A method and system for optimizing PDCP timer based on multi-dimensional perception, and storage medium

Abstract

The application relates to a multi-dimensional perception-based PDCP timer optimization method and system and a storage medium. The method comprises the following steps: collecting multi-dimensional heterogeneous data in a network environment and constructing a feature vector; mapping based on a preset rule base and / or a machine learning model to generate a recommended parameter value of a PDCP timer; updating the configuration of the PDCP timer of the target bearer with the recommended parameter value; monitoring the change of one or more preset performance indicators of the target bearer after the updated configuration takes effect; if the change degree of any preset performance indicator exceeds the corresponding quantization threshold, triggering a rollback operation to restore the configuration of the PDCP timer of the target bearer to the state before the update. The application solves the technical problems of rigid configuration, single optimization dimension and lack of safety closed loop in the prior art.

Description

Technical Field

[0001] This invention relates to the field of mobile communication technology, specifically to a PDCP timer optimization method, system, and storage medium based on multi-dimensional sensing. Background Technology

[0002] In 5G and other mobile communication networks, the PDCP layer is responsible for the integrity, order, and security of data transmission. Among these, the discard timer and the t-reordering timer are two key parameters that directly impact the user's experience. The former determines the maximum dwell time of an unacknowledged data packet in the sending buffer; setting it too short will cause unnecessary retransmissions, while setting it too long will lead to lag. The latter determines the maximum time the receiver can wait for out-of-order data packets; setting it too short will cause packet loss, while setting it too long will increase latency.

[0003] However, existing technologies generally suffer from the following problems: (1) Rigid configuration: Most solutions follow the 3GPP standard and adopt a unified static parameter configuration across the entire network, which cannot meet the differentiated requirements of different services such as eMBB (enhanced mobile broadband) and VoLTE (long-term evolution voice bearer) for latency and reliability. (2) Single optimization dimension: Existing optimization methods often rely on only single-dimensional data (such as only network load or only service type), lacking the ability to perceive multi-dimensional environmental linkages, and cannot make accurate decisions in complex wireless scenarios. (3) Lack of security closed loop: Some automated or semi-automated tuning solutions lack effective effect verification and rapid recovery mechanisms, and there is a risk of performance degradation or even network instability after parameter adjustment. Summary of the Invention

[0004] To address the technical problems in existing technologies, such as the inability of PDCP timer configuration to meet the differentiated latency and reliability requirements of various services like eMBB enhanced mobile broadband and VoLTE long-term evolution voice bearer; the lack of perception capability for multi-dimensional environmental linkage, resulting in inaccurate decision-making in complex wireless scenarios; and the risk of performance degradation or even network instability after parameter adjustment, this invention provides a PDCP timer optimization method, system, and storage medium based on multi-dimensional perception.

[0005] The technical solution of the present invention to solve the above-mentioned technical problems is as follows: A PDCP timer optimization method based on multi-dimensional sensing includes the following steps: Collect multidimensional heterogeneous data in the network environment and construct feature vectors; wherein, the multidimensional heterogeneous data includes key network performance indicators, user service types, communication environment parameters, and communication terminal capability information; The feature vector is received and mapped based on a preset rule base and / or machine learning model to generate suggested parameter values ​​for the PDCP timer. The suggested parameter values ​​are sent to the target bearer to update the configuration of the target bearer's PDCP timer; After the updated configuration of the PDCP timer for the target bearer takes effect, monitor the changes in one or more preset performance indicators of the target bearer. If any of the preset performance indicators is found to have changed beyond the corresponding quantization threshold, a rollback operation is triggered to restore the configuration of the PDCP timer carried by the target to its state before the update.

[0006] The beneficial effects of this invention are: it integrates multi-dimensional data such as KPIs, services, communication environment, and terminal capabilities, upgrading parameter decision-making from single-dimensional experience-based judgment to multi-dimensional intelligent decision-making. This better adapts to complex wireless scenarios such as weak edge coverage and high interference, significantly improving the optimization accuracy of timers. Furthermore, by introducing an automated closed-loop verification and rollback mechanism based on quantized thresholds, it effectively avoids the network risk of "getting worse with each adjustment," solving the biggest security pain point in automated network optimization and fundamentally ensuring system security and stability. Therefore, this invention solves the technical problems of rigid configuration, single optimization dimensions, and lack of security closed-loop in existing technologies.

