Network switching method, system and equipment based on multi-round simulation optimization and medium

Abstract

The invention relates to the technical field of wireless communication, and discloses a network switching method, system and device based on multi-round simulation optimization and a medium, and the method comprises the steps: obtaining real-time network state information for traffic prediction, and obtaining an overall traffic prediction result; obtaining a load value from the overall traffic prediction result, and if the load value is higher than a load threshold value, determining that the access point is a high-load access point and redirecting the traffic to an adjacent access point to obtain initial network configuration; performing multi-round switching simulation to generate a candidate sequence, if a preset load balancing condition is met, obtaining a preliminary switching scheme and calculating a reward value, and selecting the switching scheme with the highest reward value as an optimized switching scheme; executing the optimized switching scheme, and recording the state of the access point; and performing anomaly detection, if an abnormal access point exists, adjusting the prediction method and predicting the overall flow, and optimizing the network configuration according to the prediction result to obtain the final network switching configuration. According to the method, the wireless network can be efficiently and stably switched in a complex network environment.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of wireless communication, and in particular to a network switching method, system, device and medium based on multi-round simulation optimization. BACKGROUND

[0002] At present, network communication is widely used in cloud computing, Internet of Things and edge computing, intelligent transportation and other scenarios. In order to guarantee business continuity and service quality, terminals or business flows need to be frequently switched between multiple accesses, multiple links and multiple operation domains. With the expansion of network scale and the intensification of heterogeneity, the network state presents strong dynamic, strong uncertainty and strong coupling characteristics. How to realize global level collaborative optimization switching under distributed conditions has become a key challenge.

[0003] In a wireless Internet (WiFi) scenario, network switching usually relies on centralized control and static threshold judgment. Each access point collects link quality and traffic load at a fixed period and reports it to the central controller. The controller sorts the candidate links by fixed weight, selects the target access point with higher priority, and the central controller uniformly issues the switching instruction after consistency check.

[0004] However, fixed period sampling is difficult to capture instantaneous fluctuations, static threshold lacks self-adaptive ability to business types and environmental changes, fixed weight cannot reflect real-time business differences, and global state synchronization has delay. In summary, the prior art has the problem of inefficient and stable switching of wireless networks in complex network environments. SUMMARY

[0005] The present application provides a network switching method, system, device and medium based on multi-round simulation optimization to efficiently and stably switch wireless networks in complex network environments.

[0006] In a first aspect, to solve the above technical problems, the present application provides a network switching method based on multi-round simulation optimization, comprising: Obtaining real-time network state information and extracting features to obtain network state features; Each distributed access point performs partial traffic prediction according to the network state features to obtain partial traffic prediction results, and aggregates the partial traffic prediction results to obtain overall traffic prediction results; Obtaining the load value of each access point from the overall traffic prediction results, if the load value is higher than the preset load threshold, it is determined as a high load access point, and the traffic of the high load access point is redirected to a nearby access point to obtain an initial network configuration; According to the initial network configuration, a multi-round handover simulation is performed to generate a candidate sequence containing handover actions, if the candidate sequence containing handover actions meets a preset load balancing condition, a preliminary handover scheme is obtained, a reward value of the preliminary handover scheme is calculated according to a preset reward function, a handover scheme with the highest reward value is selected, and an optimized handover scheme is obtained; The optimized handover scheme is sent to each access point to obtain an optimized network configuration, and an adjusted access point state is recorded; According to the adjusted access point state, an anomaly detection is performed, if there is an abnormal access point, a prediction method is adjusted and an overall traffic is predicted to obtain an updated overall traffic prediction result; According to the updated overall traffic prediction result, the optimized network configuration is optimized to obtain a final network handover configuration.

[0007] In an optional implementation, the real-time network state information is obtained and the features are extracted to obtain network state features, including: Real-time network state information is obtained; The real-time network state information is segmented by a time window to obtain segmented traffic data; The segmented traffic data is filtered to calculate a feature vector to obtain traffic pattern features; The traffic pattern features are identified using a preset normal mode feature to obtain network state features.

[0008] In an optional implementation, each distributed access point performs partial traffic prediction according to the network state features to obtain a partial traffic prediction result, and aggregates the partial traffic prediction results to obtain an overall traffic prediction result, including: The network state features are encrypted to obtain encrypted network state features; Each distributed access point performs prediction according to the encrypted network state features to obtain a partial traffic prediction result; The partial traffic prediction results are weighted and averaged to obtain an overall traffic prediction result.

[0009] In an optional implementation, the load value of each access point is obtained from the overall traffic prediction result, if the load value is higher than a preset load threshold, the high-load access point is determined, and the traffic of the high-load access point is redirected to a nearby access point to obtain an initial network configuration, including: The load value of each access point is obtained from the overall traffic prediction result to obtain a local load value; If the local load value is higher than a preset load value threshold, it is determined that the access point is a high-load access point, and a nearby access point is determined according to a preset switching rule, and the traffic of the high-load access point is redirected to the nearby access point to obtain an initial network configuration.

[0010] In an optional embodiment, the multi-round switching simulation is performed according to the initial network configuration to generate a candidate sequence containing switching actions, if the candidate sequence containing switching actions meets a preset load balancing condition, a preliminary switching scheme is obtained, a reward value of the preliminary switching scheme is calculated according to a preset reward function, and the switching scheme with the highest reward value is selected as an optimized switching scheme, including: A multi-round traffic distribution scenario is simulated according to the initial network configuration to generate a candidate sequence containing switching actions. The load distribution of the candidate sequence is extracted and evaluated, and if the load distribution meets a preset load balancing condition, the candidate sequence is determined as a preliminary switching scheme. The reward value of the preliminary switching scheme is calculated according to a preset reward function, and the scheme with the highest reward value is selected as an optimized switching scheme.

