Cooperative computing task time delay prediction method and system based on dynamic knowledge graph

Abstract

The invention relates to the technical field of cooperative computing, and particularly discloses a cooperative computing system task time delay prediction method and system based on a dynamic knowledge graph. Aiming at the problems of heterogeneous entity association fragmentation, spatial-temporal feature segmentation and insufficient multi-source data fusion in the prior art, the method comprises the following steps: constructing a dynamic knowledge graph, disassembling equipment into entity types such as computational nodes and data sources according to functions, and uniformly managing motion state and position information through an equipment carrier; adopting Transform to encode dynamic attribute time sequence characteristics, and utilizing a heterogeneous graph attention network to generate a meta-path embedding vector; and fusing a space-time diagram convolutional network to extract space load conduction and time resource evolution characteristics, and predicting task time delay through a neural network. The problem of space-time coupling modeling is solved, artificial feature engineering is avoided, prediction precision is improved, and the method is suitable for dynamic scenes such as unmanned aerial vehicle clusters and edge calculation.

Description

Technical Field

[0001] The present invention relates to the field of collaborative computing, and in particular to a method and system for predicting collaborative computing task delays based on a dynamic knowledge graph. Background Art

[0002] In collaborative computing scenarios, task delay prediction is a core technology for ensuring efficient system operation, optimized resource allocation, and high-quality service. By predicting task execution times in advance, the system can dynamically adjust task priorities and allocate computing resources to ensure that critical tasks are completed before deadlines, avoiding system crashes or functional failures caused by accumulated delays.

[0003] Existing forecasting methods have the following fundamental flaws: 1. Fragmentation of heterogeneous entity association modeling: The complex relationships between device functional attributes (such as the GPU acceleration capabilities of edge nodes), task characteristics (such as data size), and task latency have not been structured and modeled. For example, traditional approaches simplify multifunctional devices into single "nodes," ignoring the deeper mapping relationships between computing nodes, data formats, and transmission latency. This results in predictive models failing to capture relevant key patterns.

[0004] 2. Split analysis of spatiotemporal dynamic characteristics: Spatial dimension: The cascading effects of sudden changes in the load of adjacent nodes (e.g., edge node overload → increased communication latency for surrounding nodes) are not modeled, making it impossible to predict the transmission path from "device A failure → device B's computing power consumption surges → increased latency for tasks assigned to device B"; Time dimension: This approach ignores the fragmentation of residual resources from historical tasks (e.g., discontinuous memory allocation → increased computing power consumption). It relies on simple time series statistics and cannot quantify the impact of periodic resource fluctuations (e.g., load fluctuations caused by the start and stop of industrial equipment) on task waiting and execution time.

[0005] 3. Insufficient multi-source data fusion capabilities: Data such as device logs (structured), task descriptions (text), and network topology (graph structure) are stored in isolation. Traditional machine learning requires manual feature design and is difficult to integrate spatiotemporal coupling features (such as the influencing link of "device motion trajectory → network topology change → transmission delay"), resulting in poor model generalization ability.

[0006] The above defects limit the accuracy of delay prediction and the scope of application of the delay prediction method. Summary of the Invention

[0007] In response to the above-mentioned deficiencies in the prior art, the present invention provides a collaborative computing task delay prediction method based on a dynamic knowledge graph.

[0008] In order to achieve the above-mentioned object of the invention, the technical solution adopted by the present invention is: A collaborative computing task latency prediction method based on a dynamic knowledge graph includes the following steps: S1. Collect static and dynamic information of each entity in the collaborative computing network and build a dynamic knowledge graph; S2, generates dynamic vector representation based on the dynamic knowledge graph constructed in S1; S3. Given a computing system task and computing nodes, predict the delay value based on the obtained dynamic vector representation.

[0009] Furthermore, the S1 specifically includes the following steps: S11. Collect static information of each entity in the collaborative computing network, including the CPU model, number of cores, frequency, memory type, memory size, dedicated computing chip type, and cache size of the computing device; S12, organizing the static information obtained in S11 into a knowledge graph according to the ontology model; S13. Collect dynamic information of each entity in the collaborative computing network at each moment, including real-time memory usage, real-time CPU usage, and real-time bandwidth usage, and update it into the knowledge graph as dynamic attributes; S14. Repeat S13 continuously to maintain the real-time and timeliness of the knowledge graph.

