基于PSO-SVR的植物纤维地膜抗张强度预测研究  

Tensile Strength Prediction for Plant Fiber Mulch Based on PSO-SVR

作　　者：刘环宇[1] 陈海涛[1] 闵诗尧 张颖[1] 

机构地区：[1]东北农业大学工程学院,哈尔滨150030

出　　处：《农业机械学报》2017年第4期118-124,共7页Transactions of the Chinese Society for Agricultural Machinery

基　　金："十二五"国家科技支撑计划项目(2012BAD32B02-5)

摘　　要：为快速、准确地对生产过程中植物纤维地膜抗张强度进行预测,降低生产成本,提高原料利用率,以植物纤维地膜中试平台为依托,基于粒子群算法(PSO)优化支持向量机回归(SVR)模型,结合正交试验设计L25(56)方法,以纤维打浆度、施胶剂添加量、湿强剂添加量、地膜定量、混合比作为模型输入参数,以植物纤维地膜抗张强度为输出进行模拟预测,并将模拟结果与SVR、BP、RBF智能算法模型进行对比分析。结果表明:PSO-SVR模型能够较好地表达植物纤维地膜抗张强度与模型参数间的非线性关系,并能根据输入参数快速准确地对植物纤维地膜抗张强度进行预测,测试集样本中预测值与实际值间均方误差、决定系数和均方根误差为0.117 N2、0.915、0.342 N;与其他智能算法(SVR、BP、RBF)相比,PSO-SVR算法模型具有更高的适用性与稳定性。研究结果可为生产过程中不同抄造工艺参数下植物纤维地膜抗张强度的在线监控提供参考依据。Straw fiber is a kind of huge renewable biological macromolecule material, and using crop straw as the raw material to manufacture plant fiber mulch is an ideal way of promoting comprehensive utilization of straw resource. Tensile strength of plant fiber mulch is a measure of damage caused by external stress. In order to accurately and effectively predict the tensile strength, reduce production cost and improve the utilization rate of raw materials, based on pilot-production line of plant fiber mulch, particle swarm optimization （PSO） used to optimize support vector machine regression （SVR） combined with the orthogonal test method （ L25 （56 ） ） was proposed, namely, the PSO - SVR. The production processes variables were chosen, and the PSO- SVR model was established in Matlab 201lb. The input parameters affecting plant fiber mulch tensile strength through mechanism analysis were beating degree, dosage of wet strength agent, regulator, basis weight and mixture ratio; the evaluation index was tensice strength. The results were compared in terms of prediction accuracy with three prediction models respectively based on support vector machine regression （SVR）, back propagation neural network regression （BP） and radial basis function neural network regression （RBF）. The results obtained by using the PSO - SVR model showed that the mean square error was 0. 117 N^2, the coefficient of determination was 0. 915 and the root mean square error was 0. 342 N. The punishment factor and kernel parameter of SVR can select by PSO automatically. Compared with other intelligent algorithms, such as SVR, BP and RBF, PSO- SVR algorithm possessed superior applicability and stability. Therefore, this method can better reflect the actual tensile strength of plant fiber film, which can be used as a theoretical basis for the intelligent controlling under different process conditions.

关 键 词：植物纤维地膜 抗张强度 预测模型 支持向量机回归 粒子群算法 正交试验设计 

分 类 号：S216.2[农业科学—农业机械化工程] TP391.9[农业科学—农业工程]