Digital information intelligent selection processing system for optimizing transmission path

Abstract

The invention discloses a digital information intelligent selection processing system for optimizing a transmission path. The system comprises a data acquisition module, a prediction analysis module, a path evaluation module, a path decision module, a control and switching module and a feedback learning module. The data acquisition module is used for acquiring performance parameters from a plurality of candidate transmission paths and acquiring priority and reliability requirements of applications. A predictive analysis module generates predictive context data for each path. The path evaluation module calculates a merit figure of each candidate path. And the path decision module selects an optimal transmission path or a multi-path combination on the premise of ensuring the compatibility of the network protocol. And the control and switching module realizes seamless path switching according to a decision result. And the feedback learning module corrects the weight of the model according to the switched actual transmission performance so as to realize self-adaptive optimization. According to the invention, intelligent selection and optimization of a digital information transmission path are realized, time delay, jitter and energy consumption can be obviously reduced, and transmission stability and security are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of digital information transmission, and in particular to a digital information intelligent selection and processing system for optimizing transmission paths. Background Art

[0002] With the continuous expansion of the internet and industrial control networks, the transmission of digital information across multiple physical or logical transmission paths has become the norm. Existing technologies optimize physical traffic path selection by obtaining the currently planned route, generating alternative routes, and then deciding whether to switch routes based on the additional travel time when specific trigger conditions are met. These technologies primarily target vehicle travel scenarios, making decisions based on road type and additional travel time, making them difficult to directly apply to digital information transmission environments. Additionally, some technologies for multi-path data transmission propose fusing location-aware multi-day network performance data with short-term historical data within relay devices, mapping data between clients and servers onto multiple independent communication paths based on contextual information. This technology focuses on parallel data flow splitting and lacks adaptive trade-offs for different application requirements. Other literature discloses data forwarding methods that use long-term and short-term network performance data for multi-path transmission. These methods estimate network performance and perform data reorganization by fusing historical and short-term data. While these methods consider network performance prediction, they do not provide a comprehensive evaluation formula or involve the application of reinforcement learning or dynamic windowing strategies. In addition, existing intelligent picking systems only improve logistics efficiency from aspects such as map generation, route planning and time conflict optimization, and do not involve multi-objective selection of digital information transmission.

[0003] In summary, the existing technology still lacks a digital information transmission path optimization system that can simultaneously consider real-time network status prediction, application demand preferences, adaptive weight adjustment, and smooth path switching. New technical solutions are urgently needed to solve the above problems. Summary of the Invention

[0004] Technical purpose: In response to the shortcomings of the existing technology, the present invention discloses a digital information intelligent selection and processing system for optimizing transmission paths. By integrating long-term and short-term network performance data, an improved comprehensive path merit formula, and a decision-making strategy combining dynamic windows and reinforcement learning, real-time optimization and intelligent selection of digital information transmission paths are achieved.

[0005] Technical solution: To achieve the above technical objectives, the present invention adopts the following technical solution:

[0006] A digital information intelligent selection and processing system for optimizing transmission paths, comprising:

[0007] A data acquisition module is configured to collect network performance parameters and application requirement information from at least two candidate transmission paths in real time, wherein the network performance parameters include average latency, available bandwidth, packet loss rate, jitter, energy consumption, and security risk factor;

[0008] A prediction analysis module is used to generate a predicted performance context for each candidate transmission path using time series analysis based on the short-term real-time data collected by the data acquisition module and the long-term historical data related to the candidate transmission paths;

[0009] A path evaluation module is used to calculate the merit of each candidate transmission path based on the comprehensive path merit formula;

[0010] The path decision module is used to select the transmission path or multi-path combination with the highest comprehensive merit value based on the merit value calculated by the path evaluation module and application requirements through a dynamic window evaluation algorithm and reinforcement learning strategy;

[0011] The control and switching module is used to switch the transmission path of digital information in real time according to the decision results of the path decision module during the transmission process, ensuring data sequence and reducing packet loss during the switching process;

[0012] The feedback learning module is used to adaptively update the weight parameters of the comprehensive path merit formula according to the actual transmission performance after the transmission path is switched, and feed back the updated results to the path evaluation module and the path decision module.

