Network resource scheduling method based on artificial intelligence

Abstract

The invention relates to the field of network resource scheduling, in particular to a network resource scheduling method based on artificial intelligence. The method comprises the steps of collecting resource scheduling data, and dynamically adjusting task priorities; constructing a network load prediction model and a resource demand prediction model to obtain a network load prediction value and a resource demand prediction value; calculating the comprehensive benefit of the task based on the task priority, the network load predicted value and the resource demand predicted value; and based on the comprehensive benefit of the task, a task scheduling decision is optimized through a reinforcement learning algorithm. The problem that a traditional resource scheduling method lacks adaptability to real-time network states and task demand changes is solved. A network load and resource demand prediction method generally adopts simple historical data analysis, and periodic fluctuation and a load attenuation effect are not considered; the problems of low task scheduling efficiency and unreasonable resource allocation due to the fact that scheduling decisions are mostly based on fixed rules and scheduling strategies cannot be adjusted according to real-time comprehensive benefits of tasks in the prior art are solved.

Description

Technical Field

[0001] The present invention relates to the field of network resource scheduling, and in particular to a network resource scheduling method based on artificial intelligence. Background Art

[0002] With the rapid development of information technology and network communication technology, network resource scheduling, as a key technology to improve computing resource utilization and optimize network performance, has received widespread attention. Especially in application scenarios such as cloud computing and edge computing, network resource scheduling faces increasingly complex challenges. Traditional scheduling methods mainly rely on resource allocation strategies based on static task priorities or fixed rules, which are relatively limited in dealing with dynamic changes in tasks and fluctuations in network loads. With the diversification of task requirements and the complexity of network loads, how to adjust resource allocation according to real-time status and improve resource utilization efficiency has become the focus of current technology development.

[0003] Therefore, there are some key problems that need to be solved in traditional network resource scheduling methods: first, traditional resource scheduling methods rely on static priority allocation and lack adaptability to changes in real-time network status and task requirements, resulting in inflexible resource scheduling and inefficient resource utilization; second, network load and resource demand prediction methods usually use simple historical data analysis, fail to consider periodic fluctuations and load attenuation effects, resulting in insufficient prediction accuracy, thus affecting the accuracy of resource scheduling decisions; finally, scheduling decisions are mostly based on fixed rules, lack intelligent feedback mechanisms and dynamic optimization capabilities, and cannot adjust scheduling strategies according to the real-time comprehensive benefits of tasks, resulting in inefficient task scheduling, unreasonable resource allocation, and performance improvement. Therefore, it is urgent to improve it through a new algorithm and processing technology. Summary of the invention

[0004] The present invention provides a network resource scheduling method based on artificial intelligence to solve the problems that the traditional resource scheduling method relies on static priority allocation and lacks adaptability to changes in real-time network status and task requirements, resulting in inflexible resource scheduling and inefficient resource utilization; the network load and resource demand prediction method usually adopts simple historical data analysis, fails to consider periodic fluctuations and load attenuation effects, resulting in insufficient prediction accuracy, thereby affecting the accuracy of resource scheduling decisions; scheduling decisions are mostly based on fixed rules, lack intelligent feedback mechanisms and dynamic optimization capabilities, and cannot adjust scheduling strategies according to the real-time comprehensive benefits of tasks, resulting in low task scheduling efficiency and unreasonable resource allocation.

[0005] The present invention provides a network resource scheduling method based on artificial intelligence, which specifically includes the following technical solutions:

[0006] A network resource scheduling method based on artificial intelligence comprises the following steps:

[0007] S1: Collect resource scheduling data and dynamically adjust task priorities; Based on the resource scheduling data, construct a network load prediction model and a resource demand prediction model to obtain a network load prediction value and a resource demand prediction value;

[0008] S2: Calculate the comprehensive benefit of a task based on the task priority, network load prediction value, and resource demand prediction value; Optimize the task scheduling decision through a reinforcement learning algorithm based on the comprehensive benefit of the task.

[0009] Preferably, the S1 specifically includes:

[0010] Collect resource scheduling data in the network in real time; The resource scheduling data includes the resource requirements, network load, task priority, remaining execution time, and executed time of the task.

[0011] Preferably, the S1 specifically includes:

[0012] Based on the executed time and remaining execution time of the task, introduce a priority balance coefficient to dynamically adjust the task priority.

