Vehicle state fault-tolerant estimation method based on long and short term memory neural network

Abstract

The invention discloses a vehicle state fault-tolerant estimation method based on a long-short term memory neural network. The method comprises the steps of building a one-step prediction model at any moment based on a vehicle two-degree-of-freedom dynamics model and a vehicle longitudinal kinematics model; establishing an observation model based on the measurement quantity of a left rear wheel speed sensor of the vehicle, the measurement quantity of a right rear wheel speed sensor of the vehicle and the fault-tolerant absolute vehicle speed of the whole vehicle at any moment; and combining the one-step prediction model and the observation model with the Kalman filtering theory, and estimating the longitudinal speed, the transverse speed and the yaw velocity of the vehicle at any moment. According to the method, the vehicle state is estimated by fusing the vehicle-mounted sensor and the GPS signal through the multi-sensor fusion technology, and the reconstructed absolute speed signal of the whole vehicle is solved by using the LSTM-RNN when the GPS is missing, so that the problem of estimation system fault caused by the missing of the GPS is solved, the vehicle state estimation precision is ensured, and the method has relatively strong fault tolerance.

Description

technical field [0001] The invention belongs to the technical field of system state estimation, and in particular relates to a vehicle state fault-tolerant estimation method based on a long-short-term memory neural network. Background technique [0002] In high-level autonomous vehicles, the vehicle's real-time cognition of its own state is crucial to vehicle energy efficiency and safety. Key vehicle state variables such as vehicle longitudinal velocity, vehicle lateral velocity, and vehicle yaw rate are the basis for the normal operation of vehicle safety systems such as ESP, ABS, and lateral stability control, and are an important branch of vehicle intelligence research. [0003] Due to factors such as measurement noise and inherent bias of a single sensor, it is difficult for a vehicle to accurately measure its own state through a single sensor. Therefore, multi-sensor fusion technology is developed to make full use of the advantages of multiple sensors and improve the co...

AI Technical Summary

Problems solved by technology

[0004] Aiming at the deficiencies of the above-mentioned prior art, the object of the present invention is to provide a vehicle state fault-tolerant estimation method based on long-short-term memory neural network to solve the problem of low confidence in vehicle state estimation caused by the lack of GPS signals in the prior art

Images