[0007] To address the aforementioned technical problems, this invention also provides a PDCP timer optimization system based on multi-dimensional sensing, the specific technical content of which is as follows: A PDCP timer optimization system based on multi-dimensional sensing, the system being deployed in a mobile communication base station or a centralized self-organizing network platform, comprising: A multidimensional perception engine is used to collect multidimensional heterogeneous data in the network environment and construct feature vectors; wherein, the multidimensional heterogeneous data includes key network performance indicators, user service types, communication environment parameters, and communication terminal capability information; The intelligent decision-making model is communicatively connected to the multi-dimensional perception engine and is used to receive the feature vector and map it based on a preset rule base and / or machine learning model to generate suggested parameter values ​​for the PDCP timer. The parameter sending executor is communicatively connected to the intelligent decision model and is used to send the suggested parameter values ​​to the target bearer to update the configuration of the PDCP timer of the target bearer; The closed-loop verification and rollback module is connected to the parameter sending actuator and is used to monitor the changes of one or more preset performance indicators of the target bearer after the updated PDCP timer configuration of the target bearer takes effect; if the change of any preset performance indicator exceeds the corresponding quantization threshold, a rollback operation is triggered to restore the PDCP timer configuration of the target bearer to the state before the update.

[0008] To address the aforementioned technical problems, the present invention also provides a storage medium, the specific technical content of which is as follows: A storage medium storing a computer program or computer instructions, wherein when the computer program or computer instructions are executed by a computer processor, the steps of the above-described PDCP timer optimization method based on multidimensional perception are implemented. Attached Figure Description

[0009] Figure 1 This is a flowchart of a PDCP timer optimization method based on multidimensional sensing in an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of a PDCP timer optimization system based on multidimensional sensing in an embodiment of the present invention; Figure 3 This is a flowchart of the intelligent decision-making process in an embodiment of the present invention; Figure 4 This is a closed-loop verification logic diagram in an embodiment of the present invention; Figure 5 This is a schematic diagram comparing the effects of parameter adjustment in an embodiment of the present invention. Detailed Implementation

[0010] The principles and features of the present invention are described below with reference to the accompanying drawings. The examples given are only for explaining the present invention and are not intended to limit the scope of the present invention.

[0011] In LTE and 5G NR systems, the PDCP (Packet Data Convergence Protocol) layer primarily maintains two types of core timers, located at the transmitting and receiving ends, respectively. These timers are configured by higher layers (RRC) to control packet dropping, reordering, and status report triggering, in order to balance latency and packet loss rate.

[0012] Cell-level KPIs (Key Performance Indicators) are quantitative indicators used in mobile communication networks (such as 4G LTE and 5G NR) to measure core performance aspects such as access, availability, mobility, capacity, and coverage at the smallest statistical granularity. They are used to monitor, optimize, and ensure network quality.

[0013] Example 1 like Figure 1 As shown, this embodiment of the invention provides a PDCP timer optimization method based on multi-dimensional sensing, including the following steps: S1. Collect multidimensional heterogeneous data in the network environment and construct feature vectors; wherein, the multidimensional heterogeneous data includes key network performance indicators, user service types, communication environment parameters, and communication terminal capability information; The specific steps for acquiring the multidimensional heterogeneous data include: The network key performance indicators are obtained in real time from the base station logs and statistics interface within the most recent second; wherein, the network key performance indicators include at least the PDCP layer packet loss rate, the percentage of sampling points with lag latency greater than a first preset threshold, and the physical resource block (PRB) utilization rate; the first preset threshold is preferably 200ms; By parsing the Quality of Service (QoS) flow configuration in the Resource Control (RRC) signaling, the QoS identifier or service category identifier corresponding to the user service type can be obtained. The communication environment parameters are obtained from the measurement report reported by the terminal. The communication environment parameters include at least the reference signal received power (RSRP) and the signal-to-interference-plus-noise ratio (SINR). The terminal capability information is obtained from the information reported by the terminal; the terminal capability information includes at least the maximum number of Hybrid Automatic Repeat Request (HARQ) processes and PDCP processing capacity.

[0014] S2. Receive the feature vector and map it based on a preset rule base and / or machine learning model to generate suggested parameter values ​​for the PDCP timer; Specifically, the steps include the following: The feature vector is matched with a preset rule base; wherein the rule base includes rules set according to the combination of the user service type, the communication environment parameters and the network key performance indicators, and the rules directly map to the suggested parameter values ​​of the PDCP timer; When the feature vector cannot match any rule in the rule base, the feature vector is input into the pre-trained machine learning model, which then predicts and generates the suggested parameter value.

