Multi-Source Information-based State Fusion Estimation for Distributed Electric Drive Vehicles

Abstract

To enhance the accuracy and reliability of state estimation for distributed electric drive vehicles in GNSS-denied and complex terrain environments, this paper innovatively proposes a multi-source information-based state fusion estimation method. The core of the method is a dichotomy based hybrid model data state fusion framework designed to address dual uncertainties in both GNSS signal and terrain. First, a vehicle dynamics model based Cubature Kalman Filter (CKF) is designed. Second, to tackle model mismatch in complex terrain, a vehicle state estimation method based on a Convolutional Neural Network (CNN) and Long Short Term Memory (LSTM) network is proposed. In this approach, the CNN is responsible for terrain type recognition, while the LSTM serves as the backbone network to extract nonlinear correlated features from eight types of time series data and compute the model output. Then, to improve the reliability of the estimation algorithm and reduce errors from any single method, a physics-informed trend reference is introduced, which operates independently of absolute ground truth. Using dichotomy logic, the weights of the CKF-based estimate and the CNN-LSTM-based estimator are dynamically adjusted in real time by comparing their respective deviations from this reference. To validate the effectiveness of the proposed method, experiments under four operating conditions were carried out on an eight wheel distributed electric drive vehicle. The results demonstrate that the proposed method effectively improves the estimation accuracy of longitudinal velocity and sideslip angle, as well as the environmental adaptability of the vehicle.