Real-time network handover decision system based on lightweight neural network

Abstract

The present application relates to the technical field of wireless communication and intelligent network switching, and particularly relates to a real-time network switching decision system based on a lightweight neural network, comprising: a state representation module, which is used to collect original link state data, and perform index conversion and time alignment on the original link state data to obtain a three-network state sequence; a network prediction module, which is used to predict the network quality state of the three networks; a prediction verification module, which is used to update the prediction state of the network quality prediction result according to the consistency between the change direction, change amplitude and network advantage-disadvantage relationship of each network prediction quality in adjacent prediction time windows; and a switching control module, which is used to perform link pre-establishment and service migration according to the relationship between the prediction quality improvement degree, prediction quality advantage persistence state and switching cost of the target network. The present application improves the switching accuracy of a three-network fusion terminal.

Description

Technical Field

[0001] This invention relates to the field of wireless communication and intelligent network handover technology, and in particular to a real-time network handover decision system based on a lightweight neural network. Background Technology

[0002] In existing technologies, because cellular networks, Wi-Fi and private networks use different communication standards, the physical meaning, value range and sampling period of each network link index are different. In addition, traditional network handover is mostly triggered directly based on the current signal status or fixed threshold, which makes it difficult to accurately reflect future network quality changes and handover benefits. This leads to problems such as network status judgment distortion, false handover, frequent round-trip handover and short-term service interruption in dynamic communication environments.

[0003] Chinese Patent Publication No. CN109068350A discloses a terminal autonomous network selection system and method for a wireless heterogeneous network. The method includes: the terminal using ANDSF to detect and obtain service node information to form a service node list; obtaining the reception parameters such as RSRP, RSRQ, RSSI, and SINR of each service node through pilot signal measurement and carrier sensing; inputting the reception parameters of each node into a feature learner to obtain a link quality assessment value; obtaining the load information of each node, and inputting the link quality assessment value and load information into a policy learner to determine the network node to be accessed by the terminal; after the terminal performs network access, updating and recording the network selection benefits based on the throughput feedback from the nodes.

[0004] Therefore, the terminal autonomous network selection method for the aforementioned heterogeneous wireless network mainly selects networks based on the link quality parameters and network load status obtained at the current moment. It lacks judgment on the future quality change trends of different networks and the reliability of continuous prediction results. At the same time, it does not jointly analyze the continuous state of target network quality improvement with the switching costs such as link establishment and service interruption. Therefore, in the case of rapid changes or short-term fluctuations in network status, it is difficult to identify the current network deterioration trend in advance and to judge the actual necessity of network switching. Summary of the Invention

[0005] To address this, the present invention provides a real-time network switching decision system based on a lightweight neural network. This system overcomes the limitations of existing technologies, which primarily rely on link quality parameters and network load status obtained at the current moment for network selection. These technologies lack the ability to assess the future quality trends of different networks and the reliability of continuous prediction results. Furthermore, they fail to jointly analyze the continuous improvement of the target network quality with the switching costs such as link establishment and service interruption. Consequently, in situations of rapid changes or short-term fluctuations in network status, it is difficult to identify the current network deterioration trend in advance and to determine the actual necessity of network switching.

[0006] To achieve the above objectives, the present invention provides a real-time network switching decision system based on a lightweight neural network, comprising: The status characterization module is used to collect raw link status data of cellular network, Wi-Fi and private network, and to perform index transformation and time alignment on the raw link status data to obtain the status sequence of the three networks. The network prediction module is used to input the three network state sequences into a lightweight neural network to predict the network quality state of the three networks within several prediction time windows, so as to obtain the network quality prediction results. The prediction verification module is used to determine whether the reliability of the network quality prediction results meets the requirements based on the direction and magnitude of change in the prediction quality of each network within adjacent prediction time windows and the consistency between the network's superiority and inferiority, and to update the prediction status of the network quality prediction results. The switching control module is used to determine whether the current network is necessary for switching based on the relationship between the predicted quality improvement level of the target network, the predicted quality advantage duration, and the switching cost, and to perform link pre-establishment. After the target link meets the service carrying conditions, the services carried by the current network are migrated to the target network.