[0011] In an optional embodiment, the abnormality detection is performed according to the adjusted access point state, and if there is an abnormal access point, the prediction method is adjusted and the overall traffic is predicted to obtain an updated overall traffic prediction result, including: The adjusted access point state is compared with a preset access point state template, and if the access point state is abnormal, an abnormal access point list is obtained. Real-time data of the abnormal access point list is obtained, the prediction method is adjusted to predict the overall traffic, and an updated overall traffic prediction result is obtained.

[0012] In an optional embodiment, the optimized network configuration is optimized according to the updated overall traffic prediction result to obtain a final network switching configuration, including: The contribution value of each access point is calculated according to the updated overall traffic prediction result. The updated optimized network configuration is determined according to the contribution value of each access point and distributed to each access point to obtain a final network switching configuration.

[0013] In a second aspect, the present application provides a network switching system based on multi-round simulation optimization, including: A data acquisition module is configured to acquire real-time network state information and extract features to obtain network state features. a flow prediction module, configured to perform partial flow prediction according to the network state features by each distributed access point, to obtain partial flow prediction results, and to aggregate the partial flow prediction results to obtain an overall flow prediction result; a load configuration module, configured to obtain a load value of each access point from the overall flow prediction result, to determine a high-load access point if the load value is higher than a preset load threshold, and to redirect flow of the high-load access point to a nearby access point to obtain an initial network configuration; a switching strategy generation module, configured to perform multi-round switching simulation according to the initial network configuration, to generate a candidate sequence containing switching actions, to obtain a preliminary switching scheme if the candidate sequence containing switching actions satisfies a preset load balancing condition, to calculate a reward value of the preliminary switching scheme according to a preset reward function, and to select a switching scheme with the highest reward value to obtain an optimized switching scheme; an access point state recording module, configured to issue the optimized switching scheme to each access point to obtain an optimized network configuration, and to record an adjusted access point state; an abnormality detection module, configured to perform abnormality detection according to the adjusted access point state, to adjust a prediction method and to predict an overall flow if there is an abnormal access point, and to obtain an updated overall flow prediction result; a network configuration module, configured to optimize the optimized network configuration according to the updated overall flow prediction result to obtain a final network switching configuration.

[0014] In a third aspect, the present application further provides an electronic device, comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor implements the network switching method based on multi-round simulation optimization according to any one of the above descriptions when executing the computer program.

[0015] In a fourth aspect, the present application further provides a computer readable storage medium, comprising a stored computer program, wherein the computer readable storage medium controls a device where the computer readable storage medium is located to execute the network switching method based on multi-round simulation optimization according to any one of the above descriptions when the computer program runs.

[0016] Compared with the prior art, the present application has the following beneficial effects: (1) The application generates an overall traffic prediction result by predicting part of the traffic based on the network state features of the encryption processing on the distributed access points and weighting and aggregating the prediction results of each access point. While ensuring data privacy and security, the application avoids the defect that single access point prediction is easily affected by local anomalies, making the global prediction result more stable and comprehensive, improving the identification accuracy of high-load access points, providing a solid data foundation for subsequent traffic scheduling and switching optimization, and achieving accurate description and efficient prediction ability of network running state from a global perspective.

[0017] (2) The application obtains refined network state features by obtaining real-time network state information and performing time window segmentation, feature filtering and pattern recognition, aligning, smoothing and noise removal of multi-source heterogeneous raw data, so that the input data has continuity and stability, which can effectively avoid the failure problem of static threshold method in dynamic environment. On this basis, the high-trust feature vector formed provides more accurate foundation support for traffic prediction and load identification, and realizes fine perception of complex network state.

[0018] (3) The application carries out multi-round switching simulation based on the initial network configuration, and quantitatively scores the candidate sequence with a multi-dimensional reward function, comprehensively evaluates the potential differences of different schemes in topology constraints and resource consumption, so that the selected scheme has higher executability and load balancing when it is executed. At the same time, after the strategy is executed, the real-time record and abnormal detection of the access point state can timely feedback the deviation mode to the prediction model and perform adaptive adjustment, so that the scheme can dynamically respond to fluctuations and abnormalities in operation. Through this continuous derivation process from simulation evaluation to execution feedback to prediction correction, the switching strategy naturally maintains balance, stability and efficiency in actual operation, and realizes the optimal utilization of network resources and the continuous robustness of the system. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 is a network switching method flowchart provided by the first embodiment of the application based on multi-round simulation optimization; Figure 2 is a network switching system structure diagram provided by the second embodiment of the application based on multi-round simulation optimization. DETAILED DESCRIPTION

[0020] The technical solutions in the embodiments of the application will be described clearly and completely below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, not all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the application.

[0021] Referring toFigure 1 The first embodiment of the present application provides a network switching method based on multi-round simulation optimization, comprising the following steps: S11, obtaining real-time network state information and extracting features to obtain network state features; S12, each distributed access point performs partial flow prediction according to the network state features to obtain partial flow prediction results, and aggregates the partial flow prediction results to obtain overall flow prediction results; S13, obtaining the load value of each access point from the overall flow prediction results, if the load value is higher than the preset load threshold, determining it as a high-load access point, and redirecting the flow of the high-load access point to a nearby access point to obtain an initial network configuration; S14, performing multi-round switching simulation according to the initial network configuration to generate a candidate sequence containing switching actions, if the candidate sequence containing switching actions meets the preset load balancing condition, obtaining a preliminary switching scheme, calculating the reward value of the preliminary switching scheme according to the preset reward function, selecting the switching scheme with the highest reward value to obtain an optimized switching scheme; S15, issuing the optimized switching scheme to each access point to obtain an optimized network configuration, and recording the adjusted access point state; S16, performing anomaly detection according to the adjusted access point state, if there is an abnormal access point, adjusting the prediction method and predicting the overall flow to obtain updated overall flow prediction results; S17, optimizing the optimized network configuration according to the updated overall flow prediction results to obtain a final network switching configuration.