[0013] Preferably, the comprehensive path merit formula is:

[0014] ,

[0015] in is the merit of the i-th candidate transmission path, is the available bandwidth of the i-th transmission path, d i is the average delay of the i-th transmission path, is the packet loss rate of the i-th transmission path, is the jitter of the i-th transmission path, is the energy consumption of the i-th transmission path, is the security risk coefficient of the i-th transmission path, 、 、 、 、 、 and is a non-zero weight parameter.

[0016] Preferably, the data acquisition module is further used to obtain the priority u and reliability requirement r of the application task, and the prediction analysis module adjusts the initial value of the weight parameter accordingly, so that different application types have different preferences in path selection.

[0017] Preferably, the long-term historical data integrated by the prediction and analysis module includes the average throughput, delay and packet loss rate of the transmission path in the past at least one day, and the short-term real-time data is the network performance parameters monitored in the past few seconds. The prediction and analysis module uses an exponential smoothing model to unify the long-term historical data and the short-term real-time data into predictive performance indicators.

[0018] Preferably, the dynamic window evaluation algorithm adopted by the path decision module includes: setting a sliding time window and merit threshold , when a candidate transmission path is in the nearest The average figure of merit within When the merit relative to the current transmission path exceeds a predetermined threshold, switching to the candidate transmission path is triggered; otherwise, the existing transmission path is maintained to avoid frequent switching.

[0019] Preferably, the reinforcement learning strategy adopted by the path decision module is an adaptive weight adjustment algorithm based on policy gradient, which uses real-time transmission benefits as reward signals and updates the policy network so that subsequent choices can better match the network status.

[0020] Preferably, the control and switching module establishes a new transmission connection and buffers data packets that have not been sent before switching the transmission path, and migrates data streams between the old transmission path and the new transmission path through double buffering technology to achieve seamless switching.

[0021] Preferably, the feedback learning module uses a method combining sliding average and gradient descent to adjust the weight parameters according to the deviation between the actual transmission delay and packet loss rate and the predicted value, so that the weight parameters converge adaptively over time.

[0022] A method for intelligently selecting and processing digital information for optimizing a transmission path is applied to the above-mentioned system for intelligently selecting and processing digital information for optimizing a transmission path, and specifically comprises the following steps:

[0023] S1. Using a data acquisition module to collect network performance parameters and application requirement information from multiple candidate transmission paths;

[0024] S2. Use the prediction analysis module to perform long-term and short-term fusion on the collected data to obtain the predicted performance context of each candidate transmission path;

[0025] S3. Calculate the merit of each candidate transmission path according to the comprehensive path merit formula using the path evaluation module;

[0026] S4. Utilize the path decision module to select the transmission path or multi-path combination with the highest merit using a dynamic window evaluation algorithm and reinforcement learning strategy;

[0027] S5. Using the control and switching module to switch the transmission path of the digital information in real time according to the decision result;

[0028] S6. Use the feedback learning module to update the weight parameters according to the actual transmission performance and return to step S2.

[0029] Preferably, in step S3, the initial value of the weight parameter is dynamically adjusted according to the priority and reliability requirements of the application task, so that high-priority tasks pay more attention to average delay and packet loss rate, and low-priority tasks pay more attention to energy consumption.

[0030] Beneficial effects: The digital information intelligent selection and processing system for optimizing transmission paths provided by the present invention has the following beneficial effects:

[0031] This invention simultaneously collects network performance indicators and application demand information in the data acquisition module, integrates long-term and short-term data into a prediction context in the prediction analysis module, objectively quantifies candidate paths using an improved formula in the path evaluation module, and implements adaptive path selection in the path decision module using dynamic windows and reinforcement learning. The control and switching module enables seamless switching, and the feedback learning module continuously updates the model, enabling the system to select the optimal transmission path in real time under dynamic network environments. Compared with existing technologies, this invention not only considers multiple network performance indicators but also dynamically adjusts weights based on different application requirements, significantly improving transmission efficiency and stability, reducing energy consumption, and enhancing security. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for describing the embodiments or the prior art.