[0007] Furthermore, the state characterization module is used to perform index transformation and time alignment on the link state data of the three networks according to the mapping relationship between the original link indicators of the three networks and the unified network state indicators to obtain the state sequence of the three networks.

[0008] Furthermore, the state characterization module is used to convert original link indicators with the same state meaning according to the network state meaning corresponding to the original link indicators of the three networks, so as to obtain a unified network state indicator corresponding to each network. The state characterization module is used to perform time alignment based on the sampling time of each unified network state index and the unified sampling time interval to obtain network state data with a unified time reference.

[0009] Furthermore, the state characterization module is used to determine the missing network state indicators of the three networks at each reference sampling time based on network state data with a unified time reference, and to perform equivalent conversion based on the original link indicators that have a corresponding state correlation with the missing network state indicators, or to supplement the state based on the network state indicators corresponding to adjacent reference sampling times. The state characterization module is used to arrange the unified network state indicators of the three networks after the completion of indicator conversion, time alignment and missing state supplementation according to the order of the reference sampling time, so as to obtain the state sequence of the three networks.

[0010] Furthermore, the network prediction module is used to extract network state data from several consecutive historical moments in the three-network state sequence and input it into a lightweight neural network to extract the features of each network state changing over time. The network prediction module is used to predict the network quality status of the three networks within several prediction time windows based on the time change characteristics, and to generate a prediction quality status sequence for each network in time sequence to obtain the network quality prediction result.

[0011] Furthermore, the prediction verification module is used to determine that the reliability of the network quality prediction result meets the requirements when the direction of change of the prediction quality of each network is consistent within adjacent prediction time windows, the magnitude of change is within the allowable range of change, and the network quality relationship of the three networks remains stable. The prediction verification module is used to determine that the reliability of the network quality prediction result does not meet the requirements when at least one of the following conditions is not met: the direction of the predicted quality change, the magnitude of the change, and the network quality relationship of the three networks.

[0012] Furthermore, the prediction verification module is used to update the three-network state sequence to expand the range of historical observation data when the reliability of the network quality prediction results does not meet the requirements. The prediction verification module is used to re-input the updated three-network state sequence into the lightweight neural network for network quality prediction, and replace the network quality prediction result that does not meet the reliability requirements with the newly obtained network quality prediction result to complete the prediction state update.

[0013] Furthermore, the switching control module is used to determine whether the current network needs to be switched based on the predicted quality improvement of the target network relative to the current network, the duration of the quality advantage, and the switching cost. The switching control module is used to determine whether the current network is necessary for switching when the predicted quality improvement meets the switching improvement requirements, the quality advantage continues to meet the switching duration requirements, and the switching cost can be covered; otherwise, it determines that the current network is not necessary for switching and maintains the current network's service carrying status.

[0014] Furthermore, the switching control module is used to maintain the current network to continue carrying services when it is determined that the current network needs to be switched, and to start the communication interface corresponding to the target network to perform target network discovery, access authentication and communication resource establishment, so as to form a candidate communication link with the target network.

[0015] Furthermore, the switching control module is used to obtain the connection status and service carrying status of the candidate communication link. When the candidate communication link completes access and meets the service carrying conditions, the candidate communication link is determined as the target link, and the service currently carried by the network is migrated to the target link.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: The system of the present invention performs index conversion, time alignment, and equivalent conversion or state supplementation of missing indicators on the original link state data of cellular networks, Wi-Fi, and private networks through the state characterization module, thereby eliminating the differences in the meaning, value range, and sampling period of indicators among different communication standards, forming a three-network state sequence with a unified time base and state meaning, providing a consistent data foundation for cross-network quality analysis; and the network prediction module uses a lightweight neural network deployed on the terminal side to extract the temporal change characteristics of the three-network states, predicts the network quality state within several prediction time windows, and realizes the transformation from current network state response to future network state prediction. The prediction verification module verifies the reliability of the prediction results by combining the direction and magnitude of network quality changes within adjacent prediction time windows and the relationship between network advantages and disadvantages. When the reliability is insufficient, the prediction status is updated to reduce false handovers caused by instantaneous disturbances or abnormal predictions. The handover control module judges the necessity of handover by comprehensively considering the degree of improvement in the predicted quality of the target network, the persistence of the quality advantage, and the handover cost. The target link is pre-established only when there is a handover benefit, and the service migration is completed after the target link meets the service carrying conditions. This reduces invalid handovers and frequent round-trip handovers, shortens service interruption time, improves the network handover stability and service continuity of the triple-play terminal, and improves the handover accuracy of the triple-play terminal.

