Abstract
Novel generalized predictive control (GPC) is proposed to deal with the problems of cumulative deviation of the predicted value, long control process and large noise that often encountered in conventional GPC. First, the objective function of GPC is improved based on integral time square error (ITSE) criterion, and the output expression under soft constraint is deduced. Then, the predicted value is corrected by adopting the integral weighting method, the convergence of which is demonstrated. Finally, both the simulations and hardware experiments are conducted in a practical photoelectric tracking system. It turns out that the superiority of the improved GPC in control precision, dynamic response and robustness is verified.
Introduction
Photoelectric tracking systems are the critical "eyes" of Unmanned Aerial Vehicles (UAVs), tasked with maintaining a stable lock on targets despite the vehicle's inherent vibrations and rapid movements. Engineers frequently employ Generalized Predictive Control (GPC)—a strategy that utilizes mathematical models to predict future behavior and optimize control actions. However, conventional GPC faces significant challenges in high-stakes tracking environments: cumulative prediction errors lead to drift, slow response times cause lag during fast target maneuvers, and high noise levels result in unstable imaging. This research addresses these "blind spots" by refining the GPC framework for real-world aerial tracking.
Methodology
The researchers introduced a two-pronged enhancement to the standard GPC framework to improve both speed and accuracy.
Improved Objective Function (ITSE Criterion)
In standard GPC, prediction errors are typically weighted equally across the horizon. This paper adopts the Integral Time Square Error (ITSE) criterion, introducing a weighting matrix where the importance of prediction errors increases over time. This approach forces the controller to prioritize long-term convergence, accelerating the system's response toward the set-point without inducing excessive overshoot.
Predictive Value Correction
To combat cumulative drift, an integral weighting method was implemented. Rather than relying solely on the current error, the system assigns weights to the integrals of all current and previous prediction errors. This creates an internal feedback loop for the prediction model itself, ensuring that predicted values converge more accurately with real-world outputs and dampening the effects of sensor noise.
The system was modeled around a DC torque motor, as illustrated in the block diagram above, which accounts for armature resistance ($R_a$) and torque coefficients ($K_T$).
Experimental Validation
The proposed algorithm was implemented on a hardware platform using a DSP+FPGA architecture (specifically the DSP28335) to control a 60V DC torque motor. The system operated at a 1ms sampling period to ensure real-time performance.
Validation involved testing the physical tracking system (pictured above) against simulated sinusoidal disturbances (e.g., 2° amplitude at 1 Hz) to mimic the rocking motion of a UAV.
Key Results and Discussion
The improved GPC outperformed conventional methods across all critical performance metrics:
- Response Speed: The rising time decreased from 31ms to 24ms, while the settling time improved from 42ms to 29ms.
- Tracking Precision: Under disturbance, stability accuracy improved from 0.054 mil to 0.034 mil.
- Noise Reduction: Hardware experiments demonstrated that rotational speed noise was halved, from ±0.4 °/s to ±0.2 °/s.
- Robustness: The system remained stable even when load inertia was increased by 50%, proving its ability to handle varying camera payloads.
Conclusion
By redesigning the objective function and incorporating predictive value correction, this study provides a more robust and responsive control solution for UAV photoelectric tracking. While the method introduces additional parameters that require manual tuning, the significant gains in precision and noise immunity mark a substantial improvement over standard GPC. Future research will explore the ITAE criterion and the integration of hard physical constraints to further refine system performance.
Summary updated