A multi-link intelligent switching method

Abstract

The application discloses a multi-link intelligent switching method, which comprises the following steps: starting link state monitoring through a server, periodically starting an adaptive learning mechanism after being online on a wireless link; periodically sending a probe message through each wireless link, and a client replying a response message carrying a request serial number, a client unique identifier and corresponding link index information; introducing a multi-dimensional weighted scoring mechanism to calculate the comprehensive scores of each link and sort them based on the information reported by the client and comprehensive scoring weights; selecting the link with the highest score as a primary link; when the comprehensive score of a standby link is higher than that of the current primary link, or when the primary link is not responded within a timeout period, triggering link switching to the standby link with the highest score, sending a switching instruction through a link notification mechanism, and executing switching by the client and feeding back a confirmation; after learning is completed, adjusting the comprehensive scoring weights according to the adaptive learning result, and updating the adjusted comprehensive scoring weights to link scoring calculation.

Description

Technical Field

[0001] This application relates to the field of network communication technology, and in particular to a method for intelligent switching of multiple links. Background Technology

[0002] With the development of wireless communication technology and edge computing, an increasing number of terminal devices, gateway devices, and application systems need to achieve stable data backhaul and service transmission in complex network environments. To improve connectivity continuity and resilience, multiple access methods such as cellular networks, satellite links, and private networks / Wi-Fi are often used in actual deployments. Multiple wireless links are simultaneously established between server devices and client devices to achieve link redundancy, load sharing, and fault backup. Especially in scenarios such as emergency communication, vehicle / airborne communication, industrial IoT, and remote operation and maintenance, link quality is significantly affected by factors such as coverage, interference, mobility, and network congestion. Systems often need to dynamically select and switch between multiple links to ensure service requirements are met.

[0003] In related technologies, multi-link management typically employs a heartbeat / probe mechanism combined with a primary / backup switchover strategy: link online status is determined by periodically sending probe or keep-alive messages; when the primary link times out or its quality degrades, it switches to a backup link. Some solutions further introduce threshold judgments or simple weighting based on a few indicators such as signal strength, latency, or packet loss rate to assist in link selection. However, these solutions still have shortcomings in engineering applications: link quality is often the result of multiple coupled factors; a single indicator or fixed-weight scoring cannot simultaneously consider signal quality, transmission quality, and changes in service requirements. This can easily lead to links that appear usable but are unstable being selected as primary links, or frequent switching due to indicator jitter. Furthermore, during the recovery process after a link failure, the lack of a reasonable cleaning / reconstruction strategy and control loop can result in unreliable delivery of switching commands and difficulty in clearing residual states of failed links, further reducing system stability.

[0004] Therefore, in multi-link communication and intelligent switching between server and client devices, the difficulty in comprehensively and accurately assessing multi-dimensional link quality and the insufficient reliability of link switching and failure recovery have become urgent problems to be solved. Summary of the Invention

[0005] This application provides a multi-link intelligent switching method, which aims to solve the problems in the existing technology of multi-link communication and intelligent switching between server devices and client devices, where it is difficult to comprehensively and accurately evaluate the multi-dimensional link quality and the reliability of link switching and failure recovery is insufficient.

[0006] This solution provides a multi-link intelligent handover method, the method comprising:

[0007] Step 1: The server starts link status monitoring and periodically starts the adaptive learning mechanism after detecting that the wireless link is online;

[0008] Step 2: The server periodically sends probe packets as neighbor management request packets through each wireless link, and the client replies with neighbor management response packets;

[0009] Step 3: The server performs link quality assessment based on the information reported by the client and the comprehensive scoring weight, and introduces a multi-dimensional weighted scoring mechanism to calculate the comprehensive score of each link;

[0010] Step 4: The server sorts the links according to their scores and selects the link with the highest score as the primary link;

[0011] Step 5: When the overall score of the backup link is higher than the overall score of the current primary link, or when the primary link times out and fails to respond, trigger a link switch to the backup link with the highest score.

[0012] Step 6: Send a handover command to the client through the link notification mechanism, and the client performs the link handover and responds with confirmation;

[0013] Step 7: After the adaptive learning cycle ends, adjust the comprehensive scoring weight according to the adaptive learning result, and update the adjusted comprehensive scoring weight in the link scoring calculation.

[0014] Optionally, in step 2, the neighbor management request message carries a request sequence number.

[0015] The neighbor management response message carries the request sequence number, the client's unique identifier, and the corresponding link's indicator information;

[0016] The metrics include reference signal received power, reference signal received quality, signal-to-interference-plus-noise ratio, range, and downlink packet loss rate.

[0017] The server maintains a data structure in memory corresponding to the client based on the client's unique identifier, which is used to record the client's latest comprehensive score and most recent response time on each wireless link.

[0018] Optionally, in step 2, the server sends the neighbor management request message via UDP multicast on each wireless link in the above scheme.

[0019] If the server does not receive the neighbor management response message for the corresponding link within the preset detection window, or detects continuous failures in MAC layer confirmation or a signal quality degradation rate exceeding a preset threshold, it determines that the corresponding wireless link has entered a failure state.

[0020] The server writes the failure status into a memory data structure and updates the link availability flag.

[0021] Optionally, in step 3, the server calculates latency, uplink packet loss rate, bandwidth, distance, and moving speed in the above scheme.

[0022] The delay is the round-trip time (RTT), which is calculated by the server based on the round-trip time between the probe message and the neighbor management response message, and smoothed using an exponentially weighted moving average.

[0023] The uplink packet loss rate is obtained by the server based on the number of probe messages sent and the number of neighbor management response messages received.

[0024] The bandwidth is the remaining bandwidth of the link for a specific client, which is calculated by the server based on the total number of bytes sent and received on the link within the statistical period, the number of bytes occupied by the client, and the predicted service bandwidth of the client.

[0025] The server combines the downlink packet loss rate and the uplink packet loss rate into an effective packet loss rate.

[0026] The server uses the effective packet loss rate as a packet loss indicator in the calculation of the comprehensive score;

[0027] The effective packet loss rate is used to characterize the transmission reliability in both the server-to-client and client-to-server directions.

[0028] The distance is calculated using the Haversine formula based on the coordinate latitude and longitude of the GPS / BeiDou positioning system.

[0029] The movement speed parameter is calculated based on the absolute distance and the time difference between distance acquisition.