[0017] Furthermore, the system described in this invention performs unified characterization of heterogeneous network status based on the mapping relationship between the original link indicators of the three networks and the unified network status indicators. Since the link indicators of different communication standards differ in physical meaning, value range and sampling period, direct comparison can easily lead to distortion of network status characterization. By establishing the mapping relationship between the original link indicators of each communication standard and the unified network status indicators, aligning the data of different sampling periods in time, and performing equivalent conversion or status supplementation for missing indicators, a three-network status sequence with a unified time benchmark and unified status meaning can be formed, improving the accuracy of subsequent network quality prediction and further improving the switching accuracy of the three-network converged terminal.

[0018] Furthermore, the system of the present invention determines whether the reliability of the network quality prediction results meets the requirements based on the consistency of the direction, magnitude, and network quality variation of each network prediction quality within adjacent prediction time windows. Since instantaneous occlusion, sudden interference, and network load changes may cause short-term abnormal fluctuations, a single prediction can easily misjudge instantaneous changes as continuous deterioration. By comparing the direction, magnitude, and network quality variation of changes in adjacent prediction windows, it can identify whether the prediction results are stable. For prediction results that do not meet the requirements, data is re-collected, the historical observation window is expanded, and the prediction is updated, thereby reducing the probability of erroneous handover and ping-pong handover, and further improving the handover accuracy of the triple-play terminal.

[0019] Furthermore, the system of the present invention determines whether the current network is necessary for switching based on the relationship between the predicted quality improvement degree of the target network relative to the current network, the predicted quality advantage duration, and the corresponding switching cost of the target network. Since different networks have different link establishment times, service interruption times, and resource consumption, short-term quality advantages may not be able to compensate for the switching cost. By comprehensively predicting the quality improvement degree, the advantage duration, and the switching cost, the target link is established and services are migrated in advance only when the quality improvement can cover the switching cost. This can reduce invalid switching and shorten the communication interruption time during the service migration phase, thereby further improving the switching accuracy of the triple-play terminal. Attached Figure Description

[0020] Figure 1 This is an overall structural block diagram of the real-time network switching decision system based on a lightweight neural network according to an embodiment of the present invention; Figure 2 This is a flowchart illustrating the process by which the real-time network switching decision system based on a lightweight neural network determines whether to perform state supplementation, according to an embodiment of the present invention. Figure 3 This is a flowchart illustrating the process by which the real-time network switching decision system based on a lightweight neural network in this invention determines whether to perform a predicted state update. Figure 4 This is a flowchart illustrating the process by which the real-time network switching decision system based on a lightweight neural network, according to an embodiment of the present invention, determines whether to pre-establish a target link and migrate services. Detailed Implementation

[0021] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0022] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0023] Please see Figure 1 The diagram shown is an overall structural block diagram of a real-time network switching decision system based on a lightweight neural network according to an embodiment of the present invention. The system described in this embodiment includes: The status characterization module is used to collect raw link status data of cellular network, Wi-Fi and private network, and to perform index transformation and time alignment on the raw link status data to obtain the status sequence of the three networks. A network prediction module, which is connected to the state representation module, is used to input the state sequence of the three networks into a lightweight neural network to predict the network quality state of the three networks within several prediction time windows, so as to obtain the network quality prediction result. The prediction verification module, which is connected to the network prediction module, is used to determine whether the reliability of the network quality prediction result meets the requirements based on the direction and magnitude of change of the network prediction quality within adjacent prediction time windows and the consistency between the network quality and the network quality relationship, and to update the prediction status of the network quality prediction result. The switching control module, which is connected to the prediction verification module, is used to determine whether the current network is necessary for switching based on the relationship between the degree of improvement in the prediction quality of the target network, the duration of the prediction quality advantage, and the switching cost. It also performs link pre-establishment and migrates the services carried by the current network to the target network after the target link meets the service carrying conditions.

