Method and system for intelligent allocation of mobile Internet access IP addresses based on dynamic strategy

Abstract

This invention discloses a method and system for intelligently allocating IP addresses for mobile internet access based on dynamic policies. The method comprises the following steps: S1: collecting access behavior, network status, and address pool data, constructing a multi-source feature dataset, and performing masking processing; S2: constructing a masked autoencoder model to generate a reconstructed network status representation vector; S3: applying a proximal policy optimization algorithm to generate an allocation policy, dispatching IP addresses, and recording the results and feedback; S4: calculating a policy execution reward and updating policy parameters; and S5: inputting feedback information into the masked autoencoder model for cyclic optimization and control. By integrating the masked autoencoder with the proximal policy optimization algorithm, the present invention achieves dynamic perception and continuous optimization control of intelligent IP address allocation policies for mobile internet access scenarios.

Description

Technical Field

[0001] The present invention relates to the technical field of computer networks, and in particular to a method and system for intelligently allocating mobile Internet access IP addresses based on dynamic strategies. Background Art

[0002] With the rapid development of the mobile internet, the dynamic management and allocation of IP address resources has become a critical issue that needs to be addressed in wireless communication systems. Traditional IP address allocation methods, often based on static allocation strategies or preset rules, such as first-come, first-served (FIFO) allocation by a DHCP server or MAC address mapping, are difficult to adapt to complex and changing network environments. In scenarios with frequent user access, volatile network loads, and limited address pool resources, these methods often suffer from inefficient allocation, increased probability of address conflicts, and wasted resources. This leads to reduced overall system performance and an inability to meet the demands of mobile terminals for stable connectivity and low-latency access.

[0003] Existing research methods have attempted to optimize the IP address allocation process by introducing rule engines, heuristic searches, or simple machine learning models. However, these approaches often suffer from weak generalization and inadequate state modeling capabilities. Existing methods struggle to construct high-quality state representations, particularly when dealing with high-dimensional, nonlinear, and heterogeneous access behavior and network state data from multiple sources, making it difficult to effectively support the adaptive generation of allocation policies. Furthermore, these methods generally lack effective utilization of network state feedback after policy execution, resulting in delayed policy updates and difficulty achieving long-term stable allocation performance.

[0004] Although deep reinforcement learning has been increasingly applied to network resource scheduling in recent years and demonstrates promising policy self-learning capabilities, it still faces challenges in terms of the accuracy of state input and the controllability of policy output. Existing deep policy models generally rely on direct input of raw data to model network states, failing to address missing, incomplete, or noisy data, which impacts model perception accuracy. Furthermore, the policy optimization process often lacks effective prior guidance mechanisms and policy stability constraints, which can easily lead to policy fluctuations and allocation anomalies.

[0005] Therefore, the existing technology urgently needs a new method that integrates state representation enhancement, self-supervised learning, and enhanced policy optimization. This method can extract robust network state features from multi-source data, introduce feedback loops and structural constraints into the policy generation and update process, and improve the intelligence, stability, and resource utilization efficiency of IP address allocation. This invention proposes an innovative solution to this problem and overcomes the shortcomings of the existing technology.

[0006] Therefore, how to provide a method and system for intelligently allocating mobile Internet access IP addresses based on dynamic policies is a problem that those skilled in the art need to solve urgently. Summary of the Invention

[0007] One purpose of the present invention is to propose a method and system for intelligent allocation of mobile Internet access IP addresses based on dynamic strategies. The present invention integrates a masked autoencoder model with a proximal strategy optimization algorithm to construct an intelligent control process that perceives multi-source feature data and adaptively generates and optimizes IP allocation strategies. Specifically, by introducing a multi-scale masking mechanism to process access behavior and network status data, the robustness of state modeling is improved, and by combining strategy trust interval adjustment, external condition guidance and constraint regularization mechanisms, the stable generation and efficient optimization of allocation strategies are achieved. This method has the advantages of strong state perception capability, high flexibility in strategy execution, excellent resource utilization efficiency and strong ability to adapt to dynamic network environments. It is suitable for dynamic scheduling and refined management of IP addresses in complex scenarios.

[0008] According to an embodiment of the present invention, a method for intelligently allocating mobile Internet access IP addresses based on a dynamic policy includes the following steps:

[0009] S1. Collect access behavior data, current network status information, and IP address pool usage data of mobile terminals, construct a multi-source feature data set, and perform masking processing on the multi-source feature data set to form masked input data;

[0010] S2. Build a masked autoencoder model, input the masked input data into the masked autoencoder model, perform encoding and decoding operations, and generate a reconstructed network state representation vector;

[0011] S3. Using the network state representation vector as input, applying the proximal policy optimization algorithm to generate and optimize the IP address allocation policy, controlling the address scheduler to allocate IP addresses from the target address pool to the corresponding mobile terminal according to the IP address allocation policy, and recording the allocation execution results and network state feedback information;

[0012] S4. Calculate a policy execution reward value based on the allocation execution result, and use the policy execution reward value to update policy parameters in the proximal policy optimization algorithm;

[0013] S5. Input the network status feedback information into the masked autoencoder model, adjust the encoder and decoder parameters, and cyclically execute steps S1 to S4 within a preset allocation period to achieve continuous optimization and allocation control of the dynamic IP address strategy.

[0014] Optionally, the masking processing of the multi-source feature data set refers to masking some feature dimensions in the multi-source feature data set at a masking ratio of 30% in the constructed multi-source feature data set; the masking operation includes setting the position of the selected feature to zero, marking it as an invalid value, and replacing it with a mean alternative value; the masking strategy adopted is a hybrid strategy combining random masking and feature importance-guided masking, wherein 70% of the masking positions are determined by randomly selecting feature dimensions, and the remaining 30% of the masking positions are determined according to the influence weight ranking of the features in the prior evaluation, and high-weight key features are masked first.

