一种具有全局最优的神经网络BP算法

建立了描述上半周加热、下半周绝热不均匀热流边界条件下的水平管内受迫层流与自然对流叠加的混合对流换热的数学模型。该模型考虑了管壁导热和流体的变物性，研究了不同流体（水和乙二醇水溶液）、不同热流方向对对流换热的影响。同时也进行了上半周加热、下半周绝热边界条件下的水平管内混合对流换热的实验研究。理论和实验研究的结果都表明了，重力场对水平管内流体层流对流换热的影响，为在地面重力场中进行模拟太空微重力环境中的空间辐射器的传热实验研究提供了必要的理论和实验依据。

Back propagation algorithm of neural network with global optimization

This paper presents a back propagation algorithm of neural network with global optimization to avoid stucking in local minima. It forms a new hybrid neural network which combines adaptive linear element with BP neural network. The network weight is adjusted by back propagation algorithm and the adaptive linear element input is from the sample number to the power of zero to the sample number to the power of the number minus one. The network weights obtained by this algorithm are global optimal and the mathematics proof is presented. Practical examples indicate that the method works well in avoiding stucking in local minima. The algorithm is simple and practical and it does not need a long time to get global optimization for learning. This may have practical significance in application of the back propagation algorithm.