[0015] The machine learning model is a decision tree model; the pre-set rule base includes at least one independent rule for distinguishing between video streaming eMBB service and voice VoLTE service.

[0016] S3. Send the suggested parameter values ​​to the target bearer to update the configuration of the PDCP timer of the target bearer; S4. After the updated configuration of the PDCP timer of the target bearer takes effect, monitor the changes in one or more preset performance indicators of the target bearer; The preset performance indicators include: the downlink throughput of the target bearer, the packet loss rate of the target bearer at the PDCP layer, and the percentage of sampling points where the lag latency of the target bearer is greater than a first preset threshold. The degree of change of the preset performance index is obtained by calculating the rate of change of the preset performance index in the monitoring window compared with the value before the update, and comparing the rate of change with a preset quantization threshold. If the rate of change exceeds the preset quantization threshold, a rollback operation is triggered.

[0017] S5. If the change in any of the preset performance indicators is detected to exceed the corresponding quantization threshold, a rollback operation is triggered to restore the configuration of the PDCP timer carried by the target to the state before the update.

[0018] After triggering the rollback operation, the following is also included: The feature vector corresponding to the trigger rollback operation and the suggested parameter value are combined and recorded in the policy blacklist; When data is collected again with a similarity to the feature vector of any combination in the policy blacklist that exceeds a preset similarity threshold, the generation of the suggested parameter value corresponding to triggering the rollback operation is prevented or adjusted.

[0019] The PDCP timer optimization method based on multidimensional perception in this embodiment is triggered in an event-driven or periodic scanning manner.

[0020] The present invention has the following advantages over the prior art: By identifying different 5QI services, differentiated timer configurations at the load granularity are achieved, effectively resolving the conflict between video services (e.g., 5QI=9) and voice services (e.g., 5QI=1) in terms of resource competition, and realizing differentiated improvement of service experience.

[0021] It innovatively integrates four-dimensional data of KPI, business, wireless environment and terminal capabilities, upgrading parameter decision-making from single-dimensional experience judgment to multi-dimensional intelligent decision-making. It can better adapt to complex wireless scenarios (such as weak edge coverage, high interference, etc.) and significantly improve optimization accuracy.

[0022] Introducing an automated closed-loop verification and rollback mechanism based on quantized thresholds can effectively avoid the network risk of "getting worse with each adjustment," solving the biggest security pain point in automated network optimization and fundamentally ensuring system security and stability.

[0023] Based entirely on existing 3GPP standard interfaces and data, it requires no terminal cooperation or core network modification. It can be implemented simply by upgrading the software at the base station or centralized SON platform, making it highly valuable for promotion and easy to implement in engineering projects.

[0024] Example 2: Video stuttering optimization scenario In the NR900 cell, the percentage of sampling points with stuttering latency (>200ms) for enhanced mobile broadband (eMBB, 5QI=9) services reached as high as 15%, while the physical resource block utilization rate was only 45%. The multi-dimensional perception engine collected relevant features. The intelligent decision model matched the rule: "If the service type is eMBB (5QI=9) and the stuttering latency percentage is >10% and the network load is <60%", then it output optimization parameters: adjusting the discard timer from 300ms to 100ms and the t-reordering timer from 40ms to 20ms. After the parameter distribution executor completed the configuration, the closed-loop verification module monitoring showed that the stuttering rate dropped to 3%, and the throughput fluctuation was less than 5%, indicating successful optimization.

[0025] Example 3: Optimization of packet loss at the edge of voice communication At the cell edge, a VoLTE voice service user (5QI=1) has a reference signal received power of -115dBm and an uplink PDCP packet loss rate of 8%. After the perception engine collects features, the decision model matches the rule: "If the service type is VoLTE (5QI=1) and the reference signal received power is <-110dBm and the uplink packet loss rate is >5%", then the optimization parameters are output: adjust the t-reordering timer to 40ms and keep the discard timer at 300ms. After the adjustment, the subjective voice quality score (MOS) improves by 0.8 without additional traffic loss, and the verification is successful.

[0026] Example 4 like Figure 2 As shown, this embodiment provides a PDCP timer optimization system based on multi-dimensional sensing, the specific technical content of which is as follows: The system is deployed on a mobile communication base station or a centralized self-organizing network platform, including: A multidimensional perception engine is used to collect multidimensional heterogeneous data in the network environment and construct feature vectors; wherein, the multidimensional heterogeneous data includes key network performance indicators, user service types, communication environment parameters, and communication terminal capability information.