[0015] Optionally, the S2 specifically includes:

[0016] S21. Constructing a masked input dataset of multi-source feature data ,in, is the sample size, is the feature dimension, is a set of real numbers, generating multiple masked versions of the masked input data set , where the masking ratio ;

[0017] S22. Calculate the feature correlation matrix based on the collaborative variability between each feature dimension in the masked input data set ,in, Representation characteristics and features The statistical correlation coefficient of the feature correlation matrix satisfies the correlation threshold Constructing heterogeneous graph structures based on feature pairs , where the node set Represents feature dimension, edge set Express satisfaction The connection relationship of , and the graph attention mechanism is used to calculate the structural importance score of each node;

[0018] S23. For each masked version A hybrid masking operation is performed based on the structure importance score, including the ratio The mask positions are generated by random selection and graph-guided selection to form the mask input matrix ;

[0019] S24, construct a dual encoder structure, including a main encoder and an auxiliary encoder, extract the network status feedback information and strategy execution records in the previous round of allocation cycle, and construct a historical status data set. The main encoder receives the mask input matrix , the auxiliary encoder receives historical status data ;

[0020] S25, respectively input the mask into the matrix and historical status data Input to the main encoder and auxiliary encoder, and use the fusion mechanism to calculate the potential state vector ;

[0021] S26. Constructing a conditional decoder , collect external constraint information related to the current access strategy, including network topology status, service level requirements, address pool tension, terminal device category, and construct external condition variables , the potential state vector With external condition variables Splicing, input to the decoder for conditional reconstruction;

[0022] S27. Construct multiple structurally heterogeneous parallel paths in the decoder, including a fully connected decoding path, a residual skip connection path, and an attention enhancement path, and output preliminary reconstruction results respectively. ;

[0023] S28. Constructing a strategy feedback guided fusion controller module ,in Indicates the feedback information of the previous round of strategy execution, including the strategy execution reward value , IP allocation success rate and address pool remaining rate , the fusion controller is based on the latent state vector and feedback information Jointly generate decoding path fusion weights , and introduce task attention compensation output and historical residual compensation vector , calculate the final reconstruction output ;

[0024] S29, based on the final reconstruction output Combined with the previous round of strategy allocation execution data, a feature contribution analysis module is constructed to calculate the contribution score of each feature dimension in the final reconstructed output in strategy generation. , forming a confidence vector , and use the confidence threshold Filter to meet The features of the network state representation vector , where d is the total number of feature dimensions.

[0025] Optionally, the S3 specifically includes:

[0026] S31. Network state representation vector based on the output of the masked autoencoder model , and combined with the external condition variables corresponding to the current mobile terminal , construct joint input ;

[0027] S32, the joint input Input to the policy network , output the probability distribution of IP address allocation action in the current state, the action Indicates the candidate IP address allocation scheme;

[0028] S33, the joint input Input to the value network , estimate the expected value under the current state , and use the advantage function to evaluate the pros and cons of the strategy;

[0029] S34. Calculate strategy update ratio based on importance sampling ;

[0030] S35. Constructing the advantage function using generalized advantage estimation ;

[0031] S36. Constructing a strategy objective function including dynamic clipping coefficients ;

[0032] S37, introduce the strategy trust interval adjustment module based on the masked autoencoder perception mechanism, for each assigned action Introduce upper and lower bound confidence constraints, which are expressed according to the network state vector The characteristic masking entropy value and external condition variables The key constraint indicators in the are jointly determined to construct a set of legal actions , and filter the effectiveness of the strategy output actions, and only keep those that meet:

[0033] ;

[0034] in, is a dynamic threshold, which is adjusted jointly according to the mask reconstruction error and the strategy stability. is the optimal allocation action finally selected, is the set of legal actions, Represented in the network state representation vector The probability of selecting the assigned action a under the given conditions of the external condition variable C;

[0035] S38. Use the back propagation method to update the policy network parameters respectively and value network parameters , minimize the policy objective function And maintain the stability of the strategy’s credible interval;

[0036] S39. Use the updated conditional policy network Generate the optimal action, which is the optimal allocation plan in the target IP address pool;

[0037] S310, control the address scheduler to allocate the IP address in the target address pool to the current external condition variable according to the generated optimal action the associated mobile terminal;

[0038] S311. Record the allocation execution result, including the allocated IP address, terminal identification, allocation delay, and allocation status information;

[0039] S312: Collect network status feedback information during the current allocation process, including connection success rate, address pool occupancy rate, and allocation failure rate, for use in masked autoencoder modeling and strategy update for the next allocation cycle.

[0040] Optionally, the S4 specifically includes:

[0041] S41. Collect feedback indicators after the allocation action is executed and construct the allocation execution result vector ,in, Indicates that the IP allocation is successful. represents the normalized delay required to complete the allocation, Indicates the service quality matching degree, Indicates the improvement in the target address pool availability caused by the current allocation behavior.

[0042] S42. Calculate the strategy execution reward value based on the feedback index ;

[0043] S43. Introduce a sliding window mechanism to calculate the difference between the current strategy execution reward value and the average strategy execution reward value in the historical window. , used to judge the stability of strategy behavior. When the absolute value of the difference exceeds the set threshold, the steady-state adjustment mechanism is triggered to suppress drastic parameter changes;

[0044] S44, integrating the current strategy execution reward value with the advantage function estimation result to construct a weighted advantage function, thereby improving the response sensitivity and adaptability of the strategy to the allocation result feedback during the optimization process;

[0045] S45. Use the weighted advantage function as the core input variable of the policy optimization objective function to guide the parameter update of the policy network during the policy optimization process, and limit the policy update range through the credible interval adjustment mechanism;

[0046] S46. Store the updated policy network parameter set, and write the current policy execution reward value, state representation vector, and feedback indicator into the experience replay pool for sampling during the training iteration process.

[0047] The mobile Internet access IP address intelligent allocation system based on dynamic policy according to an embodiment of the present invention includes the following modules:

[0048] The data collection module is used to collect access behavior data of mobile terminals, current network status information, and IP address pool usage data to build a multi-source feature data set;

[0049] The mask modeling module is used to perform masking on multi-source feature datasets and generate network state representation vectors through masked autoencoders;

[0050] A policy structure building module is used to initialize the policy structure based on the network state representation vector and external condition variables;

[0051] The policy optimization module is used to receive the policy structure and the network state representation vector, and use the proximal policy optimization algorithm to generate and optimize the IP address allocation policy;

[0052] Allocation execution module, used to perform IP allocation, record allocation execution results and network status feedback information;

[0053] The reward update module is used to build a reward function based on the allocation execution results, calculate the policy execution reward value, and update the policy network and value network parameters;

[0054] The model adaptation module is used to receive network status feedback information and update the masked autoencoder model parameters.