[0013] Preferably, the S1 specifically includes:

[0014] Based on the resource scheduling data, combined with the historical network load real value, introduce an exponential decay factor and a periodic adjustment factor to construct a network load prediction model, predict the network load at a future moment, and obtain a network load prediction value.

[0015] Preferably, the S1 specifically includes:

[0016] Based on the resource scheduling data, combined with the resource demand real value, introduce a weighting coefficient and a decay coefficient to construct a resource demand prediction model, predict the resource demand at a future moment, and obtain a resource demand prediction value.

[0017] Preferably, the S2 specifically includes:

[0018] Calculate the comprehensive benefit of a task through a comprehensive benefit calculation formula based on the task priority, network load prediction value, and resource demand prediction value.

[0019] Preferably, the S2 specifically includes:

[0020] When the network load prediction value and the resource demand prediction value are used as the inputs of the comprehensive benefit calculation formula, the calculated value is the predicted comprehensive benefit of the task; When the network load and the resource demand real values are used as the inputs of the comprehensive benefit calculation formula, the calculated value is the real comprehensive benefit of the task; The comprehensive benefit calculation formula is as follows:

[0021]

[0022] Among them, is the predicted comprehensive benefit of the task; α 1 and α 2 are weight coefficients; is the predicted value of resource demand at time t + 1; is the predicted value of network load at time t + 1; P t is the task priority at time t; T t is the remaining execution time of the task at time t; T max is the maximum execution time of the task; exp is the exponential function; is the decay coefficient; L t is the network load at time t.

[0023] Preferably, the S2 specifically includes:

[0024] The reinforcement learning algorithm makes a scheduling decision according to the current network state information, introduces a feedback mechanism, and updates the decision evaluation value based on the predicted comprehensive benefit of the task to obtain the updated decision evaluation value.

[0025] Preferably, the S2 specifically includes:

[0026] Based on the task priority, predicted value of resource demand, predicted value of network load, and updated decision evaluation value, combined with the current network state information, construct and maximize the objective function to make a task scheduling decision.

[0027] Preferably, the S2 specifically includes:

[0028] After executing the scheduling decision, calculate the true comprehensive benefit of the task based on the comprehensive benefit calculation formula; update the decision evaluation value based on the true comprehensive benefit of the task, and continuously monitor the usage of network resources, collect resource scheduling data, and provide support for subsequent network resource scheduling.

[0029] The beneficial effects of the technical solution of the present invention are:

[0030] 1. By combining multi-dimensional information such as the resource demand and remaining execution time of the task, the present invention dynamically adjusts the task priority. Different from the static priority allocation method, the present invention continuously updates the task priority, making the resource scheduling more flexible and intelligent, and being able to better adapt to the changes in network load and task execution, thereby improving resource utilization and task execution efficiency.

[0031] 2. The present invention designs a network load prediction model based on historical data, introducing an exponential decay factor and a periodic adjustment factor, making the network load prediction more timely and accurate. At the same time, a resource demand prediction model is constructed, which can dynamically adjust the predicted value of resource demand by combining a weighting coefficient, a decay coefficient, and a bias term, improving the stability and accuracy of the prediction. Compared with traditional single prediction methods, it can more accurately capture the changing rules of network load and resource demand, thus providing a more reliable basis for subsequent resource scheduling decisions and avoiding resource waste and unbalanced scheduling problems caused by errors in traditional predictions.

[0032] 3. The present invention introduces a reinforcement learning algorithm. Through a feedback mechanism based on the comprehensive benefit of tasks, it continuously optimizes task scheduling decisions. Different from traditional fixed-rule scheduling methods, the reinforcement learning model of the present invention can dynamically adjust decision-making strategies according to real-time network status and task requirements, maximizing the overall benefit of task scheduling and improving the flexibility and efficiency of task scheduling by continuously updating decision evaluation values. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 It is a flowchart of a network resource scheduling method based on artificial intelligence according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0034] In order to further elaborate on the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0035] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present invention belongs.

[0036] The following specifically describes the specific solution of a network resource scheduling method based on artificial intelligence provided by the present invention in conjunction with the drawings.