[0010] Furthermore, the S2 specifically includes the following steps: S21, encoding the temporal dynamic attributes of each entity in the collaborative computing network to generate a temporal feature vector; S22. Aggregate multiple neighbor information through meta-path embedding to generate embedding vectors of knowledge graph entity nodes; S23. Concatenate the time series feature vector and the embedding vector of the entity node to obtain a dynamic knowledge graph and generate a dynamic vector representation.

[0011] Furthermore, the specific encoding method in S21 is:

[0012] Where, X is a temporal dynamic sequence, P is the position code, H time is the time series feature vector.

[0013] Furthermore, the meta-path embedding in S22 adopts a heterogeneous graph attention network to calculate the node embedding vector through cross-type attention weights, and the formula is:

[0014] Where, For the The hidden vector of the layer The value of the component, is the cross-type attention weight, is the type-dependent projection matrix, For the The hidden vector of the layer The value of the component, is a nonlinear transformation function.

[0015] Furthermore, the S3 specifically includes the following steps: S31. Given a task J, assign it to a computing node N and use the subgraph g consisting of all feasible paths from J to N in the knowledge graph as the input to be predicted. S32, perform graph encoding on the subgraph g formed in S31 to form a latent vector h; S33. The latent vector h output by S32 is calculated through a neural network to obtain an estimated delay t of task J on computing node N.

[0016] A collaborative computing system task delay prediction system based on dynamic knowledge graph includes: an information perception module, a knowledge preprocessing module and a delay prediction module, wherein: The information perception module includes: Static information collection unit, used to obtain equipment hardware configuration and task characteristics; Knowledge graph initialization unit, building ontology models and organizing static information; Dynamic information collection unit, which obtains resource usage data in real time and updates the knowledge graph; The knowledge preprocessing module: Dynamic attribute encoding unit, generating time series feature vectors; Meta-path embedding unit, generating entity embedding vectors; Vector splicing unit, generating dynamic vector representation; The delay prediction module: Subgraph extraction unit, constructs the path subgraph of tasks and computing nodes; Spatiotemporal graph convolutional unit, which encodes subgraphs; Neural network prediction unit, outputs delay value.

[0017] The present invention has the following beneficial effects: This invention is based on dynamic knowledge graph technology and has multi-dimensional technical advantages: through hierarchical modeling of device functions, task characteristics and resource status, it solves the problem of fragmented associations between heterogeneous entities and supports the structured expression of dynamic spatiotemporal relationships; it integrates the spatiotemporal graph convolutional network to accurately capture the laws of spatial load conduction and temporal evolution, effectively improving the accuracy of delay prediction and energy consumption prediction; it uses the heterogeneous graph attention network (HGAT) to automatically fuse multi-source data, reduce the cost of manual feature engineering, and significantly improve the accuracy and efficiency of task delay prediction in collaborative computing systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 This is the system architecture and processing flow chart of the present invention.

[0019] Figure 2 This is a schematic diagram of the knowledge modeling of the equipment in this embodiment of the method.

[0020] Figure 3 This is an example diagram of the knowledge graph ontology structure of an embodiment of the present invention.

[0021] Figure 4 This is an example of a subgraph formed by a task-to-computing node path according to an embodiment of the present invention. DETAILED DESCRIPTION

[0022] The specific embodiments of the present invention are described below to facilitate understanding of the present invention by those skilled in the art. However, it should be clear that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the appended claims, these changes are obvious, and all inventions and creations utilizing the concepts of the present invention are protected.

[0023] A collaborative computing task delay prediction method based on dynamic knowledge graph, such as Figure 1 As shown, the following steps are included: S1. Collect static and dynamic information of each entity in the collaborative computing network and build a dynamic knowledge graph; For collaborative computing networks, we define entity types, relationship types, and entity attributes to form an ontology model (Schema). In the research and construction of collaborative computing networks, the differences and overlapping functions of different types of devices cannot be ignored. For example, smart terminals serve as data sources, providing raw data for computations and acting as computing nodes to execute tasks, demonstrating their typical multifunctionality.