[0055] The beneficial effects of the present invention are:

[0056] This invention significantly improves the adaptability and robustness of IP address allocation strategies in mobile internet environments by constructing an intelligent allocation mechanism that integrates a masked autoencoder model with a proximal policy optimization algorithm. Compared to existing traditional methods that rely on static rules or a single policy model, the multi-scale masking processing mechanism introduced in this invention effectively alleviates modeling biases caused by missing, incomplete, and noisy feature data from multiple sources, ensuring the accuracy and generalization of network state representation. At the same time, the improved model based on conditional decoders and graph structure embedding makes the encoding and decoding process more context-dependent and structurally cognitive, providing a more refined and dynamic state input foundation for policy generation.

[0057] In terms of policy optimization, this invention combines a guidance mechanism based on external conditional variables with a dynamic adjustment method for policy credibility intervals to enhance the allocation strategy's responsiveness to environmental changes and enhance its execution stability. The introduction of allocation constraint regularization further ensures the legitimacy of the policy output and the controllability of resource scheduling, effectively reducing the probability of conflicting allocations and abnormal decisions. Furthermore, after executing the policy, the system simultaneously collects allocation results and network feedback information, which are used to update the parameters of the masking model and policy network, establishing a complete perception-decision-feedback-update closed loop.

[0058] Through the above-mentioned collaborative optimization mechanism, the present invention realizes dynamic control of the entire process of intelligent IP address allocation, which not only significantly improves the accuracy and real-time performance of policy decision-making, but also enhances the system's adaptability to complex network conditions such as high concurrent access and frequent state changes. It has significant engineering application value and promotion prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0060] Figure 1 This is a flow chart of the method for intelligently allocating mobile Internet access IP addresses based on dynamic strategies proposed by the present invention;

[0061] Figure 2 This is a structural diagram of the dynamic strategy-based intelligent allocation system for mobile Internet access IP addresses proposed by the present invention. DETAILED DESCRIPTION

[0062] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0063] refer to Figure 1 , a method for intelligently allocating mobile Internet access IP addresses based on dynamic strategies, comprising the following steps:

[0064] S1. Collect access behavior data, current network status information, and IP address pool usage data of mobile terminals, construct a multi-source feature data set, and perform masking processing on the multi-source feature data set to form masked input data;

[0065] S2. Build a masked autoencoder model, input the masked input data into the masked autoencoder model, perform encoding and decoding operations, and generate a reconstructed network state representation vector;

[0066] S3. Using the network state representation vector as input, applying the proximal policy optimization algorithm to generate and optimize the IP address allocation policy, controlling the address scheduler to allocate IP addresses from the target address pool to the corresponding mobile terminal according to the IP address allocation policy, and recording the allocation execution results and network state feedback information;

[0067] S4. Calculate a policy execution reward value based on the allocation execution result, and use the policy execution reward value to update policy parameters in the proximal policy optimization algorithm;

[0068] S5. Input the network status feedback information into the masked autoencoder model, adjust the encoder and decoder parameters, and cyclically execute steps S1 to S4 within a preset allocation period to achieve continuous optimization and allocation control of the dynamic IP address strategy.

[0069] This paper, by constructing a closed-loop optimization process that incorporates a masked autoencoder model and a proximal policy optimization algorithm, achieves intelligent control of mobile terminal IP address allocation in dynamic network environments, demonstrating significant practical application. First, it collects and integrates three key data types: access behavior, network status, and address pool usage, to form a high-dimensional multi-source feature set. A masking mechanism is introduced to address data loss and noise, effectively enhancing the model's robustness and generalization capabilities. Second, by constructing an improved masked autoencoder model, it accurately generates a network state representation vector, improving the state perception accuracy of subsequent policy optimization. During the policy generation phase, a proximal policy optimization algorithm is introduced to generate high-quality allocation policies driven by access conditions and network status. Through trusted policy regulation, external condition guidance, and allocation constraint mechanisms, the stability and legitimacy of the policy output are ensured. Furthermore, by recording allocation execution results and feedback information and using this information to calculate a reward signal, policy parameters are dynamically updated, achieving iterative evolution of policy optimization. During system operation, the masked autoencoder model is inversely regulated using feedback information, establishing a collaborative optimization closed loop for policy learning and state modeling. The overall process has the advantages of data-driven, accurate perception, efficient strategy and strong adaptability, which significantly improves the utilization rate of IP address resources and the success rate of terminal access.

[0070] In this embodiment, the masking processing of the multi-source feature data set refers to masking some feature dimensions in the multi-source feature data set according to a masking ratio of 30% in the constructed multi-source feature data set; the masking operation includes setting the position of the selected feature to zero, marking it as an invalid value and replacing it with a mean value; the masking strategy adopted is a hybrid strategy combining random masking and feature importance-guided masking, wherein 70% of the masking positions are determined by randomly selecting feature dimensions, and the remaining 30% of the masking positions are determined according to the influence weight ranking of the features in the prior evaluation, and high-weight key features are masked first.

[0071] In this embodiment, S2 specifically includes:

[0072] S21. Constructing a masked input dataset of multi-source feature data ,in, is the sample size, is the feature dimension, is a set of real numbers, generating multiple masked versions of the masked input data set , where the masking ratio ;

[0073] S22. Calculate the feature correlation matrix based on the collaborative variability between each feature dimension in the masked input data set ,in, Representation characteristics and features The statistical correlation coefficient of the feature correlation matrix satisfies the correlation threshold Constructing heterogeneous graph structures based on feature pairs , where the node set Represents feature dimension, edge set Express satisfaction The connection relationship of , and the graph attention mechanism is used to calculate the structural importance score of each node;

[0074] S23. For each masked version A hybrid masking operation is performed based on the structure importance score, including the ratio The mask positions are generated by random selection and graph-guided selection to form the mask input matrix ;

[0075] S24, construct a dual encoder structure, including a main encoder and an auxiliary encoder, extract the network status feedback information and strategy execution records in the previous round of allocation cycle, and construct a historical status data set. The main encoder receives the mask input matrix , the auxiliary encoder receives historical status data ;

[0076] S25, respectively input the mask into the matrix and historical status data Input to the main encoder and auxiliary encoder, and use the fusion mechanism to calculate the potential state vector :

