A Machine Learning-Based IP Address Resource Allocation Method and System

Abstract

The present invention discloses a method and system for IP address resource allocation based on machine learning, which relates to the technical field of network resource allocation. The method includes: receiving an IP address allocation request message sent by a device to be allocated; in response to the IP address allocation request message, obtaining an IP address segment belonging to a gateway group from an IP address resource pool, dividing the IP address segment belonging to the gateway group into multiple sub-address ranges, and saving the divided multiple sub-address ranges into an address range status set; predicting the IP address requirement of the device to be allocated to obtain an IP address requirement prediction result; inputting the resource occupancy status of each sub-address range in the address range status set and the IP address requirement prediction result into a resource allocation model that has been trained and optimized, and outputting an IP address allocation policy; and allocating IP address resources to the device to be allocated according to the output IP address allocation policy. This application realizes dynamic and intelligent IP address allocation.

Description

Technical Field

[0001] This application relates to the technical field of network resource allocation. Specifically, it relates to a method and system for IP address resource allocation based on machine learning. Background Art

[0002] IP address resources refer to digital labels used to uniquely identify devices in a network in the Internet protocol. These resources are the basis of network communication, enabling data to be accurately sent to the correct destination. With the rapid development of Internet technology, the number of Internet users and terminal devices has increased sharply, and the demand for IP address resources has also grown accordingly. In particular, IPv4 address resources have gradually been exhausted and it is difficult to meet future needs.

[0003] Existing IP address resource allocation methods have several problems, including: low resource allocation efficiency: Traditional IP address allocation methods often use static allocation, lacking flexibility and resulting in low resource utilization. Lack of intelligent management: Current IP address management is mostly manual operation, lacking intelligent analysis capabilities.

[0004] In response to the above problems, no effective solution has been proposed yet. Summary of the Invention

[0005] Embodiments of this application provide a method and system for IP address resource allocation based on machine learning to solve the above technical problems.

[0006] The present application provides a method for allocating IP address resources based on machine learning, which is used for dynamic management and optimization of IP addresses in a network system, including: receiving an IP address allocation request message sent by a device to be allocated; wherein, an access point name is carried in the IP address allocation request message; in response to the IP address allocation request message, obtaining an IP address segment belonging to a gateway group from an IP address resource pool, dividing the IP address segment belonging to the gateway group into multiple sub-address ranges, and storing the divided multiple sub-address ranges into an address range status set; wherein, the gateway group is the group to which the access point name carried in the IP address allocation request message belongs; predicting the IP address demand of the device to be allocated to obtain an IP address demand prediction result; inputting the resource occupancy status of each sub-address range in the address range status set and the IP address demand prediction result into a resource allocation model that has been trained and optimized, and outputting an IP address allocation strategy; wherein, the resource allocation model is constructed based on a machine learning algorithm; allocating IP address resources to the device to be allocated according to the output IP address allocation strategy; wherein, the resource allocation model is trained and optimized based on the following steps: obtaining a historical gateway group and a set of historical IP address demands; obtaining an IP address segment belonging to the historical gateway group from the IP address resource pool, and dividing the IP address segment belonging to the historical gateway group into multiple historical sub-address ranges; determining a demand-sub-address range association decision matrix by using the Lagrangian dual decomposition method according to the multiple historical sub-address ranges and the set of historical IP address demands; constructing and training a random forest model by using the demand-sub-address range association decision matrix to adjust the model parameters of the random forest model; wherein, the model parameters include the number of decision trees, the maximum depth of the decision trees, and the minimum number of samples required for splitting each node; determining the trained and optimized random forest model as the resource allocation model.

[0007] Further, the predicting the IP address demand of the device to be allocated to obtain an IP address demand prediction result includes: collecting historical IP address usage data of the allocated devices in the network system; establishing and training and optimizing an IP address demand prediction model based on the long short-term memory neural network model according to the historical IP address usage data; wherein, the IP address demand prediction model includes an input gate layer, a forget gate layer, an update gate layer, and an output gate layer; predicting the IP address demand of the device to be allocated according to the trained and optimized IP address demand prediction model to obtain the IP address demand prediction result.

