Tilt Angle On-Line Prognosis by Using Improved Sparse LSSVR and Dynamic Sliding Window

Abstract

On-line tilt angle detection and prognosis are the important prerequisites of real-time movement controlling. This paper develops a time-effective tilt angle prognosis methodology and its evaluation prototype. First, a complementary filter is imported to extend both the superiorities sufficiently of accelerometer and gyroscope, accurate tilt angle detection then can be achieved in a wide range of frequency. Second, an improved sparse least-squares support vector regression is proposed for tilt angle prognosis, its configurable singularity spectrum technology keeps away from the high-order matrix operation through eliminating nonprincipal components among raw vectors. Third, novel composite judgment criteria and its derived dynamic sliding window mechanism are proposed to pursue scientific balance of the prediction precision and computational cost. Fourth, double chains quantum genetic algorithm is introduced as the optimization tool to search optimal parameters of the above-mentioned models. Finally, the prototype setup and the corresponding experimental results are demonstrated and discussed in detail, which promise that our proposed methodology could be potentially applied to the actual motion control systems.