An edge resource scheduling method and system for industrial internet

Abstract

The application discloses an edge resource scheduling method and system for an industrial internet, relates to the field of the industrial internet, and comprises the following steps: predicting a user request amount in a future time period through a SparseTSF model; based on the user request amount in the future time period, calculating the minimum number of servers under the premise of full service quality by using an M / M / c queuing model; and determining a predicted active server set based on energy efficiency optimization; and generating a server start-stop instruction according to the predicted active server set and a current active server set, and performing resource allocation. The application can predict future load, reduce system power consumption, optimize resource utilization and improve service quality through resource planning.

Description

Technical Field

[0001] This invention relates to the field of edge computing technology in the Industrial Internet, and more specifically to an edge resource scheduling method and system for the Industrial Internet. Background Technology

[0002] Currently, industrial internet edge systems mainly consist of a three-tier architecture: terminal devices, edge server clusters, and cloud platforms. Terminal devices (such as industrial sensors and autonomous driving terminals) generate data requests, edge server clusters process the data locally to reduce latency, and the cloud platform only processes non-real-time global data. With the rapid development of the industrial internet and edge computing, edge server clusters are widely used in scenarios with high real-time requirements, such as intelligent manufacturing, connected vehicles, and smart factories.

[0003] However, the dense deployment of edge nodes has led to a sharp increase in energy consumption, becoming a key issue restricting the sustainable development of the system. In existing technologies, most resource scheduling methods rely on real-time load monitoring and dynamic adjustment strategies, such as online decision-making methods based on reinforcement learning or heuristic greedy algorithms. While these methods can reduce energy consumption to some extent, they still have many problems. For example, existing scheduling relies on real-time decision-making, lacking the ability to predict user requests, unable to anticipate load changes, and struggling to cope with the spatiotemporal dynamics of user requests, resulting in scheduling lags. It also leads to some servers running at high loads for extended periods while others remain idle, causing a surge in energy consumption and resource waste. Secondly, existing methods lack adaptation to edge characteristics; for example, user-server proximity constraints and resource multidimensionality are not fully modeled, resulting in large fluctuations in service quality (such as response time).

[0004] Therefore, how to predict future loads, reduce system power consumption, optimize resource utilization, and improve service quality are problems that urgently need to be solved by those skilled in the art. Summary of the Invention

[0005] In view of this, the present invention provides an edge resource scheduling method and system for the Industrial Internet to solve the above problems.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: This invention discloses an edge resource scheduling method for the Industrial Internet, comprising: Predict user request volume for future time periods using the SparseTSF model; Based on the user request volume in the future time period, the minimum number of servers under the premise of full service quality is calculated using the M / M / c queuing model; and the set of predicted active servers is determined based on energy efficiency optimization. Based on the predicted set of active servers and the current set of active servers, server start / stop instructions are generated, and resources are allocated.

[0007] Furthermore, the prediction of user request volume in future time periods using the SparseTSF model specifically includes: Collect raw request volume / load data for industrial internet edge servers; The original request / load data is decomposed into multiple scales to obtain... k One time-domain feature; The decomposed temporal features are subjected to cross-period convolution to obtain higher-order features; Gated fusion is performed on the higher-order feature maps to obtain k The weights of each feature; Obtain current request / load data from the Industrial Internet, perform multi-scale decomposition to obtain k current time-domain features, and calculate the user request volume for future time periods based on the weights of each feature. .

[0008] Furthermore, the formula for the multi-scale decomposition is: ; in, express k Each time-domain feature Indicates the inverse Fourier transform. Represents the frequency domain mask matrix. Indicates Fourier transform, Λ This represents the raw request volume / load data of the edge server.

[0009] Furthermore, the formula for the cross-period convolution is: ; in, Represents higher-order features, Indicates input features, Indicates weight, For activation function, Indicates the first k The layer bias vector is shifted to make the feature distribution more easily activated by ReLU, and the ReLU function filters out effective features.

[0010] Furthermore, the formula for the gating fusion is: ; in, Indicates the first k The weights of each feature Indicates global average pooling. U and v For learnable parameters, and All of these are activation functions.

