Projects:2017s1-107 Protection of a Convoy of Ships Under Attack

Supervisors

Prof. Cheng-Chew Lim

Students

Chris Arney, Daniel Milne

Client

DST Group

Advisers

Lloyd Damp, Dr. Jijoong Kim, Dr. Arvind Rajagopalan

Introduction

Australian maritime defence vessels often lead an isolated life and must be able to protect themselves in addition to any vessels they may be escorting in a convoy, all with minimal external support.

These defence vessels often employ missiles as interceptors in order to provide protection. When an asset is under attack, there is an element of uncertainty as to what entity the inbound threat is pursuing, the maneuverability of the inbound threat and the number of threats comprise the attack. Therefore, the defence vessel must allocate the limited interceptors residing on board in an effective style to ensure that it is not hit.

This project seeks to evaluate the performance of a neural network within a simulated single entity attack. The measure of effectiveness will be if the vessel can survive this attack.

Objectives

The objective of this project is to design and develop a neural network that will reside within a simulation software package.

The neural network will interpolate and extrapolate separation distances for a particular set of missile launch conditions, these missile launch conditions refer to the angle and time parameters, to predict the separation distance between the ship and when the interceptor engages an inbound threat.

This prediction of separation distance will have a minimal steady state error when compared to modelled data for the same missile launch parameters.

Neural Network

A simple neural network has a structure that contains an input layer that connects to the real world and receives information directly from a host of sources from electronic sensors to data files. There is also an output layer that can connect directly to a myriad of devices like a secondary computer, control system or out put a data file.

Between the input and output layers there is a hidden layer, a more complex neural network may have many more. It is within this hidden layer that the nodes make decisions based on the input signals that have been received. The neural network will then decide the best course of action before relaying this onto a second hidden layer or to the output.

A neural network will allow the use of continues spaces or to be put more precisely real time.

Results

The objective for this project was to develop a neural network able to interpolate and extrapolate separation distances for a particular set of missile launch conditions, to predict the separation distance between the ship and when the interceptor engages an inbound threat. With the error between the prediction and modelled separation distances being minimal.

On completion of development of the ANN and after conducting rigorous testing, the predicted separation distances for launch parameters that had not been used to train the ANN with were plotted against the launch parameters that had been used to train with. When viewing the results, the predicted errors fall well within tolerance and the conclusion that the ANN has successfully been able to learn what the separation distance will be for any given launch parameter.

References

[1] C. Leboucher, H. Shin, S. Ménec, A. Tsourdos and A. Kotenkoff, "Optimal Weapon Target Assignment Based on an Geometric Approach*", IFAC Proceedings Volumes, vol. 46, no. 19, pp. 341-346, 2013.

[2] S. Le Ménec, K. Markham, A. Tsourdos, H. Shin and H. Piet-Lahanier, "Cooperative Allocation and Guidance for Air Defence Application", IFAC Proceedings Volumes, vol. 44, no. 1, pp. 3897-3902, 2011.

[3] H. Shin, C. Leboucher and A. Tsourdos, "Resource allocation with cooperative path planning for multiple UAVs", Proceedings of 2012 UKACC International Conference on Control, 2012.

[4] A. Benaskeur, F. Kabanza, and E. Beaudry, "CORALS: a real-time planner for anti-air defense operations." ACM Transactions on Intelligent Systems and Technology (TIST) 1.2 (2010): 13.

[5] C. Leboucher, H. Shin, P. Siarry, R. Chelouah, S. Le and A. Tsourdos, "A Two-Step Optimisation Method for Dynamic Weapon Target Assignment Problem", Recent Advances on Meta-Heuristics and Their Application to Real Scenarios, 2013.

[6] V. Mnih, K. Kavukcuoglu, D. Silver, A. Graves, I. Antonoglou, D. Wierstra and M. Riedmiller, "Playing atari with deep reinforcement learning." arXiv preprint arXiv:1312.5602 (2013).

[7] I. Bello, H. Pham, Q. Le, M. Norouzi, S. Bengio, "Neural Combinatorial Optimization with Reinforcement Learning." arXiv preprint arXiv:1611.09940 (2016).

[8] S. Maind, and P. Wankar, "Research paper on basic of Artificial Neural Network." International Journal on Recent and Innovation Trends in Computing and Communication 2.1 (2014): 96-100.

[9] E. Akbaş, "Deep Reinforcement Learning." (2016).

[10] M. Tambet, "Guest post: Demystifying deep reinforcement learning-nervana." (2016).

[11] P. Finnman and M. Winberg, "Deep reinforcement learning compared with Q-table learning applied to backgammon", Undergraduate, Kungliga Tekniska Hogskolan Royal Institute of Technology, 2016.

[12] M. Littman, "Reinforcement learning improves behaviour from evaluative feedback", Nature, vol. 521, no. 7553, pp. 445-451, 2015.

[13] Royal Australian Navy, "Guided Missile Frigate (FFG) | Royal Australian Navy", Navy.gov.au, 2017. [Online]. Available: http://www.navy.gov.au/fleet/ships-boats-craft/ffg. [Accessed: 28 Apr. 2017].

[14] Royal Australian Navy, "Standard Missile | Royal Australian Navy", Navy.gov.au, 2017. [Online]. Available: http://www.navy.gov.au/weapon/standard-missile. [Accessed: 28 Apr. 2017].

[15] Raytheon Australia Communications, "Raytheon Australia: Standard Missile-2", Raytheon.com.au, 2017. [Online]. Available: http://www.raytheon.com.au/capabilities/products/sm-2. [Accessed: 28 Apr. 2017].

[16] H. Mouton, H. Le Roux, and J. Roodt, "Applying reinforcement learning to the weapon assignment problem in air defence." Scientia Militaria: South African Journal of Military Studies 39.2 (2011): 99-116.

[17] G. Ferrari-Trecate, and M. Muselli, "A new learning method for piecewise linear regression." Artificial Neural Networks—ICANN 2002 (2002): 135-135.

[18] N. Conaway, and K. Kurtz. "Solving nonlinearly separable classifications in a single-layer neural network." Neural computation vol. 29, no. 3, pp. 861-866, 2017.