| Summary: | 博士 === 國防大學中正理工學院 === 國防科學研究所 === 95 === This dissertation employs the Continuous-time analogue Hopfield Neural Network (CHNN) to compute the temperature distribution in various linear and nonlinear heat conductions problems and presents an efficient technique in which one- and two-dimensional inverse heat conduction problems (IHCP) are analyzed using a back propagation neural network (BPNN) and a Kalman Filter-enhanced Back Propagation Neural Network (KF-B2PNN) to identify the unknown boundary conditions, and applies binary and multi-class Support Vector Machine (SVM) schemes to estimate the unknown point (or points) of application of the heat flux. The BPNN is assumed to have a three-layered structure and is trained using eight different training algorithms. This study overcomes the weak generalization capacity of BPNN when applied to the solution of non-linear function approximations by employing the Bayesian regularization algorithm. The training data for the neural network are prepared by solving forward heat conduction problems using the CHNN method. The feasibility of the proposed method is examined in a series of numerical simulations, and furthermore to verify military facilities. The performance of the KF-B2PNN scheme is shown to be better than that of a stand-alone Back Propagation Neural Network trained using the Levenberg-Marquardt algorithm. Compared with a conventional Artificial Neural Network (ANN), the results show that the current SVM-based schemes achieve a superior classification performance; particularly as the number of training samples is reduced. Furthermore, the SVM schemes have an improved computational efficiency and avoid the requirement to pre-process the heat flux signal in order to extract its features. The results show that the proposed neural network analysis method successfully solves forward heat conduction problems and is capable of predicting the unknown parameters and the unknown point (or points) in inverse problems with acceptable errors.
|
|---|