[0011] Furthermore, the calculation of the minimum number of servers required to achieve full service quality using the M / M / c queuing model specifically includes: First, determine the user request volume for the future time period. Server speed Target average queuing delay ; Then, initialize the parameters, set c=1 ; Finally, iterative calculations are performed to determine the minimum number of servers. The iterative calculation steps are as follows: Step 1: Calculate system utilization ,like If yes, proceed to step 4; otherwise, proceed to step 2. Step 2: Calculate the Erlang C formula: ; Step 3: Calculate the average queuing time: ;like Then let =c Exit; otherwise proceed to step 4. Step 4: c = c + 1 Return to step 1 in, This represents an Erlang C formula used to calculate the probability that a new request will have to wait when all servers are busy in an M / M / c queuing model. This indicates the waiting time for the request in the queue.

[0012] Furthermore, the step of determining the predicted activation server set based on energy efficiency optimization specifically involves: sorting the server set S by energy efficiency ratio and selecting the top... The most energy-efficient servers constitute the activation set The formula for the energy efficiency ratio is: ; in, In server set S, the first... j Energy efficiency ratio of the server d This indicates the number of resource dimensions (such as CPU, memory, bandwidth, etc.). Indicates the first j The first server k Class resource capacity, This represents the operating energy consumption of the j-th server. .

[0013] Furthermore, the generation of server start / stop instructions specifically involves: for each server :like and Then send a start command; if and If so, a shutdown command is sent; where This indicates the currently active set.

[0014] Furthermore, the resource allocation specifically includes: According to the proximity constraint guarantee formula: ,in Indicates user The server assigned at time t; will the user The set of servers assigned to its coverage area middle; Within each server set, a dominant resource fairness allocation algorithm is used for local resource allocation, specifically: First, calculate the dominant resource requirement for each user request using the following formula: ;in Indicates user The dominant resource demand, Indicates user For the k The demand for class resources; and according to Sort user requests in descending order; Then, users are greedily assigned to the server with the lowest current utilization that meets the capacity constraint: The capacity constraint must satisfy: ; indicates time t At that time, all were assigned to the server j users For resource types k The total demand. This constraint ensures that the server's resource utilization does not exceed 90% of its capacity at any given time, thereby avoiding performance degradation or service interruption due to excessive server load. Maintaining 10% resource redundancy helps dynamically absorb the impact of instantaneous request spikes or prediction errors. Among these, This indicates the optimal server selection. Indicates the set of active servers. Indicates user In time t For the k The demand for similar resources.

[0015] This invention also discloses an edge resource scheduling system for the Industrial Internet, comprising: Prediction layer: Predicts user request volume for future time periods using the SparseTSF model; Resource planning layer: Based on the user request volume in the future time period, the minimum number of servers under the premise of full service quality is calculated using the M / M / c queuing model; and the set of predicted active servers is determined based on energy efficiency optimization. Execution layer: Based on the predicted set of active servers and the current set of active servers, generate server start / stop instructions and allocate resources.

[0016] As can be seen from the above technical solution, compared with the prior art, the present invention provides an edge resource scheduling method and system for the Industrial Internet, with the following specific effects: (1) Significant reduction in total system energy consumption: Experimental results show that during a continuous 7-day test covering different load cycles on weekdays and weekends, the total energy consumption of this method was reduced by 78.6% compared to the baseline strategy (AlwaysOn, all servers are always on). Compared to the current advanced online scheduling method based on deep reinforcement learning (pDRL), EAPA can still achieve an energy consumption reduction of about 60%. This huge improvement is directly due to the accurate prediction capability of this invention, which enables the system to shut down redundant servers in advance, avoiding a large amount of idle energy consumption and the energy cost of frequent start-up and shutdown.

[0017] (2) Resource utilization optimization: This method achieves "on-demand supply" of server resources through prediction-driven fine-grained management. Experiments showed that the average server utilization rate increased to 48.2%, which is much higher than the AlwaysOn strategy (always on but with uneven load, utilization rate is often below 20%) and the random start-stop strategy (large fluctuations in utilization rate, averaging only ~25%). This proves that the present invention effectively solves the resource fragmentation problem of "some servers overloaded and some servers idle" that is common in edge computing, and realizes the intensive utilization of resources.

[0018] (3) Reliable Quality of Service (QoS) Guarantee: Throughout the entire testing period, the average response time of all user requests was consistently controlled within the Service Level Agreement (SLA) threshold of 100ms, achieving a 100% SLA satisfaction rate. Even during peak request periods (such as the 12th time slot, with a request volume of 2500 / minute), by pre-allocating 5 servers based on prediction, the response time could still be maintained at 100ms. This completely overturns the traditional concept that "energy saving inevitably sacrifices performance," proving that a pre-configuration mechanism based on high-precision prediction can achieve energy saving while ensuring an ultimate user experience.

