Behavior Recognition and Fuel Consumption Prediction of Tractor Sowing Operations Using Smartphone
Lili Yang,Weize Tian,Weixin Zhai,Xinxin Wang,Zhibo Chen,Long Wen,Yuanyuan Xu,Caicong Wu
DOI: https://doi.org/10.25165/j.ijabe.20221504.7454
2022-01-01
International Journal of Agricultural and Biological Engineering
Abstract:In order to qualitatively recognize the behaviors and investigate the relationship between fuel consumption and machinery driving modes of the tractor in a low-cost approach, this study proposed a method for behavior recognition and fuel consumption prediction of tractor sowing operations using a smartphone. First, three driving modes were developed for maize sowing scenarios: manual driving assisted driving and unmanned driving. While sowing, smartphone software and CAN (Controller Area Network) storage devices collected both positional data and engine operating conditions. Second, the tractor trajectory points were divided into kinematic sequences, with six driving cycle indicators built in each series based on the time window. Based on the semantic information of the kinematic sequences, the three operations of sowing, seeds filling, and turning round were well recognized. Last, a model for maize sowing fuel consumption forecast was advanced using the principal component analyses and random forest algorithm, regarding three factors: driving cycles, operating behaviors, and driving patterns. When compared to the traditional K-means algorithm, the results demonstrated that the harmonic mean of the precision and recall (F1 score) of sowing behavior recognition, seeds filling behavior recognition, and turning behavior recognition were enhanced by 2.06%, 8.99%, and 21.79%, respectively. In terms of the impacts of driving modes and operating behaviors on fuel consumption, assisted driving mode had the lowest fuel usage for both sowing and turning behavior. Therefore, assisted driving is the most fuel-efficient mode for maize sowing. Combining the three driving modes, the relative error of the fuel consumption prediction model was 0.11 L/h, with the manual driving mode having the lowest relative error at 0.09 L/h. This research method lays the foundation for the optimization of tractor operation behavior, the selection of tractor driving mode, and the fine management of tractor fuel consumption.