Search

CN-121814644-B - Internet of things communication link quality prediction and self-adaptive adjustment method based on machine learning

CN121814644BCN 121814644 BCN121814644 BCN 121814644BCN-121814644-B

Abstract

The invention belongs to the technical field of communication quality regulation and control, in particular relates to a machine learning-based internet of things communication link quality prediction and self-adaptive adjustment method, and aims to solve the problems of adjustment lag, low prediction precision and poor suitability in the prior art. The method comprises the steps of collecting and preprocessing multidimensional link original data to obtain a standardized characteristic data set, constructing a link quality association analysis rule base, mining association rules through mutual information entropy and an improved Apriori algorithm, obtaining an optimal rule base through verification and outputting quality grades, constructing a multi-objective optimization decision logic, matching the strategy base to screen an optimal adjustment strategy and executing, collecting and preprocessing and predicting data after adjustment, repeating preprocessing and predicting processes to obtain an actual value and a predicted value, calculating index variation, resetting the strategy when the index variation does not reach the standard, and supplementing data to update the rule base and the strategy base in an iterative mode. The method and the system realize accurate pre-judgment and multi-objective optimization adjustment of link quality, dynamically adapt to complex link environments, ensure stable operation of the business of the Internet of things and have strong applicability.

Inventors

  • HUANG LIN
  • ZHANG FANG

Assignees

  • 成都工业职业技术学院

Dates

Publication Date
20260512
Application Date
20260312

Claims (9)

