Solutions of non-linear partial differential equations of Burgers Huxley type are obtained using a feed-forward artificial neural network technique. Solution process requires minimization of an error function which is constructed by making use of the differential equation and the associated initial and boundary conditions. The minimization is carried out using the Quasi Newton algorithm employed through Matlab optimization toolbox. The solutions that are obtained using the technique are analytic in nature and have excellent generalization properties. Results are compared with the existing solutions to validate the utility and effectiveness of proposed procedure. Effect of variation in number of training points on solution accuracy is also studied through calculation of numerical rate of convergence. Solutions obtained using the neural network technique are analytic in nature and can be obtained directly without linearising non-linear partial differential equations.

使用前馈人工神经网络技术获得Burgers Huxley型非线性偏微分方程的解。求解过程要求最小化误差函数，该误差函数是通过利用微分方程以及相关的初始条件和边界条件构造的。最小化使用通过Matlab优化工具箱使用的Quasi Newton算法进行。使用该技术获得的解决方案本质上是分析性的，并且具有出色的泛化特性。将结果与现有解决方案进行比较，以验证所提出程序的实用性和有效性。还通过计算收敛速度的数值，研究了训练点数量变化对求解精度的影响。