[0019] (4) Enhanced system stability: Compared with strategies that rely entirely on real-time response and cause frequent server on / off cycles, this method is based on periodic prediction for scheduling, which significantly reduces the frequency of server start-up and shutdown, and enhances the working life of hardware devices and the overall stability of the system.

[0020] (5) Reduced direct electricity costs: Taking a small edge cluster of 20 servers as an example, assuming an electricity price of 1 yuan / kWh, the annual electricity cost using the AlwaysOn strategy can reach tens of thousands of yuan. Applying this method can save a considerable proportion of electricity costs each year. For large industrial parks or edge node networks deployed nationwide, this economic benefit will grow exponentially.

[0021] (6) Improved return on investment: Higher resource utilization means that the same computing tasks can be completed with fewer physical servers, or more business can be carried on existing servers, which delays the need to purchase new servers due to business growth and reduces the company's fixed asset investment (CAPEX).

[0022] (7) Reduced operation and maintenance costs: The increased automation of the system reduces the reliance on real-time manual monitoring and intervention. At the same time, the optimization of server start-up and shutdown frequency also reduces the equipment failure rate, indirectly reducing maintenance costs and downtime risks.

[0023] (8) Empowering the high-quality development of the industrial internet: This invention solves the energy consumption bottleneck problem, making it possible and sustainable to deploy low-latency and high-reliability edge computing services on a large scale in industrial sites. It provides a solid technical foundation for the digital transformation of key industries such as intelligent manufacturing and autonomous driving, and has broad social application value.

[0024] In summary, this invention can predict future loads and reduce system power consumption, optimize resource utilization, and improve service quality through resource planning. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0026] Figure 1 This is a schematic diagram of the overall process of the present invention. Detailed Implementation

[0027] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] This invention discloses an edge resource scheduling method for the Industrial Internet, comprising: Predict user request volume for future time periods using the SparseTSF model; Based on the user request volume in the future time period, the minimum number of servers under the premise of full service quality is calculated using the M / M / c queuing model; and the set of predicted active servers is determined based on energy efficiency optimization. Based on the predicted set of active servers and the current set of active servers, server start / stop instructions are generated, and resources are allocated.

[0029] Specifically, Figure 1 The core components of this invention are illustrated, where the data acquisition and preprocessing module is deployed at an edge node. This module is responsible for collecting historical load data in real time and performing preprocessing such as standardization and denoising. Upon receiving the preprocessed historical data, multi-scale spatiotemporal feature extraction and fusion are performed through three units: multi-scale decomposition, cross-period convolution, and gated fusion. In the resource planner, the prediction results are received, and the minimum number of servers required to meet QoS latency requirements is calculated based on the M / M / c queuing model. Based on the energy efficiency ratio of each server ( Select the optimal one from the server cluster. Taiwan servers form an activation set The dynamic start / stop controller receives the activation set At and generates specific server start / stop commands (e.g., start the server). Hibernation server The resource allocation executor (DRF) sends the data to the hardware via the underlying interface. Based on the dominant resource fairness algorithm, the DRF allocates the user tasks that arrive in real time. Assigned to the active server, satisfying both proximity and capacity constraints.

[0030] In one specific embodiment, predicting user request volume for a future time period using the SparseTSF model specifically includes: Collect raw request volume / load data for industrial internet edge servers; The original request / load data is decomposed into multiple scales to obtain... k One time-domain feature; The temporal features obtained from the decomposition are subjected to cross-period convolution to obtain higher-order features; Gated fusion of high-order feature maps yields k The weights of each feature; Obtain current request / load data from the Industrial Internet, perform multi-scale decomposition to obtain k current time-domain features, and calculate the user request volume for future time periods based on the weights of each feature. .

[0031] Taking a smart manufacturing park edge computing platform as an example, a multi-access edge computing (MEC) system is deployed, which includes an edge server cluster, terminal devices, network devices, and edge control nodes. All edge servers are connected to the SDN switch via network interfaces. The edge control nodes are connected to the SDN switch and all edge servers via the network out-of-band management interfaces, used to issue server start / stop commands and monitor server status. Terminal devices access the network through wireless APs, and their data traffic is guided to the currently active edge server via the SDN switch.

[0032] The system collects historical load data and inputs it into the SparseTSF model for training and prediction. A scheduling cycle of 24 minutes is used, dynamically activating 5-8 servers based on the prediction results and allocating computational tasks using a DRF (Distributed Randomization Rank) strategy. In actual operation, the system fine-tunes server status based on real-time load and automatically puts some servers to sleep when the load falls below a threshold. This method can be implemented collaboratively with software-defined networking (SDN) and edge controllers, and supports API integration with existing MEC (Multi-access Edge Computing) platforms.

