UCL Discovery
UCL home » Library Services » Electronic resources » UCL Discovery

Combat aircraft effectiveness prediction by artificial neural networks

Yann, Chee-Wha; (2003) Combat aircraft effectiveness prediction by artificial neural networks. Doctoral thesis (Ph.D), UCL (University College London). Green open access

[thumbnail of Combat aircraft effectiveness prediction by artificial neural networks.pdf] Text
Combat aircraft effectiveness prediction by artificial neural networks.pdf

Download (32MB)


This dissertation investigates the feasibility of applying Artificial Neural Networks (ANN) to problems of weapon system evaluation, and presents a Radial-Basis Function (RBF) Network as an ideal neural-net architecture to approximate the value function of a complex system such as combat aircraft, an approach perceived different from traditional MCDM problems for its intrinsic characteristics of weapon system engineering. The advantage of RBI against the traditional Multi-Layer Perceptron (MLP) is the relatively little amount of sample data required to train the network, which performs a more consistent and coherent prediction. A case study involving the selection of fighter aircraft is discussed with reference to the construction and validation of neural nets. The experimental results indicate that RBF yields a robust and simple-architecture solution, while the disadvantage is its heavy reliance on the data of a holistic score that can be treated as the output of neural nets employed to predict aircraft's effectiveness. The need for defence investment decisions of cardinal values rather than ordinal ranking motivates this study. MCDM methodologies including AMP, UTA, and PROMFETHEE II are compared with the ANN method. Correlations are examined before the MCDM analysis is performed, and a two-dimensional biplot plane is established to elucidate the information pertaining to alternatives and criteria. Among the compared MCDM methods, the logic of UTA coincides with that of RBF approach proposed in this dissertation, while UTA employs linear programming and RBF uses Gaussian transformation with regularisation theory The results show that RBF can accurately predict the numeric value of alternatives to interpret their operational effectiveness, and prevent the ill-condition when applying MLP. As to the reliability of method, the investigations of RBF satisfy both correspondence and coherence criteria, which demonstrates it as an economic and reliable tool for both alternative selection and concept design.

Type: Thesis (Doctoral)
Qualification: Ph.D
Title: Combat aircraft effectiveness prediction by artificial neural networks
Open access status: An open access version is available from UCL Discovery
Language: English
Additional information: Thesis digitised by ProQuest
Keywords: Social sciences; Applied sciences; Weapon systems
URI: https://discovery.ucl.ac.uk/id/eprint/10101425
Downloads since deposit
Download activity - last month
Download activity - last 12 months
Downloads by country - last 12 months

Archive Staff Only

View Item View Item