AI at the Edge: a Smart Gateway for Greenhouse Air Temperature Forecasting

Forecasting Air Temperature on Edge Devices with Embedded AI

“…In [3], an air temperature forecasting model based on an ANN, predicting air temperature with a RMSE of 2.5 ÷ 3.0°C, has been presented. Better results, in terms of RMSE, have been achieved with the same type of NN architecture (namely, ANN) in [14,19], reaching a RMSE equal to 0.839°C and 1.50°C, respectively. These results can be eventually justified by the fact that the last two models have been trained with a wider data set (in other words, a data set with a number of samples 4 or 64 times greater with respect to the one available in [3]).…”

“…These results can be eventually justified by the fact that the last two models have been trained with a wider data set (in other words, a data set with a number of samples 4 or 64 times greater with respect to the one available in [3]). In [27], a RNN-based model has been built using a data set of comparable size to [3] but achieving a better RMSE (equal to 0.865°C) than [19,27], but lower than [14]. Moreover, RBF networks have been adopted in [16] to build a model which outperforms all the above-mentioned papers in terms of RMSE.…”

“…In order to improve the management of greenhouses and to control their internal variables, in [19]-being the paper "AI at the Edge: a Smart Gateway for Greenhouse Air Temperature Forecasting" presented in the IEEE International Workshop on Metrology for Agriculture and Forestry (MetroAgriFor)-the authors have presented a novel approach based on the joint adoption of NNs, Edge Computing, and IoT technologies. In detail, the authors have "enriched" an IoT device, denoted as Smart GW and acting as a GW for a WSN installed in a greenhouse, in order to: (i) monitor its inner air temperature through an "edge intelligent" approach; (ii) locally forecast future inner air temperature's values; and (iii) regulate, according to the obtained results, the greenhouse air temperature (through internal actuators).…”

“…In order to extend the solution proposed in [19], in the current paper, the authors evaluate an alternative approach to build an air temperature forecasting model to be executed on the Smart GW presented in [19] (with similar purposes). In particular, rather than relying on meteorological weather conditions (collected outside the greenhouse), the aim is now on the exploitation of air temperature values collected inside the greenhouse to forecast the air temperature evolution.…”

“…In particular, rather than relying on meteorological weather conditions (collected outside the greenhouse), the aim is now on the exploitation of air temperature values collected inside the greenhouse to forecast the air temperature evolution. Moreover, the authors evaluate the performance (in terms of prediction precision and lightweightness) of the model developed following a novel approach, denoted as time series-oriented, with respect to the model proposed in [19]. Then, the authors investigate how the model's parameters-namely, the number of input variables (i.e., air temperature values), their sampling interval, and the NN architectureinfluence the model's prediction performance.…”