Scaled Conjugate Gradiant Neural Intelligence

Neural computing for passive target tracking

1. Scaled Conjugate Gradient Neural Intelligence For Motion Parameters Prediction Of Markov Chain Underwater Maneuvering Target 1 Presented by Dr. Wasiq Ali Postdoctoral Fellow, College of Underwater Acoustic Engineering Harbin Engineering University, Harbin, China December 14, 2023

2. Outline 2 Background of Research Introduction Flow Chart Problem Statement System Model Continuous Turn Target Trajectory Mathematical Modeling & Evaluation Criterion Proposed Neural Intelligence Simulation Results Conclusion References Questions & Answers

3. Background of Research 3 Since world War I, the underwater state estimation for passive tracking become one of the core issue among the research community. Substantial attentions has been directed to real time state estimation of underwater targets, and long-term efforts have been made in both theory and experiment. In the literature, passive tracking problems are usually formulated using state-space models. State-space models used with an observation equation that relates the state vector (i.e., target DOA’s and possibly motion states) to the acoustic hydrophone outputs, and a state equation that constrains the dynamic nature of the state vector.

4. Cont.. 4 The performance of the tracking algorithms using state-spaces relies heavily on how accurate the model represent the observed natural phenomena. Literature review shows that in most cases, it is important to use nonlinear and non-Gaussian state-space models despite of their computational complexity. In the last two decades, a lot of Bayesian filtering approaches have been proposed from the research community for solving nonlinear state estimation problems.

5. Introduction 5 The problem of target tracking using passive measurements can be formulated as bearings-only tracking (BOT), which is a technology of determining the state of a target solely through measurements obtained from the signals originated from the target. In recent years, BOT has gotten a lot of interest because of its significance and wide uses in a variety of practical applications like aircraft surveillance, underwater SONAR tracking, navigation, Radar and guidance, etc., in which a real-time state of object is found by using only the bearings passive measurements. In BOT the major goal is to efficiently estimate the dynamics of the single or multi moving targets by means of noise-corrupted passive measurements received on single or multi observation platforms.

6. Cont.. 6 Due to the inherent nonlinearity between the passive measurements and the state of the target, the design of a reliable and robust algorithm for BOT is still a challenge. B earings-only passive target tracking is a typical nonlinear filtering problem that is usually related to target motion analysis (TMA).

7. Flow Chart of Research 7

8. Problem Statement 8

9. Motion Parameters Prediction System Model 9 A two-dimensional Cartesian coordinates system model for accurate motion prediction of a moving target is developed. The aim of this study is to accurately track the parameters of an underwater moving object with different prediction algorithms and compare their performances. The measurements are only the bearings (angles) of the target with respect to orientation of the angular observers. In this proposed passive motion prediction model, target movements are assumed in continuous turning trajectory.

10. Continuous Turning Target Trajectory 10

11. Mathematical Modeling & Evaluation Criterion 11

12. Cont.. 12 The assessment criterion for deep neural intelligence methodology involves formulating a minimal mean square error (MSE) among the real and predicted motion parameters of turning object at each time instant †. This paper presents an analysis of the reliability and robustness exhibited by the neural intelligence technique. As a consequence, the MSE formula for SCGNI, IMMEKF, and IMMUKF is derived separately for each individual Monte Carlo simulation as: In the mentioned MSE function, actual motion parameters of the turning object are denoted as . On the other hand, approximated motion parameters of the object, obtained through the utilization of SCGNI and Bayesian filtering approaches, are expressed by . The total number of data points are denoted by P, with a value of 200 in simulations, whereas †=1 represents the starting data point. The computation of motion parameter errors for the turning trajectory occurs at each time step.

13. Proposed Neural Intelligence 13 A novel application of deep learning based neural intelligence is proposed for efficient real time state estimation of underwater maneuvering object. The intelligent strategy utilizes the capabilities of Scaled Conjugate Gradient Neural Intelligence (SCGNI) to estimate the dynamics of underwater object that adhere to discrete-time Markov chain. C ontinuous coordinated turning trajectory of the underwater passive maneuvering object is modeled for analyzing the performance of neural computing paradigm. Real time position, velocity and turn rate of turning maneuvering object are computed by neural computing.

14. Flow Chart of SCGNI 14

15. Architecture of SCGNI 15 Scaled Conjugate Gradient Neural Intelligence is applied for predicting future values by efficiently incorporating previous data for efficient motion parameters prediction of maneuvering target.

16. Simulation Results 16 This section of the study provides a concise explanation of the simulation findings for the proposed smart computing based on SCGNI. The results include real-time motion parameters estimates, location error, velocity divergence, rotation estimates, error histogram, and regression study.  Five distinct scenarios in simulations are performed and the performance parameter is measured Gaussian noise standard deviation. This parameter is systematically tuned within the range of 0.01 to 1 radian.

17. State and Turn Rate Estimates for Measurement Noise Standard Deviation λ = 0.05 radian 17

18. Position and Velocity Error for Measurement Noise Standard Deviation λ = 0.05 radian 18

19. Error Histogram and Regression Analysis of SCGNI 19

20. Mean Square Error Graph for Measurement Noise Standard Deviation λ = 0.05 radian 20

21. Conclusion 21 In order to accurately approximate the motion parameters of dynamic aquatic Markov chain object, the ability and effectiveness of the SCGNI paradigm is thoroughly investigated. For a passive turning target, state estimations, position errors, velocity errors, rotation errors, deviation histograms, and regression analyses of SCGNI are calculated and compared to traditional nonlinear and multiple models Kalman filter variants like IMMEKF and IMMUKF. As an evaluation criterion, measurement noise standard deviation is used, and its magnitudes are changed inside simulations in order to investigate the pattern of existing and proposed methodologies. The simulation results show that effectiveness of SCGNI soft computing is significantly superior than multi model Kalman filters for predicting the motion parameters of an undersea navigational vehicle.

22. References 22 Ali, Wasiq , Yaan Li, Muhammad Asif Zahoor Raja, Wasim Ullah Khan, and Yigang He. "State Estimation of an Underwater Markov Chain Maneuvering Target Using Intelligent Computing." Entropy 23, no. 9 (2021): 1124. Ali, Wasiq , Wasim Ullah Khan, Muhammad Asif Zahoor Raja, Yigang He, and Yaan Li. "Design of Nonlinear Autoregressive Exogenous Model Based Intelligence Computing for Efficient State Estimation of Underwater Passive Target." Entropy 23, no. 5 (2021): 550. Ali, Wasiq, Yaan Li, Nauman Ahmed, Jun Su, and Muhammad Asif Zahoor Raja. Performance Analysis of Bayesian Filtering and Smoothing Algorithms for Underwater Passive Target Tracking. Journal of Control, Automation and Electrical Systems. 31, no. 6 (2020): 1400-1411.

23. Questions & Answers 23 In the end I am so much thankful to all team members for sparing your precious time for this talk. I will feel more pleasure if you people give me your valuable suggestion and if you have any queries you can ask.