[0008] Further, the update gate layer is used to update the cell state, updating the old cell state to the current cell state ; ; wherein, ; t represents the current moment; represents the input weight of the update gate layer; represents the bias of the update gate layer; represents the state structure of the forget gate layer, represents the input vector of the IP address demand prediction model at the current moment; represents the activation vector of the update gate layer.

[0009] Further, the determining the demand-segment address interval association decision matrix according to the multiple historical segment address intervals and the historical IP address demand set by using the Lagrangian dual decomposition method includes: defining the objective function as: ; where, and are both index variables used to index elements in the set; represents the index of the historical IP address demand; represents the index of the historical segment address interval; represents the historical IP address demand set and the historical segment address interval the degree of association; is a prediction function used to represent the historical IP address demand and the historical segment address interval the possibility of association; defining the constraint condition as ; where, the constraint condition is used to represent that each historical IP address demand is associated with at most historical segment address intervals ;

[0010] Construct the Lagrangian function ; where, , are Lagrangian multipliers; is the optimal solution to be solved; find the first optimal Lagrangian multiplier and the second Lagrangian multiplier to maximize the dual function ; determine the optimal solution according to the first optimal Lagrangian multiplier and the second Lagrangian multiplier; construct the demand-segment address interval association decision matrix according to the optimal solution.

[0011] Further, after allocating IP address resources to the device to be allocated according to the output IP address allocation policy, the method further includes: monitoring the actual usage of the IP address resources allocated to the device to be allocated, and collecting real-time feedback data; wherein, the real-time feedback data includes the usage rate of the allocated IP address resources, the disconnection time of the device, the joining time of the device to the network, and the real-time change data of network traffic; evaluating the IP address allocation policy according to the collected real-time feedback data; and using the evaluation result and the real-time feedback data to adjust the IP address allocation policy.

[0012] Further, the historical IP address usage data includes IP address allocation data, device type, network traffic pattern, time period activity, geographical location, and connection duration.

[0013] Further, after allocating IP address resources to the device to be allocated according to the output IP address allocation policy, the method further includes: updating the resource occupancy status of each sub-address range in the address range status set.

[0014] Further, allocating IP address resources to the device to be allocated according to the output IP address allocation policy includes: obtaining the service quality requirements of the network system and the load balancing requirements of the network system; and allocating IP address resources to the device to be allocated according to the output IP address allocation policy, the service quality requirements of the network system, and the load balancing requirements of the network system.

[0015] The present application provides an IP address resource allocation system based on machine learning, which is used for dynamic management and optimization of IP addresses in a network system, including an allocation request message receiving module, an address range division module, an IP address demand prediction module, an IP address allocation policy determination module, and an IP address resource allocation module; wherein, the allocation request message receiving module is configured to receive an IP address allocation request message sent by a device to be allocated; wherein, the access point name is carried in the IP address allocation request message; the address range division module is configured to, in response to the IP address allocation request message, obtain an IP address segment belonging to a gateway group from an IP address resource pool, divide the IP address segment belonging to the gateway group into a plurality of sub-address ranges, and save the divided plurality of sub-address ranges into an address range status set; wherein, the gateway group is the group to which the access point name carried in the IP address allocation request message belongs; the IP address demand prediction module is configured to predict the IP address demand of the device to be allocated to obtain an IP address demand prediction result; the IP address allocation policy determination module is configured to input the resource occupancy status of each sub-address range in the address range status set and the IP address demand prediction result into a resource allocation model that has been trained and optimized, and output an IP address allocation policy; wherein, the resource allocation model is constructed based on a machine learning algorithm; the IP address resource allocation module is configured to allocate IP address resources to the device to be allocated according to the output IP address allocation policy; wherein, the resource allocation model is trained and optimized based on the following steps: obtaining a historical gateway group and a historical IP address demand set; obtaining an IP address segment belonging to the historical gateway group from the IP address resource pool, and dividing the IP address segment belonging to the historical gateway group into a plurality of historical sub-address ranges; determining a demand-sub-address range association decision matrix by using the Lagrangian dual decomposition method according to the plurality of historical sub-address ranges and the historical IP address demand set; constructing and training a random forest model by using the demand-sub-address range association decision matrix to adjust the model parameters of the random forest model; wherein, the model parameters include the number of decision trees, the maximum depth of the decision trees, and the minimum number of samples required for splitting each node; and determining the trained and optimized random forest model as the resource allocation model.