[0024] Specifically, cellular networks include 4G LTE networks and 5G NR networks.

[0025] Specifically, Wi-Fi includes Wi-Fi 5 networks, Wi-Fi 6 / 6E networks, and Wi-Fi 7 networks; Specifically, private networks include PDT digital trunking private networks and TETRA digital trunking private networks.

[0026] In this embodiment, the system uses a state characterization module to perform index conversion, time alignment, and equivalent conversion or state supplementation of missing indicators on the original link state data of cellular networks, Wi-Fi, and private networks. This eliminates differences in the meaning, value range, and sampling period of indicators among different communication standards, forming a three-network state sequence with a unified time base and state meaning, providing a consistent data foundation for cross-network quality analysis. The network prediction module uses a lightweight neural network deployed on the terminal side to extract the temporal change characteristics of the three-network states, predicting the network quality state within several prediction time windows, thus realizing the transformation from current network state response to future network state prediction. The prediction verification module... By combining the direction and magnitude of network quality changes and the relationship between superior and inferior networks within adjacent prediction time windows, the reliability of the prediction results is verified, and the prediction status is updated when the reliability is insufficient, reducing false handovers caused by instantaneous disturbances or abnormal predictions. The handover control module comprehensively judges the necessity of handover by considering the degree of improvement in the predicted quality of the target network, the persistence of the quality advantage, and the handover cost. The target link is pre-established only when there is a handover benefit, and the service migration is completed after the target link meets the service carrying conditions. This reduces invalid handovers and frequent round-trip handovers, shortens service interruption time, improves the network handover stability and service continuity of the triple-play terminal, and improves the handover accuracy of the triple-play terminal.

[0027] Specifically, the state characterization module is used to perform index transformation and time alignment on the link state data of the three networks according to the mapping relationship between the original link indicators of the three networks and the unified network state indicators to obtain the state sequence of the three networks.

[0028] Specifically, raw link metrics include signal strength, transmission latency, and service transmission status.

[0029] Specifically, unified network status indicators include signal reception status, transmission latency status, and service carrying status.

[0030] Specifically, the state characterization module is used to convert original link indicators with the same state meaning according to the network state meaning corresponding to the original link indicators of the three networks, so as to obtain unified network state indicators corresponding to each network. The state characterization module is used to perform time alignment based on the sampling time of each unified network state index and the unified sampling time interval to obtain network state data with a unified time reference.

[0031] Please see Figure 2 The diagram shown is a logical flowchart illustrating the process of determining whether to perform state supplementation in an embodiment of the present invention. In this embodiment, based on the original link indicators corresponding to the network state meaning of the missing network state indicators, it is determined whether to supplement the missing network state indicators. The state characterization module is used to determine the missing network state indicators of the three networks at each reference sampling time based on network state data with a unified time reference, and to perform equivalent conversion based on the original link indicators that have a corresponding state correlation with the missing network state indicators, or to supplement the state based on the network state indicators corresponding to adjacent reference sampling times. The state characterization module is used to arrange the unified network state indicators of the three networks after the completion of indicator conversion, time alignment and missing state supplementation according to the order of the reference sampling time, so as to obtain the state sequence of the three networks.

[0032] Specifically, the three-network status sequence includes the cellular network status sequence, the Wi-Fi status sequence, and the private network status sequence.

[0033] In this embodiment, an equivalent conversion is performed, including... Determine the meaning of the missing network state indicators at the baseline sampling time, and query whether there are original link indicators that can characterize the same or related network state meanings for the same network at the baseline sampling time. If they exist, the original link indicators are converted to the representation range of the unified network status indicators based on the mapping relationship between the original link indicators and the unified network status indicators, combined with the indicator value range and state change direction of the corresponding communication standard, so as to obtain the equivalent status data corresponding to the missing network status indicators.

