|
| 基于支持向量回归的叶片挤压出料温度预测研究 |
| Prediction of Discharge Temperature for Blade Extrusion Based on Support Vector Regression |
| |
| DOI：10.3969/j.issn.1674-6457.2022.08.002 |
| 中文关键词: 叶片精锻 挤压 数据挖掘 质量预测 |
| 英文关键词: blade precision forging extrusion data mining (DM) quality prediction |
| 基金项目: |
| Author Name | Affiliation | | DAI Nan | School of Materials, Northwestern Polytechnical University, Xi'an 710072, China | | YU Xin-hong | School of Materials, Northwestern Polytechnical University, Xi'an 710072, China | | GUO Jia-xin | School of Materials, Northwestern Polytechnical University, Xi'an 710072, China | | YU Qi-yan | Xi'an Xiangxun Technology Limited Liability Company, Xi'an 710068, China |
|
| Hits: 1582 |
| Download times: 880 |
| 中文摘要: |
| 目的 针对传统叶片精锻过程中存在的质量波动较大的问题，提出一种预测传统叶片精锻质量的方法。方法 以GH4169转子叶片精锻的挤压工序为研究对象，以叶片精锻挤压过程中的160组模拟结果为数据集(以其中的120组数据为训练集、其余40组数据为测试集)，对截面点的出料温度进行预测。首先利用特定的数据预处理手段及特征工程提升模型的预测精度及泛化能力，建立SVR预测模型，然后基于网格搜索算法和粒子群优化算法(GS–PSO)对预测模型中参数C和γ进行调节，得到优化后的模型，最后将测试数据集带入优化后的模型中，将预测值与真实值进行对比。结果 单一的SVR模型预测效果不佳，利用GS–PSO算法优化后，模型自适应度由0.007 8左右降到0.005 2左右，模型收敛快且优化效果显著。30颗粒子迭代50次的最终优化结果为:C=425.432 8，γ=1.883 2，模型优化后的预测值与实际值之间拟合度较好，每组数据的预测误差都远小于10%。结论 经GS–PSO优化的出料温度SVR预测模型在测试集中有较好的预测效果，满足行业数据预测要求的一般标准，对传统叶片精锻过程指标预测具有较好的参考意义。 |
| 英文摘要: |
| The work aims to propose a way to predict the quality of traditional blade precision forging, so as to solve the large quality fluctuation of traditional blade precision forging. With the extrusion process of GH4169 rotor blade precision forging as the research object, 160 groups of simulation results of the precision forging extrusion process of blade were used as data sets, in which 120 groups of data were used as training sets, and the remaining 40 groups of data were used as test sets to predict the discharge temperature of the section point. First, specific data preprocessing methods and characteristics of the engineering model prediction accuracy and generalization ability were used to establish the SVR forecasting model. Then, based on grid search algorithm and particle swarm optimization algorithm (PSO), the forecast model parameter C and γ were adjusted to obtain the optimized model. At last, the test data was set into the optimized model. The predicted value was compared with the true value. The prediction effect of single SVR model was not good. After optimization with GS-PSO algorithm, the self-fitness of the model decreased from about 0.007 8 to about 0.005 2, with fast convergence and significant optimization effect. The final optimization result of 50 iterations for 30 particles was C=425.432 8, γ=1.883 2. The fitted degree between the optimized predicted values and the actual values is good, and the prediction error of each group of data is much less than 10%. The prediction results show that the SVR prediction model optimized by GS-PSO has a good prediction effect in the test set, which meets the general standard requirements of industry data prediction, and has a good reference significance for the prediction of traditional blade precision forging process indexes. |
|
View Full Text
View/Add Comment Download reader |
| Close |
|
|
|
渝公网安备 50010702501719号