[0024] To achieve more accurate and efficient modeling, for hardware devices, the abstraction of entity types is no longer limited to physical form, but closely around the functions that the device undertakes in the collaborative computing network. For entities with multiple functions, they are disassembled into multiple independent objects according to different functions for separate processing. For example, a smart camera with computing power can be disassembled into a computing node responsible for computing tasks and a data source providing data.

[0025] The network environment of different devices is affected by the gateway or router connected. At the same time, considering the device form, state difference (such as the high-speed motion state of vehicles, unmanned aerial vehicles) will have an important influence on task scheduling, and a special "device carrier" entity type is designed. This entity is used to record key information such as the form, motion state, spatial position and energy supply mode of the device (the data transmission cost between each functional entity object on the same device carrier is 0). For example, as shown in Figure 2 The center server will be abstracted as a subgraph composed of entities such as "software warehouse, data receiver, computing node and carrier device". When making task offloading decisions, instead of considering the unmanned aerial vehicle and the center server independently, all related entities are considered as a whole, and the matching between each entity and the task is considered uniformly. The characteristics of the task, the resource status of each entity, the performance parameters and the correlation between each other are comprehensively evaluated to make more reasonable and efficient task offloading strategies, optimize resource allocation and improve system performance. The main entity types and core attributes are as follows in Table 1.

[0026] Table 1 Entity types and core attributes

[0027] The knowledge graph ontology structure (Schema) formed by the relationship between entities is as shown in Figure 3 S2, generating dynamic vector representation based on the dynamic knowledge graph constructed in S1; In this embodiment, the following steps are specifically included: S21, encoding the time sequence dynamic attributes of each entity in the collaborative computing network to generate a time sequence feature vector; The time sequence attributes of the entity node (a vector composed of multiple attribute values of the entity node at each time in a time window) are encoded into time sequence features through Seq2Seq or Transformer. The time sequence X=[x1,...,xT] is encoded through Transformer:

[0028] wherein,​X is a temporal dynamic sequence, P is the position code, H time is the time series feature vector S22. Aggregate multiple neighbor information through meta-path embedding to generate embedding vectors of knowledge graph entity nodes; Multi-hop neighbor information is aggregated through HGAT (Heterogeneous Graph Attention Network) to generate embedding vectors of knowledge graph entity nodes.

[0029] Metapath Φ: ,node The embedding of is calculated as:

[0030] in is the cross-type attention weight, W τ is a type-dependent projection matrix.

[0031] S23. Concatenate the time series feature vector and the embedding vector of the entity node to obtain a dynamic knowledge graph and generate a dynamic vector representation.

[0032] S3. Given a computing system task and computing nodes, predict the delay value based on the obtained dynamic vector representation.

[0033] In this embodiment, the steps are as follows: S31. Given a task J, assign it to a computing node N and use the subgraph g consisting of all feasible paths from J to N in the knowledge graph as the input to be predicted. S32, perform graph encoding on the subgraph g formed in S31 to form a latent vector h; The specific process of ST-GCN is as follows: define a sliding time window consisting of T time steps, and subgraph g is t i Corresponding to a dynamic picture ; Use GCN for each For spatial feature extraction, each layer of GCN is calculated as:

[0034] Where A is the normalized adjacency matrix, For the l Layer node feature matrix (initial value ), is the spatial convolution weight, and σ is the activation function.

[0035] Spatial feature output for T consecutive time steps Apply one-dimensional convolution or temporal convolution (TCN) to obtain the latent vector h.

[0036] S33. The latent vector h output by S32 is calculated through a neural network to obtain an estimated delay t of task J on computing node N.

[0037] This embodiment also provides a collaborative computing system task delay prediction system based on a dynamic knowledge graph, including: an information perception module, a knowledge preprocessing module and a delay prediction module, wherein: The information perception module includes: Static information collection unit, used to obtain equipment hardware configuration and task characteristics; Knowledge graph initialization unit, building ontology models and organizing static information; Dynamic information collection unit, which obtains resource usage data in real time and updates the knowledge graph; The knowledge preprocessing module: Dynamic attribute encoding unit, generating time series feature vectors; Meta-path embedding unit, generating entity embedding vectors; Vector splicing unit, generating dynamic vector representation; The delay prediction module: Subgraph extraction unit, constructs the path subgraph of tasks and computing nodes; Spatiotemporal graph convolutional unit, which encodes subgraphs; Neural network prediction unit, outputs delay value.