[0033] Figure 1 It is the overall block diagram of the system of the present invention;

[0034] Figure 2 is a flow chart of the method of the present invention;

[0035] Figure 3 Schematic diagram of the performance curve of the present invention. DETAILED DESCRIPTION

[0036] The present invention will be described more clearly and completely below by way of a preferred embodiment in conjunction with the accompanying drawings, but the present invention is not limited to the scope of the embodiment.

[0037] like Figure 1As shown, a digital information intelligent selection and processing system for optimizing transmission paths includes:

[0038] The data acquisition module is used to collect network performance parameters and application demand information from at least two candidate transmission paths in real time. The network performance parameters include average delay, available bandwidth, packet loss rate, jitter, energy consumption and security risk factor.

[0039] This module, deployed at the entry point or client device of a candidate transmission path, collects in real time the average latency (e.g., obtained by measuring round-trip time, in milliseconds), available bandwidth (in megabits per second), packet loss rate (in dimensionless percentage), jitter (defined as the standard deviation of latency variation, in milliseconds), energy consumption (in joules per bit, measured by the energy monitoring module), and security risk factor (based on historical security event statistics, ranging from 0 to 1) of each candidate transmission path. The module also records the priority (expressed as an integer) and reliability requirement (ranging from 0 to 1) of the application task, allowing subsequent models to adjust weights based on different application requirements. The collected data is synchronized to the predictive analysis module at fixed intervals.

[0040] The prediction analysis module is used to generate the predicted performance context of each candidate transmission path by using time series analysis based on the short-term real-time data collected by the data acquisition module and the long-term historical data related to the candidate transmission path.

[0041] This module uses an improved exponential smoothing model to fuse historical data (long-term data) from at least one day in the past with real-time data (short-term data) from the past few seconds to calculate the prediction delay. , predicted bandwidth , predicted packet loss rate , predict jitter , predict energy consumption , predict safety risk factor Indicators such as , form the prediction performance context. Assume that the historical data sequence is , the real-time data series is , then the predicted value It can be expressed as ,in is the fusion coefficient, ranging from 0 to 1, determined by experiments. The prediction analysis module also and reliability requirements Give the initial weight parameter set 、 、 、 、 For example, for high priority and high reliability tasks, The initial values ​​of and γ are large to emphasize low latency and high bandwidth, while for low-priority tasks that are sensitive to energy consumption, The initial value of is large to emphasize energy consumption.

[0042] The path evaluation module is used to calculate the merit of each candidate transmission path according to the comprehensive path merit formula. The comprehensive path merit formula is:

[0043] ,

[0044] in is the merit of the i-th candidate transmission path, is the available bandwidth of the i-th transmission path, d i is the average delay of the i-th transmission path, is the packet loss rate of the i-th transmission path, is the jitter of the i-th transmission path, is the energy consumption of the i-th transmission path, is the security risk coefficient of the i-th transmission path, 、 、 、 、 、 and is a non-zero weight parameter, where Used to prevent the logarithmic function from being undefined when the bandwidth is zero. Used to adjust delay sensitivity, 、 、 、 and The impact of bandwidth, packet loss rate, jitter, energy consumption and security risk factor on the comprehensive merit is adjusted separately. By using a combination of logarithmic and power functions, the influence of extreme values ​​can be weakened and each indicator can be placed within a comparable scale range.

[0045] The path decision module is used to select the transmission path or multi-path combination with the highest comprehensive merit value through a dynamic window evaluation algorithm and reinforcement learning strategy based on the merit value calculated by the path evaluation module and application requirements.