[0034] In this embodiment, state supplementation is performed, including... When there are no original link metrics available for equivalent conversion at the reference sampling time, obtain the network status metrics of the same network at adjacent reference sampling times before and after that reference sampling time. The network state at the missing time is determined by the direction and degree of change of the adjacent network state indicators, and the corresponding supplementary network state indicators are determined accordingly. If only the valid network state index exists at the previous reference sampling time, it is supplemented based on the network state at the previous time and its continuous changes.

[0035] In this embodiment, the system performs unified characterization of heterogeneous network status based on the mapping relationship between the original link indicators of the three networks and the unified network status indicators. Since the link indicators of different communication standards differ in physical meaning, value range and sampling period, direct comparison can easily lead to distortion of network status characterization. By establishing the mapping relationship between the original link indicators of each communication standard and the unified network status indicators, aligning the data of different sampling periods in time, and performing equivalent conversion or status supplementation for missing indicators, a three-network status sequence with a unified time base and unified status meaning can be formed, improving the accuracy of subsequent network quality prediction and further improving the switching accuracy of the three-network converged terminal.

[0036] Specifically, the network prediction module is used to extract network state data from several consecutive historical moments in the three-network state sequence and input it into a lightweight neural network to extract the features of each network state changing over time. The network prediction module is used to predict the network quality status of the three networks within several prediction time windows based on the time change characteristics, and to generate a prediction quality status sequence for each network in time sequence to obtain the network quality prediction result.

[0037] In this embodiment, the lightweight neural network can be ShuffleNetV2, MobileNetV3, or TinyML network, with ShuffleNetV2 being the preferred embodiment.

[0038] Specifically, the prediction time window is a series of continuous time intervals formed by dividing the future time range according to a certain time length, starting from the current time.

[0039] Specifically, network quality prediction results include predicted quality status, predicted trends, and predicted quality status sequences.

[0040] Please see Figure 3 The diagram shows the logical flowchart of the process for determining whether to update the prediction state in an embodiment of the present invention. In this embodiment, the reliability of the network quality prediction result is determined based on the direction and magnitude of change in the prediction quality of each network within adjacent prediction time windows, as well as the consistency between the network's superiority and inferiority relationships. The prediction verification module is used to determine that the reliability of the network quality prediction result meets the requirements when the direction of change of the prediction quality of each network is consistent within adjacent prediction time windows, the change range is within the allowable change range, and the network quality relationship of the three networks remains stable. The prediction verification module is used to determine that the reliability of the network quality prediction result does not meet the requirements when at least one of the following conditions is not met: the direction of the predicted quality change, the magnitude of the change, and the network quality relationship of the three networks.

[0041] The directions of change in network prediction quality include improved prediction quality, decreased prediction quality, and prediction quality remaining essentially unchanged.

[0042] The magnitude of change in network prediction quality refers to the degree to which the prediction quality of the same network changes between adjacent prediction time windows, and is used to characterize the magnitude of improvement or decrease in network quality.

[0043] Network quality relationship refers to the relative quality relationship between cellular networks, Wi-Fi, and private networks within the same prediction time window, based on the predicted network quality status. It is used to characterize the relative communication performance between each network.

[0044] It is understandable that network quality typically exhibits a certain degree of evolutionary continuity over a continuous time scale. When a lightweight neural network makes a stable prediction of the actual trend of network changes, the direction of quality change for the same network within adjacent prediction time windows is usually consistent, and the magnitude of change is within a continuously acceptable range. At the same time, the relative superiority or inferiority relationship between different networks does not change frequently. If the direction of change reverses, the magnitude of change changes abruptly, or the relative superiority or inferiority relationship of networks changes frequently, it indicates that the prediction results may be affected by instantaneous occlusion, sudden interference, network load fluctuations, or insufficient historical observation data, making it difficult to reflect the continuous trend of network quality changes. By jointly judging the direction of change, the magnitude of change, and the relative superiority or inferiority relationship of networks, we can verify whether the network quality prediction results have a stable basis for change from two aspects: the temporal continuity of a single network and the stability of the relative relationship of the three networks. This avoids performing network switching based solely on the accidental prediction results of a single prediction time window, thereby improving the accuracy of the reliability judgment of network quality prediction results.

