Active Disturbance Rejection, Dynamic Load Forecasting, and Noise-Aware Estimation for Electric Tractor Power Take-Off Drives

High-power electric tractor power take-off (PTO) drives face severe dynamic torque disturbances from heavy soil interaction, periodic tool cutting, and abrupt obstacle impacts, posing significant challenges to precision rotational speed regulation and implement load monitoring. We propose an integrated dual-channel framework that decouples high-bandwidth motor-side speed regulation from output-side load estimation and forecasting. Electromechanical drive parameters are anchored by 1739 s of field CAN telemetry from a high-power electric tractor, driving a flexible multi-rotor simulation testbed with torsional compliance. Motor-side speed stabilization is achieved via an extended state observer-based active disturbance rejection controller (ESO–ADRC), while implement-side load monitoring incorporates timestamp-aware continuous-discrete Kalman filtering, a physical multiscale jump detector, and causal autoregressive forecasting under realistic CAN communication constraints. In benchmark evaluations across 10% load steps, continuous broadband disturbances, and rotary tillage cutting spectra, ESO–ADRC restricts maximum speed deviations to 0.166% (maintaining the speed strictly within the ±1% regulation envelope throughout the transient, compared with a 1.107% drop and 0.454 s recovery time for conventional PI) and maintains steady tracking error within 0.282%. Across held-out operational trajectories under 10 ms CAN transmission with 5% packet loss and quantization, the timestamp-aware estimator reduces 50 ms load root-mean-square error (RMSE) by 48.3%, 22.6%, and 16.2% relative to standard moving-average filtering. Under stochastic packet jitter and reordering, causal torque-balance estimation reduces full-trace RMSE from 49.40 to 37.17 N m. Furthermore, a physical multiscale jump detector integrating 137 causal momentum and variability features elevates the transient shock detection F1 score to 0.853 nominally and to 0.772 under jitter/reordering (with median detection latencies of 22–30 ms, compared with <0.020 for a random baseline). These results demonstrate that decoupling motor disturbance rejection from output load estimation ensures robust rotational stability while providing reliable predictive telemetry for electrified agricultural powertrains.