Optimal configuration strategy for temperature sensors in solar greenhouse based on HSIC

Abstract: Greenhouse has been one of the most widely passive solar energy applications for vegetable production in North China. Among them, the internal temperature is one of the most concerned environmental parameters in a greenhouse for crop growth. However, there is a non-uniform temporal and spatial distribution of temperature in the greenhouse. It is necessary to optimize the configuration with a limited number of temperature sensors for cost-saving and efficient management. In this study, an optimal configuration strategy was proposed for the temperature sensors in the greenhouse using the Hilbert Schmidt independence criterion (HSIC). The temperature was measured to reduce the redundancy of temperature sensors for the rich information. A Chinese solar greenhouse was taken as the research object, which located in Taian, Shandong Province, China. Specifically, the length, the width, the north wall, and the ridge height were 70, 10, 3.5, and 5 m, respectively. The testing period was four weeks during the coldest season from Dec. 9, 2020, to Jan. 6, 2021. Firstly, 22 temperature sensors were divided into two groups to monitor the temperature distribution in north-south vertical and east-west horizontal directions. A total of 7880 × 22 samples were collected in four weeks at a sampling interval of 5 min. A systematic analysis was made to clarify the non-uniformity of temperature spatial distribution in the greenhouse in the necessity of multi-sensors configuration, according to the monitoring data. Secondly, the different sensors were quantitatively evaluated for relative independence using HSIC. The sorting priority of sensor configuration was then proposed to maximize the comprehensive independent coefficients of sensors. Meanwhile, the sensor quantity was also selected to consider the constraints of sensor redundancy and information gain rate. Thirdly, the sorting and selection constituted the configuration strategy for the temperature sensors. As such, the sorting of sensors along the north-south vertical direction was ranked as the S6, S9, S3, S8, S2, S4, S13, S12, S1, S7, S14, S10, S11, and S5, according to the configuration strategy. With the increase of selected sensors according to sensors selection sequence, both the Root Mean Square Errors (RMSE) of average temperature values between the selected and all sensors and the information gain rates of the nth temperature sensor all decreased rapidly. When the number of selected sensors according to sensors selection sequence was 3, the RMSE of average temperature values between the selected and all sensors was 0.26 ℃ and the information gain rates of the third sensor is 4.51%. At the same time, the sorting of sensors along the east-west horizontal direction was H6, H5, H2, H3, H8, H1, H4, and H7. Similarly, with the increase of selected sensors, the RMSE of average temperature values and the information gain rates also decreased rapidly. When the number of selected sensors according to sensors selection sequence was 3, the RMSE of average temperature values between the selected and all sensors was 0.30 ℃ and the information gain rates of the third sensor is 7.31%. The results showed that the RMSE of average temperature and information gain rate were less than the thresholds (0.5 ℃ and 10%), respectively, fully meeting the monitoring requirements of the greenhouse temperature field, when the sensor combination was selected (S6, S9, and S3 sensors in the north-south vertical direction, and H6, H5, and H2 in the east-west horizontal direction). Finally, the temperature data for one week from Jan.10 to Jan. 17, 2021, was sampled to evaluate the effectiveness of the proposed strategy. The validation showed that the RMSE and information gain rate were 0.24 ℃, and 6.70%, respectively, along the north-south vertical direction, and 0.33 ℃, and 9.47%, respectively, along the east-west horizontal direction, all of which were meeting the threshold requirements. Consequently, the overall trend and information abundance of the greenhouse temperature field were optimized with the three sensors S6, S9, and S3 in the north-south vertical direction, while H6, H5, and H2 in the east-west horizontal direction, respectively, indicating better adaptability and effectiveness of the strategy. The optimal configuration strategy of temperature sensors can greatly contribute to accurately monitoring the overall trend of temperature for the higher abundance of temperature information with a limited number of sensors. The finding can provide a useful theoretical reference and technical support for the environmental monitoring and control of solar greenhouses.