[0045] Specifically, the prediction verification module is used to update the three-network state sequence to expand the range of historical observation data when the reliability of the network quality prediction results does not meet the requirements. The prediction verification module is used to re-input the updated three-network state sequence into the lightweight neural network for network quality prediction, and replace the network quality prediction result that does not meet the reliability requirements with the newly obtained network quality prediction result to complete the prediction state update.

[0046] In this embodiment, the predicted state update includes: When the reliability of the network quality prediction results does not meet the requirements, the target network and the corresponding prediction time window that caused the prediction anomaly are identified, and the three-network status sequence before the prediction time window is extracted. At the same time, the latest link status data of cellular network, Wi-Fi and private network are collected. The newly added data is converted in terms of indicators, time aligned and missing status is supplemented, and then supplemented into the original three-network status sequence according to the baseline sampling time. Based on the duration of the predicted anomaly, the historical observation range is expanded, and network state data of several consecutive historical sampling times are reselected from the updated three-network state sequence. This ensures that the re-input data simultaneously covers the network state change process before, during, and after the predicted anomaly, and forms updated network state time series data. The updated network status time series data is re-input into the lightweight neural network on the terminal side to predict the network quality status of cellular network, Wi-Fi and private network in several subsequent prediction time windows. The direction of change, magnitude of change and network quality relationship of each network prediction quality are re-determined to obtain the updated network quality prediction results. The consistency judgment of adjacent prediction time windows is performed again on the updated network quality prediction results. If the direction of change is consistent, the magnitude of change is within the required range, and the network quality relationship remains stable, the original prediction results are replaced with the updated network quality prediction results and the network switching necessity judgment is entered. If the reliability requirements are still not met, the current network service carrying status is maintained, and the prediction is updated again after waiting for new link status data.

[0047] In this embodiment, the system determines whether the reliability of the network quality prediction results meets the requirements based on the consistency of the direction, magnitude, and network quality relationship of the changes in the prediction quality of each network within adjacent prediction time windows. Since instantaneous occlusion, sudden interference, and changes in network load may cause short-term abnormal fluctuations, a single prediction may easily misjudge instantaneous changes as continuous deterioration. By comparing the direction, magnitude, and network quality relationship of changes in adjacent prediction windows, the system can identify whether the prediction results are stable. For prediction results that do not meet the requirements, data is re-collected, the historical observation window is expanded, and the prediction is updated, thereby reducing the probability of erroneous handover and ping-pong handover, and further improving the handover accuracy of the triple-play terminal.

[0048] Please see Figure 4 The diagram shown is a logical flowchart illustrating the process of determining whether to perform target link pre-establishment and service migration according to an embodiment of the present invention. In this embodiment, the necessity of switching is determined based on the relationship between the predicted quality improvement level of the target network, the duration of the predicted quality advantage, and the switching cost. The switching control module is used to determine whether the current network needs to be switched based on the predicted quality improvement of the target network relative to the current network, the duration of the quality advantage, and the switching cost. The switching control module is used to determine whether the current network is necessary for switching when the predicted quality improvement meets the switching improvement requirements, the quality advantage continues to meet the switching duration requirements, and the switching cost can be covered; otherwise, it determines that the current network is not necessary for switching and maintains the current network's service carrying status.

[0049] Specifically, the predicted degree of quality improvement includes the degree of improvement in signal reception, transmission delay, and service carrying capacity.

[0050] Specifically, predicting the sustainability of a quality advantage includes the number of consecutive windows of quality advantage, the expected duration of the quality advantage, and the trend of change in the quality advantage.

[0051] Specifically, the cost of switching to the target network includes link establishment time, expected service interruption time, and pre-establishment additional resource consumption.