[0027] The intelligent decision-making model is communicatively connected to the multi-dimensional perception engine and is used to receive the feature vector and map it based on a preset rule base and / or machine learning model to generate suggested parameter values ​​for the PDCP timer. The parameter sending executor is communicatively connected to the intelligent decision model and is used to send the suggested parameter values ​​to the target bearer to update the configuration of the PDCP timer of the target bearer; The closed-loop verification and rollback module is connected to the parameter sending actuator and is used to monitor the changes of one or more preset performance indicators of the target bearer after the updated PDCP timer configuration of the target bearer takes effect; if the change of any preset performance indicator exceeds the corresponding quantization threshold, a rollback operation is triggered to restore the PDCP timer configuration of the target bearer to the state before the update.

[0028] The multi-dimensional perception engine specifically obtains data from the base station's KPI statistics interface, RRC signaling parsing module, physical layer measurement report processing unit, and terminal capability information storage unit; the parameter distribution executor completes parameter configuration through the PDCP protocol stack management interface inside the base station or the X2AP interface or Xn interface between base stations.

[0029] The multi-dimensional perception engine collects and integrates four types of heterogeneous data sources in real time: cell-level KPIs, user service types, wireless environment, and terminal capabilities. Cell-level KPIs include the percentage of sampling points with latency >200ms, uplink and downlink PDCP packet loss rate, and PRB utilization rate; User service types include obtaining standard 5QI / QCI values ​​(e.g., 5QI=1 for VoLTE, 5QI=9 for eMBB) by parsing the QoS Flow configuration in the RRC signaling. The wireless environment includes RSRP, SINR, and mobility speed periodically reported by the UE; Terminal capabilities include the maximum number of HARQ processes from the UE Capability IE and PDCP processing capabilities.

[0030] By using the aforementioned four-dimensional data as a joint input for PDCP parameter optimization, the limitations of traditional single-dimensional perception that relies solely on load or packet loss are overcome.

[0031] The intelligent decision-making model receives structured feature vectors (such as [5QI, RSRP, SINR, uplink packet loss rate, stuttering latency percentage, PRB utilization]) and outputs the target discardTimer and t-Reordering values.

[0032] The `discardTimer` parameter is started by the PDCP sender during configuration (only applicable to DRB) and is used to limit the maximum dwell time of unacknowledged PDCP SDUs (e.g., unsuccessful RLC scheduling) in the buffer. If the timeout occurs, the SDU is discarded to prevent buffer overflow or latency exceeding limits. Its value is configured by RRC (e.g., `sd-DiscardTimer`, common values ​​are 5ms, 10ms, ..., 1000ms, or infinity).

[0033] t-Reordering: Maintained by the PDCP receiver, it is used to wait for out-of-order PDCP PDUs (such as those caused by RLC retransmission or air interface packet loss); if a timeout occurs, the missing packet is considered unrecoverable, and the receive window is pushed forward (skipping the hole). Its value is configured by RRC (e.g., t-Reordering, typically ranging from 0-200 ms, unit: ms).

[0034] like Figure 3 As shown, the intelligent decision-making process of the intelligent decision-making model is as follows: Data Acquisition: Read KPI, signaling, and measurement reports from the gNodeB log buffer for the most recent second; Feature extraction: Construct feature vectors [5QI, RSRP, SINR, uplink packet loss rate, stuttering latency percentage, PRB utilization]; Rule matching: Query the predefined rule base (e.g., IF 5QI==1 AND RSRP<-110 AND uplink packet loss rate>5% THEN t-Reordering=40ms, discardTimer=300ms); if no matching rule is found and ML mode is enabled, input the prediction parameters of the lightweight decision tree model; ML mode represents the Machine Learning deployment / service pattern or Machine Learning Design Patterns.

[0035] Machine Learning Deployment / Service Model: This refers to the way a trained model provides prediction services in a production environment, such as online inference (real-time API), batch inference, streaming inference, edge deployment, or serverless model services. It often involves model packaging (ONNX / TorchScript), inference engines (Triton / TorchServe), version control, and A / B testing.

[0036] Machine Learning Design Patterns (ML Design Patterns) refer to reusable engineering solutions for data processing, feature engineering, training processes, and model deployment, such as Pipeline, Feature Store, ShadowServing, Checkpoints, and Ensemble, which aim to improve the maintainability and production reliability of ML projects.