[0037] Refer to the attached Figure 1 , which shows a flowchart of a network resource scheduling method based on artificial intelligence provided by an embodiment of the present invention. The method includes the following steps:

[0038] S1: Collect resource scheduling data and dynamically adjust task priorities; based on the resource scheduling data, construct a network load prediction model and a resource demand prediction model to obtain a network load prediction value and a resource demand prediction value;

[0039] Before implementing network resource scheduling, it is first necessary to collect resource scheduling data in the network in real time. Specifically, it is necessary to obtain data such as the resource requirements, network load, task priority, remaining execution time, and executed time of each task. The resource requirements of a task reflect the consumption of resources such as computing, memory, and bandwidth. The network load represents the current usage status of the network bandwidth. The task priority is assigned based on the urgency of the task or other policies. The remaining execution time is the time required for the task to be completed as predicted, and the executed time represents the time that the task has consumed;

[0040] Then, based on the resource scheduling data, the priority of the task is dynamically adjusted to make the task scheduling more intelligent and efficient;

[0041] The formula for adjusting the priority of a task is as follows:

[0042]

[0043] where P t is the task priority at time t, representing the adjusted task priority; is the priority balance coefficient, which is set according to the specific implementation scenario; P orig is the task priority before adjustment; U t is the executed time of the task at time t; T t is the remaining execution time of the task at time t; N is the number of all tasks in the network; j is the task index variable; U t，j is the executed time of the jth task at time t; T t，j is the remaining execution time of the jth task at time t; w 1 and w 2 are weight coefficients, which are used to balance the influence degree of the executed time and the remaining execution time of the task on the task priority, and are set according to the specific implementation scenario;

[0044] Next, by constructing a network load prediction model, the network load at future times is predicted in advance, so as to understand the usage situation of network resources at future times in advance, and resource scheduling decisions are made according to the prediction results. The network load prediction model takes into account the influence of the true value of the historical network load on the predicted value of the network load at future times, and introduces an exponential decay factor to simulate the decay effect. In addition, in order to better reflect the periodic changes of the network load (such as day and night changes, etc.), a periodic adjustment factor is introduced to capture the influence of the periodic fluctuations of the network load. Through the above design, the network load prediction model can comprehensively and accurately predict the network load, obtain the network load prediction value, and use the network load prediction value to optimize the resource scheduling decision to improve the utilization efficiency of resources;

[0045] The formula for the network load prediction model is as follows:

[0046]

[0047] Among them, is the predicted value of network load at time t + 1; t is a time variable; ζ is a constant term, reflecting the basic level of network load, set according to the expert experience method; p is the number of lag periods, defining the number of historical moments used to calculate the predicted value of network load, indicating that the true values of network load in the past p moments are used for prediction, set according to the specific implementation scenario; i is the moment index; β i is the weighting coefficient of the true value of network load at time t - f, obtained through the training of the network load prediction model. Model training is a well-known method for those skilled in the art and will not be elaborated here. β 0 is the interpretation of β i when f = 0; L t is the network load at time t; L t-f is the true value of network load at time t - f; θ is the parameter of the exponential decay factor, set according to the expert experience method; e -θi is the exponential decay factor of historical load; sin(ωt) is a periodic adjustment factor, used to simulate the possible periodic changes in network load; ω is the frequency parameter, used to control the rate of periodic change, set according to the expert experience method; ∈ t+1 is the error at time t + 1, reflecting the prediction error or random fluctuation, set as white noise (i.e., a random error with a mean of 0 and a constant variance), set according to the specific implementation scenario;

[0048] Finally, a resource demand prediction model is constructed, introducing a weighting coefficient and a decay coefficient, so that the predicted value of resource demand can be dynamically adjusted according to the timeliness of the true value of resource demand. To further improve the accuracy of resource demand prediction, a bias term is introduced and optimized in combination with the logarithmic function. The bias term is used to correct the bias in resource demand prediction, making the resource demand prediction result more in line with the actual demand; while the logarithmic function helps to smooth the influence of the true value of historical resource demand, improving the stability and accuracy of resource demand prediction, making it more flexible and accurate to adapt to the changes in actual network resource scheduling requirements;

[0049] The formula of the resource demand prediction model is as follows:

[0050]

[0051] Among them, is the predicted value of resource demand at time t + 1; m is the number of historical moments, indicating how many past time points are referred to for predicting future resource demand, set according to the specific implementation scenario; is the moment index variable; is The weighted coefficient of the true value of resource demand at a moment, obtained through training the resource demand prediction model, λ 0 is at time of explanation; is the decay coefficient, used to control the decay speed of the influence of the true value of resource demand on the predicted value of resource demand over time, and is set according to the expert experience method; is the true value of resource demand at time ; D t is the resource demand of the task at time t; δ is the bias term, obtained through training the resource demand prediction model.