[0038] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0039] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0040] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0041] Specific embodiments are used in the present invention to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core ideas. At the same time, for those skilled in the art, according to the ideas of the present invention, there may be changes in the specific implementation methods and application scopes. In summary, the contents of this specification should not be understood as limiting the present invention.

[0042] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and it should be understood that the scope of protection of the present invention is not limited to such specific descriptions and embodiments. Those skilled in the art can make various other specific variations and combinations based on the technical teachings disclosed in the present invention without departing from the essence of the present invention, and such variations and combinations are still within the scope of protection of the present invention.

Claims

1. A collaborative computing task delay prediction method based on dynamic knowledge graph, characterized by: The steps include: S1. Collect static and dynamic information of each entity in the collaborative computing network and build a dynamic knowledge graph; S2, generates dynamic vector representation based on the dynamic knowledge graph constructed in S1; S3. Given a computing system task and computing nodes, predict the delay value based on the obtained dynamic vector representation.

2. The collaborative computing task delay prediction method based on dynamic knowledge graph according to claim 1 is characterized in that: The S1 specifically includes the following steps: S11. Collect static information of each entity in the collaborative computing network, including the CPU model, number of cores, frequency, memory type, memory size, dedicated computing chip type, and cache size of the computing device; S12, organizing the static information obtained in S11 into a knowledge graph according to the ontology model; S13. Collect dynamic information of each entity in the collaborative computing network at each moment, including real-time memory usage, real-time CPU usage, and real-time bandwidth usage, and update it into the knowledge graph as dynamic attributes; S14. Repeat S13 continuously to maintain the real-time and timeliness of the knowledge graph.

3. The collaborative computing task delay prediction method based on dynamic knowledge graph according to claim 1 is characterized in that: The S2 specifically includes the following steps: S21, encoding the temporal dynamic attributes of each entity in the collaborative computing network to generate a temporal feature vector; S22. Aggregate multiple neighbor information through meta-path embedding to generate embedding vectors of knowledge graph entity nodes; S23. Concatenate the time series feature vector and the embedding vector of the entity node to obtain a dynamic knowledge graph and generate a dynamic vector representation.

4. The collaborative computing task delay prediction method based on dynamic knowledge graph according to claim 3 is characterized in that: The specific encoding method in S21 is: Where, X is a temporal dynamic sequence, P is the position code, H time is the time series feature vector.

5. The collaborative computing task delay prediction method based on dynamic knowledge graph according to claim 3 is characterized in that: The meta-path embedding in S22 adopts a heterogeneous graph attention network to calculate the node embedding vector through cross-type attention weights. The formula is: Where, For the The hidden vector of the layer The value of the component, is the cross-type attention weight, is the type-dependent projection matrix, For the The hidden vector of the layer The value of the component, is a nonlinear transformation function.

6. The collaborative computing task delay prediction method based on dynamic knowledge graph according to claim 1 is characterized in that: The S3 specifically includes the following steps: S31. Given a task J, assign it to a computing node N and use the subgraph g consisting of all feasible paths from J to N in the knowledge graph as the input to be predicted. S32, perform graph encoding on the subgraph g formed in S31 to form a latent vector h; S33. The latent vector h output by S32 is calculated through a neural network to obtain the estimated delay t of task J on computing node N.

7. A collaborative computing system task delay prediction system based on dynamic knowledge graph, characterized by: include: Information perception module, knowledge preprocessing module and delay prediction module, among which: The information perception module includes: Static information collection unit, used to obtain equipment hardware configuration and task characteristics; Knowledge graph initialization unit, building ontology models and organizing static information; Dynamic information collection unit, which obtains resource usage data in real time and updates the knowledge graph; The knowledge preprocessing module: Dynamic attribute encoding unit, generating time series feature vectors; Meta-path embedding unit, generating entity embedding vectors; Vector splicing unit, generating dynamic vector representation; The delay prediction module: Subgraph extraction unit, constructs the path subgraph of tasks and computing nodes; Spatiotemporal graph convolution unit, which encodes subgraphs; Neural network prediction unit outputs delay value.

Images