[0030] Optionally, in step 3, the server performs a two-way weighted fusion of the metrics that can be obtained bidirectionally by the server and the client.

[0031] The server weights the client measurement value and the server measurement value according to a preset fixed fusion weight to obtain the weighted value of the corresponding indicator.

[0032] The server participates in the calculation of the overall score based on the weighted values ​​from both ends.

[0033] Optionally, in the above scheme, step 3 further includes: the server performing hard index determination on the reference signal received power, signal-to-interference-plus-noise ratio and bandwidth;

[0034] When any one of the reference signal received power, the signal-to-interference-plus-noise ratio, or the bandwidth is lower than the corresponding threshold, the server records the overall score of the link as 0;

[0035] The server terminates the subsequent score calculation for that link after setting the overall score to 0.

[0036] Optionally, in the above scheme, step 3 further includes:

[0037] The server performs normalization processing on the indicator information to obtain the normalized value of each indicator.

[0038] The server uses exponential decay normalization for latency, packet loss rate, distance, and movement speed, making the normalized value more sensitive to the decline when the indicators deteriorate.

[0039] The server calculates the comprehensive score based on the normalized value and weighted according to the comprehensive scoring weight.

[0040] Optionally, in step 5, the server performs anti-jitter determination before triggering link switching.

[0041] When the overall score of the target link is higher than the overall score of the current primary link, and the difference between the overall score of the target link and the overall score of the current primary link exceeds a preset threshold, and this condition is met multiple times consecutively, the server triggers a switchover.

[0042] In the above scheme, optionally, the link notification mechanism in step 6 is that the server sends a link notification message through all the wireless links currently established with the client, the client sends a link response message to confirm after receiving the message, and the server sends an acknowledgment response message to the client through the corresponding link after receiving the link response message.

[0043] When the link notification message indicates that a specified link should be cleaned, the client clears the link state cache and restarts the link establishment process;

[0044] When the server detects that a client's neighbor has timed out, it first sends a link cleaning notification to the client and accumulates the number of timeouts for that link. When the accumulated number of timeouts exceeds a set threshold, it performs state cleaning and reconstruction on the entire link.

[0045] Optionally, in the above scheme, step 7, the adaptive learning includes:

[0046] The server counts the number of messages for each service type and calculates the proportion of each service type during the learning period.

[0047] The comprehensive score weight is calculated by weighting the services according to their respective proportions, based on the preset scoring weights for each type of service. The service types include at least one or more of the following: voice calls, video conferencing, online video, industrial protocol reporting, web browsing, and file downloading.

[0048] Compared with the prior art, this application has at least the following beneficial effects:

[0049] Based on further analysis and research of existing technical problems, this application recognizes that existing technologies suffer from difficulties in comprehensively and accurately assessing multi-dimensional link quality and insufficient reliability in link switching and failure recovery during multi-link communication and intelligent switching between server and client devices. This application addresses these issues by having the server initiate link status monitoring and uniformly trigger adaptive learning after the wireless link goes online. This transforms link management and decision-making from a decentralized terminal-side or rule-triggered mode to centralized server-side control, ensuring a unified and continuous state awareness of multiple links between the server and client devices. Furthermore, this application involves the server periodically sending probe messages on each wireless link, with the client returning response messages carrying the request sequence number, the client's unique identifier, and corresponding link indicator information. The server introduces multi-dimensional weighting based on the information reported by the client and the adaptive learning results. The scoring mechanism calculates a comprehensive score for each link between the server and client devices and sorts them by score. This transforms the judgment of link quality from a single indicator or fixed threshold into a quantifiable and comparable comprehensive evaluation process, enabling more accurate selection of the highest-quality link as the primary link. Simultaneously, when the quality of the backup link is higher than the primary link, this application triggers a link switch and sends a switch command to the client, which then confirms the switch. This creates a perceptible control loop, logically improving the determinism of the switch execution. Finally, by updating the comprehensive scoring weights after adaptive learning and using them for subsequent link scoring calculations, the link evaluation strategy can be continuously adjusted according to changes in the operating environment and business. This provides an overall improvement in areas such as the difficulty in comprehensively and accurately evaluating multi-dimensional link quality and the insufficient stability of link switching, effectively solving the related technical problems existing in the background technology. Attached Figure Description

[0050] Figure 1 A flowchart illustrating a multi-link intelligent handover method provided in one embodiment of this application;

[0051] Figure 2 A schematic diagram illustrating the topology of a server and a client provided in one embodiment of this application;

[0052] Figure 3 This is a schematic diagram of the interaction process between the server and the client provided in one embodiment of this application. Detailed Implementation

[0053] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0054] In one embodiment, such as Figure 1 As shown, a multi-link intelligent handover method is provided, including the following steps:

[0055] Step 1: The server starts link status monitoring and periodically starts the adaptive learning mechanism after detecting that the wireless link is online;

[0056] Step 2: The server periodically sends probe packets as neighbor management request packets through each wireless link, and the client replies with neighbor management response packets;

[0057] Step 3: The server performs link quality assessment based on the information reported by the client and the comprehensive scoring weight, and introduces a multi-dimensional weighted scoring mechanism to calculate the comprehensive score of each link;

[0058] Step 4: The server sorts the links according to their scores and selects the link with the highest score as the primary link;

[0059] Step 5: When the overall score of the backup link is higher than the overall score of the current primary link, or when the primary link times out and fails to respond, trigger a link switch to the backup link with the highest score.

[0060] Step 6: Send a handover command to the client through the link notification mechanism, and the client performs the link handover and responds with confirmation;

[0061] Step 7: After the adaptive learning cycle ends, adjust the comprehensive scoring weight according to the adaptive learning result, and update the adjusted comprehensive scoring weight in the link scoring calculation.

[0062] like Figure 2 As shown, the server can establish wireless link connections with multiple clients respectively.

[0063] like Figure 3 As shown, the server can establish multiple wireless links with the same client simultaneously (e.g., multiple cellular links / multiple carrier links / multiple access links). After link status monitoring is enabled, the server first completes link enumeration and online detection: when a wireless link is detected to be online, an adaptive learning mechanism is started, and a link detection and link quality assessment loop is entered.

