基于粒子群优化支持向量机的纱线质量预测OACSTPCD

In light of the insufficient prediction accuracy in yarn quality during complex spinning processes and the demand for large datasets,this study presents a small sample yarn quality prediction method by optimizing Support Vector Machines(SVR)using Particle Swarm Optimization(PSO).Firstly,the original dataset samples are preprocessed using grey relational analysis,and they are ranked based on their degrees of correlation.Combining this with prior knowledge,the primary cotton fiber features are selected.Secondly,for the small sample prediction problem,SVR prediction models with different kernel functions,including linear,polynomial,Gaussian,and adaptive bandwidth RBF kernels,are established.Finally,PSO algorithm is employed to identify the hyperparameters of the SVR models,including the regularization coefficient and bandwidth adjustment parameter.Additionally,a linearly decreasing inertia weight strategy is designed to enhance the overall optimization capability of the PSO algorithm.The simulation results show that the PSO-optimized Gauss kernel SVR model has a good prediction effect on different yarn quality indexes,and its average relative error is less than 2%.It is concluded that the PSO-optimized Gauss kernel SVR model has a low prediction error for yarn quality index and a good adaptability.