[0046] This module first records the merit obtained by the path evaluation module in the sliding time window. Calculate the average merit of each candidate transmission path in the sliding window. When the average merit of a candidate transmission path exceeds the set threshold , and when the merit of the selected path improves by more than the difference Δ relative to the current path, the module determines that the transmission path has a clear advantage and issues a switching command to the control and switching module. To further improve decision-making, the path decision module constructs a reinforcement learning model based on policy gradients. The model uses the merit vector as the state, the selection of a transmission path as the action, and the actual transmission benefit (e.g., a linear combination of the effective transmission volume and the latency reduction per unit time) as the reward signal. Through continuous training, the model adjusts the output weight parameters so that the selected path strategy under different network conditions maximizes the long-term cumulative reward, thus achieving adaptive optimization.

[0047] The control and switching module is used to switch the transmission path of digital information in real time according to the decision results of the path decision module during the transmission process, ensuring data sequence and reducing packet loss during the switching process.

[0048] Upon receiving a handover instruction, the module first establishes a new transmission connection on the target transmission path and uses double buffering to cache unsent data packets. Once the new connection is established and stable, the module migrates the cached data to the new transmission path and closes the old connection, ensuring correct data order and no data loss during the handover. This module also monitors latency and packet loss during the handover process and provides feedback to the feedback learning module.

[0049] The feedback learning module is used to adaptively update the weight parameters of the comprehensive path merit formula according to the actual transmission performance after the transmission path is switched, and feed back the updated results to the path evaluation module and the path decision module.

[0050] This module continuously receives the real transmission indicators from the control and switching module and the data acquisition module, compares them with the predicted values, and adjusts the weight parameters by combining the sliding average and gradient descent methods. Specifically, let the deviation between the real indicator and the predicted indicator be , then the weight parameter is Update, which is the weight vector to be optimized. is the learning rate, is the loss function (mean square error or other appropriate functions can be used), It is the gradient of the loss function L with respect to the weight w. The updated weight parameters are returned to the prediction analysis module and the path evaluation module to form a closed-loop learning.

[0051] like Figure 2 As shown, the present invention also discloses a method for intelligently selecting and processing digital information for optimizing a transmission path, which is applied to the above-mentioned system for intelligently selecting and processing digital information for optimizing a transmission path, and specifically includes the following steps:

[0052] S1. Use the data acquisition module to collect network performance parameters and application requirement information from multiple candidate transmission paths.

[0053] The data acquisition module concurrently monitors M candidate transmission paths maintained by the system. For the i-th transmission path, it collects the following six network performance parameters at a configurable sampling period (typically 5ms): available bandwidth (estimated using link idle slot detection or the BBR-pacing algorithm, measured in Mbps), average latency (calculated based on a moving average of the RTT, measured in milliseconds), packet loss rate (calculated by the ACK / NAK ratio within the sampling window, measured in percentages), jitter (RTT sample standard deviation, measured in milliseconds), link energy consumption (measured by port power consumption sensors and converted to throughput, measured in J / bit), and security risk factor (output by the SD-WAN security agent based on historical security event statistics, ranging from 0 to 1). The module also reads the priority u and reliability requirement r of the current application task and writes all collected metric values, along with timestamps, into a ring buffer (BUF) in the format ((timestamp, pathID, key, value)), providing raw data for the subsequent predictive analysis module.

[0054] S2. Use the prediction analysis module to perform long-term and short-term fusion on the collected data to obtain the predicted performance context of each candidate transmission path.

[0055] After obtaining the time series data of each path from the BUF, the prediction analysis module processes it according to long windows (such as 60 seconds) and short windows (such as 1 second). The ARIMA-GARCH model is used to extract trends and fluctuations for the long window data, and the GRU network is used to perform transient extrapolation for the short window data. The two results are then fused and smoothed through the Kalman filter. After fusion, the module outputs the future Δt of each path. pred (about 500 ms) prediction vector ,The prediction vector, together with the priority u and reliability requirement r, constitutes the predicted performance context of the transmission path i, which is written into the shared parameter server for subsequent evaluation and decision-making.

