Research On Key Technologies Of Intelligent Decision Based On Machine Learning

This paper studies the typical interference sensing technologies in communication systems,employes the supervised learning method to identify interference signals,and uses interference analysis and channel information to complete intelligent decision optimization and other key technologies.Based on the basic theory of interference perception,the thesis analyzes the basic theory and performance of typical interference perception,and establishes a detection model for typical interference.Using the typical characteristics of different interference signals in different change domains,it combines with some methods such as low-complexity machine learning methods to improve the interference recognition rate.Based on a typical communication system: DVB-S2 system,this paper analyzes the signal-to-interference-and-noise ratio estimation method,forms an intelligent decision-making optimization model based on interference analysis,and verifies the performance of the algorithm model through simulation.The paper first studies interference sensing technology based on energy detection.Simulation analysis of typical interference signal models,and research on signal detection algorithms such as matched filtering,energy detection,etc.The energy detection algorithm is selected as the detection method by comparing multiple detection algorithms,comprehensive performance and complexity;This paper analyzes and compares two methods of setting interference detection thresholds,Continuous Mean Excision(CME)and Forward Consecutive Mean Excision(FCME).The FCME algorithm with better performance is selected for subsequent simulation analysis.From the perspective of time domain and frequency domain,different interferences were detected and the simulation performance was verified.The paper then studies the interference recognition and perception technology based on offline supervised learning.By extracting the typical characteristics of the interference signal,the problems caused by the large amount of data in the original interference data and the existence of redundancy are avoided,and the differences between different interferences under different typical characteristics are simulated and compared.This paper studies the interference recognition technology based on decision tree and back propagation(BP)neural network,analyzes its theoretical algorithm,and focuses on the recognition performance of two learning methods.The paper analyzes the performance of decision tree interference recognition,identifies problems with poor performance and artificial thresholds based on decision tree algorithms,and uses a combination of methods to improve performance.The paper analyzes the number of nodes in the hidden layer of the BP neural network.If the number of nodes is too small,the network fitting ability is poor;if the number of nodes is too large,it will easily lead to overfitting.In order to obtain a reliable and stable model with strong generalization ability,the paper uses the cross-validation method to analyze the average prediction error of the model with different hidden layer matrix numbers,and uses the simulation results to select the array with the smallest average prediction error.The paper simulates the performance of interference recognition based on the combination of decision tree and BP neural network.The results show that the performance of the two methods is greatly improved compared with the single learning method.This paper then studies the intelligent anti-jamming decision-making technology based on neural network and optimization algorithm.Based on the project background,using a typical communication system DVB-S2 system,8 communication links were selected for subsequent simulation.The typical theoretical model of intelligent decision-making systems and key parameters affecting performance were analyzed.This paper focuses on the signal-to-interference-and-noise ratio estimation techniques that affect decision-making performance,and compares four different estimation techniques through simulation: Split-symbolmoments estimator(SSME),Maximum Likelihood(ML),Cyclostationary(CS),and Second Moment and Fourth Moment Mehthod(M2M4)estimation techniques,Both non-data-assisted SSME and CS methods need to calculate all symbols of the received sequence,and both use oversampling technology.The computational complexity is greater than that of data-assisted ML and M2M4,Simulation results show that under the condition of similar complexity,the estimation performance of ML is better than M2M4;The paper completes the construction of an anti-jamming decision model based on the Artificial Bee Colony Algorithm,designing the relevant parameters of the learning algorithm such as fitness function,and testing the performance of the model in a typical communication system through simulation to verify that the most reliable(the lowest bit error rate)and the most reliable bit rate High-efficiency(maximum information transmission rate)communication,and achieve the minimum transmission power decision strategy while ensuring reliability and effectiveness.The paper compares and analyzes the performance of intelligent decision technology of neural networks and bee colony algorithms.The results show that both algorithms can achieve the goal of maximizing the average information transmission rate and transmitting power as little as possible when the bit error rate meets the system requirements.When not in the training set,the neural network is better in terms of bit error rate performance.The thesis finally summarizes the research results of the full text and points out the follow-up research directions.