[0037] Parameter generation: Output target discardTimer (in ms) and t-Reordering (in ms); Execution: The parameters are written to the PDCP context structure corresponding to the target bearer via X2AP or the internal message bus, and the changes take effect immediately.

[0038] Its rule matching uses two modes: Rule-first mode: Pre-built expert rule base (e.g., "IF 5QI=1 AND RSRP<-110dBm AND uplink packet loss rate>5%, t-Reordering=40ms"); ML fallback mode: When there are no matching rules, a lightweight decision tree model is invoked to predict the parameters.

[0039] It balances low complexity (rules) and high precision (ML), and all inputs are data that can be collected through 3GPP standard interfaces, without the need for additional signaling.

[0040] The parameter distribution executor writes new parameters to the corresponding UE's PDCP context structure at the Bearer granularity via the gNodeB's internal API or X2AP interface, ensuring independent configuration for different service flows. gNodeB, short for NextGeneration NodeB, is a key device in 5G networks responsible for transmitting and receiving wireless signals and connecting user mobile phones and the core network.

[0041] To resolve the experience conflicts caused by a "one-size-fits-all" approach, a differentiated timer strategy is adopted for VoLTE and eMBB services within the same cell.

[0042] The closed-loop verification and rollback module starts a 10-second monitoring window after the parameters take effect, continuously comparing the changes of three indicators: downlink throughput (compared to the average of the previous 10 seconds), PDCP packet loss rate, and the percentage of sampling points with stuttering latency greater than 200ms.

[0043] Triggering conditions: If any metric deteriorates beyond a threshold (e.g., throughput decrease rate ≥ 15%, packet loss rate increase ≥ 5%), the system will automatically roll back to the previous version and add the failed strategy to the blacklist.

[0044] By introducing a quantitative threshold-driven automatic rollback mechanism, the industry risk of "getting worse with each adjustment" is fundamentally avoided.

[0045] like Figure 4 As shown, the closed-loop verification logic of the closed-loop verification and rollback module is as follows: After the parameters are successfully sent, the system immediately starts a 10-second monitoring window; During this period, the rate of change of three key indicators was continuously calculated: (1) downlink throughput (compared to the average of the previous 10 seconds); (2) PDCP layer packet loss rate; (3) the percentage of sampling points with stuttering latency >200ms; If any indicator meets the "deterioration condition", the optimization is deemed to have failed. The deterioration condition includes a throughput decrease of ≥15%, a packet loss rate increase of ≥5%, and a stuttering rate increase of ≥8%. In other words, if any one of the following conditions is met, it is deemed to be a "deterioration condition".

[0046] The system immediately sends a rollback command to the parameter sending actuator to restore the PDCP timer carried by the target to the previous version configuration; At the same time, the feature vector and parameter combination of this failure will be written into the policy blacklist database to prevent repeated attempts in the short term.

[0047] By introducing a quantitative threshold-driven automatic rollback mechanism, the industry pain point of traditional automated optimization, which "only adjusts without verifying and gets worse with each adjustment", is addressed, thus ensuring network stability.

[0048] This embodiment operates in an event-driven and / or periodic scanning manner, employing the following steps: reading KPIs, signaling, and measurement reports from the gNodeB log buffer within the most recent second; constructing standardized feature vectors, filling missing fields with moving averages; querying a predefined rule base; if a match is found, parameters are directly output; otherwise, entering ML mode; the ML model outputs discardTimer (in milliseconds) and t-Reordering (in milliseconds); after the parameters are written to the PDCP entity, the closed-loop verification module is immediately initiated. This ultimately forms a closed-loop optimized logic for perception, decision-making, execution, and verification, with a total latency of less than 50ms, meeting 5G real-time requirements.

[0049] In other embodiments, the multi-dimensional perception engine data of this invention can be uploaded to the OSS / MDAF platform, where a centralized AI large model (such as a Transformer or LSTM network) replaces the local "intelligent decision-making model" for parameter calculation. The calculation results are then sent to the base station for execution.

[0050] This solution relies on high-bandwidth backhaul and centralized computing power, resulting in high latency (seconds). It is suitable for non-real-time weekly or monthly network optimization, but cannot meet the dynamic adjustment requirements of the millisecond-level service start-up instant that this invention addresses.

