Data-driven sensor placement for greenhouse microclimate measurement: A POD-Greedy reduced-basis approach

In large modern greenhouses, accurate measurement of the internal temperature field plays an important role in greenhouse system regulation and crop growth management. Due to the significant spatial heterogeneity of greenhouse temperature, ignoring the rationality of the number and spatial distribution of sensor nodes may easily lead to either insufficient measurement accuracy or excessive sensing investment, which will both have a negative impact on economic benefits of crop planting. Different from the traditional sensor placement, this paper fully considers spatiotemporal heterogeneity of the greenhouse temperature field, and extract the dominant features by proper orthogonal decomposition (POD), based on which the optimal node positions of sensors are efficiently derived through Greedy algorithm. Further, by using these limited sensor nodes, the whole environmental information of greenhouse could be accurately reconstructed and estimated through combining POD-linear stochastic estimation (POD-LSE). To validate the proposed methodology, a field experiment is conducted in a commercial greenhouse (15.8m×23.6m×3.6m) located in Eastern China. The internal air temperature distribution is monitored over a two-week period during summer using a network of 40 wireless sensors. Results show that, optimized sensor layouts containing only three to six sensors could reconstruct original 40 nodes’ temperature field with errors spanning from 0.368 to 0.315 (average RMSE). Further, by utilizing limited sensor nodes, the temperature field over a subsequent two-day period could also be estimated with high spatial resolution, and the best estimation accuracy for this period reaches 0.721 (six sensor nodes), which is better than previous reported methods.