[0064] During the link probing phase, the server periodically sends probe packets to each wireless link, serving as neighbor management request packets. Upon receiving the probe packets, the client replies with a neighbor management response packet. The neighbor management response packet carries at least the request sequence number, the client's unique identifier, and the corresponding link's metrics. In implementation, the probe period can be configured to a fixed period, with a default value of once per second. The server can record the most recent probe transmission time and the most recent response arrival time for each link. The comprehensive scoring parameters include: Reference Signal Received Power (RSRP), Reference Signal Received Quality (RSRQ), Signal-to-Interference-plus-Noise Ratio (SINR), Latency (RTT), Packet Loss Rate (PLR), Bandwidth (BW), Distance (Dist), and Speed ​​(Spe). The server needs to obtain the corresponding metrics information from each client. The reference signal received power, reference signal received quality, SNR, distance information, and downlink packet loss rate in the comprehensive scoring parameters are exchanged through the neighbor management response packet. The remaining information is calculated by calculating the packet loss rate of the neighbor management request packet and the packet reception interval. Among them, RSRP, SINR and bandwidth are all subject to hard index judgment. If any one of them is lower than its respective minimum threshold or demand threshold, the link is considered unavailable to prevent low-quality links from being included in the path selection.

[0065] During the link quality assessment phase, the server calculates a comprehensive score for each link based on information reported by the client and a multi-dimensional weighted scoring mechanism, thereby evaluating the links. The multi-dimensional indicators cover aspects such as wireless quality, transmission quality, and resource capabilities. Before the initial adaptive learning phase ends, the server uses preset weight parameters to synthesize the indicators across all dimensions into a comparable score. After adaptive learning concludes, the weight parameters are updated. Subsequently, the weights are updated periodically, and the server uses the updated weight parameters to synthesize the indicators across all dimensions into a comparable score. The server then sorts all link scores and selects the link with the highest score as the primary link for service delivery.

[0066] During the link switching phase, when the overall score of the backup link is higher than that of the current primary link, or when the primary link times out and fails to respond, the server triggers a link switch, selecting the backup link with the highest overall score as the new primary link. The server sends a switching command to the client through a link notification mechanism. The client executes the link switch and returns a response confirmation to the server, enabling the server to confirm the result of the switching action and complete the status update.

[0067] The server maintains an independent link timeout detection mechanism for each client's uniquely identified device. When a link fails to receive a response from a neighbor within a preset detection window (e.g., 2-3 consecutive detection cycles), or exhibits anomalies such as consecutive MAC layer ACK failures or a sharp decline in signal quality, the link is deemed to have entered a failed state. When the failed link becomes the primary link, the server automatically selects the highest-scoring backup link as the new primary link based on the latest link score output from the link scoring module, and immediately triggers the link notification process to ensure that clients complete link switching in the shortest possible time and maintain service continuity.

[0068] Based on real-time link scoring results, the server comprehensively evaluates each wireless link of the client. When it finds that the currently used link is no longer optimal, it actively triggers the link notification mechanism to instruct the client to switch to the link with the highest score, so as to ensure that communication is always kept on the best link.

[0069] During the adaptive learning phase, the server generates a comprehensive scoring weight based on actual operational data within the learning cycle, and updates this weight into the link scoring calculation after the learning process ends. In subsequent loops, the link score will continuously iterate using the updated weight parameters, thereby adjusting the scoring strategy according to changes in business structure and environment.

[0070] Through the above scheme, this embodiment realizes link detection and indicator reporting driven by the server. By introducing multi-dimensional weighted scoring and ranking the link scores, it can perform a quantifiable and comparable comprehensive quality assessment of each link between the server device and the client device, thereby more accurately selecting the optimal primary link and completing the switchover in the event of link degradation / timeout. At the same time, it forms a control closed loop by combining link notification and response confirmation, and updates the comprehensive scoring weight through adaptive learning, so that the link selection strategy can be adaptively adjusted with changes in the operating environment and business, thereby improving the stability and reliability of multi-link communication.

[0071] In this embodiment, in step 2, the neighbor management request message carries a request sequence number;

[0072] The neighbor management response message carries the request sequence number, the client's unique identifier, and the corresponding link's indicator information;

[0073] The metrics include reference signal received power, reference signal received quality, signal-to-interference-plus-noise ratio, distance, and downlink packet loss rate.

[0074] The server maintains a data structure in memory corresponding to the client based on the client's unique identifier, which is used to record the client's latest comprehensive score and most recent response time on each wireless link.

[0075] In one implementation, to ensure a one-to-one correspondence between probes and responses, the server defines probe messages as "neighbor management request messages" and sets a request sequence number in each request message. Neighbor management response messages carry the same request sequence number field in the response, enabling the server to accurately match the response to the corresponding request and avoid mismatches during parallel probing of multiple links.

[0076] In this embodiment, in step 2, the server sends the neighbor management request message via UDP multicast on each wireless link;

[0077] If the server does not receive the neighbor management response message for the corresponding link within the preset detection window, or detects continuous failures in MAC layer confirmation or a signal quality degradation rate exceeding a preset value, it determines that the corresponding wireless link has entered a failure state.

[0078] The server writes the failure status into a memory data structure and updates the link availability flag.

[0079] In one implementation, the server sends a neighbor management request message via UDP multicast on the corresponding wireless link. UDP multicast can reduce the probing overhead in multi-client scenarios and improve the broadcast / distribution efficiency of probing messages on the link side. After receiving the multicast probe, the client still sends back a neighbor management response message, which can be unicast back to the receiving port specified by the server.

[0080] The server sets a preset detection window (e.g., 2-3 consecutive detection cycles). If no neighbor management response message is received for a link within the detection window, or if MAC layer confirmation fails consecutively or the signal quality degradation rate exceeds a preset threshold, the link is determined to be in a failed state. The server writes this failed state into the aforementioned memory data structure and updates the link availability flag (e.g., setting it to unavailable) so that the link can be removed or its priority reduced during subsequent scoring. The preset threshold can be set by the administrator; for example, the preset threshold can be 6 dB / s.

[0081] In this embodiment, the server periodically (e.g., once per second) sends UDP multicast probe messages as neighbor management request messages on each wireless link. After receiving the neighbor management request message, the client immediately reports its request sequence number, client unique identifier, and the link's indicator information (reference signal received power, reference signal received quality, signal-to-interference-plus-noise ratio, distance, and downlink packet loss rate) as a neighbor management response and link keep-alive.

[0082] This embodiment sends probes via UDP multicast and sets a detection window for failure determination. This application can efficiently complete link liveness detection in a multi-link environment and promptly feed back the failure status to the unified state management structure, reducing the situation where a link is unavailable but still participates in the evaluation, thereby improving the real-time performance of link evaluation and the reliability of handover triggering.

