fusion feature vectors; The set of the load time series sub-component feature pairing interactive fusion feature vectors is cascaded to obtain the resource load time series coordination feature.

8. The task scheduling method based on cloud-edge collaboration as claimed in claim 7, characterized in that: The hyperbolic space distance measurement factor between the time series sub-component feature vector of each edge resource load and the time series sub-component feature vector of each cloud resource load is calculated as follows: Extracting the edge resource load timing sub-component feature vector at a preset position from the set of edge resource load timing sub-component feature vectors; Extracting a cloud resource load timing sub-component feature vector at a preset position from the set of cloud resource load timing sub-component feature vectors; Calculate the difference vector between the edge resource load timing sub-component feature vector at the preset position and the cloud resource load timing sub-component feature vector at the preset position, and perform square processing based on a norm of the difference vector to obtain a norm representation of the semantic difference of the load timing sub-component; Calculate the square of a norm of a characteristic vector of a time series subcomponent of the edge resource load at the preset position to obtain a first square result; perform subtraction processing on a first constant and the first square result according to corresponding positions to obtain a norm representation of a time series subcomponent of the edge resource load; Calculate the square of a norm of a characteristic vector of a cloud resource load timing subcomponent at the preset position to obtain a second square result; perform subtraction processing on a second constant and the second square result according to corresponding positions to obtain a norm representation of a cloud resource load timing subcomponent; Calculate the multiplication between the one-norm representation of the edge resource load timing subcomponent and the one-norm representation of the cloud resource load timing subcomponent, use the multiplication result as the divisor, use the one-norm semantic difference of the load timing subcomponent as the dividend, perform division processing, and obtain the one-norm semantic implicit association representation of the load timing; The load time series one-norm semantic implicit association representation is multiplied by a third constant, the product is added to a fourth constant, and the result of the addition is processed by an inverse hyperbolic cosine function to obtain the hyperbolic space distance measurement factor.

9. The task scheduling method based on cloud-edge collaboration as claimed in claim 8, characterized in that: The task to be allocated is split according to the resource load timing coordination feature to obtain a first power calculation subtask to be allocated and a second power calculation subtask to be allocated, and the first power calculation subtask to be allocated and the second power calculation subtask to be allocated are respectively scheduled to the edge end and the cloud end, specifically: Inputting the resource load time series coordination feature into a classifier-based task splitting module to obtain a first power calculation subtask to be allocated and a second power calculation subtask to be allocated; The first power calculation subtask to be allocated and the second power calculation subtask to be allocated are allocated to the edge and the cloud, respectively.

10. A task scheduling device based on cloud-edge collaboration, characterized in that: It includes a task acquisition module, a queue acquisition module, a feature extraction module, a feature interaction module and a scheduling module; among which, The task acquisition module is used to acquire the task to be assigned from the edge end, and the task to be assigned is a power calculation task; The queue acquisition module is used to acquire the time queue of the resource load value of the edge end and the time queue of the resource load value of the cloud end; The feature extraction module is used to extract the time series features of the edge resource load based on the time series of the edge resource load value; and to extract the time series features of the cloud resource load based on the time series of the cloud resource load value; The feature interaction module is used to perform detailed feature interaction based on spatial distance measurement based on the edge resource load time series characteristics and the cloud resource load time series characteristics to obtain resource load time series collaborative characteristics; The scheduling module is used to split the task to be allocated according to the resource load timing coordination characteristics to obtain a first power calculation subtask to be allocated and a second power calculation subtask to be allocated, and schedule the first power calculation subtask to be allocated and the second power calculation subtask to be allocated to the edge end and the cloud end respectively, so that the edge end and the cloud end respectively execute the first power calculation subtask to be allocated and the second power calculation subtask to be allocated.

Images