[0077] ;

[0078] in, is the fusion weight coefficient, is the main encoder function, is the auxiliary encoder function;

[0079] Latent state vector The practical significance of the formula lies in constructing a latent state representation that fuses current masked input information with historical feedback data to enhance comprehensive modeling capabilities for network states. In actual IP address allocation scenarios, relying solely on current network observation data can easily overlook the accumulated behavioral characteristics and policy adjustments during system evolution. Therefore, the formula introduces a dual encoder mechanism, inputting the current cycle's masked input data and the previous cycle's allocation records, network feedback, and other historical state information into the primary encoder and auxiliary encoder, respectively. The fusion mechanism combines the outputs of the two encoders to generate a latent state representation with greater temporal continuity and semantic integrity, thereby providing more accurate and context-sensitive state input for subsequent policy optimization and address allocation. This mechanism retains the important characteristics of the current network state while incorporating the potential impact of historical behavior on system state evolution. This significantly improves the model's perception and robustness in dynamically changing scenarios, facilitating the construction of more reasonable and stable IP address allocation strategies.

[0080] S26. Constructing a conditional decoder , collect external constraint information related to the current access strategy, including network topology status, service level requirements, address pool tension, terminal device category, and construct external condition variables , the potential state vector With external condition variables Splicing, input to the decoder for conditional reconstruction;

[0081] S27. Construct multiple structurally heterogeneous parallel paths in the decoder, including a fully connected decoding path, a residual skip connection path, and an attention enhancement path, and output preliminary reconstruction results respectively. , the construction of multiple structurally heterogeneous parallel paths refers to inputting the latent state vector into three decoding paths with different structures to improve the model's expressive power and reconstruction accuracy. First, the fully connected decoding path performs layer-by-layer nonlinear transformation on the latent state vector through a multi-layer perceptron to achieve basic reconstruction function; secondly, the residual skip connection path introduces the residual information output by the previous layer at each layer to alleviate the gradient vanishing problem in the deep decoding process and enhance the efficiency of feature transfer; finally, the attention enhancement path uses a multi-head attention mechanism to give different dimensions in the latent state vector differentiated weights, highlighting the feature dimensions that are important to the current task and outputting more focused reconstruction results. The three paths are processed in parallel and finally converge their respective preliminary reconstruction results;

[0082] S28. Constructing a strategy feedback guided fusion controller module ,in Indicates the feedback information of the previous round of strategy execution, including the strategy execution reward value , IP allocation success rate and address pool remaining rate , the fusion controller is based on the latent state vector and feedback information Jointly generate decoding path fusion weights , and introduce task attention compensation output and historical residual compensation vector , calculate the final reconstruction output :

[0083] ;

[0084] in, Decode channel output for task compensation, is the vth conditional decoder, and is the weight coefficient;

[0085] Final reconstruction output The practical significance of this formula lies in dynamically regulating the output performance of the masked autoencoder by fusing policy feedback with multi-path decoding results, thereby improving the accuracy and adaptability of the reconstructed network state representation. In the scenario of intelligent IP address allocation, network states are complex and changeable, and external conditions are dynamically changing. A single decoding path often fails to fully capture the multi-dimensional feature reconstruction requirements. Therefore, this step designs a fusion control mechanism consisting of a task compensation channel, a conditional decoding path, and a historical residual compensation vector. During the decoding phase, based on historical feedback from policy execution, the fusion weights of each path are dynamically generated, guiding the information complementarity between different paths. By introducing task attention compensation and historical residual information, the decoding results are not only more sensitive to policy behavior changes, but also improve the model's ability to recover key features. The final reconstructed vector output more comprehensively reflects the current network state and its evolutionary trends, providing high-confidence input for subsequent policy generation and allocation control. This mechanism enhances the representation power and task alignment performance of the masked autoencoder, serving as an important bridge between modeling and decision-making, with good generalization and practical value.

[0086] S29, based on the final reconstruction output Combined with the previous round of strategy allocation execution data, a feature contribution analysis module is constructed to calculate the contribution score of each feature dimension in the final reconstructed output in strategy generation. , forming a confidence vector , and use the confidence threshold Filter to meet The features of the network state representation vector , where d is the total number of feature dimensions. The feature contribution analysis module is constructed to calculate the contribution score of each feature dimension in the final reconstructed output in the strategy generation. Specifically, the feature contribution analysis module matches the final reconstructed output with the previous round of strategy allocation execution data, adopts a sensitivity analysis method or a gradient-based backpropagation mechanism to evaluate the influence of each feature dimension in generating strategy actions, and quantifies its impact on strategy selection by perturbing each feature dimension one by one and observing the change range of the strategy output, thereby assigning a contribution score to each feature, and finally forming a confidence vector with all the contribution scores.

[0087] This paper constructs a masked autoencoder model that integrates a multi-scale masking mechanism, graph structure embedding, a dual encoder architecture, a conditional decoder, and a policy feedback-guided mechanism. This model achieves deep modeling and enhanced representation of complex multi-source feature data, significantly improving the accuracy and generalization of network state representation. First, it employs multi-version masking and graph structure-guided masking strategies, combining inter-feature correlations to construct a heterogeneous graph structure. This enhances the model's understanding of feature structure dependencies and effectively suppresses redundant feature interference. Second, it introduces a primary-secondary dual encoder architecture to parallel model the masked input of the current cycle and historical feedback information, generating a more time-aware latent state representation through a fusion mechanism. In the decoder design, conditional variables are constructed by collecting external network conditions to drive the parallel reconstruction of multiple heterogeneous paths, comprehensively improving the model's context-sensitive adaptability. Furthermore, a feedback fusion controller incorporates metrics such as policy execution rewards and resource status to dynamically adjust the decoding path fusion weights, ensuring that the output is more consistent with actual network scheduling requirements. Finally, a feature contribution analysis module filters key policy features to ensure that the output network state representation vector is highly interpretable and policy-driven. This improved scheme enables the masked autoencoder model to have the advantages of structural adaptation, feedback perception and state refined expression, providing stable and high-quality input support for the strategy module.