[0052] S2: Based on the task priority, predicted network load, and predicted resource demand, calculate the comprehensive benefit of the task; based on the comprehensive benefit of the task, optimize the task scheduling decision through the reinforcement learning algorithm.

[0053] Based on the task priority, predicted network load, and predicted resource demand, use the comprehensive benefit calculation formula to calculate the comprehensive benefit of the task. When the predicted network load and predicted resource demand are used as the inputs of the comprehensive benefit calculation formula, the predicted comprehensive benefit of the task is calculated; when the true network load and true resource demand are used as the inputs of the comprehensive benefit calculation formula, the true comprehensive benefit of the task is calculated;

[0054] The comprehensive benefit calculation formula is as follows:

[0055]

[0056] where, is the predicted comprehensive benefit of the task, because the predicted network load and predicted resource demand α 1 、α 2 are the weight coefficients, used to balance the influence of the predicted resource demand, predicted network load, and task priority, and are set according to the expert experience method; is the predicted value of resource demand at time t + 1; is the predicted value of network load at time t + 1; P t is the task priority at time t; T t is the remaining execution time of the task at time t; T max is the maximum execution time of the task, set according to the expert experience method; exp is the exponential function; is the decay coefficient, used to adjust the influence of the network load on the predicted comprehensive benefit of the task, and is set according to the expert experience method;

[0057] Use the reinforcement learning algorithm to make task scheduling decisions. The reinforcement learning algorithm makes scheduling decisions based on the current network state information (task priority, resource requirements, and network load), and continuously adjusts the decision evaluation value (Q value) through a feedback mechanism. During the above process, the decision evaluation value is updated based on the predicted comprehensive benefit of the task to judge the quality of the current decision. By continuously updating the decision evaluation value, the reinforcement learning can dynamically optimize the scheduling strategy according to the predicted comprehensive benefit of the task, gradually improve the decision-making effect, and achieve optimal resource allocation;

[0058] The update formula for the decision evaluation value calculated based on the predicted comprehensive benefit of the task is as follows:

[0059]

[0060] Where, Q(s t , a t ) is the decision evaluation value at time t in state s t and action a t , representing the total return expected to be obtained by executing action a t when in state s t , and is the updated decision evaluation value; is the decision evaluation value before update; s t is the state at time t, which is a set of variables describing the situation of the network at a certain moment, including information such as network load, resource requirements of tasks, task priority, current time, etc., and is set according to the expert experience method; a t is the action selected at time t, representing the decision or behavior taken under state s t ; is the predicted comprehensive benefit of the task; γ is the discount factor, which reflects the trade-off between the current decision and future rewards, and is set according to the specific implementation scenario; is the maximum decision evaluation value of all possible actions a′ under the state s t+1 at the future time t + 1; a′ is the possible action at the future time t + 1;

[0061] Based on the task priority, predicted resource requirement value, predicted network load value, and updated decision evaluation value, finally make a task scheduling decision. The goal in the task scheduling decision process is to select the optimal scheduling action according to the current network state information. By calculating the objective function, evaluate the effects of each possible scheduling action, and select the action that can maximize the objective function. The objective function comprehensively considers factors such as task priority, resource requirements, and network load, and reflects the comprehensive benefit of task scheduling. Finally, the selected optimal action is the most suitable resource allocation plan for the task at the current moment, thereby optimizing resource usage and improving performance;

[0062] The scheduling decision formula is as follows:

[0063]

[0064] Among them, Decision t is the scheduling decision at time t; represents that in the current state s t select the action a that maximizes the objective function ; t is the weight coefficient of the resource requirement and task priority of the task, which is set according to the expert experience method; is the influence weight coefficient of network load on the scheduling decision, which is set according to the expert experience method.

[0065] After executing the scheduling decision, calculate the true comprehensive benefit of the task and use it to update the decision evaluation value to increase the effectiveness of the scheduling decision;

[0066] The update formula of the decision evaluation value calculated based on the true comprehensive benefit of the task is as follows:

[0067]

[0068] Among them, is the updated decision evaluation value; is the learning rate, which is set according to the expert experience method; R total is the actual comprehensive benefit of the task, using the true network load value corresponding to the network load prediction value and the true resource requirement value corresponding to the resource requirement prediction value to calculate; is the discount factor, which is set according to the expert experience rule; is the maximum Q value of all possible actions a' in the state s t+1 at the future time t + 1.