  1. 1. The method for predicting the quality of the communication link of the Internet of things and adaptively adjusting the communication link of the Internet of things based on machine learning is characterized by comprising the following steps of: S1, acquiring original data in a link communication process in real time, wherein the original data comprises physical layer parameters, link layer parameters and environment interference parameters, and preprocessing the acquired original data to obtain a standardized link characteristic data set; S2, constructing a link quality association analysis rule base, generating by mining the historical association relation between the core features and the link quality evaluation indexes, inputting the standardized link feature data set into the link quality association analysis rule base, and outputting the link quality grade in a future preset time window; s3, constructing a multi-objective optimization decision logic, matching a preset adjustment strategy library with the aim of minimizing the link packet loss rate, reducing the transmission delay and maximizing the bandwidth utilization rate, screening out an optimal self-adaptive adjustment strategy adapting to the current link state, and executing link parameter adjustment; S4, carrying out real-time data acquisition on the link after the adjustment, and repeatedly executing the step S1 and the step S2 to obtain an actual value and a predicted value of the link quality after the adjustment; And S5, calculating the variation of the link quality evaluation indexes before and after adjustment, returning to the step S3 if the variation does not reach the preset optimization threshold, re-matching the adjustment strategy based on the new link state, and supplementing the adjusted link data to the original data set at the same time, and periodically updating the link quality association analysis rule base and the adjustment strategy base.
  2. 2. The machine learning-based internet of things communication link quality prediction and adaptive adjustment method according to claim 1, wherein the specific process of constructing the link quality association analysis rule base in step S2 is as follows: S21, selecting a standardized link characteristic data set, supplementing historical operation data of an Internet of things communication link in a corresponding period, wherein the historical operation data comprises actual values of packet loss rate, transmission delay, bandwidth utilization rate and signal to noise ratio which are in one-to-one correspondence with core characteristics, and constructing a complete historical sample set; S22, dividing each core feature obtained by screening into three high, medium and low feature regions according to numerical distribution by adopting a self-adaptive threshold dividing algorithm based on the aligned historical sample set to form a feature region set; according to preset link quality grade dividing standards, mapping the actual values of packet loss rate, transmission delay, bandwidth utilization rate and signal to noise ratio to four link quality grades with high quality, good quality, general quality and poor quality respectively, and establishing a mapping relation table of link quality evaluation indexes and quality grades; S23, calculating the association strength of the quality grades corresponding to each core characteristic interval and each type of link quality evaluation index through mutual information entropy based on a mapping relation table of the link quality evaluation index and the quality grades, reserving a strong association characteristic interval with the association strength larger than a preset threshold value to obtain a strong association sample subset, adopting an improved Apriori association rule algorithm, combining the multi-core characteristic intervals in the strong association sample subset as a front part, setting the quality grade and the change trend of the corresponding link quality evaluation index as a rear part, setting a minimum support degree and a minimum confidence degree threshold value, and mining a high-frequency and high-reliability association rule meeting the threshold value requirement to form a candidate association rule set; S24, performing rule conflict detection and de-duplication processing on the candidate association rule set, combining the actual operation scene of the communication link of the Internet of things with the rule to be optimized with the confidence coefficient within a preset threshold value interval, supplementing the constraint condition of the environmental interference parameter, and optimizing the rule expression precision; S25, selecting a verification sample subset in the aligned historical sample set, inputting core feature data of the verification sample into an initial link quality association analysis rule base, obtaining a link quality prediction grade through rule matching, comparing the link quality prediction grade with an actual quality grade corresponding to the verification sample, calculating the matching accuracy of each rule, removing invalid rules with accuracy lower than a preset qualification threshold, supplementing corresponding feature combination samples for rules with accuracy reaching standards but insufficient coverage, and re-mining and optimizing to finally obtain an optimal link quality association analysis rule base meeting the link quality prediction requirement.
  3. 3. The machine learning-based internet of things communication link quality prediction and adaptive adjustment method according to claim 2, wherein the specific process of obtaining the strongly correlated sample subset in step S23 is as follows: S231, extracting different sections of a single core feature from a feature section set, namely a feature section variable X, X= { X 1 ,x 2 ,...,x n }, wherein n is the number of sections of the single core feature, and n=3 corresponds to high, medium and low sections; extracting a quality grade corresponding to a certain type of link quality evaluation index from a mapping relation table of the link quality evaluation index and the quality grade, and marking the quality grade as a quality grade variable Y, Y= { Y 1 ,y 2 ,y 3 ,y 4 }, wherein the quality grade variable Y, Y= { Y 1 ,y 2 ,y 3 ,y 4 }, the quality grade, the good grade, the general grade and the poor grade are respectively corresponding to the quality grade, the good grade, the poor grade and the good grade; s232, adopting mutual information entropy to quantify the association strength of the characteristic interval variable X and the quality grade variable Y: ; Wherein X is any value of a characteristic interval variable X, namely a certain interval of a single core characteristic, and Y is any value of a quality grade variable Y, namely a certain link quality grade; P (X, Y) is the joint probability that the feature interval X and the quality grade Y occur simultaneously, P (X) is the edge probability that the feature interval X occurs, and P (Y) is the edge probability that the quality grade Y occurs; S233, strong correlation characteristic interval screening and sample subset construction, namely presetting a correlation strength threshold T, comparing the calculated correlation strength value with the threshold T, reserving a core characteristic interval with the correlation strength value being more than T, eliminating weak correlation characteristic intervals with the correlation strength value being less than or equal to T, eliminating samples containing the weak correlation characteristic intervals from aligned historical sample sets, and forming a strong correlation sample subset by combining the strong correlation characteristic intervals and corresponding quality levels of the rest samples.
  4. 4. The machine learning-based internet of things communication link quality prediction and adaptive adjustment method according to claim 2, wherein the specific process of obtaining the candidate association rule set in step S23 is as follows: S234, taking a strongly correlated sample subset as a mining data source, defining a term set as a single value or a combined value of a core feature interval, wherein the term set only containing the single core feature interval is a 1-term set; s235, traversing a strong correlation sample subset, counting occurrence frequencies of all 1-item sets, calculating the support degree of each 1-item set, reserving 1-item sets with the support degree more than or equal to S min , forming frequent 1-item sets, and eliminating non-frequent 1-item sets with the support degree less than S min ; Removing subsets containing non-frequent (k-1) -item sets in the candidate k-item sets in the pruning stage, repeating the step S235 until new frequent k-item sets cannot be generated, and summarizing all the frequent item sets to form a high-frequency item set; S236, generating and screening association rules, namely splitting each frequent item set in a high-frequency item set to generate all possible rule forms, calculating the confidence coefficient of each rule, reserving the association rules with the confidence coefficient not less than C min , eliminating the low-reliability rules with the confidence coefficient less than C min , performing redundancy detection on the generated rules, eliminating the redundancy rules, and classifying and sorting the remaining rules according to the link quality evaluation index type to form candidate association rule sets.