[0016] Based on the embodiments provided in this application, an IP address allocation request message sent by a device to be allocated is received; wherein, the access point name is carried in the IP address allocation request message; in response to the IP address allocation request message, an IP address segment belonging to the gateway group is obtained from the IP address resource pool, the IP address segment belonging to the gateway group is divided into multiple sub-address ranges, and the divided multiple sub-address ranges are saved to the address range status set; wherein, the gateway group is the group to which the access point name carried in the IP address allocation request message belongs; the IP address demand of the device to be allocated is predicted to obtain an IP address demand prediction result; the resource occupancy status of each sub-address range in the address range status set and the IP address demand prediction result are input into the trained and optimized resource allocation model, and an IP address allocation strategy is output; wherein, the resource allocation model is constructed based on a machine learning algorithm; according to the output IP address allocation strategy, IP address resources are allocated to the device to be allocated. Thus, it is realized to predict future IP address demands using a machine learning algorithm and achieve dynamic and intelligent IP address allocation. By this method, the utilization rate of IP address resources can be improved, the dynamic changes of the network system can be adapted, at the same time, network management can be simplified, operation costs can be reduced, and greater convenience and economic benefits can be brought to users and operators. Brief Description of the Drawings

[0017] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, form a part of this application, and the illustrative embodiments and descriptions thereof are used to explain this application and do not constitute an improper limitation of this application. In the drawings:

[0018] Figure 1 It is a flowchart of an optional IP address resource allocation method based on machine learning according to an embodiment of this application.

[0019] Figure 2 It is a flowchart of another optional IP address resource allocation method based on machine learning according to an embodiment of this application.

[0020] Figure 3 It is a structural diagram of an optional IP address resource allocation system based on machine learning according to an embodiment of this application.

[0021] The realization, functional features and advantages of the objectives of the present invention will be further described with reference to the embodiments and the drawings. Detailed Embodiments

[0022] In order to enable those skilled in the art of this technology to better understand the solution of this application, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the drawings in the embodiments of this application.

[0023] Optionally, as Figure 1As shown in the figure, the present application provides an IP address resource allocation method based on machine learning for dynamic management and optimization of IP addresses in a network system, including: S101, receiving an IP address allocation request message sent by a device to be allocated; wherein, the access point name is carried in the IP address allocation request message; S102, in response to the IP address allocation request message, obtaining an IP address segment belonging to the gateway group from the IP address resource pool, dividing the IP address segment belonging to the gateway group into multiple sub-address ranges, and saving the divided multiple sub-address ranges into the address range status set; wherein, the gateway group is the group to which the access point name carried in the IP address allocation request message belongs; S103, predicting the IP address demand of the device to be allocated to obtain an IP address demand prediction result; S104, inputting the resource occupancy status of each sub-address range in the address range status set and the IP address demand prediction result into the resource allocation model that has been trained and optimized, and outputting an IP address allocation policy; wherein, the resource allocation model is constructed based on a machine learning algorithm; S105, allocating IP address resources to the device to be allocated according to the output IP address allocation policy.

[0024] Among them, the resource allocation model is trained and optimized based on the following steps: obtaining a historical gateway group and a historical IP address demand set; obtaining an IP address segment belonging to the historical gateway group from the IP address resource pool, and dividing the IP address segment belonging to the historical gateway group into multiple historical sub-address ranges; determining a demand-sub-address range association decision matrix by using the Lagrangian dual decomposition method according to the multiple historical sub-address ranges and the historical IP address demand set; constructing and training a random forest model by using the demand-sub-address range association decision matrix to adjust the model parameters of the random forest model; wherein, the model parameters include the number of decision trees, the maximum depth of the decision trees, and the minimum number of samples required for splitting each node; determining the trained and optimized random forest model as the resource allocation model.