[0083] In this embodiment, in step 3, the server calculates latency, uplink packet loss rate, bandwidth, distance, and moving speed;

[0084] The delay is the round-trip time (RTT), which is calculated by the server based on the round-trip time between the probe message and the neighbor management response message, and smoothed using an exponentially weighted moving average.

[0085] The uplink packet loss rate is obtained by the server based on the number of probe messages sent and the number of neighbor management response messages received.

[0086] The bandwidth is the remaining bandwidth of the link for a specific client, which is calculated by the server based on the total number of bytes sent and received on the link within the statistical period, the number of bytes occupied by the client, and the predicted service bandwidth of the client.

[0087] The server combines the downlink packet loss rate and the uplink packet loss rate to form the effective packet loss rate.

[0088] The server uses the effective packet loss rate as a packet loss indicator in the calculation of the comprehensive score;

[0089] The effective packet loss rate is used to characterize the transmission reliability in both the server-to-client and client-to-server directions.

[0090] The distance is calculated using the Haversine formula based on the coordinate latitude and longitude of the GPS / BeiDou positioning system.

[0091] The movement speed parameter is calculated based on the absolute distance and the time difference between distance acquisition.

[0092] Among them, uplink / downlink packet loss are respectively , The combined effective packet loss rate is: .

[0093] Uplink packet loss is the packet loss rate of the message sent by the server to client i, and downlink packet loss is the packet loss rate of the message sent by client i to the server. The downlink packet loss rate is obtained by obtaining data through the neighbor management response message.

[0094] Where t represents the sampled data at time t (which can also be understood as the time when the formula is calculated).

[0095] j represents the j-th link from the server's perspective;

[0096] i represents the i-th client from the server's perspective;

[0097] : The downlink packet loss rate between the server and the i-th client on the j-th link at time t;

[0098] The uplink packet loss rate between the server and the i-th client on the j-th link at time t;

[0099] The combined effective packet loss rate of the server and the i-th client at time t on the j-th link. A smaller packet loss rate indicates a more stable link. The packet loss rate is divided into uplink packet loss rate and downlink packet loss rate. The uplink packet loss rate is the packet loss situation of the message sent by the server to the client i through the j-th link. The downlink packet loss rate is the packet loss situation of the message when the client i replies to the server through the j-th link. The server sends a neighbor management request message to the client i. If the server does not receive a corresponding neighbor management response message from the client i within a specified time, it is considered to have lost packets. The specified time refers to the detection window (such as 2 to 3 consecutive detection cycles) in which no neighbor response is received.

[0100] After receiving a neighbor management request message, the client immediately replies with a neighbor management response message. If the client does not receive a corresponding ACK message from the server within a specified time, it is considered a packet loss. The specified time refers to the period within the detection window (e.g., 2 to 3 consecutive detection cycles) during which no neighbor response is received.

[0101] The packet loss rate is the number of lost packets divided by the total number of sent packets in the previous data collection period.

[0102] The downlink packet loss rate is recorded by the client and fed back to the server through the neighbor management response message.

[0103] In this embodiment, in step 3, the server performs a two-way weighted fusion of the metrics that can be obtained bidirectionally by the server and the client;

[0104] The server weights the client measurement value and the server measurement value according to a preset fixed fusion weight to obtain the weighted value of the corresponding indicator.

[0105] The server participates in the calculation of the overall score based on the weighted values ​​from both ends.

[0106] For metrics that can be obtained bidirectionally, the performance parameters of the i-th neighbor link are calculated by weighting the server-side and client-side measurements to obtain a comprehensive bidirectional weighted value:

[0107]

[0108]

[0109]

[0110] like Any value below the RSRP threshold, or or If any value falls below the SINR threshold, the overall link score for this calculation is recorded as 0, and subsequent calculations are terminated.

[0111] in, Let J be the reference signal received power of the server at time t on the j-th link;

[0112] : The reference signal received power of the i-th client at time t on the j-th link;

[0113] : The weighted reference signal received power of the server and the i-th client at time t on the j-th link;

[0114] The quality of the reference signal received by the server at time t on the j-th link;

[0115] The quality of the reference signal received by the i-th client at time t on the j-th link;

[0116] The weighted reference signal reception quality between the server and the i-th client at time t on the j-th link;

[0117] The signal-to-interference-plus-noise ratio (SIR) of the server at time t on the j-th link;

[0118] The signal-to-interference-plus-noise ratio (SIR) of the i-th client on the j-th link at time t;

[0119] The weighted signal-to-interference-plus-noise ratio (SINNR) between the server and the i-th client at time t on the j-th link.

[0120] Metrics collected by a single device can only reflect the link reception status on its own side, and may be affected by factors such as its own hardware performance and local interference, thus having a limited perspective. Calculating the average value from both ends and fusing the monitoring data from both ends of the link can offset the local interference and hardware deviations on the single device side, and more objectively and comprehensively reflect the overall transmission quality of the link.

[0121] Taking RSRP as an example, RSRPj RSRP represents the received power of the reference signal on the j-th link of the server. j,i RSRP represents the received signal power of the j-th link for the i-th client. j and RSRP j,i Equally important in communication, any value that is too low will prevent business data from being successfully delivered. This indicates "risk appetite," characterizing the level of attention paid to volatility risk during the scoring phase of the process. Let's take an example to illustrate, and The range of values ​​and the mode of operation of are related to same. The default value is 0.5; the value range is between 0 and 1, excluding 0 and 1. The larger the value, the higher the weighted reference signal received power. The more the fusion result leans towards the server-side RSRP; The smaller the value, the more it favors the client-side RSRP.

[0122] RTT takes round-trip measurements and smooths the stationary value using the EWMA exponentially weighted moving average function. You can adjust the smoothing speed (fast / slow): .

[0123] In the formula, The latency sample value between the server and the i-th client on the j-th link at time t;

[0124] : The historical smoothed latency value between the server and the i-th client on the j-th link at time t-1;

[0125] The current smoothed latency value between the server and the i-th client on the j-th link at time t.

[0126] is the time delay smoothing coefficient, used to adjust the smoothing speed, and t is the current timestamp.

[0127] Calculate the latency of the j-th link for the i-th client.