[0022] In step S11, the real-time network state information is obtained and the features are extracted to obtain network state features, comprising: Obtaining real-time network state information; Segmenting the real-time network state information with a time window to obtain segmented flow data; Filtering the segmented flow data to calculate a feature vector to obtain flow pattern features; Using a preset normal mode feature to identify the mode of the flow pattern features to obtain network state features.

[0023] It should be noted that the network state information is collected in real time by the sniffer deployed in the distributed access points of the router and the switch, including timestamp, source IP, target IP, port number and packet size. The network state information is divided into multiple time slices with 10 seconds as a time window, each slice containing all traffic indicators in the time period, including access point total traffic, message quantity, port receiving and sending traffic, port packet number and packet size, to obtain segmented traffic data. Among them, the 10-second time window can quickly reflect the instantaneous fluctuation of the network state, support traffic prediction and load scheduling in time, avoid excessive data volume and noise interference caused by too short window, and cover the typical short-term traffic fluctuation period, which is conducive to calculating the average load, peak value and variance, etc. Characteristics, forming stable and reliable segmented traffic data, providing high credibility input basis for subsequent traffic feature extraction, pattern recognition and intelligent switching.

[0024] In order to ensure the effectiveness of the segmented traffic data, network packets with data volume less than 50 bytes are filtered, which are usually heartbeat packets, probe packets or fragmented transmission data, which have limited contribution to network traffic analysis and are easy to produce high-frequency noise. The threshold of 50 bytes is set because most real business traffic data packets exceed this byte number, and packets below the threshold are mainly connection maintenance or fragmented packets, which can significantly reduce the impact of abnormal fluctuations on statistical characteristics, improve the stability of traffic pattern extraction and the accuracy of prediction model, and thus provide reliable and high-credibility data basis for subsequent traffic analysis and intelligent switching.

[0025] At the same time, there may still be some small control packets with important business significance between 40 and 50 bytes, such as domain name resolution request and network time synchronization message, so an exception detection mechanism is introduced when performing threshold filtering. Specifically, the header information of the data packet is identified, and when it is detected that the protocol field of the small packet belongs to DNS, NTP, DHCP, ICMP, IKE / ISAKMP or ARP, it is not discarded directly, but marked and retained.

[0026] The principal component analysis algorithm is used to calculate the flow feature vector for the filtered segmented flow information. Specifically, first, the mean centering processing is performed on each flow index of each time slice, that is, the average value of the index in all time slices is calculated, and the value of the index of each time slice is subtracted from the average value, so that the data is centered on zero. Subsequently, the variance normalization processing is performed on the centered data, that is, the value of each index is divided by the standard deviation of the index in all time slices, so that indexes of different dimensions and value ranges have the same weight in subsequent analysis. After the standardization processing is completed, the data of each time slice is combined into a matrix form, and the correlation between the indexes is calculated. Specifically, for each pair of indexes, first, the values in all time slices are taken out, the deviations of the two indexes relative to the respective average values are calculated, and then the product of the two index deviations is averaged in all time slices to obtain the correlation of the index pair. Then, the eigenvalues and corresponding eigenvectors of the correlation matrix are calculated in sequence, and each eigenvalue reflects the variance in the corresponding direction. Among all the eigenvalues, the direction with a single eigenvalue contribution to the total variance exceeding 5% is selected as the principal component, and the contribution rates are sorted from high to low; then the variance contribution of each principal component is accumulated, and the principal components with a cumulative variance contribution of 80% are selected to form the reduced feature vector. The projection value of the data of each time slice in the principal component direction constitutes the final feature vector, and the flow pattern feature is obtained.

[0027] Among all the eigenvalues, the direction with a single eigenvalue contribution to the total variance exceeding 5% is selected as the principal component, and the contribution rates are sorted from high to low; then the variance contribution of each principal component is accumulated, and the principal components with a cumulative variance contribution of 80% are selected to form the reduced feature vector. The projection value of the data of each time slice in the principal component direction constitutes the final feature vector, and the flow pattern feature is obtained.

[0028] The absolute difference value is calculated between the traffic pattern feature and the preset normal mode feature in each dimension, then all absolute difference values are divided by the corresponding value of the normal mode feature, and finally multiplied by 100 to obtain the deviation percentage. The average deviation percentage is calculated based on all obtained deviation percentages, and the deviation value of the traffic pattern feature and the normal mode feature is obtained. The preset normal mode feature is obtained based on historical network operation data and typical business scenario statistics. In order to comprehensively cover the traffic features of the network under different load conditions and avoid deviation caused by short-term observation, the traffic feature vectors within a week are collected when the network is in a stable and abnormal interference-free state, and the mean and variance of each index in the time dimension are calculated to form a feature vector representing the normal operation mode of the network. The network traffic usually has obvious periodicity and daily regularity, for example, there are differences in load modes between weekdays and weekends, and between daytime and nighttime. Therefore, the complete one-week data is selected for calculation, which can include peak period, trough period and business fluctuation, so as to ensure that the normal mode feature calculated by statistics can fully reflect the overall behavior of the network in the typical operation period.