  5. 5. The machine learning-based internet of things communication link quality prediction and adaptive adjustment method according to claim 2, wherein the specific process of step S25 is as follows: s251, randomly extracting samples from the aligned historical sample set according to a preset proportion to form a verification sample subset, and reserving the residual samples as a follow-up rule optimization supplementary data source; S252, extracting core feature data from each verification sample in the verification sample subset, repeating a standardized processing flow to obtain standard verification feature data, inputting the standard verification feature data into a generated initial link quality association analysis rule base, traversing association rules corresponding to link quality evaluation indexes in the link quality association analysis rule base by adopting a feature interval precise matching algorithm, determining a target rule completely matched with the core feature interval combination of the verification sample, and outputting a link quality prediction grade corresponding to the target rule; s253, extracting an actual link quality grade corresponding to each verification sample from a verification sample subset, comparing the link quality prediction grade with the actual quality grade one by one, and recording a matching result of each standard rule, wherein the matching result is divided into two types of matching success and matching failure; S254, based on the comparison result, counting the matching success times and the total matching times of each association rule respectively, and quantifying the validity of the single rule by adopting an accuracy index; S255, presetting an accuracy qualification threshold A min , comparing the accuracy of each rule with A min , eliminating invalid rules with accuracy < A min , reserving valid rules with accuracy not less than A min , marking rules with accuracy not less than A min but total matching times less than the preset threshold as rules to be optimized, providing basis for re-mining of subsequent supplementary samples, and finally forming a rule set with validity verification.
  6. 6. The machine learning-based internet of things communication link quality prediction and adaptive adjustment method according to claim 1, wherein the specific process of step S3 is as follows: S31, extracting a link quality prediction result, wherein the link quality prediction result comprises a packet loss rate prediction value L pred , a transmission delay prediction value D pred , a bandwidth utilization rate prediction value U pred and a signal to noise ratio prediction value S pred , and determining a quality grade and a core short board index of a current predicted link by combining a set link quality grade division standard; s32, constructing multi-objective optimization decision logic by taking an analytic prediction result and a quantized communication requirement as inputs, wherein a core optimization objective is to minimize the link packet loss rate, reduce the transmission delay and maximize the bandwidth utilization rate, a signal-to-noise ratio predicted value S pred is taken as a constraint condition, S pred ≥S1 min ,S1 min is the lowest signal-to-noise ratio threshold of stable link communication, and a weight method is adopted to convert the multi-objective into a single-objective optimization function; S33, calling a preset adjustment strategy library, wherein the strategies comprise modulation and demodulation mode switching, transmission power adjustment, communication frequency band switching and data packetization strategy adjustment; S34, substituting each strategy in the candidate adjustment strategy set into an optimization objective function F to calculate an F value corresponding to each strategy, introducing a strategy execution cost coefficient C, including energy consumption cost and switching delay cost, correcting the optimization objective function to be F ' =FX (1+C), wherein the energy consumption cost is quantized according to the adjustment amplitude of transmission power, the switching delay cost is quantized according to the time consumption of modulation mode/frequency band switching, selecting the strategy with the minimum corrected F ' value as the optimal self-adaptive adjustment strategy, and preferentially selecting the strategy with the minimum execution cost coefficient C if a plurality of strategies with the same F ' value exist; S35, the gateway of the Internet of things analyzes the optimal self-adaptive adjustment strategy into a standardized control instruction, the control instruction is issued to the corresponding terminal of the Internet of things through the secure communication channel, the terminal starts the parameter configuration module after receiving the instruction, and the link parameter adjustment is completed according to a preset time sequence.
  7. 7. The machine learning-based internet of things communication link quality prediction and adaptive adjustment method according to claim 6, wherein the specific process of step S4 is as follows: s41, carrying out real-time data acquisition on the adjusted communication link of the Internet of things to form an adjusted original data set, and preprocessing to obtain an adjusted standardized link characteristic data set; S42, extracting actual values of link quality evaluation indexes from the standardized link characteristic data after adjustment, wherein the actual values comprise packet loss rate after adjustment, transmission delay after adjustment, bandwidth utilization rate after adjustment and signal-to-noise ratio after adjustment; s43, inputting the standardized link characteristic data into an optimal link quality association analysis rule base, repeating the link quality analysis prediction flow to obtain an adjusted link quality prediction value, and checking the prediction precision of the link quality association analysis rule base in the adjusted link state.
  8. 8. The method for predicting and adaptively adjusting the quality of an internet of things communication link based on machine learning according to claim 7, wherein the process of verifying the prediction accuracy of the link quality association analysis rule base in the adjusted link state in step S43 is as follows: S431, calculating an accuracy evaluation index, namely selecting an average absolute error, a root mean square error and a prediction accuracy as core accuracy evaluation indexes, and respectively quantifying the prediction deviation and the overall prediction reliability of the single-class link quality index; S432, presetting an accuracy evaluation threshold standard, wherein the accuracy evaluation threshold standard comprises an MAE threshold, an RMSE threshold and a prediction accuracy threshold; S433, when all three precision indexes meet the corresponding threshold requirements, judging that the prediction precision of the link quality association analysis rule base meets the standard in the adjusted link state, and then continuing to be used for the subsequent link quality prediction, when any precision index does not meet the threshold requirements, marking that the suitability of the link quality association analysis rule base is insufficient, and in the iterative optimization flow, preferentially supplementing the adjusted data serving as a core sample to a data set, and pertinently optimizing the characteristic association logic and interval division standard of the link quality association analysis rule base.
  9. 9. The machine learning-based internet of things communication link quality prediction and adaptive adjustment method according to claim 7, wherein the specific process of step S5 is as follows: S51, extracting a link quality predicted value before adjustment as a comparison reference, and respectively calculating the variation of each core link quality evaluation index, wherein the variation comprises a packet loss rate variation delta L, a transmission delay variation delta D, a bandwidth utilization variation delta U and a signal-to-noise ratio variation delta S; S52, judging that the adjustment effect meets the standard if the positive variation of all the core indexes is more than or equal to the corresponding optimization threshold value, judging that the adjustment effect does not meet the standard if the positive variation of any core index is less than the corresponding optimization threshold value or the negative variation occurs, immediately returning to the step S3, re-executing the steps S31-S35 based on the adjusted link state data, matching a new optimal self-adaptive adjustment strategy and executing until the adjustment effect meets the standard; S53, supplementing the adjusted original data set, the adjusted actual value and the adjusted predicted value of the quality index to the original data set in the step S1 to form an expanded original data set, updating the standardized link characteristic data set at the same time, and periodically starting an iterative optimization flow.

