Search

CN-121985386-A - Three-network fusion intelligent switching system and method based on lightweight neural network

CN121985386ACN 121985386 ACN121985386 ACN 121985386ACN-121985386-A

Abstract

The invention relates to the technical field of communication data processing, in particular to a three-network fusion intelligent switching system and method based on a lightweight neural network. According to the invention, the future network quality is predicted by arranging the lightweight neural network model at the terminal, so that the switching decision is converted from passive response to prospective judgment, and the stability evaluation and self-adaptive correction are carried out on the prediction result through the reliability correction mechanism, so that abnormal fluctuation interference is restrained, and further, an accurate switching strategy is generated based on the corrected decision index, and finally, the accurate switching strategy is efficiently executed through the firmware layer, thereby effectively solving the problems that network deterioration cannot be avoided in a prospective manner, the switching decision response speed is low and the reliability is poor due to the fact that the delay judgment on the current state is relied on and the model calculation efficiency is insufficient.

Inventors

  • LIU WEIJIE

Assignees

  • 北京中元易尚科技有限公司

Dates

Publication Date
20260505
Application Date
20260203

Claims (10)

  1. 1. Three-network integration intelligent switching system based on lightweight neural network, characterized by comprising: the data acquisition module is used for acquiring network quality parameters corresponding to the cellular network, the wireless local area network and the private network at the terminal side in real time; the prediction module is deployed in the hardware acceleration unit at the terminal side, is used for constructing the characteristic input of the historical time sequence based on the network quality parameter, and is input into a preset lightweight neural network model to respectively predict the network quality of the cellular network, the wireless local area network and the private network in a future time window, and outputs a corresponding prediction quality index; The reliability correction module is used for generating a reliability quantification factor used for reflecting the prediction stability according to the change characteristics of the prediction quality index in a plurality of continuous prediction periods, and carrying out joint correction or inhibition correction on the prediction quality index based on the change trend judgment result of the reliability quantification factor in the future time window so as to obtain a decision index used for switching judgment; The decision control module is used for generating a network switching strategy based on the numerical interval distribution and the relative ordering relation of the decision index and combining the current service; and the firmware execution module is used for generating a corresponding protocol stack control instruction based on the network switching strategy and executing the protocol stack control instruction through the hardware acceleration unit so as to complete the switching among the cellular network, the wireless local area network and the private network.
  2. 2. The three-network convergence intelligent switching system based on the lightweight neural network as claimed in claim 1, wherein the data acquisition module comprises: a parameter obtaining unit, configured to obtain signal strength, round trip delay, available bandwidth, and packet loss rate from driving layers of the cellular network, the wireless local area network, and the private network, respectively; and the time alignment unit is used for performing time stamp alignment and sampling period unified processing on the signal strength, round trip delay, available bandwidth and packet loss rate of different network sources so as to form the network quality parameters.
  3. 3. The three-network converged intelligent switching system based on a lightweight neural network of claim 2, wherein the prediction module comprises: A feature construction unit for constructing the network quality parameter into a feature matrix including a plurality of historical time points based on a preset time sliding window; and the model calculation unit is used for calling the lightweight neural network model, calculating the feature matrix and outputting the predicted quality indexes of the corresponding cellular network, the wireless local area network and the private network in the future time window.
  4. 4. The three-network convergence intelligent switching system based on the lightweight neural network as claimed in claim 3, wherein the reliability correction module comprises: A fluctuation analysis unit for calculating a fluctuation amplitude and a dispersion of the predictive quality index in a continuous preset analysis number of predictive periods; a reliability calculation unit to generate the corresponding reliability quantization factor based on the fluctuation amplitude and dispersion; And the correction control unit is used for carrying out joint correction on the predicted quality index when the reliability quantization factor meets a preset stability condition or carrying out inhibition correction on the predicted quality index when the stability condition is not met so as to obtain the decision index.
  5. 5. The three-network converged intelligent switching system based on a lightweight neural network of claim 4, wherein the decision control module comprises: The interval judging unit is used for judging whether the cellular network, the wireless local area network or the special network meets a switching trigger condition according to the falling condition of the decision index in a preset decision interval, and generating a corresponding candidate network set; The sorting comparison unit is used for sorting the corresponding decision indexes in the candidate network set from large to small, marking the first decision index as a main selection index and marking the second decision index as an alternative index; And the strategy selection unit is used for combining the current service type, the difference relation between the main selection index and the alternative index to generate a single-network switching strategy, a multi-network concurrency strategy or a pre-switching strategy so as to obtain the network switching strategy.
  6. 6. The three-network convergence intelligent switching system based on the lightweight neural network as claimed in claim 5, wherein the policy selection unit comprises: a single-network switching generation subunit, configured to generate the single-network switching policy when the primary selection index is higher than a preset first switching threshold, and a difference between the primary selection index and the alternative index is greater than a preset difference threshold; A multi-network concurrency generation subunit, configured to generate the multi-network concurrency policy when both the primary selection index and the alternative index are higher than a preset second concurrency threshold, and the difference value is less than or equal to the preset difference value threshold; and the pre-switching generation subunit is used for generating the pre-switching strategy when the main selection index is higher than a preset third pre-switching threshold value and the delay sensitivity level of the current service meets a preset pre-switching condition.
  7. 7. The three-network converged intelligent switching system based on a lightweight neural network of claim 6, wherein the policy selection unit further comprises: An index extraction subunit configured to extract an allowable maximum delay of a service from a service data packet at the terminal side; And the level calculating subunit is used for calculating the corresponding time delay sensitivity level according to a preset level dividing rule based on the allowable maximum time delay.
  8. 8. The three-network converged intelligent switching system based on a lightweight neural network of claim 7, wherein the firmware execution module comprises: A policy parsing unit, configured to parse the network handover policy to determine a target network, a handover type, and a priority of handover execution; The instruction generating unit is used for generating a hardware control instruction sequence matched with a network communication protocol stack of the target network based on the target network, the switching type and the priority, so as to obtain the protocol stack control instruction; The drive execution unit is used for calling a special drive interface of the hardware acceleration unit and executing the protocol stack control instruction so as to control the radio frequency front end and the baseband processing unit corresponding to the terminal side to finish the switching operation to the target network.
  9. 9. The three-network convergence intelligent switching system based on the lightweight neural network as claimed in claim 8, wherein the instruction generation unit comprises: The protocol matching subunit is used for determining the corresponding network communication protocol stack type according to the network type of the target network and selecting a control instruction template matched with the network communication protocol stack type from a preset protocol instruction template set; The parameter filling subunit is used for filling the configurable parameters in the control instruction template based on the priority of the switching execution and the current network running state of the terminal so as to generate a specific protocol stack control instruction; And the timing sequence arrangement subunit is used for sequencing and combining the protocol stack control instructions according to the dependency relationship among the protocol stack control instructions to form the protocol stack control instructions which can be sequentially executed.
  10. 10. The three-network fusion intelligent switching method based on the lightweight neural network is applied to the three-network fusion intelligent switching system based on the lightweight neural network, and is characterized by comprising the following steps: acquiring network quality parameters corresponding to a cellular network, a wireless local area network and a private network in real time at a terminal side; Constructing characteristic input comprising a historical time sequence based on the network quality parameters, inputting the characteristic input into a preset lightweight neural network model to respectively predict network quality of the cellular network, the wireless local area network and the private network in a future time window, and outputting a corresponding prediction quality index; Generating a reliability quantization factor for reflecting the prediction stability according to the change characteristics of the prediction quality index in a plurality of continuous prediction periods, and carrying out joint correction or inhibition correction on the prediction quality index based on the change trend judgment result of the reliability quantization factor in the future time window so as to obtain a decision index for switching judgment; generating a network switching strategy by combining the current service based on the numerical interval distribution and the relative ordering relation of the decision index; and generating a corresponding protocol stack control instruction based on the network switching strategy, and executing the protocol stack control instruction through the hardware acceleration unit to finish switching among the cellular network, the wireless local area network and the private network.