[0051] In some other embodiments, the alternative to terminal-assisted perception reporting is as follows: Instead of relying on base station-side KPIs and measurement reports, the UE directly measures application-layer QoE (such as video stuttering and voice MOS) by extending terminal-side functions and reports it to the base station via specific signaling (such as PDCP Status Report extension or MAC CE).

[0052] While this solution offers high sensing accuracy, it relies on hardware and software modifications to the terminal chip, resulting in slow standardization and significant challenges in widespread adoption. The passive sensing solution adopted in this invention, located on the base station side, requires no terminal cooperation, offering stronger compatibility and practicality for deployment in existing networks.

[0053] In other embodiments, reinforcement learning-based online training alternatives are as follows: In the intelligent decision-making model, instead of adopting the rule-first, lightweight ML fallback mode preferred in this invention, an online reinforcement learning algorithm is used to allow the model to directly interact with the environment in the live network environment for trial and error training.

[0054] like Figure 5 As shown, after parameter adjustments to this system, a histogram is used to quantitatively compare the user experience before and after optimization, allowing operations personnel to intuitively evaluate the results: before the adjustment, sampling points with latency >200ms accounted for 15%, which decreased to 5% after the adjustment, and the throughput curve remained stable, proving that the optimization was effective and safe. The horizontal axis represents the end-to-end latency (ms) of a single business interaction, and the vertical axis represents the number of sampling points.

[0055] Before adjustment: In NR900 cells, the percentage of sampling points with stuttering latency (>200ms) for eMBB services (5QI=9) reached 15% (i.e., 45 / 300 sampling points). After adjustment: The system shortens discardTimer from 300ms to 100ms and t-Reordering from 40ms to 20ms; After optimization, the number of sampling points longer than 200ms decreased to 5% (15 / 300), and the total throughput fluctuation was less than 5%, proving that there were no side effects; Standardized effect visualization templates can be provided and directly integrated into the OSS alarm dashboard to support the large-scale promotion of optimization strategies.

[0056] Example 5 This embodiment provides a storage medium storing a computer program or computer instructions. When the computer program or computer instructions are executed by the computer's processor, the steps of the above-described PDCP timer optimization method based on multidimensional perception are implemented.

[0057] The storage medium can be an internal storage unit of any data processing device described in any of the foregoing embodiments, such as a hard disk or memory. The storage medium can also be an external storage device of any data processing device, such as a plug-in hard disk, smart memory card, SD card, flash memory card, etc., mounted on the device. Furthermore, the storage medium can include both internal storage units and external storage devices of any data processing device. The computer-readable storage medium is used to store the computer program and other programs and data required by the data processing device, and can also be used to temporarily store data that has been output or will be output.

[0058] Example 6 This embodiment provides a computer, including a memory and one or more processors. The memory stores executable code. When the one or more processors execute the executable code, they implement the steps of the PDCP timer optimization method based on multidimensional perception in Embodiment 1.

[0059] The memory can be an internal storage unit of any data processing device described in any of the foregoing embodiments, such as a hard disk or RAM. The memory can also be an external storage device of any data processing device, such as a plug-in hard disk, smart memory card, SD card, flash memory card, etc., mounted on the device. Furthermore, the memory can include both internal storage units and external storage devices of any data processing device. The memory is used to store the computer program and other programs and data required by the data processing device, and can also be used to temporarily store data that has been output or will be output.

[0060] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the concept and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A PDCP timer optimization method based on multi-dimensional sensing, characterized in that, Includes the following steps: Collect multidimensional heterogeneous data in the network environment and construct feature vectors; wherein, the multidimensional heterogeneous data includes key network performance indicators, user service types, communication environment parameters, and communication terminal capability information; The feature vector is received and mapped based on a preset rule base and / or machine learning model to generate suggested parameter values ​​for the PDCP timer. The suggested parameter values ​​are sent to the target bearer to update the configuration of the target bearer's PDCP timer; After the updated configuration of the PDCP timer for the target bearer takes effect, monitor the changes in one or more preset performance indicators of the target bearer. If any of the preset performance indicators is found to have changed beyond the corresponding quantization threshold, a rollback operation is triggered to restore the configuration of the PDCP timer carried by the target to its state before the update.

