Design of Nonlinear Autoregressive Exogenous Model Based Intelligence Computing for Efficient State Estimation of...

Area of Science:

• Underwater acoustics
• Artificial intelligence
• Signal processing

Background:

• Real-time motion parameter extraction for underwater passive targets typically relies on nonlinear filtering techniques.
• Existing methods often associate nonlinear passive measurements with linear target kinetics within a state-space framework.
• Improving tracking accuracy and minimizing position error for dynamic passive objects remains a challenge.

Purpose of the Study:

• To present an intelligent computing paradigm utilizing a nonlinear autoregressive exogenous (NARX) feedback neural network for accurate state estimation of underwater passive targets.
• To leverage deep learning strengths for enhanced feature estimation and reduced position error in dynamic passive object tracking.
• To evaluate the performance of the NARX-based supervised learning approach in bearings-only tracking scenarios.

Main Methods:

• Development of a deep learning model based on a nonlinear autoregressive exogenous (NARX) feedback neural network.
• Application of NARX-based supervised learning for estimating the real-time state of a passive moving object following a semi-curved path.
• Performance evaluation through Monte Carlo simulations under six different standard deviation scenarios of white Gaussian measurement noise.

Main Results:

• The proposed NARX feedback neural network scheme effectively estimates the real-time state of underwater passive targets.
• Root mean square error (RMSE) in rectangular coordinates was computed to quantify the accuracy of the estimated versus real positions.
• The NARX model demonstrated superior state estimation capabilities compared to conventional nonlinear filters like the spherical radial cubature Kalman filter and unscented Kalman filter.

Conclusions:

• Intelligent computing using NARX-based deep learning offers a capable alternative for underwater passive target state estimation.
• The NARX model shows significant potential in improving tracking accuracy and minimizing position errors in challenging underwater environments.
• The study validates the effectiveness of the proposed NARX feedback neural network over traditional filtering algorithms for bearings-only tracking.