[0128] Latency differs from parameters such as RSRP. RSRP and similar parameters are obtained by the client sending neighbor management response messages to the server, which include metrics such as RSRP and RSRQ. RTT, on the other hand, is calculated by the server based on the time difference between sending a neighbor management request message and a neighbor management response message, rather than from the message content itself.

[0129] Here, is the time delay smoothing coefficient, where, The default value is 0.2; the value range is between 0 and 1, excluding 0 and 1. The larger the value of α, the more sensitive the RTT estimate is to the current sample and the faster the response, but the greater the fluctuation; rtt The smaller the value, the smoother and more stable the estimation, but the slower the response to changes in the link.

[0130] In this embodiment, step 3 further includes: the server performing hard index determination on the reference signal received power, signal-to-interference-plus-noise ratio, and bandwidth;

[0131] When any one of the reference signal received power, the signal-to-interference-plus-noise ratio (SIR), or the bandwidth is lower than the corresponding threshold, the server records the overall score of the link as 0. Specifically, when the reference signal received power is lower than -105dBm, the SIR is lower than 5dB, or the available bandwidth is lower than 1Mbps, the link is determined to have insufficient communication capability, and the overall score of the link is recorded as 0.

[0132] The server terminates the subsequent score calculation for that link after setting the overall score to 0.

[0133] In this embodiment, step 3 further includes:

[0134] The server performs normalization processing on the indicator information to obtain the normalized value of each indicator.

[0135] The server uses exponential decay normalization for latency, packet loss rate, distance, and movement speed, making the normalized value more sensitive to the decline when the indicators deteriorate.

[0136] The server calculates the comprehensive score based on the normalized value and weighted according to the comprehensive scoring weight.

[0137] Distance parameters can be calculated using the Haversine formula based on coordinates (latitude and longitude) from positioning systems such as GPS / BeiDou. The greater the distance, the longer the signal propagation path, increasing the possibility of attenuation and interference. At the same time, the energy required to maintain the link is higher. Distance scores can reflect potential link degradation risks, enhancing the predictability and robustness of the scoring.

[0138] The movement speed parameter can be calculated based on the absolute distance and the time difference between distance acquisition. A higher speed indicates a greater likelihood of link fluctuations, dynamically reflecting link stability and identifying links prone to fluctuations.

[0139]

[0140] In the formula, : The device distance between the server and the i-th client at time t;

[0141] : The inter-device distance between the server and the i-th client at the previous sampling time;

[0142] Let v be the device movement speed of the server and the i-th client at time t; v is the device movement speed. Device movement causes dynamic changes in the distance between devices, which in turn causes fluctuations in link signal strength and stability. The faster the movement speed, the more drastic the changes in link state, and the higher the risk of communication interruption. By dividing the distance difference corresponding to two different timestamps by the time difference, the average movement speed can be obtained, quantifying the movement characteristics of the device and predicting the future stability of the link. The absolute value of the distance difference is taken because speed is a scalar, focusing only on the speed of movement and not the direction of movement, ensuring the rationality of the calculation results. The introduction of this indicator extends link evaluation from "static status quo" to "dynamic prediction," improving the foresight of link selection.

[0143] For the i-th client The coordinates of the two moments can be calculated using technologies such as GPS (data acquisition is not involved here). The client's movement speed can be calculated by dividing the difference in displacement in two-dimensional or three-dimensional space by the time difference.

[0144] The remaining bandwidth parameter in the aforementioned indicator information can be calculated based on the current total bandwidth of the device and the total number of bytes sent and received over a period of time. A larger remaining bandwidth indicates smoother transmission of service packets.

[0145]

[0146] in This represents the time difference between two calculations. (A link score calculation is performed for each neighbor management response message received); BW used,j Indicates t in link j last The average bandwidth occupied within the ~t time window, when the time difference is small (around 1 second), BW used,j It can also be expressed as the bandwidth currently used at any given moment. (BW) used,j,i This represents the average bandwidth used by client i on link j. (BW is used when link j is not the primary link used by client i.) used,j,i (0); BW used,j Subtract BW used,j,i This represents the remaining used bandwidth after removing the bandwidth used by client i. (BW) res,j,i This represents the remaining bandwidth on link j based on client i. Here, α is the bandwidth smoothing coefficient, with a default value of 0.2 and a value range of 0 to 1 (excluding 0 and 1). A larger α results in more sensitive bandwidth prediction and faster response, but also greater fluctuations; a smaller α results in smoother and more stable prediction, but slower response to sudden traffic changes.

[0147] Let J be the total number of bytes of messages sent and received by the j-th link of the server at time t.

[0148] : The total number of bytes of packets corresponding to the j-th link of the server at the previous sampling time;

[0149] The total number of bytes of messages exchanged between the server and the i-th client on the j-th link at time t.

[0150] : The total number of bytes of the message between the server and the i-th client on the j-th link at the previous sampling time;

[0151] The total bandwidth of the j-th link of the server at time t;

[0152] Let be the historical smoothed bandwidth requirement value of the server and the i-th client at time t-1.

[0153] : The smoothed predicted bandwidth requirement of the server and the i-th client at time t.

[0154] This represents the service bandwidth usage of the i-th client device during the current time period. The bandwidth of the i-th client device in the next period is predicted by using an exponentially weighted moving average. This refers to the service bandwidth occupied by client i on the primary link. When the primary link is the j-th link, and same.

[0155] like If the bandwidth is below the threshold, the overall link score is recorded as 0, and subsequent calculations are terminated.

[0156] Normalization function calculation:

[0157]

[0158] in, The reference signal received power score between the server and the i-th client on the j-th link at time t;

[0159] The signal-to-interference-plus-noise ratio (SIR) score between the server and the i-th client on the j-th link at time t;

[0160] The quality score of the reference signal received by the server and the i-th client on the j-th link at time t;

[0161] The latency score between the server and the i-th client on the j-th link at time t;

[0162] For chip measurement capabilities, the chip provides specific values, for example:

[0163] RSRP min -156dBm, RSRP max It is -31dBm.

[0164] RSRQ min -43dB, RSRQ max It is 20dB.

[0165] SINR min -23dB, SINR max It is 40dB.

[0166] Where, α rt It is the RTT attenuation / sensitivity coefficient, used to control the strength of the influence of the smoothing delay value on the normalization result; α rt The default value is 0.01, and the range of values ​​is α. rt >0, and in engineering practice, it is usually taken as 0.001 to 0.05.