Description

Three-network fusion intelligent switching system and method based on lightweight neural network Technical Field The invention relates to the technical field of communication data processing, in particular to a three-network fusion intelligent switching system and method based on a lightweight neural network. Background With the deep fusion and wide deployment of heterogeneous access technologies such as a high-speed cellular network, a new generation wireless local area network, various special wireless networks and the like, a user terminal is facing a complex communication environment with coexisting multi-system networks, signal coverage interweaving and dynamic change, and meanwhile, near-harsh requirements on the time delay, the reliability and the continuity of network connection are provided by ultra-high-definition video streaming, real-time interactive application and critical task services. Under the background, the traditional network switching mechanism depending on fixed threshold and hysteresis response is difficult to meet the requirements because the network switching mechanism cannot predict quality trend, is easy to cause service interruption and resource waste, and the traditional intelligent scheme based on cloud or complex neural network is limited by terminal computing power, model complexity and system integration cost, so that efficient and real-time decision and execution are difficult to realize at the actual terminal side, and the core challenges of improving seamless experience and reliable service of heterogeneous networks are formed together. CN113630830A discloses a network switching method, device and equipment based on a multi-attribute fuzzy neural network, wherein the method comprises the steps of obtaining a candidate switching access point set of terminal equipment, inputting motion attribute information of the terminal equipment and network attribute information of each candidate switching access point into a trained first target fuzzy neural network if the candidate switching access point set of the terminal equipment comprises a plurality of candidate switching access points, obtaining switching probability of each candidate switching access point, determining the candidate switching access point with the largest switching probability as a target switching access point of the terminal equipment, training the first target fuzzy neural network based on a plurality of groups of first history switching sample data which are updated in real time and have completed switching, determining whether the target switching access point meets the target switching condition of the terminal equipment or not if the candidate switching access point set of the terminal equipment comprises one candidate switching access point, and judging whether the target switching access point meets the target switching condition of the terminal equipment or not by the current network switching access point. Therefore, the prior art has the following problems that the adopted fuzzy neural network essentially carries out classification decision on the current network state based on historical data, lacks active prediction capability on the future quality change trend of the network, cannot realize prospective operation to avoid business interruption risks, is highly dependent on specific combination modeling of motion attributes and network attributes, is easy to obviously reduce in decision accuracy and reliability when the motion mode of a terminal or the network environment characteristics exceed the range of the historical training data, and can generate higher calculation delay and power consumption on terminal equipment with limited resources, and the real-time switching decision and execution of millisecond level are difficult to support. Disclosure of Invention Therefore, the invention provides a three-network fusion intelligent switching system and method based on a lightweight neural network, which are used for predicting future network quality by arranging a lightweight neural network model at a terminal side and introducing a reliability correction mechanism based on prediction stability to solve the problems that network deterioration cannot be avoided in advance, switching decision response speed is low and reliability is poor due to the fact that the current state is dependent on hysteresis judgment and model calculation efficiency is insufficient in the prior art. In order to achieve the above object, in one aspect, the present invention provides a three-network convergence intelligent switching system based on a lightweight neural network, including: the data acquisition module is used for acquiring network quality parameters corresponding to the cellular network, the wireless local area network and the private network at the terminal side in real time; the prediction module is deployed in the hardware acceleration unit at the terminal side, is used for constructing the characteristic