[0052] Understandably, the necessity of network handover depends not only on whether the target network is superior to the current network at a given predicted time, but also on whether the quality advantage is significant enough, how long it lasts, and the costs incurred in establishing links, interrupting services, and consuming resources to complete the handover. If the predicted quality improvement of the target network is small or the advantage is short-lived, even if the target network is temporarily superior to the current network, the actual communication improvement may not be enough to compensate for the costs incurred during the handover process, easily leading to ineffective handover or frequent round-trip handovers. However, when the target network has a continuous and significant quality improvement, and this improvement can cover the corresponding handover costs within the predicted duration, it indicates that a stable and meaningful improvement in communication performance can be obtained after the handover. By jointly considering the predicted quality improvement, the duration of the quality advantage, and the handover costs, handover can be avoided by triggering it solely based on the instantaneous network superiority / inferiority relationship, thereby improving the rationality of network handover decisions and the continuity of services.

[0053] Specifically, the switching control module is used to maintain the current network to continue carrying services when it is determined that the current network needs to be switched, and to start the communication interface corresponding to the target network to perform target network discovery, access authentication and communication resource establishment, so as to form a candidate communication link with the target network.

[0054] Specifically, the switching control module is used to obtain the connection status and service carrying status of the candidate communication link. When the candidate communication link completes access and meets the service carrying conditions, the candidate communication link is determined as the target link, and the services currently carried by the network are migrated to the target link.

[0055] Specifically, this involves pre-establishing the target link and migrating services, including: After determining that the current network needs to be switched, the current network continues to carry existing services, and the corresponding communication interface is started according to the communication standard of the target network. The target network discovery, access request, identity authentication and communication resource application are performed to establish a candidate communication link between the terminal and the target network. At the same time, the data transmission status of the current network is maintained to avoid premature interruption of services before the target link is ready. The connection status, transmission latency, and service carrying capacity of candidate communication links are obtained to determine whether the candidate communication links meet the carrying conditions corresponding to the current service. When a candidate communication link completes access authentication and can meet the transmission requirements of the current service, the candidate communication link is identified as the target link, and a forwarding relationship between the current service and the target link is established. In accordance with service continuity requirements, the data streams, voice streams, or video streams currently carried by the network are gradually migrated to the target link. During the migration process, the current network and the target link are kept in parallel for a short period of time, and the data reception status and service transmission status of the target link are monitored. After the target link can stably carry the service, the corresponding service transmission of the current network is terminated and its communication resources are released, thus completing the network switchover. If the target link experiences access failure, abnormal transmission status, or fails to meet service carrying conditions during pre-establishment or service migration, the service migration will be stopped and the current network will continue to carry the service. The link pre-establishment will be re-executed once the target network status meets the requirements.

[0056] Specifically, the services carried by the network include voice services, data services, and video services.

[0057] In this embodiment, the system determines whether the current network is necessary for switching based on the relationship between the predicted quality improvement of the target network relative to the current network, the duration of the predicted quality advantage, and the corresponding switching cost of the target network. Since different networks have different link establishment times, service interruption times, and resource consumption, short-term quality advantages may not be able to compensate for the switching cost. By comprehensively predicting the quality improvement, the duration of the advantage, and the switching cost, the system can establish the target link and migrate services in advance only when the quality improvement can cover the switching cost. This can reduce invalid switching and shorten the communication interruption time during the service migration phase, thereby further improving the switching accuracy of the triple-play terminal.

[0058] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A real-time network switching decision system based on a lightweight neural network, characterized in that, include: The status characterization module is used to collect raw link status data of cellular network, Wi-Fi and private network, and to perform index transformation and time alignment on the raw link status data to obtain the status sequence of the three networks. The network prediction module is used to input the three network state sequences into a lightweight neural network to predict the network quality state of the three networks within several prediction time windows, so as to obtain the network quality prediction results. The prediction verification module is used to determine whether the reliability of the network quality prediction results meets the requirements based on the direction and magnitude of change in the prediction quality of each network within adjacent prediction time windows and the consistency between the network's superiority and inferiority, and to update the prediction status of the network quality prediction results. The switching control module is used to determine whether the current network is necessary for switching based on the relationship between the predicted quality improvement level of the target network, the predicted quality advantage duration, and the switching cost, and to perform link pre-establishment. After the target link meets the service carrying conditions, the services carried by the current network are migrated to the target network.

