Kalman Filter and Neural Network Algorithms Based Observer Estimation Benchmark

This study examines three different methods for predicting vehicle speed: Extended Kalman Filter (EKF), Multi-Layer Perceptron (MLP), and Long Short-Term Memory (LSTM). EKF models the movement state of the vehicle by linearizing non-linear systems. This method is particularly effective in reducing prediction errors and increasing model accuracy. MLP, trained with supervised learning, is a neural network architecture that produces an output from various inputs. An MLP model, trained with data collected from the vehicle, has been evaluated comparatively with other methods in this study. LSTM is successful in processing time series data and can model long-term dependencies and transient effects when working with time-dependent data like vehicle dynamics. This study analytically compares the linearization capability of EKF and the ability of MLP and LSTM to model complex patterns and temporal dependencies.