[0025] Based on the embodiments provided in this application, an IP address allocation request message sent by a device to be allocated is received; wherein, the access point name is carried in the IP address allocation request message; in response to the IP address allocation request message, an IP address segment belonging to the gateway group is obtained from the IP address resource pool, the IP address segment belonging to the gateway group is divided into multiple sub-address ranges, and the divided multiple sub-address ranges are saved to the address range status set; wherein, the gateway group is the group to which the access point name carried in the IP address allocation request message belongs; the IP address demand of the device to be allocated is predicted to obtain an IP address demand prediction result; the resource occupancy status of each sub-address range in the address range status set and the IP address demand prediction result are input into the resource allocation model that has been trained and optimized, and an IP address allocation policy is output; wherein, the resource allocation model is constructed based on a machine learning algorithm; according to the output IP address allocation policy, IP address resources are allocated to the device to be allocated. Thus, it realizes the analysis of network traffic and usage patterns using machine learning algorithms, predicts future IP address demands, and achieves dynamic and intelligent IP address allocation. Through this method, the utilization rate of IP address resources can be improved, the dynamic changes of the network can be adapted, the network management can be simplified, the operation cost can be reduced, and greater convenience and economic benefits can be brought to users and operators.

[0026] Further, as Figure 2 shown, predicting the IP address demand of the device to be allocated to obtain an IP address demand prediction result includes: S201, collecting historical IP address usage data of the allocated devices in the network system; S202, based on the long short-term memory neural network model, establishing and training and optimizing the IP address demand prediction model according to the historical IP address usage data; wherein, the IP address demand prediction model includes an input gate layer, a forget gate layer, an update gate layer, and an output gate layer; S203, according to the IP address demand prediction model that has been trained and optimized, predicting the IP address demand of the device to be allocated to obtain an IP address demand prediction result.

[0027] Based on the embodiments provided in this application, by using a long short-term memory neural network model, long-term dependencies in time series data can be captured, thereby improving the prediction accuracy of IP address requirements; by accurately predicting IP address requirements, IP address resources can be planned and allocated more effectively, reducing resource waste; the collection and analysis of historical IP address usage data help to understand network usage patterns, thus enabling more scientific network planning; the automated IP address requirement prediction model reduces manual intervention and improves the efficiency of network management; the LSTM model can adapt to the dynamic changes in network requirements, enabling the prediction model to timely reflect the latest network usage trends; by reducing unnecessary IP address allocations, network operation costs can be reduced; reasonable IP address allocation can avoid network congestion and improve the overall performance of the network; timely meeting the IP address requirements of devices can enhance the network experience of end users; the prediction results can provide decision-making support for network expansion and upgrade; the prediction model can be adjusted and expanded according to the growth of the network scale to adapt to changing requirements; by predicting future IP address requirements, network administrators can flexibly adjust network configurations to adapt to different usage scenarios; the collected historical data and prediction results can be used for further analysis to promote data-driven network management.

[0028] Furthermore, the update gate layer is used to update the cell state, updating the old cell state to the current cell state ; wherein, t represents the current time; represents the input weight of the update gate layer; represents the bias of the update gate layer; represents the state structure of the forget gate layer, represents the input vector of the IP address requirement prediction model at the current time; represents the activation vector of the update gate layer.