[0029] If the deviation value exceeds 10%, it is judged as an abnormal state, and the current traffic pattern feature is recorded as the network state feature. The deviation value threshold of 10% is set based on the statistical analysis of historical network operation data and the comprehensive consideration of business traffic feature rules. When the network is in a stable and abnormal interference-free state, the deviation percentage is calculated based on the traffic features of different time slices within a week and the normal mode feature. The results show that the deviation values of most time slices are less than 10%, indicating that the network load fluctuation and traffic pattern are within the normal range. When the deviation value exceeds 10%, it means that the current traffic feature deviates significantly from the normal mode in multiple indexes, exceeding the historical normal fluctuation range, and has a high probability of corresponding abnormal traffic or sudden event. Therefore, setting 10% as the threshold can effectively distinguish between normal fluctuation and abnormal state, ensure that the deviation value judgment is sensitive to abnormal changes, and will not misjudge due to slight fluctuation, thereby providing a reliable basis for subsequent network state evaluation, traffic prediction and intelligent switching.

[0030] In step S12, the distributed access points perform partial traffic prediction based on the network state feature to obtain partial traffic prediction results, and aggregate the partial traffic prediction results to obtain an overall traffic prediction result, including: The network state feature is encrypted to obtain an encrypted network state feature; Each distributed access point performs prediction based on the encrypted network state feature to obtain partial traffic prediction results; The partial traffic prediction results are weighted and averaged to obtain an overall traffic prediction result.

[0031] It should be noted that, in order to prevent the network state feature from being intercepted or tampered with during transmission, the TLS protocol is used to encrypt the network state feature to generate an encrypted network state feature. Specifically, a TLS secure connection is established between the distributed access points, and the TLS secure connection uses the RSA algorithm for key exchange and identity authentication. In the implementation process of the RSA algorithm, a pair of RSA public and private keys are first generated, and the public key is sent to the distributed access points through a digital certificate; the access point encrypts the generated symmetric encryption session key using the public key after receiving the public key; finally, the corresponding RSA private key is used to decrypt the ciphertext to recover the session key, thereby completing the secure key exchange. Subsequently, the network state feature is encrypted using the key before being sent, converting the original feature vector into ciphertext, ensuring that even if the data is intercepted during transmission, it cannot be parsed or tampered with. The receiving end uses the corresponding key to decrypt the ciphertext after establishing a TLS connection, restoring the original network state feature for subsequent traffic prediction and intelligent switching calculation, thereby realizing end-to-end secure transmission of network state data.

[0032] The encrypted network state feature is distributed to each distributed access point, and each access point locally uses a federated learning algorithm to perform local training to generate an overall traffic prediction model. Specifically, first, an initial weight is generated using a uniform distribution and the overall traffic prediction model is initialized, and the global model parameters are distributed to each access point; the access point uses its own encrypted network state feature to perform multiple rounds of gradient optimization to update the model weight locally, and only uploads the encrypted weight update to the center without transmitting the original data to ensure privacy security. The weight updates uploaded by each access point are weighted and aggregated to form new global model parameters, which are again distributed to the access points, and the cycle is iterated until the mean absolute error is less than 0.01 or the maximum number of iterations is reached, which is 100 rounds. The threshold of 0.01 is selected for the mean absolute error because when the error is controlled within 0.01, the accuracy of the prediction model is sufficient to support traffic scheduling and intelligent switching, and further reducing the threshold has limited performance improvement but significantly increases computational overhead; the upper limit of 100 iterations is selected to balance the convergence speed and resource consumption, and also refers to mainstream federated learning practices, which can converge to the target accuracy in most cases within 50-80 rounds, and setting 100 rounds can ensure that the model converges sufficiently and avoid unnecessary computational waste.

[0033] The model structure adopted by each distributed access point is a feedforward neural network with two hidden layers, the input is the dimension-reduced traffic feature vector, and the output is the traffic prediction value in the future period of time. Each distributed access point obtains a corresponding weight coefficient according to the local training data amount used divided by the total training data amount, and the partial traffic prediction results of each access point are combined to obtain a global traffic prediction result by weighted averaging of the access point weights. Specifically, based on the prediction results of each access point, a corresponding weight coefficient is obtained according to the training data amount used by each access point divided by the total training data amount, and the missing partial prediction results of the access points are filled by averaging the prediction values before and after. For the local prediction result of each access point, the prediction value is multiplied by the access point weight and stacked one by one to complete the weighted calculation and obtain the overall traffic prediction result.

[0034] In step S13, the load value of each access point is obtained from the overall traffic prediction result, and if the load value is higher than the preset load threshold, the high-load access point is determined, and the traffic of the high-load access point is redirected to the adjacent access point to obtain the initial network configuration, including: The load value of each access point is obtained from the overall traffic prediction result to obtain the local load value. If the local load value is higher than the preset load value threshold, the high-load access point is determined, and the adjacent access point is determined according to the preset switching rule, and the traffic of the high-load access point is redirected to the adjacent access point to obtain the initial network configuration.

[0035] It should be noted that after the overall traffic prediction model is trained and converged, the model is used to predict the overall traffic trend of the distributed network, and the global prediction result is decomposed and mapped on demand in combination with the state feature data of each distributed access point to calculate the predicted load value of each distributed access point. Specifically, first, the historical traffic of each point is summed to obtain the historical total traffic, then the historical traffic of each access point is divided by the historical total traffic to obtain the historical traffic proportion as the allocation ratio, and then the allocation ratio is multiplied by the overall traffic prediction result to obtain the local predicted load value of each access point. If the load value exceeds 20% of the normal load of the access point, the access point is determined to be a high-load access point. The threshold of 20% is determined by a large number of historical running samples and stress test experiments. In multiple experiments, when the load of the access point exceeds 20% of the normal benchmark value, the average processing delay and packet loss rate increase significantly, and the resource utilization efficiency decreases significantly, so 20% is selected as a reasonable threshold for high-load determination to balance the sensitivity and tolerance of abnormal detection.

