• 研究简报 •

### 地采暖木地板释热温度场的BP神经网络预测

1. 1. 山东建筑大学热能工程学院 济南 250101;
2. 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:2018-07-20 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）；山东建筑大学校内博士基金（X18006Z）。

### Prediction of Thermal Released Field by the Wood Flooring for Ground with Heating System Based on BP Network

Zhou Shiyu1, Du Guangyue2, Chu Xin2, Liu Xiaoping2, Zhou Yucheng2

1. 1. School of Thermal Engineering, Shandong Jianzhu University Jinan 250101;
2. School of Information and Electrical Engineering, Shandong Jianzhu University Jinan 250101
• Received:2018-04-02 Revised:2018-07-20 Online:2018-11-25 Published:2018-12-04

Abstract: [Objective] Based on the temperature data of the finite points obtained from the test, BP neural network is used to predict and analyze the temperature field in the closed cavity in the time and space dimensions, so as to provide the theoretical and data support for the thermal storage performance analysis of wood floor.[Method] Based on the equipment for testing the thermal released field by the wood floor which was developed by the author's research group, the temperature data in the closed cavity is acquired and divided into training and testing sets of neural network. In the time dimension, the coordinates and the temperature values of the first three time nodes are taken as inputs and the temperature value of the fourth time node is defined as the output. In this process, the previous 80 sets of temperature data are defined as training set while the next 28 sets of temperature data are defined as testing set. In the space dimension, temperature data of 4/5 of the total sensors are chosen as training set while the remaining 1/5 of the total sensors are defined as testing set. The BP networks of time and space are constructed based on the training set, respectively. Furthermore, the constructed models could be validated according to the testing set.[Result] In the time dimension, the computed errors and R2 are as following: mean relative error(MRE)=0.269 2%,maximum relative error(MAE)=5.916 0%,mean square error(MSE)=0.422 4%,fitting degree(R2)=0.998 7. In the space dimension, MRE=0.364 2%,MAE=4.781 8%,MSE=0.521 9%,R2=0.998 5.[Conclusion] The prediction result derived by BP neural network are adequately reliable, demonstrating that this method can effectively obtain the continuous and complete temperature field inside the measurement cavity of the wood floor, thus a new theoretical support is provided for the analysis and calculation of the thermal storage performance of the wood floor.