[0029] Based on the embodiments provided in this application, the update gate layer dynamically determines which information should be retained in the cell state, which information should be updated or forgotten, enabling the model to capture the dynamic changes in the data. The update gate layer helps the long short-term memory neural network model capture long-term dependencies, which is particularly important for sequence prediction tasks such as IP address demand prediction; by effectively updating the cell state, the model can more accurately predict future IP address demands; the model can adapt to changes in time series data and respond quickly to changes in network conditions; the calculation formula of the update gate layer ensures the smoothness and continuity of state transitions, helping to improve the prediction stability of the model; by reasonably updating the cell state, the model can avoid overfitting to the training data and improve the generalization ability; the working mechanism of the update gate layer can help researchers and network administrators understand how the model makes predictions based on historical data; the model can flexibly process sequence data of different lengths and adapt to different prediction tasks; the update gate layer enables the model to update the cell state in real time for real-time prediction; even in the presence of noise in the input data, the update gate layer can help the model maintain stable prediction performance; the update gate layer helps fuse feature information from different time steps into the cell state to improve the prediction accuracy; by adjusting the parameters of the update gate layer, the model can be optimized to adapt to the prediction task; for complex time series data, the update gate layer can help the model identify and understand complex patterns in the data.

[0030] Further, according to multiple historical sub-address intervals and the historical IP address demand set, the Lagrangian dual decomposition method is used to determine the demand-sub-address interval association decision matrix, including: defining the objective function as: ; where and are both index variables used to index elements in the set; represents the index of the historical IP address demand; represents the index of the historical sub-address interval; represents the historical IP address demand set and the historical sub-address interval ; is a prediction function used to represent the possibility of association between the historical IP address demand and the historical sub-address interval ; defining the constraint condition as ; where the constraint condition is used to represent that each historical IP address demand is associated with at most historical sub-address intervals ; constructing the Lagrangian function ; where , is the Lagrange multiplier; is the optimal solution to be solved; find the first optimal Lagrange multiplier and the second Lagrange multiplier to maximize the dual function ; determine the optimal solution according to the first optimal Lagrange multiplier and the second Lagrange multiplier; construct a demand-segmented address interval association decision matrix according to the optimal solution.

[0031] Based on the embodiments provided in this application, use the integrated learning method to construct multiple decision trees, and improve the stability and accuracy of prediction through the voting mechanism; random forests can handle a large amount of data and features, reduce the risk of overfitting, and improve the generalization ability of the model; optimizing model parameters can improve the performance of the model and make it better adapt to specific data sets and application scenarios; the association decision matrix provides a clear guidance for resource allocation, making the allocation decision more scientific and reasonable; this process can ensure that the optimal solution of resource allocation is found under the premise of meeting the constraint conditions; the optimal solution provides an optimal strategy for resource allocation under all constraint conditions.

[0032] Further, after allocating IP address resources to the device to be allocated according to the output IP address allocation policy, the method further includes: monitoring the actual usage of the IP address resources allocated to the device to be allocated and collecting real-time feedback data; wherein, the real-time feedback data includes the usage rate of the allocated IP address resources, the disconnection time of the device, the joining time of the device to the network, and the real-time change data of network traffic; evaluating the IP address allocation policy according to the collected real-time feedback data; using the evaluation result and the real-time feedback data to adjust the IP address allocation policy.

[0033] Further, the historical IP address usage data includes IP address allocation data, device type, network traffic pattern, time period activity, geographical location, and connection duration.

[0034] Further, after allocating IP address resources to the device to be allocated according to the output IP address allocation policy, the method further includes: updating the resource occupancy status of each segmented address interval in the address interval status set.

[0035] Further, combine the IP address allocation policy output by the resource allocation model with the adjustment result of the network administrator to optimize the IP address allocation policy.

[0036] Further, allocating IP address resources to the device to be allocated according to the output IP address allocation policy includes: obtaining the service quality requirements of the network system and the load balancing requirements of the network system; allocating IP address resources to the device to be allocated according to the output IP address allocation policy, the service quality requirements of the network system, and the load balancing requirements of the network system.