[0036] The non-high-load access points directly connected to the high-load access point are traversed to determine the neighboring access points. The neighboring access points are determined according to the preset switching rule, which refers to the access points that are directly connected to the high-load access point in the network or are logically adjacent to the high-load access point in terms of network topology and transmission quality, and are dynamically selected by obtaining real-time state characteristics (including bandwidth, delay, and load) through the monitoring function of the distributed access points.

[0037] In an implementation, in a 5G base station network, the neighboring access points are determined by calculating the current bandwidth utilization and end-to-end delay, and the preset switching rule is that the bandwidth utilization is less than 80% and the end-to-end delay is less than 50 ms, and the temporary access points are dynamically identified and determined. The 80% threshold is based on analysis of historical data, which shows that when the link bandwidth utilization continuously exceeds 80%, the risk of congestion and dramatic increase in queuing delay will significantly increase, and therefore 20% of bandwidth is reserved to provide a buffer area for sudden traffic peaks. The 50 ms threshold is because the perception and decision-making requirements of actual scenarios are extremely strict, and the end-to-end delay must be controlled within tens of milliseconds, and therefore the 50 ms threshold is set. The 50 ms threshold is also a commonly used and strict QoS target.

[0038] The excess load is divided by the number of neighboring access points, and is sequentially redirected and allocated to each neighboring access point to obtain an initial network configuration. The normal load value of an access point is obtained based on historical monitoring data of the access point in a long-term stable running state. Specifically, under the condition that no abnormal fluctuations occur in the network, the traffic and processing data of the access point in a complete one-week period are collected, the mean and standard deviation of the load of the access point are calculated, and the mean is taken as the normal load reference value of the access point. If the real-time load value of the access point exceeds 20% of the reference value, the access point is determined to be a high-load access point. The selection of complete one-week data for calculation can simultaneously include peak periods, trough periods, and business fluctuation conditions, so as to ensure that the normal mode characteristics obtained by statistical calculation can fully reflect the overall behavior of the network in a typical running period.

[0039] In step S14, a candidate sequence containing switching actions is generated according to the initial network configuration, and if the candidate sequence containing switching actions satisfies a preset load balancing condition, a preliminary switching scheme is obtained. The reward value of the preliminary switching scheme is calculated according to a preset reward function, the switching scheme with the highest reward value is selected, and an optimized switching scheme is obtained, including: A plurality of rounds of traffic distribution scenarios are simulated according to the initial network configuration to generate a candidate sequence containing switching actions. The load distribution of the candidate sequence is extracted and evaluated, and if the load distribution satisfies a preset load balancing condition, the candidate sequence is determined to be a preliminary switching scheme. According to a preset reward function, a reward value of the preliminary switching scheme is calculated, and a scheme with the highest reward value is selected as an optimized switching scheme.

[0040] It should be noted that the initial network configuration is used as the initial load distribution, and the traffic allocation is simulated for multiple rounds. Specifically, a genetic algorithm is used for multiple rounds of iterative optimization to search for the optimal traffic switching scheme. The population size is set to 50 individuals to balance the computational efficiency and search space diversity; the maximum number of iterations is set to 100 generations, and if the optimal solution does not improve significantly for 20 consecutive generations (with an improvement threshold of 1%, i.e., when the improvement is less than or equal to 1%, it is not significantly improved), the optimization is terminated in advance, taking into account the optimization effect and computational overhead; the tournament selection method is used to retain high fitness individuals and maintain population diversity, and the tournament size is set to 3; binary crossover is simulated for crossover operation, with a crossover probability of 0.8 and a distribution index of 5 to effectively explore the solution space; polynomial mutation is used for mutation operation, with a mutation probability of 0.1 and a distribution index of 10 to introduce random perturbations and avoid premature convergence; the top 2 individuals with the highest fitness are retained in each generation to ensure that the optimal solution is not lost. During the optimization process, the historical traffic average value is used for distribution simulation. The historical traffic average value is calculated based on traffic data within a week, which can include peak periods, low periods, and business fluctuations, thereby ensuring that the statistical calculation can fully reflect the overall behavior of the network in a typical operating period.

[0041] On the basis of the initial load distribution, the traffic value allocated to each access point is added to obtain the allocated load distribution. It is determined whether the load value of each access point exceeds or is less than 20% of the average value of the load in the past week. The reason for selecting 20% as the threshold is that when the actual load of an access point exceeds 20% of its normal load level, it is easy to cause overload risks such as processing delay and queue congestion; and when the actual load of an access point is less than 20% of the normal load, it means that the resource utilization is insufficient, which may cause idle computing power and overall efficiency decline. Therefore, the 20% threshold can effectively avoid performance degradation caused by access point overload, and prevent waste caused by uneven resource allocation, ensuring a balance between stability and utilization. If it exceeds or is less than 20% of the average value of the load in the past week, the current candidate sequence is discarded; if it does not exceed or is not less than 20% of the average value of the load in the past week, it meets the requirements, and the current candidate sequence is recorded to obtain a preliminary switching scheme.