2. The real-time network switching decision system based on lightweight neural networks according to claim 1, characterized in that, The state characterization module is used to perform index transformation and time alignment on the link state data of the three networks according to the mapping relationship between the original link indicators of the three networks and the unified network state indicators to obtain the state sequence of the three networks.

3. The real-time network switching decision system based on a lightweight neural network according to claim 2, characterized in that, The state characterization module is used to convert original link indicators with the same state meaning according to the network state meaning corresponding to the original link indicators of the three networks, so as to obtain the unified network state indicators corresponding to each network. The state characterization module is used to perform time alignment based on the sampling time of each unified network state index and the unified sampling time interval to obtain network state data with a unified time reference.

4. The real-time network switching decision system based on a lightweight neural network according to claim 3, characterized in that, The state characterization module is used to determine the missing network state indicators of the three networks at each reference sampling time based on network state data with a unified time reference, and to perform equivalent conversion based on the original link indicators that have a corresponding state correlation with the missing network state indicators, or to supplement the state based on the network state indicators corresponding to adjacent reference sampling times. The state characterization module is used to arrange the unified network state indicators of the three networks after the completion of indicator conversion, time alignment and missing state supplementation according to the order of the reference sampling time, so as to obtain the state sequence of the three networks.

5. The real-time network switching decision system based on a lightweight neural network according to claim 4, characterized in that, The network prediction module is used to extract network state data from several consecutive historical moments in the three-network state sequence and input it into a lightweight neural network to extract the features of each network state changing over time. The network prediction module is used to predict the network quality status of the three networks within several prediction time windows based on the time change characteristics, and to generate a prediction quality status sequence for each network in time sequence to obtain the network quality prediction result.

6. The real-time network switching decision system based on a lightweight neural network according to claim 5, characterized in that, The prediction verification module is used to determine that the reliability of the network quality prediction result meets the requirements when the direction of change of the prediction quality of each network is consistent within adjacent prediction time windows, the change range is within the allowable change range, and the network quality relationship of the three networks remains stable. The prediction verification module is used to determine that the reliability of the network quality prediction result does not meet the requirements when at least one of the following conditions is not met: the direction of the predicted quality change, the magnitude of the change, and the network quality relationship of the three networks.

7. The real-time network switching decision system based on a lightweight neural network according to claim 6, characterized in that, The prediction verification module is used to update the three-network status sequence to expand the range of historical observation data when the reliability of the network quality prediction results does not meet the requirements. The prediction verification module is used to re-input the updated three-network state sequence into the lightweight neural network for network quality prediction, and replace the network quality prediction result that does not meet the reliability requirements with the newly obtained network quality prediction result to complete the prediction state update.

8. The real-time network switching decision system based on a lightweight neural network according to claim 7, characterized in that, The switching control module is used to determine whether the current network needs to be switched based on the predicted quality improvement of the target network relative to the current network, the duration of the quality advantage, and the switching cost. The switching control module is used to determine whether the current network is necessary for switching when the predicted quality improvement meets the switching improvement requirements, the quality advantage continues to meet the switching duration requirements, and the switching cost can be covered; otherwise, it determines that the current network is not necessary for switching and maintains the current network's service carrying status.

9. The real-time network switching decision system based on a lightweight neural network according to claim 8, characterized in that, The switching control module is used to maintain the current network to continue carrying services when it is determined that the current network needs to be switched, and to start the communication interface corresponding to the target network to perform target network discovery, access authentication and communication resource establishment, so as to form a candidate communication link with the target network.

10. The real-time network switching decision system based on a lightweight neural network according to claim 9, characterized in that, The switching control module is used to obtain the connection status and service carrying status of the candidate communication link. When the candidate communication link completes access and meets the service carrying conditions, the candidate communication link is determined as the target link, and the service currently carried by the network is migrated to the target link.

Images