[0037] Optionally, as Figure 3 shown, this application provides an IP address resource allocation system based on machine learning, which is used for dynamic management and optimization of IP addresses in a network system, including an allocation request message receiving module 301, an address range division module 302, an IP address demand prediction module 303, an IP address allocation policy determination module 304, and an IP address resource allocation module 305; among them, the allocation request message receiving module 301 is used to receive an IP address allocation request message sent by a device to be allocated; among them, the IP address allocation request message carries an access point name; the address range division module 302 is used to, in response to the IP address allocation request message, obtain an IP address segment belonging to a gateway group from the IP address resource pool, divide the IP address segment belonging to the gateway group into multiple sub-address ranges, and save the divided multiple sub-address ranges to an address range status set; among them, the gateway group is the group to which the access point name carried in the IP address allocation request message belongs; the IP address demand prediction module 303 is used to predict the IP address demand of the device to be allocated to obtain an IP address demand prediction result; the IP address allocation policy determination module 304 is used to input the resource occupancy status of each sub-address range in the address range status set and the IP address demand prediction result into a resource allocation model that has been trained and optimized, and output an IP address allocation policy; among them, the resource allocation model is constructed based on a machine learning algorithm; the IP address resource allocation module 305 is used to allocate IP address resources to the device to be allocated according to the output IP address allocation policy.

[0038] Among them, the resource allocation model is trained and optimized based on the following steps: obtaining a historical gateway group and a historical IP address demand set; obtaining an IP address segment belonging to the historical gateway group from the IP address resource pool, and dividing the IP address segment belonging to the historical gateway group into multiple historical sub-address ranges; according to the multiple historical sub-address ranges and the historical IP address demand set, using the Lagrangian dual decomposition method to determine a demand-sub-address range association decision matrix; using the demand-sub-address range association decision matrix to construct and train a random forest model to adjust the model parameters of the random forest model; among them, the model parameters include the number of decision trees, the maximum depth of the decision trees, and the minimum number of samples required for splitting each node; determining the trained and optimized random forest model as the resource allocation model.

[0039] Based on the embodiments provided in this application, through the allocation request message receiving module, address range division module, IP address demand prediction module, IP address allocation policy determination module, and IP address resource allocation module, it is possible to predict future IP address demands using machine learning algorithms and achieve dynamic and intelligent IP address allocation. By this method, the utilization rate of IP address resources can be improved, the dynamic changes of the network system can be adapted, while simplifying network management, reducing operating costs, and bringing greater convenience and economic benefits to users and operators.

[0040] It should be noted that in this application, the embodiments implemented on the side of the IP address resource allocation system based on machine learning can be referred to each other with the embodiments implemented on the side of the IP address resource allocation method based on machine learning, and this application will not elaborate on them one by one.

[0041] The above are only the preferred embodiments of the present invention, and do not limit the patent scope of the present invention accordingly. Any equivalent structural or equivalent process transformations made using the content of the specification and drawings of the present invention, or directly or indirectly applied in other related technical fields, are similarly included in the patent protection scope of the present invention.