[0167] α rt The larger the value, the more sensitive the smoothing delay is to the penalty on the score (it decreases faster); α rt The smaller the value, the more gradual the impact of the smoothing delay value on the score.

[0168] Because when α rt When =0.01,

[0169] When the smoothing delay value is 20ms, the result is: 0.81;

[0170] When the smoothing delay value is 40ms, the result is: 0.67;

[0171] When the smoothing delay value is 70ms, the result is: 0.49;

[0172] When the smoothing delay value is 100ms, the result is: 0.36;

[0173] When α rt When the value is 0.01, a smoothed latency value of less than 20ms corresponds to a normalized score greater than 0.81, indicating excellent network quality. When the smoothed latency value is between 20ms and 70ms, the score is approximately 0.49 to 0.81, and the network quality gradually declines from good to an edge state. When the smoothed latency value is greater than 70ms, the score is less than 0.49, indicating high network latency and relatively poor quality.

[0174]

[0175] Score the packet loss rate between the server and the i-th client on the j-th link at time t;

[0176] Where, β plr The default value is 10, and the value range is between 5 and 20. B plr This is the PLR ​​packet loss rate attenuation / sensitivity coefficient;

[0177] β plr The larger the value of β, the stronger the penalty for packet loss on the score. plr When the score is 10, minor packet loss is not excessively penalized, while high packet loss is quickly reduced in score.

[0178] When β plr When =10,

[0179] When the effective packet loss rate is 1%, the result is 0.90;

[0180] When the effective packet loss rate is 5%, the result is 0.61;

[0181] When the effective packet loss rate is 10%, the result is 0.37;

[0182] When β plr When the effective packet loss rate is 10, the normalized score is approximately 0.90, indicating that the network packet loss has a small impact and the quality is excellent. When the effective packet loss rate is 5%, the score is approximately 0.61, indicating that the network has some packet loss but is still usable and the quality is at a medium level. When the effective packet loss rate is 10%, the score drops to approximately 0.37, indicating that the packet loss is relatively serious, the network quality is poor, and it may affect the service experience.

[0183] in, α is the distance score between the server and the i-th client at time t; d The default value is 0.00005. This value needs to be adjusted according to different application scenarios. In the current environment, α is used for long-distance communication such as intelligent transportation and emergency communication. d The default value is 0.00005. In the current environment, α is used for short-range communication such as industrial IoT. d The default value is 0.01. α d Distance attenuation / sensitivity coefficient

[0184] When α d When = 0.00005,

[0185] When the spacing is 1000m, the result is 0.95;

[0186] When the spacing is 5000m, the result is 0.78;

[0187] When the spacing is 10000m, the result is 0.61;

[0188] When the spacing is 20,000m, the result is 0.37;

[0189] A normalized score of approximately 0.95 corresponds to a spacing of 1km, indicating that the impact of distance factors is relatively small and the risk is low under short-distance conditions; the score is approximately 0.78 when the spacing is 5km, indicating that the risk increases with distance but is still within an acceptable range; the score is approximately 0.61 when the spacing is 10km, indicating that there are certain risks in long-distance communication and the quality begins to decline; the score drops to approximately 0.37 when the spacing is 20km, indicating that the impact of distance factors on communication reliability is significant in ultra-long-distance scenarios, and the network quality is relatively poor.

[0190] Where, α v The default value is 0.05, and the value range is 0.01 to 0.1; α v This refers to the velocity decay / sensitivity coefficient. Rate the movement speed of the server and the i-th client at time t.

[0191] When α v When =0.05,

[0192] When the moving speed is 1 m / s, the result is 0.95;

[0193] When the moving speed is 5 m / s, the result is 0.78;

[0194] When the moving speed is 10m / s, the result is 0.61;

[0195] When the moving speed is 20m / s, the result is 0.37;

[0196] A normalized score of approximately 0.95 corresponds to a movement speed of 1 m / s, indicating that low-speed movement has little impact on the network and carries low risk. A score of approximately 0.78 corresponds to a movement speed of 5 m / s, indicating that the potential risk caused by movement increases but is still within an acceptable range. A score of approximately 0.61 corresponds to a movement speed of 10 m / s, indicating that network stability begins to decline and the risk of movement becomes significant. A score of approximately 0.37 corresponds to a movement speed of 20 m / s, indicating that high-speed movement may lead to link instability and relatively poor network quality.

[0197] This embodiment uses exponential normalization for calculation, and the result ranges from 0 to 1. When the time delay, packet loss rate, distance, and movement speed are smaller, the result is more biased towards 1.

[0198]

[0199] This is the bandwidth score for the server and the i-th client on the j-th link at time t.

[0200] The calculation is performed when the "remaining" bandwidth is greater than or equal to 0 and less than the bandwidth requirement of client i (bandwidth * β). β represents the "safety margin," which prevents the remaining bandwidth from being too high when it is the same as the required bandwidth, thus preventing the link from being unable to cope with traffic bursts. An empirical value for β can be 1.5.

[0201] The score is 1 when the "remaining" bandwidth is greater than or equal to the "required bandwidth", or when the "required bandwidth" is 0.

[0202] in, Using these as empirical boundaries, RTT, PLR, Dist, and Spe use exponential attenuation to make them more sensitive to degradation. In bandwidth calculations, β is greater than 1 to prevent link quality degradation caused by sudden bursts of service bandwidth. and When the scores are the same, the BW score will be 1, which will make the link unable to handle traffic bursts. β can be used for traffic burst control, where β > 1.

[0203] Finally, the score of the i-th neighbor is calculated using a scoring algorithm:

[0204]

[0205] Wherein, the default weight value w RSRP =0.15、w RSRQ =0.10、w SINR =0.15、w RTT =0.15、w PLR =0.15、w Dist =0.10、w V =0.05、w BW =0.15. w RSRP w RSRQ w SINR w RTT w PLR w Dist w V w BW The weighted average of the link quality metrics—reference signal received power, reference signal received quality, signal-to-interference-plus-noise ratio, latency, packet loss rate, distance, moving speed, and bandwidth—quantifies the impact of each metric on the final link quality score. All weights sum to 1. The default weight values ​​are configured based on typical communication link service scenarios and can be dynamically adjusted according to actual application needs.