[0088] In this embodiment, S3 specifically includes:

[0089] S31. Network state representation vector based on the output of the masked autoencoder model , and combined with the external condition variables corresponding to the current mobile terminal , construct joint input ;

[0090] S32, the joint input Input to the policy network , output the probability distribution of IP address allocation action in the current state, the action Indicates the candidate IP address allocation scheme;

[0091] S33, the joint input Input to the value network , estimate the expected value under the current state , and uses the advantage function to evaluate the quality of the strategy. The expected value under the current state is obtained by inputting the joint input, including the feature combination formed by splicing the network state representation vector generated by the masked autoencoder and the corresponding external conditional variable, into the value network for forward propagation calculation. The value network predicts the long-term return that the state may obtain under the current strategy based on the input, thereby providing a numerical reference for strategy evaluation and update, and further combining the advantage function to quantitatively evaluate the quality of the strategy and provide optimization guidance;

[0092] S34. Calculate strategy update ratio based on importance sampling :

[0093] ;

[0094] in, is the IP address allocation action output by the policy network at time step t, is the network state representation vector generated by the masked autoencoder model at time step t, is the external condition variable at time step t, For the current policy network, the parameters are When the network state vector and external condition variables Next, output action The probability of is the parameter before updating;

[0095] Policy update ratio The practical significance of this formula lies in the introduction of an importance sampling mechanism that dynamically measures the probability difference between the current policy and the old policy in specific allocation actions, thereby providing a robust estimation basis for the policy update process. In practical IP address intelligent allocation tasks, the policy network undergoes continuous iterative updates, and the action distribution generated by each policy may vary slightly or significantly. If left uncontrolled, this can lead to instability in the policy training process or performance degradation. By constructing the probability ratio of the output of a specific action by the current policy and the old policy under the same state and external conditions, this formula effectively captures the degree of drift during the policy update process and serves as an importance correction term in policy optimization. This correction ratio not only helps adjust the gradient weights of each sample in the objective function but also ensures that the update direction is more consistent with the trajectory of the old policy, improving the stability of policy iteration and sample utilization efficiency. This mechanism is a core policy convergence control method in proximal policy optimization methods and is particularly suitable for distribution-sensitive and state-variable communication resource allocation scenarios. It effectively avoids unreasonable IP allocation behavior caused by rapid policy drift and improves the reliability and adaptability of model decisions.

[0096] S35. Constructing the advantage function using generalized advantage estimation :

[0097] ;

[0098] in, is the discount factor, is the control factor for generalized dominance estimation, is the temporal difference error at time step t+l, is the temporal difference error at time step t, is the expected value under the next state and conditional variables, is the network state representation vector generated by the masked autoencoder model at time step t+1, is the external condition variable at time step t+1;

[0099] Advantage function The practical significance of this formula lies in the introduction of a generalized advantage estimation mechanism, which enables a smoother and more stable assessment of the merits of the intelligent IP address allocation strategy under specific conditions, thereby improving the optimization performance of the policy network. In practical applications, single-step temporal difference errors are often susceptible to short-term feedback fluctuations, leading to unstable policy updates. However, generalized advantage estimation, by integrating weighted accumulation of multi-step temporal difference errors and value predictions of future states, effectively integrates short-term and long-term information, providing a more robust advantage function. This mechanism not only reduces the jitter caused by high-variance estimation during policy training, but also enables the policy model to more accurately determine the value difference between a particular IP allocation action and the current state, thereby more specifically optimizing the probability of selecting high-value actions. Combining network state representation with external conditional variables, generalized advantage estimation enables policy review and evaluation over time, helping to improve policy learning efficiency and allocation effectiveness in dynamic environments. It is a key technical component supporting the stable convergence of policy reinforcement learning modules.

[0100] S36. Constructing a strategy objective function including dynamic clipping coefficients :

[0101] ;

[0102] in, represents the expected operation on all samples at time step t, To find the minimum value, is the cropping operation, where The trust boundary coefficient of the dynamically adjusted strategy:

[0103] ;

[0104] in, and is the coefficient hyperparameter, is the network state representation vector The element variance of is the indicator function related to service quality in the external condition variable C;

[0105] Strategy objective function The practical significance of this formula lies in constructing an objective function that integrates the policy update objective with a dynamic trust interval adjustment mechanism to guide the stable optimization of IP address allocation policies. In policy reinforcement learning, policy network updates are prone to large policy drift due to gradient fluctuations or drastic state changes, impacting model performance and allocation stability. This formula sets a dynamic clipping coefficient to limit the policy probability ratio, ensuring that the policy update amplitude at each step remains within a controllable range and avoiding sudden policy changes. Furthermore, this clipping threshold is adaptively adjusted based on the variance of the network state representation and the quality of service requirements of the external conditional variables, enabling the trust interval to perceive state complexity and changes in external constraints. While optimizing policy returns, the overall objective function also incorporates a trust adjustment term, enabling the update process to balance performance improvement with policy security and robustness. This design makes the policy optimization process more flexible, precise, and adaptable, making it particularly suitable for mobile internet access scenarios with frequent dynamic changes and complex constraints. It is a key mechanism for improving the intelligence and reliability of the system's IP address allocation.

[0106] S37, introduce the strategy trust interval adjustment module based on the masked autoencoder perception mechanism, for each assigned action Introduce upper and lower bound confidence constraints, which are expressed according to the network state vector The characteristic masking entropy value and external condition variables The key constraint indicators in the are jointly determined to construct a set of legal actions , and filter the effectiveness of the strategy output actions, and only keep those that meet:

[0107] ;

[0108] in, is a dynamic threshold, which is adjusted jointly according to the mask reconstruction error and the strategy stability. is the optimal allocation action finally selected, is the set of legal actions, Represented in the network state representation vector The probability of selecting the assigned action a under the given conditions of the external condition variable C;

[0109] S38. Use the back propagation method to update the policy network parameters respectively and value network parameters , minimize the policy objective function And maintain the stability of the strategy’s credible interval;

[0110] S39. Use the updated conditional policy network Generate the optimal action, which is the optimal allocation plan in the target IP address pool;

[0111] S310, control the address scheduler to allocate the IP address in the target address pool to the current external condition variable according to the generated optimal action The associated mobile terminal, the IP address in the target address pool is assigned to the current external condition variable The associated mobile terminal refers to the address scheduler, after receiving the optimal action output by the policy network, selecting the corresponding available IP address from the target address pool according to the allocation scheme indicated by the action, and completing the precise matching based on the mobile terminal characteristics identified in the external condition variable;

[0112] S311. Record the allocation execution result, including the allocated IP address, terminal identification, allocation delay, and allocation status information;

[0113] S312: Collect network status feedback information during the current allocation process, including connection success rate, address pool occupancy rate, and allocation failure rate, for use in masked autoencoder modeling and strategy update for the next allocation cycle.

