关键词: 模型驱动, 田间数据, 压缩采集, 适用性评价

Abstract: The energy consumption of data transmission is effectively reduced when model-driven in-situ data compressive gathering method is employed. The mechanism was elaborated on about several data-prediction models, such as autoregressive integrated moving average model (ARIMA), support vector regression (SVR), derivative-based prediction (DBP) and clock work recurrent neural networks (CW-RNN). Moreover, their setting strategies of training parameters and error thresholds were explored based on air temperature, soil moisture, fruit growth and wind speed. Based on the gathered data error, data transmission ratio, model update and prediction overhead, the applicability of data-prediction models was evaluated using entropy weight technique for order preference by similarity to and ideal solution (TOPSIS). It was demonstrated that the optimal training parameters were relevant to model mechanism and data objects, and it was feasible to automatically acquire error thresholds based on previously gathered data. The model applicability was dependent on data objects, computation resources of nodes and network bandwidth. More specifically, Constant model had the greatest applicability, followed by DBP. ARIMA model was suitable to scenarios with limited bandwidth, but abundant computation resources. On the contrary, SVR model could be applied to scenarios with high bandwidth, but limited computation resources.

Key words: model-driven, in-situ data, compressive gathering, applicability evaluation