[0033] In a specific embodiment, the formula for multi-scale decomposition is: ; in, express k Each time-domain feature Indicates the inverse Fourier transform. Represents the frequency domain mask matrix. Indicates Fourier transform, Λ This represents the raw request / load data of the edge server. Multi-scale decomposition transforms the raw request / load data of the edge server into frequency domain features using Fourier transform. A frequency domain mask matrix is ​​used to preserve key frequency bands and filter noise. Then, an inverse Fourier transform restores the filtered frequency domain data to the time domain.

[0034] In one specific embodiment, the formula for cross-period convolution is: ; in, Represents higher-order features, The multi-scale decomposition yields the first... k Each time-domain feature Indicates weight, For activation function, Indicates the first k Layer bias vectors are used to shift the feature distribution, making it more susceptible to ReLU activation. Cross-period convolutions extract periodic features of industrial requests (such as shift cycles in smart manufacturing) through convolutional layers (input features, weights), and use the ReLU activation function to filter effective features.

[0035] In one specific embodiment, the gating fusion formula is: ; in, Indicates the first k The weights of each feature Indicates global average pooling. U and v For learnable parameters, and All of these are activation functions.

[0036] In one specific embodiment, the minimum number of servers required to achieve full service quality is calculated using the M / M / c queuing model, specifically including: First, determine the user request volume for the future time period. Server speed Target average queuing delay ; Then, initialize the parameters, set c=1 ; Finally, iterative calculations are performed to determine the minimum number of servers. The iterative calculation steps are as follows: Step 1: Calculate system utilization ,like If yes, proceed to step 4; otherwise, proceed to step 2. Step 2: Calculate the Erlang C formula: ; Step 3: Calculate the average queuing time: ;like Then let =c Exit; otherwise proceed to step 4. Step 4: c = c + 1 Return to step 1 in, This represents an Erlang C formula used to calculate the probability that a new request will have to wait when all servers are busy in an M / M / c queuing model. This indicates the waiting time for the request in the queue.

[0037] In one specific embodiment, determining the predicted activation server set based on energy efficiency optimization specifically involves: sorting the server set S by energy efficiency ratio and selecting the top... The most energy-efficient servers constitute the activation set The formula for energy efficiency ratio is: ; in, In server set S, the first... j Energy efficiency ratio of the server d This indicates the number of resource dimensions (such as CPU, memory, bandwidth, etc.). Indicates the first j The first server k Class resource capacity, Indicates the first j Power consumption of the server .

[0038] In one specific embodiment, generating server start / stop instructions involves: for each server :like and Then send a start command; if and If so, a shutdown command is sent; where This indicates the currently active set.

[0039] In one specific embodiment, resource allocation includes: According to the proximity constraint guarantee formula: ,in Indicates user In time t The assigned server; will the user The set of servers assigned to its coverage area middle; Within each server set, a dominant resource fairness allocation algorithm is used for local resource allocation, specifically: First, calculate the dominant resource requirement for each user request using the following formula: ;in Indicates user The dominant resource demand, Indicates user For the k The demand for class resources; and according to Sort user requests in descending order; Then, users are greedily assigned to the server with the lowest current utilization that meets the capacity constraint: The capacity constraint must satisfy: This indicates that at time t, all servers assigned to the server... j users For resource types k The total demand. This constraint ensures that the server's resource utilization does not exceed 90% of its capacity at any given time, thereby avoiding performance degradation or service interruption due to excessive server load. Maintaining 10% resource redundancy helps dynamically absorb the impact of instantaneous request spikes or prediction errors. Among these, This indicates the optimal server selection. Indicates the set of active servers. Indicates user In time t For the k The demand for similar resources.

[0040] During resource allocation execution, the SDN controller determines based on... Update the server table to route newly arriving user requests to the activation server.

[0041] This invention also discloses an edge resource scheduling system for the Industrial Internet, comprising: Prediction layer: Predicts user request volume for future time periods using the SparseTSF model; Resource planning layer: Based on the user request volume in the future time period, the minimum number of servers is calculated using the M / M / c queuing model to ensure full service quality; and the set of servers to be activated is determined based on energy efficiency optimization. Execution layer: Based on the predicted set of active servers and the current set of active servers, generate server start / stop instructions and allocate resources.

