On-line modeling of tractor pitch angle based on ARMA

Background: The working quality and efficiency of agricultural machinery are largely dependent on the bottom-layer flatness of paddy fields. There are various kinds of profiling mechanism which are designed for the attitude adjustment of agricultural implement, however, response rate is still limited to the control and performing system.

Methods: An on-line modelling approach based on the time-series theory was put forward as a method to predict attitudes of tractor within 1-2 seconds. An AHRS (Attitude and heading reference system, Mti-300) was installed on the tractor to obtain attitudes in real time. The steps to build a time-series model mainly include: data processing, model identification as well as parameters estimation. By taking the difference to the non-stationary data acquired in the field, a set of stationary time-series was created for modelling. The AIC (Akaike information criterion) method was adopted to determine the model order, and the RLS (Recursive least square) algorithm was applied in model parameter-estimated on­ line. The estimated model is considered adequate to make prediction if residuals of the model are free from autocorrelation. The performance of the model can be validated by making the comparison between predicting values and sensor measurements with RMSE (Root mean square error).

Results: AR(10) model was found to follow the dynamic trend of tractor roll angle (10Hz) better, thus a 10-step (1s) prediction was conducted utilizing Matlab, and then continuing with the same loop iteration, a set of 30s prediction was finished with the RMSE less than 2°.

Discussions: The accuracy of the prediction satisfies most of attitude adjustment of tractors and the algorithm can be used in more agricultural situations.

Conclusion: The algorithm was tested and simulated merely with the roll angle data, and the prediction of three-axis attitude needs further study.