[0042] The preset reward function calculates a reward value according to the load balancing degree of the switching scheme. Specifically, the average load value of each distributed access point is calculated, and the load value of each access point is subtracted from the average load value to obtain an access point deviation. The access point deviations are accumulated to calculate a global load deviation total. The dimension of the global load deviation total is consistent with the load value of the access point, that is, in units of load traffic Mbps. The global load deviation total is inverted and multiplied by 100 to convert it to a percentage, which is used as the reward value. The switching scheme with the highest reward value is selected as the optimized switching scheme.

[0043] In step S15, the optimized switching scheme is issued to each access point to obtain an optimized network configuration, and the adjusted access point state is recorded.

[0044] It should be noted that after the optimized switching scheme is determined, the scheme is issued to each distributed access point. Each access point dynamically adjusts the subsequent received traffic according to the proportion parameter allocated to it in the scheme, that is, receives, forwards, or processes the traffic according to the allocated proportion, thereby achieving global load balancing. The access point continuously collects and stores access point state information during the execution of the switching strategy, in order to track and evaluate the execution effect of the access point. The access point state information includes processing delay, response time, throughput, and bandwidth utilization.

[0045] In step S16, the adjusted access point state is used for anomaly detection. If there is an abnormal access point, the prediction method is adjusted and the overall traffic is predicted to obtain an updated overall traffic prediction result, including: The adjusted access point state is compared with a preset access point state template. If the access point state is abnormal, an abnormal access point list is obtained; Real-time data of the abnormal access point list is obtained, the prediction method is adjusted to predict the overall traffic, and an updated overall traffic prediction result is obtained.

[0046] It should be noted that when calculating the adjusted access point state, it is compared with the preset access point state template item by item, and a deviation value is calculated. If the deviation value exceeds 10%, it is determined that the access point has a state anomaly, and it is included in the abnormal access point list. Specifically, the preset access point state template is obtained by averaging the multi-dimensional state information collected during the historical normal operation stage, and can objectively reflect the baseline operation level of the access point. The deviation value is calculated as follows: the absolute difference value of the current access point state and the template value is calculated item by item, and the average value of the access point state template corresponding item is taken as the relative deviation rate. If the relative deviation rate is greater than 10%, it indicates that the running state of the access point has deviated from the normal level.

[0047] It is worth mentioning that the threshold of 10% is determined by combining a large amount of running experience and statistical analysis results. Practice shows that when the access point deviation exceeds the threshold, the probability of performance degradation or abnormal failure significantly increases, so the threshold can ensure the detection sensitivity while reducing the misjudgment rate.

[0048] When the access point state is determined to be abnormal, the real-time data of the abnormal access point is collected and incorporated into the historical data set of the corresponding access point, and the partial traffic prediction model of the access point is re-iterated and trained until the average absolute error of the validation set is less than 0.01 or reaches the maximum iteration number of 100 rounds. The threshold of 0.01 is selected as the average absolute error because when the error is controlled within 0.01, the accuracy of the prediction model is sufficient to support traffic scheduling and intelligent switching, and further reducing the threshold has limited performance improvement but significantly increases the computational overhead. The upper limit of 100 iterations is selected by considering the balance between convergence speed and resource consumption, and referring to the mainstream federated learning practice. In most cases, it can converge to the target accuracy within 50-80 rounds, and setting 100 rounds can ensure that the model converges sufficiently and avoid unnecessary waste of calculation. After the partial traffic prediction model converges, the prediction results of each access point are aggregated as weighted factors according to the traffic allocation proportion of each access point in the scheduling scheme. Specifically, the allocation proportion of each access point is taken as the weight, and the weighted sum of the local prediction results is calculated to obtain the updated overall traffic prediction result.

[0049] In step S17, the updated overall traffic prediction result is used to optimize the optimization network configuration to obtain a final network switching configuration, including: According to the updated overall traffic prediction result, the contribution value of each access point is calculated by weighting; According to the contribution value of each access point, an updated optimization network configuration is determined and distributed to each access point to obtain a final network switching configuration.

[0050] It should be noted that after generating the updated overall traffic prediction result, the traffic prediction value of each access point is first extracted from the overall traffic prediction result. The ratio of the access point's traffic prediction value to the total overall traffic prediction value is then calculated to obtain the initial contribution value of each access point, which characterizes the relative role of that access point in the overall service traffic carrying capacity. Based on this, the initial contribution value is multiplied by the preset access point traffic allocation ratio in the optimized switching scheme to obtain the product factor combining the prediction result and the policy allocation. The preset access point traffic allocation ratio is derived from 100 rounds of traffic allocation simulation experiments. By evaluating the results of each round of simulation, the best-performing optimized switching scheme is selected. The access point traffic allocation ratio corresponding to this optimized switching scheme is the preset access point traffic allocation ratio. Using this ratio as the access point traffic allocation ratio achieves stable and relatively optimal results in all subsequent simulations and actual scheduling. Subsequently, the product factor corresponding to each access point is normalized to the sum of the product factors of all access points. That is, the product factor of a single access point is divided by the sum of the product factors of all access points, which is used as the updated access point traffic allocation ratio to obtain the final network handover configuration.

[0051] In summary, this invention conducts multiple rounds of handover simulations based on the initial network configuration, recording the load distribution, handover cost, and latency changes of each candidate scheme. A multi-dimensional reward function is used to quantify and score the candidate sequences, ensuring that the optimal scheme balances load balancing and resource consumption. After policy execution, access point status is continuously collected and compared with a preset template. Deviation patterns are identified through anomaly detection, and the results are fed back to the prediction model for adaptive adjustment, enabling candidate schemes to dynamically correct themselves to cope with network fluctuations and abnormal access points. Multiple rounds of simulation reveal topological constraints and cumulative costs, the reward function unifies multi-objective measurement standards, and execution feedback and anomaly correction form a closed-loop optimization, naturally harmonizing the handover strategy in terms of executability, load balancing, and long-term stability. Finally, the handover scheme, verified through multi-level simulation, quantitative evaluation, and closed-loop correction, is applied, achieving efficient and stable network handover in complex network environments.