Claims

1. A machine learning-based IP address resource allocation method for dynamic management and optimization of IP addresses in a network system, characterized in that: include: Receiving an IP address allocation request message sent by a device to be allocated; wherein the IP address allocation request message carries an access point name; In response to the IP address allocation request message, obtain an IP address segment belonging to a gateway group from an IP address resource pool, divide the IP address segment belonging to the gateway group into a plurality of subdivided address intervals, and save the divided plurality of subdivided address intervals to an address interval state set; wherein the gateway group is the group to which the access point name carried in the IP address allocation request message belongs; Predicting the IP address demand of the device to be allocated to obtain an IP address demand prediction result; Input the resource occupancy status of each subdivided address interval in the address interval status set and the IP address demand prediction result into the trained and optimized resource allocation model, and output the IP address allocation strategy; wherein the resource allocation model is constructed based on a machine learning algorithm; Allocate IP address resources to the device to be allocated according to the output IP address allocation policy; The resource allocation model is trained and optimized based on the following steps: Get the historical gateway group and historical IP address demand set; Acquire the IP address segment belonging to the historical gateway group from the IP address resource pool, and divide the IP address segment belonging to the historical gateway group into a plurality of historical subdivided address intervals; Determine a demand-segmented address interval association decision matrix using Lagrange dual decomposition method according to the multiple historical segmented address intervals and the historical IP address demand set; Using the demand-segmented address interval association decision matrix, a random forest model is constructed and trained to adjust model parameters of the random forest model; wherein the model parameters include the number of decision trees, the maximum depth of the decision trees, and the minimum number of samples required for each node to split; Determine the random forest model that has been trained and optimized as the resource allocation model; The predicting of the IP address demand of the device to be allocated to obtain the IP address demand prediction result includes: Collect historical IP address usage data of allocated devices in the network system; Based on the long short-term memory neural network model, an IP address demand prediction model is established and trained to optimize the IP address demand prediction model according to the historical IP address usage data; wherein the IP address demand prediction model includes an input gate layer, a forget gate layer, an update gate layer and an output gate layer; According to the trained and optimized IP address demand prediction model, the IP address demand of the device to be allocated is predicted to obtain the IP address demand prediction result; The update gate layer is used to update the cell status and convert the old cell status to Status t-1 Update to the current cell status Status t ; Status t =σ(W f ×[h t ,x t ]+b f )×Status t-1 +σ(Update t ×Status t-1 ) Where σ represents the sigmoid function; t represents the current time; W f represents the input weight of the update gate layer; b f represents the bias of the update gate layer; h t represents the state structure of the forget gate layer, x t represents the input vector of the IP address demand prediction model at the current moment; Update t represents the activation vector of the update gate layer; The method of determining a demand-segmented address interval association determination matrix using the Lagrange dual decomposition method according to the multiple historical segmented address intervals and the historical IP address demand set includes: The objective function is defined as: Where i and j are both indicator variables used to index elements in a set; i represents the index of historical IP address demand; j represents the index of historical subdivided address intervals; x ij The historical IP address demand set D = {d1, d2, ..., d m } and the historical subdivision address interval S = {s1, s2, ..., s n }; p ij is a prediction function used to represent the historical IP address demand d i With historical subdivision address intervals j Possibility of association; Define the constraints as The constraint condition is used to represent each historical IP address requirement d i Associated with at most k historical segment address intervals s j ; Construct the Lagrangian function L(X, ω i , μ ij ); Among them, ω i , μ ij is the Lagrange multiplier; X is the optimal solution to be solved; Find the first optimal Lagrangian multiplier and the second Lagrangian multiplier so that the dual function g(ω i , μ ij )=MinimizeL(X,ω i , μ ij )maximize; Determining the optimal solution according to the first optimal Lagrangian multiplier and the second Lagrangian multiplier; According to the optimal solution, the demand-subdivided address interval association determination matrix is ​​constructed.

2. The method for allocating IP address resources based on machine learning according to claim 1, characterized in that: After allocating IP address resources to the device to be allocated according to the output IP address allocation policy, the method further includes: Monitor the actual usage of the IP address resources allocated to the device to be allocated and collect real-time feedback data; wherein the real-time feedback data includes the usage rate of the allocated IP address resources, the time when the device is disconnected, the time when the device joins the network, and the real-time change data of the network traffic; Evaluating the IP address allocation strategy based on the collected real-time feedback data; The evaluation results and the real-time feedback data are used to adjust the IP address allocation strategy.

3. The method for allocating IP address resources based on machine learning according to claim 1, characterized in that: Historical IP address usage data includes IP address allocation data, device type, network traffic patterns, time period activity, geographic location, and connection duration.

4. The method for allocating IP address resources based on machine learning according to claim 1, characterized in that: After allocating IP address resources to the device to be allocated according to the output IP address allocation policy, the method further includes: The resource occupation status of each subdivided address interval in the address interval status set is updated.

5. The method for allocating IP address resources based on machine learning according to claim 1, characterized in that: The allocating IP address resources to the device to be allocated according to the output IP address allocation policy includes: Obtaining the quality of service requirements of the network system and the load balancing requirements of the network system; According to the output IP address allocation strategy, the service quality requirement of the network system and the load balancing requirement of the network system, IP address resources are allocated to the device to be allocated.