2. The PDCP timer optimization method based on multi-dimensional sensing according to claim 1, characterized in that, The specific steps for acquiring the multidimensional heterogeneous data include: The key network performance indicators are obtained in real time from base station logs and statistical interfaces; wherein, the key network performance indicators include at least PDCP layer packet loss rate, percentage of sampling points with lag latency greater than a first preset threshold, and physical resource block utilization rate. By parsing the quality of service flow configuration in the resource control signaling, the quality of service identifier or service category identifier corresponding to the user service type can be obtained. The communication environment parameters are obtained from the measurement report reported by the terminal. The communication environment parameters include at least the reference signal received power and the signal-to-interference-plus-noise ratio. The terminal capability information is obtained from the information reported by the terminal; the terminal capability information includes at least the maximum number of processes for Hybrid Automatic Repeat Request and PDCP processing capability.

3. The PDCP timer optimization method based on multi-dimensional sensing according to claim 2, characterized in that, Based on a pre-defined rule base and / or machine learning model, a mapping is performed to generate suggested parameter values ​​for the PDCP timer, including the following steps: The feature vector is matched with a preset rule base; wherein the rule base includes rules set according to the combination of the user service type, the communication environment parameters and the network key performance indicators, and the rules directly map to the suggested parameter values ​​of the PDCP timer; When the feature vector cannot match any rule in the rule base, the feature vector is input into the pre-trained machine learning model, which then predicts and generates the suggested parameter value.

4. The PDCP timer optimization method based on multi-dimensional sensing according to claim 3, characterized in that, The machine learning model is a decision tree model; the pre-set rule base includes at least one independent rule for distinguishing between video streaming services and voice services.

5. The PDCP timer optimization method based on multi-dimensional sensing according to claim 1, characterized in that, The preset performance indicators include: the downlink throughput of the target bearer, the packet loss rate of the target bearer at the PDCP layer, and the percentage of sampling points where the lag latency of the target bearer is greater than a first preset threshold. The degree of change of the preset performance index is obtained by calculating the rate of change of the preset performance index in the monitoring window compared with the value before the update, and comparing the rate of change with a preset quantization threshold. If the rate of change exceeds the preset quantization threshold, a rollback operation is triggered.

6. The PDCP timer optimization method based on multi-dimensional sensing according to claim 1, characterized in that, After triggering the rollback operation, the following is also included: The feature vector corresponding to the trigger rollback operation and the suggested parameter value are combined and recorded in the policy blacklist; When data is collected again with a similarity to the feature vector of any combination in the policy blacklist that exceeds a preset similarity threshold, the generation of the suggested parameter value corresponding to triggering the rollback operation is prevented or adjusted.

7. The PDCP timer optimization method based on multi-dimensional sensing according to claim 1, characterized in that, The optimization method is triggered in an event-driven or periodic scanning manner.

8. A PDCP timer optimization system based on multi-dimensional sensing, used to implement the method as described in any one of claims 1 to 7, characterized in that, The system is deployed on a mobile communication base station or a centralized self-organizing network platform, including: A multidimensional perception engine is used to collect multidimensional heterogeneous data in the network environment and construct feature vectors; wherein, the multidimensional heterogeneous data includes key network performance indicators, user service types, communication environment parameters, and communication terminal capability information; The intelligent decision-making model is communicatively connected to the multi-dimensional perception engine and is used to receive the feature vector and map it based on a preset rule base and / or machine learning model to generate suggested parameter values ​​for the PDCP timer. The parameter sending executor is communicatively connected to the intelligent decision model and is used to send the suggested parameter values ​​to the target bearer to update the configuration of the PDCP timer of the target bearer; The closed-loop verification and rollback module is connected to the parameter sending actuator and is used to monitor the changes of one or more preset performance indicators of the target bearer after the updated PDCP timer configuration of the target bearer takes effect; if the change of any preset performance indicator exceeds the corresponding quantization threshold, a rollback operation is triggered to restore the PDCP timer configuration of the target bearer to the state before the update.

9. The PDCP timer optimization system based on multi-dimensional sensing according to claim 8, characterized in that, The multi-dimensional perception engine specifically obtains data from the base station's KPI statistics interface, RRC signaling parsing module, physical layer measurement report processing unit, and terminal capability information storage unit; the parameter distribution executor completes parameter configuration through the PDCP protocol stack management interface inside the base station or the X2AP interface or Xn interface between base stations.

10. A storage medium, characterized in that, The storage medium stores a computer program or computer instructions, which, when executed by the computer's processor, implement the steps of the PDCP timer optimization method based on multidimensional perception as described in any one of claims 1 to 7.

Images