[0052] Reference Figure 2 The second embodiment of the present invention provides a network handover system based on multi-round simulation optimization, comprising: The data acquisition module is used to acquire real-time network status information and extract features to obtain network status features; The traffic prediction module is used for each distributed access point to perform partial traffic prediction based on the network status characteristics, obtain partial traffic prediction results, and aggregate the partial traffic prediction results to obtain the overall traffic prediction result. The load configuration module is configured to obtain a load value of each access point from the overall traffic prediction result, determine the high-load access point if the load value is higher than a preset load threshold, and redirect the traffic of the high-load access point to a nearby access point to obtain an initial network configuration. The switching strategy generation module is configured to generate a candidate sequence containing switching actions by performing multi-round switching simulation according to the initial network configuration, obtain a preliminary switching scheme if the candidate sequence containing switching actions meets a preset load balancing condition, calculate a reward value of the preliminary switching scheme according to a preset reward function, and select the switching scheme with the highest reward value to obtain an optimized switching scheme. The access point state recording module is configured to distribute the optimized switching scheme to each access point to obtain an optimized network configuration and record an adjusted access point state. The anomaly detection module is configured to perform anomaly detection according to the adjusted access point state, and adjust the prediction method and predict the overall traffic to obtain an updated overall traffic prediction result if there is an abnormal access point. The network configuration module is configured to optimize the optimized network configuration according to the updated overall traffic prediction result to obtain a final network switching configuration.

[0053] It should be noted that the network switching system based on multi-round simulation optimization provided by the embodiments of the present application is used to perform all process steps of the network switching method based on multi-round simulation optimization of the above embodiments, and the working principles and advantages of the two are one-to-one corresponding, thus not being repeated.

[0054] The embodiments of the present application further provide an electronic device. The electronic device includes a processor, a memory, and a computer program stored in the memory and executable on the processor, for example, a data acquisition program. The processor implements the steps in the above various network switching methods based on multi-round simulation optimization when executing the computer program, for example Figure 1 The processor implements the functions of each module in the above various device embodiments when executing the computer program, for example, a data acquisition module.

[0055] For example, the computer program can be divided into one or more modules, which are stored in the memory and executed by the processor to complete the present application. The one or more modules can be a series of computer program instruction segments capable of completing a specific function, which are used to describe the execution process of the computer program in the electronic device.

[0056] The electronic device can be a desktop computer, a notebook computer, a palm computer, a smart tablet, and the like. The electronic device can include, but is not limited to, a processor, a memory. Those skilled in the art can understand that the above components are only examples of the electronic device, and do not constitute a limitation on the electronic device, and can include more or fewer components than the above, or combine certain components, or different components, for example, the electronic device can also include an input / output device, a network access device, a bus, and the like.

[0057] The processor can be a central processing unit (CPU), and can also be other general-purpose processors, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, and the like. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, and the like. The processor is the control center of the electronic device, and connects various parts of the electronic device through various interfaces and lines.

[0058] The memory can be used to store computer programs and modules, and the processor can realize various functions of the electronic device by running or executing the computer programs and modules stored in the memory, and calling data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store operating systems, at least one application required by a function (such as a sound playing function, an image playing function, and the like), and the like; and the data storage area can store data created according to the use of the electronic device (such as audio data, a phone book, and the like), and the like. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, for example, a hard disk, a memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one disk storage device, a flash memory device, or other volatile solid-state memory device.

[0059] The modules integrated in the electronic device can be stored in a computer readable storage medium if they are implemented in the form of software function units and sold or used as independent products. Based on this understanding, all or part of the processes in the above-mentioned embodiment methods can also be completed by a computer program instructing related hardware. The computer program can be stored in a computer readable storage medium. When the computer program is executed by a processor, the steps of the above-mentioned various method embodiments can be implemented. The computer program includes computer program code, which can be in the form of source code, object code, executable files or some intermediate forms, etc. The computer readable medium can include any entity or device, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal and software distribution medium, etc. that can carry the computer program code. It should be noted that the content included in the computer readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer readable medium does not include electrical carrier signals and telecommunication signals.

[0060] It should be noted that the above-described device embodiments are only schematic, and the units described as separate components can or can not be physically separated, and the components shown as units can or can not be physical units, i.e. they can be located in one place or distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the embodiment. In addition, the connection relationship between the modules in the device embodiment provided by the present application indicates that there is a communication connection between them, which can be implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement it without creative labor.

[0061] The above-described specific embodiments further illustrate the purpose, technical solutions and beneficial effects of the present application. It should be understood that the above-described specific embodiments are only examples of the present application and do not limit the protection scope of the present application. It is particularly pointed out that any modification, equivalent replacement, improvement, etc. made by those skilled in the art within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A network handover method based on multi-round simulation optimization, characterized in that, include: Real-time wireless network status information is acquired and features are extracted to obtain network status features; Each distributed access point performs partial traffic prediction based on the network state characteristics to obtain partial traffic prediction results, and aggregates the partial traffic prediction results to obtain the overall traffic prediction result. The load value of each access point is obtained from the overall traffic prediction result. If the load value is higher than the preset load threshold, it is determined to be a high-load access point, and the traffic of the high-load access point is redirected to a nearby access point to obtain the initial network configuration. Based on the initial network configuration, perform multiple rounds of handover simulation to generate a candidate sequence containing handover actions. If the candidate sequence containing handover actions satisfies the preset load balancing conditions, a preliminary handover scheme is obtained. The reward value of the preliminary handover scheme is calculated according to the preset reward function. The handover scheme with the highest reward value is selected to obtain the optimized handover scheme. The optimized switching scheme is distributed to each access point to obtain the optimized network configuration, and the adjusted access point status is recorded. Anomaly detection is performed based on the adjusted access point status. If an abnormal access point is found, the prediction method is adjusted and the overall traffic is predicted to obtain an updated overall traffic prediction result. Based on the updated overall traffic prediction results, the optimized network configuration is optimized to obtain the final network switching configuration.