[0206]

[0207] In addition to traditional link scoring parameters such as reference signal power, reference reception quality, and signal-to-interference-plus-noise ratio (SNR), this embodiment further introduces indicators such as device distance, device movement speed, and remaining link bandwidth to comprehensively characterize the link status. Specifically, device distance can be jointly evaluated based on reference signal power, reference reception quality, and SNR to derive a more stable and interference-resistant link; device movement speed reflects the future stability of the link; significant device movement may lead to rapid link attenuation; and remaining link bandwidth directly reflects the link's ability to meet service transmission demands.

[0208] Based on this, the system supports an adaptive learning mechanism: by statistically analyzing and learning the business characteristics within a certain time range, the weight of each indicator is dynamically calculated, thereby generating a link scoring result that is more in line with the current business environment and needs, and realizing intelligent optimization for different business scenarios.

[0209] In this embodiment, in step 5, the server performs anti-jitter determination before triggering link switching;

[0210] When the overall score of the target link is higher than the overall score of the current primary link, and the difference between the overall score of the target link and the overall score of the current primary link exceeds a preset threshold, and this condition is met multiple times consecutively, the server triggers a switchover.

[0211] In this embodiment, the target link is another link that is compared with the main link in terms of comprehensive score. The default value of the preset threshold is 0.1, and the default value can be satisfied 5 times consecutively.

[0212] In this embodiment, the link notification mechanism in step 6 involves the server sending link notification messages through all currently established wireless links with the client, the client receiving the messages and sending a link response message to confirm, and the server receiving the link response message and sending an acknowledgment response message to the client through the corresponding link.

[0213] When the link notification message indicates that a specified link should be cleaned, the client clears the link state cache and restarts the link establishment process;

[0214] When the server detects that a client's neighbor has timed out, it first sends a link cleaning notification to the client and accumulates the number of timeouts for that link. When the accumulated number of timeouts exceeds a set threshold, it performs state cleaning and reconstruction on the entire link.

[0215] In this embodiment, step 7, the adaptive learning includes:

[0216] The server counts the number of messages for each service type and calculates the proportion of each service type during the learning period.

[0217] The comprehensive score weight is calculated by weighting the services according to their respective proportions, based on the preset scoring weights for each type of service. The service types include at least one or more of the following: voice calls, video conferencing, online video, industrial protocol reporting, web browsing, and file downloading.

[0218] In this embodiment, when the server determines that a client needs to perform link switching or link cleaning operations, it sends link notification messages through all currently established wireless links with the client to ensure reliable delivery of the notification even under link quality fluctuations. The notification content includes information such as the identifier of the target link or the link to be cleaned. After receiving the link response, the server sends an ACK message to the corresponding client through the corresponding link. This message is only useful for the client to collect downlink packet loss rate. The downlink packet loss rate is used by the server to calculate the "effective packet loss rate".

[0219] Upon receiving the link notification, the client will immediately send a link response message for confirmation and perform corresponding operations according to the notification instructions (such as switching to the designated primary link or clearing designated link resources). After successful notification confirmation, the server updates the link status accordingly to ensure that the link management information of both parties remains synchronized.

[0220] Link cleaning module: For the client, when it receives the "clean the specified link" notification message from the server, it will immediately call the link cleaning module to perform state cleaning operations on the link, including clearing the link state cache, resetting the link parameters, and restarting the link establishment process, so that the link is rejoined to the system in a clean state, thereby effectively avoiding continuous communication anomalies caused by link state disorder, parameter drift or cache residue.

[0221] For the server, since the same wireless link may serve multiple clients simultaneously, when a client's neighbor times out, the server does not immediately clean the entire link. Instead, it prioritizes sending a link cleaning notification to that client. Simultaneously, the server records the number of neighbor timeouts for that link. Only when the cumulative number of timeouts exceeds a set threshold (e.g., 10 times) does the server invoke its own link cleaning module to clean and rebuild the link's status, thus preventing the entire link from being mistakenly cleaned due to a single client's anomaly.

[0222] The adaptive learning mechanism in this embodiment is a periodic learning mechanism. Within a specified learning period, the business traffic is parsed, analyzed, and statistically analyzed. After the learning is completed, the indicator weights are calculated based on the learning data and updated in real time to the link scoring module.

[0223] During the learning process, the system filters traffic based on IP address, port, and protocol type within each sampling period, extracting valid business transmission and reception packets for statistical analysis. After the entire self-learning process is complete, this mechanism can dynamically generate adapted scoring weight algorithms based on different business environments, thereby better reflecting the actual link status. Specific business packet scoring standards are shown in Table 1.

[0224] Table 1

[0225]

[0226] After adaptive learning, firstly, based on the number of each type of business message in the statistics, let the number of the m-th type of business message be... The total number of messages during the learning period was Then the proportion of the m-th type of business in all businesses is:

[0227]

[0228] Subsequently, the proportion of business messages will be... Scoring criteria corresponding to this business type

[0229] Combining these factors, we obtain the final comprehensive score weight for the k-th indicator:

[0230]

[0231] in, This represents the weight of the m-th type of service on the k-th indicator. The reason this embodiment aims to predict future weights is that traffic patterns emerge after repeated learning over a long period. For example, when users use home routers, traffic is low from 9 AM to 6 PM on weekdays, high from 6 PM to 11 PM, low from 11 PM to 7 AM the next day, and low from 7 AM to 9 AM. On weekends, traffic is higher during the day and lower at night. Furthermore, it can be predicted that during weekday low traffic periods, industrial protocol (IoT control protocol) traffic is higher, while during evening peak traffic periods, video traffic, voice calls, and gaming traffic account for a higher proportion.

[0232] There is a regular pattern of traffic usage with a 7-day cycle. Therefore, it is believed that after a period of learning, the learning model can predict information such as traffic volume and traffic share in the future, and fine-tune the learning model and the predicted results based on the learning results.

[0233] The above method can calculate the overall scoring weight during the learning period, but this weight is mainly based on historical message statistics and cannot accurately reflect future business trends. To improve the predictive ability of the overall scoring weight for the next hour, the system models and predicts weight changes based on recurrent neural networks or long short-term memory networks.

[0234] During data collection, the system records the proportion of various business messages and their corresponding comprehensive scoring weights within each learning cycle as samples, and organizes them into time-series data in chronological order, such as by hour, day, or week. Based on this, the system uses the historical samples to train a recurrent neural network or long short-term memory network model, enabling it to learn the correlation between the historical message proportions and comprehensive scoring weights.

