Energy management optimization for de-icing robot based on adaptive Kalman filter algorithm

Abstract

Aiming at the critical issues of severe power consumption fluctuations, sensor noise interference, and inaccurate range prediction faced by de-icing robots in the complex operational environments of high-voltage transmission lines, this paper proposes a physics-informed and deeply coupled energy management optimization strategy. First, the electromechanical system architecture of the robot is systematically constructed, and a refined dynamic energy consumption model is established based on mechanical principles to reveal the nonlinear coupling mechanism among slope, ice thickness, and operational speed. Key physical parameters were identified using empirical data from a laboratory prototype in a cold chamber (−15°C), ensuring the mechanistic transparency and physical coherence of the energy sensing process. Second, an Adaptive Kalman Filter (AKF) algorithm with step-wise quasi-linearization features is employed, utilizing the Sage-Husa operator for real-time online correction of system noise statistical characteristics. Theoretical analysis demonstrates the global observability and convergence stability of the estimator in non-stationary environments. Simulation and semi-physical validation results show that under a standardized evaluation protocol involving intense electromagnetic interference and mutation loads, the Root Mean Square Error (RMSE) of the state-of-charge (SOC) estimation for the proposed AKF is as low as 0.005, significantly outperforming standard KF, LSTM, and GPR models in terms of convergence and robustness. Finally, a hierarchical energy management strategy centered on the solution for optimal economic speed is designed. Task-level quantitative evaluation reveals that this strategy increases the effective range of the robot by 28.5% and improves the task success rate from 72% to 95%, effectively mitigating the risk of robots becoming stranded on lines due to power exhaustion. The single-step average computation time of the algorithm on an embedded controller is only 0.85 ms, fully demonstrating its practical engineering value for ensuring power grid maintenance safety and improving operational efficiency.