[0114] This paper significantly improves the intelligence and resource scheduling efficiency of dynamic IP address allocation by constructing a policy generation and execution process that integrates a masked autoencoder perception mechanism with a proximal policy optimization architecture. First, it achieves joint modeling of network state representation vectors and external conditional variables, enhancing the context-awareness of policy generation. Combined with collaborative evaluation of the policy network and value network, it effectively enhances the ability to discern the quality of actions. Second, it employs importance sampling and generalized advantage estimation strategies to accurately calculate update ratios and advantage functions, enhancing the stability and convergence efficiency of policy optimization. In terms of objective function construction, a dynamic clipping boundary based on state variance and quality of service constraints is introduced to achieve adaptive control of the policy update amplitude and avoid policy oscillation. A credible interval adjustment mechanism is constructed, integrating masked reconstruction error with policy confidence indicators to further constrain the action output space, effectively eliminating illegal or unstable allocation schemes and enhancing system robustness. Finally, a closed-loop allocation-feedback-update system is constructed, enabling efficient response to mobile terminal access requests and dynamic resource balancing management. The overall approach takes into account accuracy, flexibility and scalability, and can significantly improve IP resource utilization and terminal access success rate in complex access scenarios.

[0115] In this embodiment, the S4 specifically includes:

[0116] S41. Collect feedback indicators after the allocation action is executed and construct the allocation execution result vector ,in, Indicates that the IP allocation is successful. represents the normalized delay required to complete the allocation, Indicates the service quality matching degree, Indicates the improvement in the target address pool availability caused by the current allocation behavior.

[0117] S42. Calculate the strategy execution reward value based on the feedback index , defined as follows:

[0118] ;

[0119] in, 、 、 、 is the preset weight coefficient;

[0120] Strategy execution reward value The practical significance of this formula lies in establishing a comprehensive reward evaluation function for immediate feedback during the policy optimization process, quantifying the effectiveness of each IP address allocation action. By incorporating multiple feedback metrics, including allocation success, required latency, quality of service matching, and improved address pool resource utilization efficiency, this formula constructs an immediate reward value that reflects the overall performance of policy execution. These metrics are weighted and integrated using preset weight coefficients, allowing the reward function to flexibly adjust the importance of each metric based on application needs. This mechanism not only considers the direct effectiveness of the allocation action itself, but also balances system resource optimization and service experience, providing clear and targeted feedback for subsequent updates to policy network parameters. In dynamic environments, this multi-metric reward design helps enhance the sensitivity and adaptability of the policy, improving the overall performance of the intelligent allocation system under multi-task constraints. It is a key technical foundation for the practical application of reinforcement learning in complex scenarios.

[0121] S43. Introduce a sliding window mechanism to calculate the difference between the current strategy execution reward value and the average strategy execution reward value in the historical window. , used to judge the stability of strategy behavior. When the absolute value of the difference exceeds the set threshold, the steady-state adjustment mechanism is triggered to suppress drastic parameter changes;

[0122] S44, linearly combining the current strategy execution reward value and the advantage function estimation result according to a preset weighting coefficient to construct a weighted advantage function, thereby improving the response sensitivity and adaptability of the strategy to the allocation result feedback during the optimization process;

[0123] S45. Use the weighted advantage function as the core input variable of the policy optimization objective function to guide the parameter update of the policy network during the policy optimization process, and limit the policy update range through the credible interval adjustment mechanism;

[0124] S46. Store the updated policy network parameter set, and write the current policy execution reward value, state representation vector, and feedback indicator into the experience replay pool for sampling during the training iteration process.

[0125] By introducing a feedback-driven policy reward evaluation and update mechanism, this paper achieves efficient linkage between IP address allocation strategies and actual allocation results, effectively enhancing the dynamic adaptability and stability of policy optimization. First, an allocation execution result vector is constructed to comprehensively quantify key indicators such as allocation success rate, latency, service quality matching, and resource utilization improvement, providing a multi-dimensional basis for reward generation. A real-time policy execution reward function is designed based on a weighted summation approach, integrating multiple feedback dimensions to ensure the accuracy and controllability of the reward signal. A sliding window mechanism is introduced to analyze policy fluctuation trends using the difference between current and historical reward values. Steady-state control is automatically triggered when the policy changes drastically, avoiding overfitting and model oscillation. Furthermore, a weighted advantage signal is constructed by integrating the real-time reward with the advantage function, improving the policy's sensitivity and responsiveness to feedback. The weighted advantage signal is further introduced into the main policy optimization process, combined with a credible interval constraint mechanism to ensure both efficiency and robustness in the update process. Finally, a feedback memory mechanism is constructed through an experience replay pool to continuously accumulate allocation experience samples, supporting long-term policy learning and improving generalization capabilities. The overall mechanism ensures that the strategy can continuously evolve and be robustly optimized, achieving efficient and reliable IP resource scheduling in complex access scenarios.

[0126] In this embodiment, the S5 specifically includes:

[0127] After each round of IP address allocation is completed, real-time network status feedback information, including connection success rate, address pool occupancy rate, and allocation failure rate, is input into the masked autoencoder model as the basis for model updates. By adjusting the encoder and decoder parameters, the model can more accurately reconstruct the representation vector reflecting the current network status. This update process is performed periodically within the set allocation cycle, allowing the model to continuously adapt to dynamic changes in the network environment. At the same time, relying on the updated state representation vector, the generated IP address allocation strategy will be more timely and accurate, thereby achieving dynamic optimization and closed-loop control of the IP allocation strategy, improving the resource allocation efficiency and stability of the system in complex mobile access environments.