[0056] S3. Calculate the merit of each candidate transmission path according to the comprehensive path merit formula using the path evaluation module.

[0057] In step S3, the initial values ​​of the weight parameters are dynamically adjusted based on the priority and reliability requirements of the application tasks, so that high-priority tasks place greater emphasis on average latency and packet loss rate, while low-priority tasks place greater emphasis on energy consumption. After the module calculation is completed, all merits are normalized and sorted from largest to smallest, and the results are written to the decision queue.

[0058] S4. Use the path decision module to adopt a dynamic window evaluation algorithm and reinforcement learning strategy to select the transmission path or multiple transmission path combinations with the highest merit.

[0059] Path decision module maintenance length The sliding window saves the last three Ranking, and jointly run a deep reinforcement learning agent based on DQN. The agent takes the predicted performance context as state input, the action space is {single path, parallel multi-path, hard switching}, and the reward function is defined as

[0060]

[0061] in The same source as the parameters in the comprehensive path merit formula. Smooth switching strategy: When the same transmission path is continuous Ranked first and score increment Maintain the status quo; when Enable MPTCP parallel multi-transmission path when A hard switch is performed when In one embodiment, .

[0062] S5. Use the control and switching module to switch the transmission path of digital information in real time according to the decision results.

[0063] The control and switching module accurately sends the transmission path switching instruction through the SD-WAN control plane interface according to the decision result of step S4: For parallel or soft switching, the "make-before-break" method is adopted to first establish a session on the new transmission path and cache unsent data packets, and then disconnect the old transmission path; for hard switching, after the new transmission path completes the TCP three-way handshake, an RST packet is immediately sent to terminate the old connection. During the switching process, the control and switching module monitors the packet disorder rate. If it exceeds 10 −3 FEC frames are automatically injected to ensure service continuity.

[0064] S6. Use the feedback learning module to update the weight parameters according to the actual transmission performance and return to step S2.

[0065] Feedback learning module in cycles s collects actual transmission indicators And the corresponding score , calculate the immediate reward , and based on

[0066]

[0067] Perform TD learning on DQN. At the same time, the module adjusts the weight parameters in the comprehensive path merit formula according to the formula Perform online gradient updates:

[0068]

[0069] The updated weights are simultaneously sent to the prediction analysis module and the path evaluation module, and the system returns to step S2 to form a closed-loop adaptive optimization process.

[0070] like Figure 3 The figure shows a comparison curve of the three key performance indicators of the present invention over time / batch. The solid line represents the decay trend of the average delay over time; the dotted line represents the decay trend of the link energy consumption over time; and the dot-dash line represents the decay trend of the packet loss rate (%) over time. The curve shows that after the method of the present invention undergoes online prediction and dynamic switching, all three indicators are significantly optimized.

[0071] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A digital information intelligent selection and processing system for optimizing transmission paths, characterized in that: include: A data acquisition module is configured to collect network performance parameters and application requirement information from at least two candidate transmission paths in real time, wherein the network performance parameters include average latency, available bandwidth, packet loss rate, jitter, energy consumption, and security risk factor; A prediction analysis module is used to generate a predicted performance context for each candidate transmission path using time series analysis based on the short-term real-time data collected by the data acquisition module and the long-term historical data related to the candidate transmission paths; A path evaluation module is used to calculate the merit of each candidate transmission path based on the comprehensive path merit formula; The path decision module is used to select the transmission path or multi-path combination with the highest comprehensive merit value based on the merit value calculated by the path evaluation module and application requirements through a dynamic window evaluation algorithm and reinforcement learning strategy; The control and switching module is used to switch the transmission path of digital information in real time according to the decision results of the path decision module during the transmission process, ensuring data sequence and reducing packet loss during the switching process; The feedback learning module is used to adaptively update the weight parameters of the comprehensive path merit formula according to the actual transmission performance after the transmission path is switched, and feed back the updated results to the path evaluation module and the path decision module.