[0069] After completing the task scheduling decision and updating the decision evaluation value, continuously monitor the usage of network resources and collect resource scheduling data to provide support for subsequent network resource scheduling.

[0070] In summary, a network resource scheduling method based on artificial intelligence is completed.

[0071] The order of the invention embodiments is only for description and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0072] ​Each embodiment in this specification is described in a progressive manner. For the same or similar parts among the embodiments, reference can be made to each other, and the differences between each embodiment and other embodiments are emphasized.

[0073] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than limiting them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of each embodiment of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. A network resource scheduling method based on artificial intelligence, characterized in that: The following steps are involved: S1: Collect resource scheduling data and dynamically adjust task priorities; Based on the resource scheduling data, a network load prediction model and a resource demand prediction model are constructed to obtain network load prediction values ​​and resource demand prediction values; S2: Calculate the comprehensive benefits of the task based on task priority, network load prediction value, and resource demand prediction value; Based on the comprehensive benefits of the task, optimize the task scheduling decision through the reinforcement learning algorithm.

2. The network resource scheduling method based on artificial intelligence according to claim 1, characterized in that: The S1 specifically includes: The resource scheduling data in the network is collected in real time; the resource scheduling data includes the resource requirements of the task, network load, task priority, remaining execution time and elapsed execution time.

3. The network resource scheduling method based on artificial intelligence according to claim 2 is characterized in that: The S1 specifically includes: Based on the executed time and remaining execution time of the task, a priority balance coefficient is introduced to dynamically adjust the priority of the task.

4. The network resource scheduling method based on artificial intelligence according to claim 3 is characterized in that: The S1 specifically includes: Based on resource scheduling data and combined with the actual value of historical network load, an exponential decay factor and a periodic adjustment factor are introduced to build a network load prediction model to predict the network load at future times and obtain the network load prediction value.

5. The network resource scheduling method based on artificial intelligence according to claim 3 is characterized in that: The S1 specifically includes: Based on resource scheduling data and the actual value of resource demand, weighting coefficients and attenuation coefficients are introduced to build a resource demand prediction model to predict resource demand at future times and obtain the predicted value of resource demand.

6. The network resource scheduling method based on artificial intelligence according to claim 1, characterized in that: The S2 specifically includes: Based on the task priority, network load forecast value and resource demand forecast value, the comprehensive benefit of the task is calculated using the comprehensive benefit calculation formula.

7. The network resource scheduling method based on artificial intelligence according to claim 6, characterized in that: The S2 specifically includes: When the predicted network load value and the predicted resource demand value are used as the input of the comprehensive benefit calculation formula, the predicted comprehensive benefit of the task is calculated; when the actual network load and resource demand values ​​are used as the input of the comprehensive benefit calculation formula, the actual comprehensive benefit of the task is calculated; the comprehensive benefit calculation formula is as follows: in, is the predicted comprehensive benefit of the task; α1 and α2 are weight coefficients; is the predicted resource demand value at time t+1; is the predicted network load value at time t+1; P t is the task priority at time t; T t is the remaining execution time of the task at time t; T max is the maximum execution time of the task; exp is the exponential function; is the attenuation coefficient; L t is the network load at time t.

8. The network resource scheduling method based on artificial intelligence according to claim 7, characterized in that: The S2 specifically includes: The reinforcement learning algorithm makes scheduling decisions based on current network status information, and introduces a feedback mechanism to update the decision evaluation value based on the predicted comprehensive benefits of the task to obtain an updated decision evaluation value.

9. The network resource scheduling method based on artificial intelligence according to claim 8, characterized in that: The S2 specifically includes: Based on the task priority, resource demand prediction value, network load prediction value and updated decision evaluation value, combined with the current network status information, the objective function is constructed and maximized to make task scheduling decisions.

10. The network resource scheduling method based on artificial intelligence according to claim 9, characterized in that: The S2 specifically includes: After executing the scheduling decision, the real comprehensive benefit of the task is calculated based on the comprehensive benefit calculation formula; based on the real comprehensive benefit of the task, the decision evaluation value is updated, and the use of network resources is continuously monitored, and resource scheduling data is collected to provide support for subsequent network resource scheduling.

Images