2. The network handover method based on multi-round simulation optimization according to claim 1, characterized in that, The process of acquiring real-time network status information and extracting features to obtain network status features includes: Obtain real-time network status information; The real-time network status information is segmented using a time window to obtain segmented traffic data; After filtering the segmented traffic data, feature vectors are calculated to obtain traffic pattern features; The network state features are obtained by performing pattern recognition on the traffic pattern features using preset normal pattern features.

3. The network handover method based on multi-round simulation optimization according to claim 1, characterized in that, Each distributed access point performs partial traffic prediction based on the network state characteristics to obtain partial traffic prediction results, and aggregates the partial traffic prediction results to obtain an overall traffic prediction result, including: The network state features are encrypted to obtain encrypted network state features; Each distributed access point makes predictions based on the encrypted network state characteristics to obtain partial traffic prediction results; The overall traffic prediction result is obtained by weighting and averaging the partial traffic prediction results.

4. The network handover method based on multi-round simulation optimization according to claim 1, characterized in that, The process of obtaining the load value of each access point from the overall traffic prediction result, and determining a high-load access point if the load value is higher than a preset load threshold, and redirecting the traffic of the high-load access point to a nearby access point to obtain the initial network configuration, includes: Based on the overall traffic prediction results, the load values ​​of each access point are obtained to obtain the local load values; If the local load value is higher than the preset load value threshold, it is determined to be a high-load access point, and a nearby access point is determined according to the preset switching rules. The traffic of the high-load access point is redirected to the nearby access point to obtain the initial network configuration.

5. The network handover method based on multi-round simulation optimization according to claim 1, characterized in that, The process involves performing multiple rounds of handover simulation based on the initial network configuration to generate a candidate sequence containing handover actions. If the candidate sequence containing handover actions satisfies a preset load balancing condition, a preliminary handover scheme is obtained. The reward value of the preliminary handover scheme is calculated based on a preset reward function, and the handover scheme with the highest reward value is selected to obtain an optimized handover scheme, including: Simulate multiple rounds of traffic allocation scenarios based on the initial network configuration, and generate a candidate sequence containing switching actions; Extract the load distribution of the candidate sequence and evaluate it. If the load distribution meets the preset load balancing conditions, then determine the candidate sequence as a preliminary switching scheme. The reward value of the initial switching scheme is calculated according to the preset reward function, and the scheme with the highest reward value is selected as the optimized switching scheme.

6. The network handover method based on multi-round simulation optimization according to claim 1, characterized in that, The step involves performing anomaly detection based on the adjusted access point status. If an abnormal access point is found, the prediction method is adjusted and the overall traffic is predicted to obtain an updated overall traffic prediction result, including: The adjusted access point status is compared with the preset access point status template. If the access point status is abnormal, a list of abnormal access points is obtained. Obtain real-time data of the abnormal access point list, adjust the prediction method to predict the overall traffic, and obtain the updated overall traffic prediction result.

7. The network handover method based on multi-round simulation optimization according to claim 1, characterized in that, The step of optimizing the network configuration based on the updated overall traffic prediction results to obtain the final network switching configuration includes: Based on the updated overall traffic prediction results, the contribution value of each access point is calculated using a weighted average. Based on the contribution value of each access point, the updated optimized network configuration is determined and distributed to each access point to obtain the final network switching configuration.

8. A network handover system based on multi-round simulation optimization, characterized in that, include: The data acquisition module is used to acquire real-time network status information and extract features to obtain network status features; The traffic prediction module is used for each distributed access point to perform partial traffic prediction based on the network status characteristics, obtain partial traffic prediction results, and aggregate the partial traffic prediction results to obtain the overall traffic prediction result. The load configuration module is used to obtain the load value of each access point from the overall traffic prediction result. If the load value is higher than the preset load threshold, it is determined to be a high-load access point, and the traffic of the high-load access point is redirected to a nearby access point to obtain the initial network configuration. The handover strategy generation module is used to perform multiple rounds of handover simulation based on the initial network configuration, generate a candidate sequence containing handover actions, and if the candidate sequence containing handover actions meets the preset load balancing conditions, a preliminary handover scheme is obtained. The reward value of the preliminary handover scheme is calculated according to the preset reward function, and the handover scheme with the highest reward value is selected to obtain the optimized handover scheme. The access point status recording module is used to distribute the optimized switching scheme to each access point, obtain the optimized network configuration, and record the adjusted access point status. An anomaly detection module is used to perform anomaly detection based on the adjusted access point status. If an abnormal access point exists, the prediction method is adjusted and the overall traffic is predicted to obtain an updated overall traffic prediction result. The network configuration module is used to optimize the optimized network configuration based on the updated overall traffic prediction results to obtain the final network switching configuration.

9. An electronic device, characterized in that, The method includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor, when executing the computer program, implements the network switching method based on multi-round simulation optimization as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform the network handover method based on multi-round simulation optimization as described in any one of claims 1 to 7.

Images