• 论文与研究报告 •

### 基于LS-SVM的采暖木质地板自然对流温度场建模

1. 1. 山东建筑大学信息与电气工程学院 济南 250101;
2. 山东建筑大学热能工程学院 济南 250101;
3. 辽宁科技大学电子与信息工程学院 鞍山 114051
• 收稿日期:2018-04-02 修回日期:2018-09-11 出版日期:2018-11-25 发布日期:2018-12-04
• 基金资助:
泰山学者优势特色学科人才团队（2015162）。

### Modeling of Natural Convection Temperature Field Produced by Wood Floor with Geothermal System Based on LS-SVM

Cao Zhengbin1, Zhou Shiyu2, Liu Xiaoping1, Gao Chuang3, Du Guangyue1, Zhou Yucheng1

1. 1. School of Information and Electrical Engineering, Shandong Jianzhu University Jinan 250101;
2. School of Thermal Engineering, Shandong Jianzhu University Jinan 250101;
3. School of Electrics and Information Engineering, University of Science and Technology Liaoning Anshan 114051
• Received:2018-04-02 Revised:2018-09-11 Online:2018-11-25 Published:2018-12-04

Abstract: [Objective] In order to study the distribution of temperature field of wood floor after storage of heat, in this paper, wooded heating source was placed on the bottom of the closed cylindrical cavity, and the prediction model of temperature field caused by natural convection is established by using the least squares support vector machine(LS-SVM), which will provide the data base for the subsequent inversion of the heat storage characteristics of wood floor with geothermal system.[Method] Firstly, the boundary conditions of the physical model of a cylindrical cavity and the initial temperature of the wood sample are given. Meanwhile, according to the law of fluid mechanics, the mass, momentum and energy conservation equations of heat transfer in heat field are established and solved by computational fluid dynamics(CFD) software. Secondly, the data obtained by CFD is normalized, and according to the initial temperature of wood samples, the normalized data is divided into two parts:training set and prediction set. There are 5 481 data in the training set, and the forecast set is 8 groups with 609 data in each group. The training set is used to train the LS-SVM model, the optimization algorithm of drosophila is used to optimize the kernel function parameter σ and regularization parameter γ of the model,and the optimal combination of the parameters is(1.0×1010,0.06).Finally, the optimized model is used to predict the temperature of each prediction set, and the result are compared with the prediction result of the BP-ANN method.[Result] The result showed that when using LS-SVM to predict different sample temperature field, the maximum fitting degree is 0.998 9, the minimum is 0.996 8; the average relative error, the maximum relative error and mean square error of the maximum values are 0.25%, 2.6% and 0.31%, respectively; the minimum values are 0.054%, 0.84% and 0.12%, respectively; the longest modeling time is 12.93 s, and the shortest is 12.72 s. Whereas, when the BP-ANN is used to predict the temperature field, the maximum fitting degree is 0.999 7, and the minimum fit degree is 0.998 3; the average relative error, the maximum relative error and the mean square error maximum value are 0.47%, 5.43% and 0.63%, respectively; the minimum values are 0.21%, 2.08% and 0.33% respectively; the longest modeling time is 107.15 s, and the shortest is 106.23 s. In contrast, it can be seen that the fitting error of LS-SVM to predict different temperature fields is smaller than that of BP-ANN, the fitting degree is similar to that of BP-ANN, the modeling and prediction time is much less than those of BP-ANN.[Conclusion] It is feasible to model the heat storage temperature field of the wood floor with geothermal system by using the LS-SVM method. Because this method is suitable for small sample modeling and has good generalization, it has obvious advantages for modeling and prediction of temperature field in the experimental environment with limited data. It can provide guidance for subsequent work of retrieving heat storage mechanism of wood floor with geothermal system.