Font Size: a A A

Application Of Improved PSO Algorithm In Machine Learning

Posted on:2021-02-15Degree:MasterType:Thesis
Country:ChinaCandidate:Q LuoFull Text:PDF
GTID:2428330647951517Subject:Computer technology
Abstract/Summary:PDF Full Text Request
PSO has been widely used in the field of optimization.As a hot research field at present,machine learning has the characteristics of covering a wide range of knowledge and is the most core part of artificial intelligence.This article takes the improved PSO as a starting point,and then tries to combine it with machine learning technology,and uses the optimized model to perform actual case experiments to verify its effectiveness.The specific content of this article includes three aspects,respectively for:1)A particle swarm optimization algorithm-DCOPSO using a new mutation strategy is proposed.In the mutation strategy,the global optimal particle will be dimensionally reversed learning mutation of the center of gravity.Dimensional variation reduces interdimensional interference,and guides particles to fly to better positions by updating the global optimal position,while enhancing the diversity of the population..2)Use DCOPSO to optimize the parameters of the BP neural network in the initialization process,and use the optimized weights and thresholds to train the model.In this paper,the demand for shared bicycles in Seoul is used as the forecast target to verify the model.The experiment compares the three models of BP,PSO-BP and DCOPSO-BP.The results show that the BP neural network optimized by the optimized algorithm can be A relatively small loss can be achieved,which can quickly converge and increase the speed of model training.The DCOPSO algorithm is also better than the PSO algorithm.3)Use DCOPSO to optimize the hyperparameters of the SVM model.In this article,we take the Gaussian kernel function as an example.The optimization targets are C and ?.The traditional hyperparameter search methods include Bayesian optimization or network partitioning.The data set used in the experimental data is the clinical characteristics of various relevant indicators of breast cancer patients.There are 64 patients in total,and 52 healthy people are the control samples.Comparing the three models of SVM,PSO-SVM and DCOPSO-SVM,from the experimental results,the optimization algorithm can be used to find a hyperparameter that makes the model better,and the hyperparameter classification result that the DCOPSO algorithm can find the best.
Keywords/Search Tags:Swarm Intelligence, Particle Swarm Optimizition, BP neural network, SVM
PDF Full Text Request
Related items