2. A digital information intelligent selection and processing system for optimizing transmission paths according to claim 1, characterized in that: The comprehensive path merit formula is: ， in is the merit of the i-th candidate transmission path, is the available bandwidth of the i-th transmission path, d i is the average delay of the i-th transmission path, is the packet loss rate of the i-th transmission path, is the jitter of the i-th transmission path, is the energy consumption of the i-th transmission path, is the security risk coefficient of the i-th transmission path, 、 、 、 、 、 and is a non-zero weight parameter.

3. The digital information intelligent selection and processing system for optimizing transmission paths according to claim 1, characterized in that: The data acquisition module is further used to obtain the priority u and reliability requirement r of the application task, and the prediction analysis module adjusts the initial value of the weight parameter accordingly, so that different application types have different preferences in path selection.

4. The digital information intelligent selection and processing system for optimizing transmission paths according to claim 1, characterized in that: The long-term historical data integrated by the prediction and analysis module includes the average throughput, latency and packet loss rate of the transmission path in the past day or more. The short-term real-time data is the network performance parameters monitored in the past few seconds. The prediction and analysis module uses an exponential smoothing model to unify the long-term historical data and the short-term real-time data into predictive performance indicators.

5. The digital information intelligent selection and processing system for optimizing transmission paths according to claim 1, characterized in that: The dynamic window evaluation algorithm used by the path decision module includes: setting a sliding time window and merit threshold , when a candidate transmission path is in the nearest The average figure of merit within When the merit relative to the current transmission path exceeds a predetermined threshold, switching to the candidate transmission path is triggered; otherwise, the existing transmission path is maintained to avoid frequent switching.

6. The digital information intelligent selection and processing system for optimizing transmission paths according to claim 1, characterized in that: The reinforcement learning strategy adopted by the path decision module is an adaptive weight adjustment algorithm based on policy gradient. This algorithm uses real-time transmission benefits as reward signals and updates the policy network so that subsequent choices can better match the network status.

7. The digital information intelligent selection and processing system for optimizing transmission paths according to claim 1, characterized in that: The control and switching module establishes a new transmission connection and caches data packets that have not yet been sent before switching the transmission path. It migrates data streams between the old and new transmission paths through double buffering technology to achieve seamless switching.

8. The digital information intelligent selection and processing system for optimizing transmission paths according to claim 1, characterized in that: The feedback learning module uses a method combining sliding average and gradient descent to adjust the weight parameters according to the deviation between the actual transmission delay and packet loss rate and the predicted value, so that the weight parameters converge adaptively over time.

9. A digital information intelligent selection and processing method for optimizing transmission paths, characterized in that: A digital information intelligent selection and processing system for optimizing a transmission path as claimed in any one of claims 1 to 8, comprising the following steps: S1. Using a data acquisition module to collect network performance parameters and application requirement information from multiple candidate transmission paths; S2. Use the prediction analysis module to perform long-term and short-term fusion on the collected data to obtain the predicted performance context of each candidate transmission path; S3. Calculate the merit of each candidate transmission path according to the comprehensive path merit formula using the path evaluation module; S4. Utilize the path decision module to select the transmission path or multi-path combination with the highest merit using a dynamic window evaluation algorithm and reinforcement learning strategy; S5. Using the control and switching module to switch the transmission path of the digital information in real time according to the decision result; S6. Use the feedback learning module to update the weight parameters according to the actual transmission performance and return to step S2.

10. The method for intelligent selection and processing of digital information for optimizing transmission paths according to claim 9, characterized in that: In step S3, the initial value of the weight parameter is dynamically adjusted according to the priority and reliability requirements of the application task, so that high-priority tasks pay more attention to average delay and packet loss rate, and low-priority tasks pay more attention to energy consumption.

Images