[0042] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0043] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An edge resource scheduling method for the Industrial Internet, characterized in that, include: Predict user request volume for future time periods using the SparseTSF model; Based on the user request volume in the future time period, the minimum number of servers required to meet the service quality requirements is calculated using the M / M / c queuing model. And the set of servers to be activated is determined based on energy efficiency optimization; Based on the predicted set of active servers and the current set of active servers, server start / stop instructions are generated, and resources are allocated. The calculation of the minimum number of servers under the premise of full service quality using the M / M / C queuing model specifically includes: First, determine the user request volume for the future time period. Server speed Target average queuing delay ; Then, initialize the parameters, set c=1 ; Finally, iterative calculations are performed to determine the minimum number of servers. The iterative calculation steps are as follows: Step 1: Calculate system utilization ,like If yes, proceed to step 4; otherwise, proceed to step 2. Step 2: Calculate the Erlang C formula: ; Step 3: Calculate the average queuing time: ;like Then let =c Exit; otherwise proceed to step 4. Step 4: c = c + 1 Return to step 1; in, This represents an Erlang C formula used to calculate the probability that a new request will have to wait when all servers are busy in an M / M / c queuing model. Indicates the waiting time of the request in the queue; The process of determining the set of servers to be activated based on energy efficiency optimization specifically involves: sorting the server set S by energy efficiency ratio and selecting the top... The most energy-efficient servers constitute the activation set The formula for the energy efficiency ratio is: ； in, In server set S, the first... j Energy efficiency ratio of the server d Indicates the number of resource dimensions. Indicates the first j The first server k Class resource capacity, Indicates the first j The operating energy consumption of the server; The server start / stop command is specifically as follows: for each server :like and Then send a start command; if and If so, a shutdown command is sent; where This indicates the currently active set.

2. The edge resource scheduling method for the Industrial Internet according to claim 1, characterized in that, The prediction of user request volume in future time periods using the SparseTSF model specifically includes: Collect raw request volume / load data for industrial internet edge servers; The original request / load data is decomposed into multiple scales to obtain... k One time-domain feature; The decomposed temporal features are subjected to cross-period convolution to obtain higher-order features; Gated fusion is performed on the higher-order feature maps to obtain k The weights of each feature; Obtain current request / load data from the Industrial Internet, perform multi-scale decomposition to obtain k current time-domain features, and calculate the user request volume for future time periods based on the weights of each feature. .

3. The edge resource scheduling method for the Industrial Internet according to claim 2, characterized in that, The formula for the multi-scale decomposition is: ； in, Indicates the first k Each time-domain feature Indicates the inverse Fourier transform. Represents the frequency domain mask matrix. Indicates Fourier transform, Λ This represents the raw request volume / load data of the edge server.

4. The edge resource scheduling method for the Industrial Internet according to claim 2, characterized in that, The formula for the cross-period convolution is: ； in, Represents higher-order features, Indicates input features, Indicates weight, For activation function, Indicates the first k Layer bias vector.

5. The edge resource scheduling method for the Industrial Internet according to claim 2, characterized in that, The formula for the gating fusion is: ； in, Indicates the first k The weights of each feature, Indicates global average pooling. U and v For learnable parameters, and All of these are activation functions.

6. The edge resource scheduling method for the Industrial Internet according to claim 1, characterized in that, The resource allocation process specifically includes: According to the proximity constraint guarantee formula: ,in Indicates user In time t The assigned server; will the user The set of servers assigned to its coverage area middle; Within each server set, a dominant resource fairness allocation algorithm is used for local resource allocation, specifically: First, calculate the dominant resource requirement for each user request using the following formula: ;in Indicates user The dominant resource demand, Indicates user Demand for the k-th type of resource; and according to Sort user requests in descending order; Then, users are greedily assigned to the server with the lowest current utilization that meets the capacity constraint: The capacity constraint must satisfy: ; indicates time t At that time, all were assigned to the server j users For resource types k The total demand; of which, This indicates the optimal server selection. Indicates the set of active servers. Indicates user In time t For the k The demand for similar resources.

7. An edge resource scheduling system for the Industrial Internet, employing the edge resource scheduling method for the Industrial Internet as described in any one of claims 1 to 6, characterized in that, include: Prediction layer: Predicts user request volume for future time periods using the SparseTSF model; Resource planning layer: Based on the user request volume in the future time period, use the M / M / c queuing model to calculate the minimum number of servers under the premise of full service quality; And the set of servers to be activated is determined based on energy efficiency optimization; Execution layer: Based on the predicted set of active servers and the current set of active servers, generate server start / stop instructions and allocate resources.

Images