[0128] refer to Figure 2 ,The mobile Internet access IP address intelligent allocation system based on dynamic ,strategies includes the following modules:

[0129] The data collection module is used to collect access behavior data of mobile terminals, current network status information, and IP address pool usage data to build a multi-source feature data set;

[0130] The mask modeling module is used to perform masking on multi-source feature datasets and generate network state representation vectors through masked autoencoders;

[0131] A policy structure building module is used to initialize the policy structure based on the network state representation vector and external condition variables;

[0132] The policy optimization module is used to receive the policy structure and the network state representation vector, and use the proximal policy optimization algorithm to generate and optimize the IP address allocation policy;

[0133] Allocation execution module, used to perform IP allocation, record allocation execution results and network status feedback information;

[0134] The reward update module is used to build a reward function based on the allocation execution results, calculate the policy execution reward value, and update the policy network and value network parameters;

[0135] The model adaptation module is used to receive network status feedback information and update the masked autoencoder model parameters.

[0136] Example 1:

[0137] To verify the feasibility of this invention, we applied it to an IPv6 campus wireless network deployed on a university campus. During peak hours (8:00 AM to 10:00 AM and 8:00 PM to 10:00 PM), student devices (such as mobile phones, tablets, and laptops) frequently access the network, leading to congestion in IP address allocation and multiple reconnection failures for some devices, severely impacting the use of the teaching platform and the online exam experience. Traditional static IP address allocation strategies cannot adapt to real-time changes in terminal behavior, resulting in wasted resources and reduced service quality.

[0138] To this end, we implemented the "Dynamic Policy-Based Intelligent IP Address Allocation Method for Mobile Internet Access," proposed in this paper. This method focuses on modeling access status using a masked autoencoder and introduces a proximal policy optimization algorithm to achieve dynamic allocation adjustments. In experimental scenarios, the system collected multi-source data, including access behavior logs, network topology changes, terminal category labels, and address pool usage. A 30% masking ratio was used to process feature inputs. A graph-guided masking mechanism was constructed to explore structural correlations between features, enhancing the accuracy of state modeling.

[0139] During the policy optimization process, the model automatically reconstructs the masked autoencoder state representation every 30 minutes and, based on external variables such as the current terminal's service level and address pool occupancy, outputs the optimal allocation action within a dynamic trust interval. The system records changes in key indicators before and after implementation and selects 10 representative sets of terminal access data for comparative analysis.

[0140] Table 1 Comparison of IP address allocation performance between traditional strategy and dynamic strategy of the present invention

[0141]

[0142] As can be seen from the data in Table 1, the proposed method achieved an average IP address allocation success rate of 98.8% across 10 samples, significantly outperforming the 93.9% average of traditional methods. In particular, in samples 1, 2, and 6, the proposed solution achieved a 100% success rate, demonstrating exceptional stability. This demonstrates that the proposed method, by enhancing state perception through masked autoencoders and combining them with a proximal policy optimization mechanism to effectively regulate IP allocation decisions, can significantly reduce the probability of allocation failure.

[0143] In terms of average allocation latency, the proposed method takes an average of 37.1 milliseconds, nearly 32% less than the 54.6 milliseconds of the traditional method. For example, in Sample 5, the proposed method completes allocation in just 33 milliseconds, while the traditional method takes 59 milliseconds. This demonstrates that the proposed method improves allocation efficiency through optimized policy network structure and dynamic action screening mechanisms, providing strong real-time response capabilities.

[0144] In terms of service quality matching, the method of the present invention achieved an average of 91.5% across all samples, exceeding the 81.3% of the traditional solution. In samples 8 and 10, the matching degrees of the present invention were 94% and 96%, respectively, far exceeding the 75% and 79% of the traditional method. This demonstrates that the present invention can better identify and match the service requirements of terminal devices, demonstrating its adaptability to heterogeneous devices.

[0145] The proposed method improves address pool availability by an average of 7.1%, while the traditional method averages only 2.4%. For example, in Sample 9, the improvement is as high as 9.5%, far exceeding the traditional method's 3.2%. This demonstrates that the proposed method, through its dynamic allocation strategy and state feedback mechanism, can effectively alleviate IP address resource shortages and improve resource utilization efficiency.

[0146] In summary, from the multi-dimensional performance comparison results, it can be seen that the present invention is superior to traditional methods in terms of allocation success rate, allocation delay, service matching and resource utilization efficiency, verifying its significant practical value and promotion prospects in complex mobile network environments.

[0147] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A method for intelligently allocating mobile Internet access IP addresses based on dynamic policies, characterized in that: The steps include: S1. Collect access behavior data, current network status information, and IP address pool usage data of mobile terminals, construct a multi-source feature data set, and perform masking processing on the multi-source feature data set to form masked input data; S2. Build a masked autoencoder model, input the masked input data into the masked autoencoder model, perform encoding and decoding operations, and generate a reconstructed network state representation vector; S3. Using the network state representation vector as input, applying the proximal policy optimization algorithm to generate and optimize the IP address allocation policy, controlling the address scheduler to allocate IP addresses from the target address pool to the corresponding mobile terminal according to the IP address allocation policy, and recording the allocation execution results and network state feedback information; S4. Calculate a policy execution reward value based on the allocation execution result, and use the policy execution reward value to update policy parameters in the proximal policy optimization algorithm; S5. Input the network status feedback information into the masked autoencoder model, adjust the encoder and decoder parameters, and cyclically execute steps S1 to S4 within a preset allocation period to achieve continuous optimization and allocation control of the dynamic IP address strategy.

2. The method for intelligently allocating mobile Internet access IP addresses based on dynamic policies according to claim 1, wherein: The masking processing of the multi-source feature data set refers to masking some feature dimensions in the constructed multi-source feature data set according to a masking ratio of 30%; the masking operation includes setting the position of the selected feature to zero, marking it as an invalid value and replacing it with a mean replacement value; the adopted masking strategy is a hybrid strategy combining random masking and feature importance-guided masking, wherein 70% of the masking positions are determined by randomly selecting feature dimensions, and the remaining 30% of the masking positions are determined according to the influence weight ranking of the features in the prior evaluation, and high-weight key features are masked first.