6. A machine learning-based IP address resource allocation system, the system implementing the method as claimed in claim 1, the system is used for dynamic management and optimization of IP addresses in a network system, characterized in that: It includes an allocation request message receiving module, an address interval division module, an IP address demand prediction module, an IP address allocation strategy determination module and an IP address resource allocation module; wherein, The allocation request message receiving module is used to receive an IP address allocation request message sent by a device to be allocated; wherein the IP address allocation request message carries an access point name; The address interval division module is used to obtain the IP address segment belonging to the gateway group from the IP address resource pool in response to the IP address allocation request message, divide the IP address segment belonging to the gateway group into multiple subdivided address intervals, and save the divided multiple subdivided address intervals to the address interval state set; wherein the gateway group is the group to which the access point name carried in the IP address allocation request message belongs; The IP address demand prediction module is used to predict the IP address demand of the device to be allocated and obtain an IP address demand prediction result; The IP address allocation strategy determination module is used to input the resource occupancy status of each subdivided address interval in the address interval status set and the IP address demand prediction result into the resource allocation model that has been trained and optimized, and output the IP address allocation strategy; wherein the resource allocation model is constructed based on a machine learning algorithm; The IP address resource allocation module is used to allocate IP address resources to the device to be allocated according to the output IP address allocation policy; The resource allocation model is trained and optimized based on the following steps: Get the historical gateway group and historical IP address demand set; Acquire the IP address segment belonging to the historical gateway group from the IP address resource pool, and divide the IP address segment belonging to the historical gateway group into a plurality of historical subdivided address intervals; Determine a demand-segmented address interval association decision matrix using Lagrange dual decomposition method according to the multiple historical segmented address intervals and the historical IP address demand set; Using the demand-segmented address interval association decision matrix, a random forest model is constructed and trained to adjust model parameters of the random forest model; wherein the model parameters include the number of decision trees, the maximum depth of the decision trees, and the minimum number of samples required for each node to split; Determine the random forest model that has been trained and optimized as the resource allocation model; The IP address demand of the device to be allocated is predicted to obtain the IP address demand prediction result, including: Collect historical IP address usage data of allocated devices in the network system; Based on the long short-term memory neural network model, an IP address demand prediction model is established and trained to optimize the IP address demand prediction model according to the historical IP address usage data; wherein the IP address demand prediction model includes an input gate layer, a forget gate layer, an update gate layer and an output gate layer; According to the trained and optimized IP address demand prediction model, the IP address demand of the device to be allocated is predicted to obtain the IP address demand prediction result; The update gate layer is used to update the cell status and convert the old cell status to Status t-1 Update to the current cell status Status t ; Status t =σ(W f ×[h t ,x t ]+b f )×Status t-1 +σ(Update t ×Status t-1 ) Where σ represents the sigmoid function; t represents the current time; W f represents the input weight of the update gate layer; b f represents the bias of the update gate layer; h t represents the state structure of the forget gate layer, x t represents the input vector of the IP address demand prediction model at the current moment; Update t represents the activation vector of the update gate layer; The method of determining a demand-segmented address interval association determination matrix using the Lagrange dual decomposition method according to the multiple historical segmented address intervals and the historical IP address demand set includes: The objective function is defined as: Where i and j are both indicator variables used to index elements in a set; i represents the index of historical IP address demand; j represents the index of historical subdivided address intervals; x ij The historical IP address demand set D = {d1, d2, ..., d m } and the historical subdivision address interval S = {s1, s2, ..., s n }; p ij is a prediction function used to represent the historical IP address demand d i With historical subdivision address intervals j Possibility of association; Define the constraints as The constraint condition is used to represent each historical IP address requirement d i Associated with at most k historical segment address intervals s j ; Construct the Lagrangian function L(X, ω i , μ ij ); Among them, ω i , μ ij is the Lagrange multiplier; X is the optimal solution to be solved; Find the first optimal Lagrangian multiplier and the second Lagrangian multiplier so that the dual function g(ω i , μ ij )=MinimizeL(X,ω i , μ ij )maximize; Determining the optimal solution according to the first optimal Lagrangian multiplier and the second Lagrangian multiplier; According to the optimal solution, the demand-subdivided address interval association determination matrix is ​​constructed.

Images