Description

Internet of things communication link quality prediction and self-adaptive adjustment method based on machine learning Technical Field The invention belongs to the technical field of communication quality regulation and control, and particularly relates to a machine learning-based internet of things communication link quality prediction and self-adaptive adjustment method. Background Along with the rapid development of the internet of things technology, the internet of things terminal is widely applied to multiple fields such as industrial control, intelligent security, environmental monitoring and the like, and the link communication quality directly determines the data transmission reliability, the real-time performance and the operation stability of the internet of things system. The communication link of the Internet of things is easily affected by factors such as complex environment interference (including electromagnetic interference, temperature and humidity change and shielding object influence), terminal movement, service load fluctuation and the like, so that the problems of link packet loss rate rise, transmission delay increase, bandwidth utilization unbalance and the like are caused, and data transmission interruption can be caused when serious, so that the normal development of the service of the Internet of things is affected. In order to ensure the link communication quality, in the prior art, a link quality monitoring and passive adjustment strategy is mostly adopted, namely, an adjustment mechanism is triggered after the link quality is found to be deteriorated by collecting link parameters in real time, and the adjustment hysteresis exists in the mode, so that the risk of link quality degradation is difficult to avoid in advance. Some technologies attempt to introduce association rules or simple predictive models to achieve advanced adjustment, but still suffer from a number of disadvantages: Firstly, a link quality association analysis rule base is constructed to divide feature intervals depending on manual experience, the suitability of the link quality association analysis rule base for complex link features is poor, a multi-dimensional parameter combination scene is difficult to cover, and the prediction precision is insufficient; secondly, the association strength mining lacks a quantification means, weak association characteristics are easy to keep, and the reliability of interference rule generation is improved; thirdly, the self-adaptive adjustment strategy lacks multi-objective optimization logic, is difficult to balance core indexes such as packet loss rate, time delay, bandwidth utilization rate and the like, does not consider the strategy execution cost, and causes unbalanced adjustment effect and economy, and fourthly, the link quality association analysis rule base and the strategy base update mechanism are stiff, the adjusted link state cannot be dynamically adapted, and the prediction and adjustment capability is degraded after long-term operation. In addition, in the prior art, the link quality prediction and self-adaptive adjustment have insufficient cooperativity, the prediction result cannot accurately guide the adjustment strategy screening, the reverse verification and the iterative optimization of a prediction model or a link quality association analysis rule base are also lacking after adjustment, a 'prediction-adjustment' disjoint closed loop is formed, and the high-precision and high-reliability requirements of the internet of things service on the link quality are difficult to meet. Therefore, the invention aims to provide a link quality control method capable of accurately excavating link characteristic association rules, realizing advanced prediction and multi-objective optimization adjustment and dynamic iterative optimization, and solving the technical problems of hysteresis, poor suitability, poor cooperativity and the like in the prior art. Disclosure of Invention The invention aims to provide a machine learning-based internet of things communication link quality prediction and self-adaptive adjustment method, which is used for solving the technical problems of hysteresis, poor adaptability and insufficient cooperativity in the prior art. In order to solve the technical problems, the invention adopts the following technical scheme: the Internet of things communication link quality prediction and self-adaptive adjustment method based on machine learning comprises the following steps: S1, acquiring original data in a link communication process in real time, wherein the original data comprises physical layer parameters, link layer parameters and environment interference parameters, and preprocessing the acquired original data to obtain a standardized link characteristic data set; S2, constructing a link quality association analysis rule base, generating by mining the historical association relation between the core features and the link quality evaluation ind