3. The method for intelligently allocating mobile Internet access IP addresses based on dynamic policies according to claim 1, wherein: The S2 specifically includes: S21. Constructing a masked input dataset of multi-source feature data ,in, is the sample size, is the feature dimension, is a set of real numbers, generating multiple masked versions of the masked input data set , where the masking ratio ; S22. Calculate the feature correlation matrix based on the collaborative variability between each feature dimension in the masked input data set ,in, Representation characteristics and features The statistical correlation coefficient of the feature correlation matrix satisfies the correlation threshold Constructing heterogeneous graph structures based on feature pairs , where the node set Represents feature dimension, edge set Express satisfaction The connection relationship of , and the graph attention mechanism is used to calculate the structural importance score of each node; S23. For each masked version A hybrid masking operation is performed based on the structure importance score, including the ratio The mask positions are generated by random selection and graph-guided selection to form the mask input matrix ; S24, construct a dual encoder structure, including a main encoder and an auxiliary encoder, extract the network status feedback information and strategy execution records in the previous round of allocation cycle, and construct a historical status data set. The main encoder receives the mask input matrix , the auxiliary encoder receives historical status data ; S25, respectively input the mask into the matrix and historical status data Input to the main encoder and auxiliary encoder, and use the fusion mechanism to calculate the potential state vector ; S26. Constructing a conditional decoder , collect external constraint information related to the current access strategy, including network topology status, service level requirements, address pool tension, terminal device category, and construct external condition variables , the potential state vector With external condition variables Splicing, input to the decoder for conditional reconstruction; S27. Construct multiple structurally heterogeneous parallel paths in the decoder, including a fully connected decoding path, a residual skip connection path, and an attention enhancement path, and output preliminary reconstruction results respectively. ; S28. Constructing a strategy feedback guided fusion controller module ,in Indicates the feedback information of the previous round of strategy execution, including the strategy execution reward value , IP allocation success rate and address pool remaining rate , the fusion controller is based on the latent state vector and feedback information Jointly generate decoding path fusion weights , and introduce task attention compensation output and historical residual compensation vector , calculate the final reconstruction output ; S29, based on the final reconstruction output Combined with the previous round of strategy allocation execution data, a feature contribution analysis module is constructed to calculate the contribution score of each feature dimension in the final reconstructed output in strategy generation. , forming a confidence vector , and use the confidence threshold Filter to meet The features of the network state representation vector , where d is the total number of feature dimensions.

4. The method for intelligently allocating mobile Internet access IP addresses based on dynamic policies according to claim 3, characterized in that: The S3 specifically includes: S31. Network state representation vector based on the output of the masked autoencoder model , and combined with the external condition variables corresponding to the current mobile terminal , construct joint input ; S32, the joint input Input to the policy network , output the probability distribution of IP address allocation action in the current state, the action Indicates the candidate IP address allocation scheme; S33, the joint input Input to the value network , estimate the expected value under the current state , and use the advantage function to evaluate the pros and cons of the strategy; S34. Calculate strategy update ratio based on importance sampling ; S35. Constructing the advantage function using generalized advantage estimation ; S36. Constructing a strategy objective function including dynamic clipping coefficients ; S37, introduce the strategy trust interval adjustment module based on the masked autoencoder perception mechanism, for each assigned action Introduce upper and lower bound confidence constraints, which are expressed according to the network state vector The characteristic masking entropy value and external condition variables The key constraint indicators in the are jointly determined to construct a set of legal actions , and filter the effectiveness of the strategy output actions, and only keep those that meet: ； in, is a dynamic threshold, which is adjusted jointly according to the mask reconstruction error and the strategy stability. is the optimal allocation action finally selected, is the set of legal actions, Represented in the network state representation vector The probability of selecting the assigned action a under the given conditions of the external condition variable C; S38. Use the back propagation method to update the policy network parameters respectively and value network parameters , minimize the policy objective function And maintain the stability of the strategy’s credible interval; S39. Use the updated conditional policy network Generate the optimal action, which is the optimal allocation plan in the target IP address pool; S310, control the address scheduler to allocate the IP address in the target address pool to the current external condition variable according to the generated optimal action the associated mobile terminal; S311. Record the allocation execution result, including the allocated IP address, terminal identification, allocation delay, and allocation status information; S312: Collect network status feedback information during the current allocation process, including connection success rate, address pool occupancy rate, and allocation failure rate, for use in masked autoencoder modeling and strategy update for the next allocation cycle.

5. The method for intelligently allocating mobile Internet access IP addresses based on dynamic policies according to claim 1, characterized in that: The S4 specifically includes: S41. Collect feedback indicators after the allocation action is executed and construct the allocation execution result vector ,in, Indicates that the IP allocation is successful. represents the normalized delay required to complete the allocation, Indicates the service quality matching degree, Indicates the improvement in the target address pool availability caused by the current allocation behavior. S42. Calculate the strategy execution reward value based on the feedback index ; S43. Introduce a sliding window mechanism to calculate the difference between the current strategy execution reward value and the average strategy execution reward value in the historical window. , used to judge the stability of strategy behavior. When the absolute value of the difference exceeds the set threshold, the steady-state adjustment mechanism is triggered to suppress drastic parameter changes; S44, integrating the current strategy execution reward value with the advantage function estimation result to construct a weighted advantage function, thereby improving the response sensitivity and adaptability of the strategy to the allocation result feedback during the optimization process; S45. Use the weighted advantage function as the core input variable of the policy optimization objective function to guide the parameter update of the policy network during the policy optimization process, and limit the policy update range through the credible interval adjustment mechanism; S46. Store the updated policy network parameter set, and write the current policy execution reward value, state representation vector, and feedback indicator into the experience replay pool for sampling during the training iteration process.

6. A system for intelligently allocating mobile Internet access IP addresses based on dynamic policies, applied to the method for intelligently allocating mobile Internet access IP addresses based on dynamic policies according to any one of claims 1 to 5, characterized in that: Includes the following modules: The data collection module is used to collect access behavior data of mobile terminals, current network status information, and IP address pool usage data to build a multi-source feature data set; The mask modeling module is used to perform masking on multi-source feature datasets and generate network state representation vectors through masked autoencoders; A policy structure building module is used to initialize the policy structure based on the network state representation vector and external condition variables; The policy optimization module is used to receive the policy structure and the network state representation vector, and use the proximal policy optimization algorithm to generate and optimize the IP address allocation policy; Allocation execution module, used to perform IP allocation, record allocation execution results and network status feedback information; The reward update module is used to build a reward function based on the allocation execution results, calculate the policy execution reward value, and update the policy network and value network parameters; The model adaptation module is used to receive network status feedback information and update the masked autoencoder model parameters.

Images