[0235] When predicting future comprehensive scoring weights, the system first obtains prediction results for multiple future time periods based on the trained model, and then further corrects them by combining the actual calculation results of the current learning period. Specifically, the comprehensive scoring weight of the current learning period is compared with the comprehensive scoring weight of the corresponding time point of the previous week in the prediction period. Based on the difference, the prediction result for the next hour is adjusted. This maintains long-term consistency while enhancing the adaptability to short-term business fluctuations and sudden traffic changes, thereby improving the accuracy of comprehensive scoring weight prediction and the reliability of link scoring.

[0236] Through the aforementioned collaborative learning and prediction mechanism, once adaptive learning is complete, the system can dynamically determine the comprehensive scoring weight that best suits the current business environment by combining historical patterns, current business status, and future trend predictions. Ultimately, the link score is calculated based on the comprehensive scoring weight and the normalized values ​​of each link indicator, enabling the link switching function to make decisions based on more accurate and forward-looking link scoring results, thereby achieving more stable and reasonable path selection and network operation assurance.

[0237] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

Claims

1. A multi-link intelligent handover method, characterized in that, The method includes: Step 1: The server starts link status monitoring and periodically starts the adaptive learning mechanism after detecting that the wireless link is online; Step 2: The server periodically sends probe packets as neighbor management request packets through each wireless link, and the client replies with neighbor management response packets; Step 3: The server performs link quality assessment based on the information reported by the client and the comprehensive scoring weight, and introduces a multi-dimensional weighted scoring mechanism to calculate the comprehensive score of each link; Step 4: The server sorts the links according to their scores and selects the link with the highest score as the primary link; Step 5: When the overall score of the backup link is higher than the overall score of the current primary link, or when the primary link times out and fails to respond, trigger a link switch to the backup link with the highest score. Step 6: Send a handover command to the client through the link notification mechanism, and the client performs the link handover and responds with confirmation; Step 7: After the adaptive learning cycle ends, adjust the comprehensive scoring weight according to the adaptive learning result, and update the adjusted comprehensive scoring weight in the link scoring calculation.

2. The method according to claim 1, characterized in that, In step 2, the neighbor management request message carries a request sequence number; The neighbor management response message carries the request sequence number, the client's unique identifier, and the corresponding link's indicator information; The metrics include reference signal received power, reference signal received quality, signal-to-interference-plus-noise ratio, range, and downlink packet loss rate. The server maintains a data structure in memory corresponding to the client based on the client's unique identifier, which is used to record the client's latest comprehensive score and most recent response time on each wireless link.

3. The method according to claim 1, characterized in that, In step 2, the server sends the neighbor management request message via UDP multicast on each wireless link. If the server does not receive the neighbor management response message for the corresponding link within the preset detection window, or detects continuous failures in MAC layer confirmation or a signal quality degradation rate exceeding a preset threshold, it determines that the corresponding wireless link has entered a failure state. The server writes the failure status into a memory data structure and updates the link availability flag.

4. The method according to claim 2, characterized in that, In step 3, the server calculates latency, uplink packet loss rate, bandwidth, distance, and moving speed. The delay is the round-trip time (RTT), which is calculated by the server based on the round-trip time between the probe message and the neighbor management response message, and smoothed using an exponentially weighted moving average. The uplink packet loss rate is obtained by the server based on the number of probe messages sent and the number of neighbor management response messages received. The server combines the downlink packet loss rate and the uplink packet loss rate into an effective packet loss rate. The server uses the effective packet loss rate as a packet loss indicator in the calculation of the comprehensive score; The effective packet loss rate is used to characterize the transmission reliability in both the server-to-client and client-to-server directions. The bandwidth is the remaining bandwidth of the link for a specific client, which is calculated by the server based on the total number of bytes sent and received on the link within the statistical period, the number of bytes occupied by the client, and the predicted service bandwidth of the client. The distance is calculated using the Haversine formula based on the coordinate latitude and longitude of the GPS / BeiDou positioning system. The movement speed parameter is calculated based on the absolute distance and the time difference between distance acquisition.

5. The method according to claim 1, characterized in that, In step 3, the server performs a two-way weighted fusion of the metrics that can be obtained bidirectionally by the server and the client. The server weights the client measurement value and the server measurement value according to a preset fixed fusion weight to obtain the weighted value of the corresponding indicator. The server participates in the calculation of the overall score based on the weighted values ​​from both ends.

6. The method according to claim 1, characterized in that, Step 3 further includes: the server performing hard index determination on the reference signal received power, signal-to-interference-plus-noise ratio and bandwidth; When any one of the reference signal received power, the signal-to-interference-plus-noise ratio, or the bandwidth is lower than the corresponding threshold, the server records the overall score of the link as 0; The server terminates the subsequent score calculation for that link after setting the overall score to 0.

7. The method according to claim 2, characterized in that, Step 3 also includes: The server performs normalization processing on the indicator information to obtain the normalized value of each indicator. The server uses exponential decay normalization for latency, packet loss rate, distance, and movement speed, making the normalized value more sensitive to the decline when the indicators deteriorate. The server calculates the comprehensive score based on the normalized value and weighted according to the comprehensive scoring weight.

8. The method according to claim 1, characterized in that, In step 5, the server performs anti-jitter determination before triggering link switching; When the overall score of the target link is higher than the overall score of the current primary link, and the difference between the overall score of the target link and the overall score of the current primary link exceeds a preset threshold, and this condition is met multiple times consecutively, the server triggers a switchover.

9. The method according to claim 1, characterized in that, The link notification mechanism in step 6 involves the server sending link notification messages through all currently established wireless links with the client. After receiving the messages, the client sends a link response message to confirm. After receiving the link response message, the server sends an acknowledgment response message to the client through the corresponding link. When the link notification message indicates that a specified link should be cleaned, the client clears the link state cache and restarts the link establishment process; When the server detects that a client's neighbor has timed out, it first sends a link cleaning notification to the client and accumulates the number of timeouts for that link. When the accumulated number of timeouts exceeds a set threshold, it performs state cleaning and reconstruction on the entire link.

10. The method according to claim 1, characterized in that, In step 7, the adaptive learning includes: The server counts the number of messages for each service type and calculates the proportion of each service type during the learning period. The comprehensive score weight is calculated by weighting the services according to their respective proportions, based on the preset scoring weights for each type of service. The service types include at least one or more of the following: voice calls, video conferencing, online video, industrial protocol reporting, web browsing, and file downloading.

Images