• 论文与研究报告 •

### 基于TWSVM和模糊的木质采暖地板蓄热温度预测模型

1. 山东建筑大学信息与电气工程学院 济南 250101
• 收稿日期:2018-04-02 修回日期:2018-06-20 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）。

### Temperature Prediction Model of Heat Storage of Wooden Flooring for Ground with Heating System Based on TWSVM and Fuzzy

Cao Zhengbin, Liu Xiaoping, Du Guangyue, Chu Xin, Liu Dawei, Zhou Yucheng

1. School of Information and Electrical Engineering, Shandong Jianzhu University Jinan 250101
• Received:2018-04-02 Revised:2018-06-20 Online:2018-11-25 Published:2018-12-04

Abstract: [Objective] With regard to the heat storage characteristics of wooden flooring, a temperature prediction model based on twin support vector machine(TWSVM) and fuzzy algorithm was established for the distribution of temperature field of wooden flooring heat storage, which provides an effective analysis method for further research on heat storage of wooden flooring.[Method] Firstly, the wood samples which are heated to the set temperature are pushed into the detection cavity of the heating floor storage performance testing instrument for free heat dissipation, and the temperature sensor array with multi-layer annular distribution characteristics in the instrument is used to dynamically record the temperature data on the time dimension and spatial dimension, then filtering noise reduction and normalization processing are carried out. Secondly, aiming at the problem that too much modeling data will lead to the rapid expansion of the computational complexity of the twin support vector machine, the test data is evenly partitioned in the paper, and the verification set sample are randomly extracted from each block data, and the remaining ones are training samples. Based on the different training samples, the TWSVM model is trained, and the generalization performance of the model is verified by the verification set in the corresponding samples. The kernel function parameters σ, penalty parameters γ and relaxation factors ξ of the model are optimized by the grid search method. Finally, based on the fuzzy principle, the Gaussian function is constructed for the input space of the test sample.The prediction result of all models are superimposed with the corresponding membership weight values to generate the final training result of the model.[Result] The result show that when TWSVM and fuzzy method are used to predict the temperature values of different samples in time dimension, the maximum fitting degree is 99.59%, the minimum is 98.92%, the longest and shortest modeling time is 186.90 s and 64.39 s, respectively. When the temperature values in spatial dimension are predicted, the maximum fitting degree is 99.23%, the minimum is 98.96%, the longest and shortest modeling time is 274.37 s and 93.30 s, respectively.[Conclusion] Because the TWSVM method involves matrix inversion in the computation process, it is only suitable for processing small data samples. In this experiment, because of the large amount of temperature data needed to study the storage characteristics of wooden flooring, there is a limitation to directly use TWSVM to model the experimental data. After introducing the fuzzy method, the temperature data are divided into several small training samples in time and spatial respectively, then each training sample is modeled and trained by TWSVM, respectively. According to fuzzy rules, the final prediction result is determined by the membership value of each temperature point on the fuzzy function. The above method can improve the adaptation range of TWSVM modeling, and give full play to its fast and generalization advantages.