Research on the Course Structure of Mechanical Engineering Using Neural Network

Mechanical engineering course system was taken as the research object, based on QS World University rankings relevant parameters, in order to cultivate innovative talents, used modern mathematical analysis method, established a three layer BP neural network analysis model, errors have been lumped in -0.6~+0.5. Based on this model, some Chinese universities, which were not listed in the QS rankings system, were quantitatively calculated. The results of comparative analysis show that the structure of the mechanical course system in universities is an important influence factor to the rankings of the QS system. This study is useful for the mechanical engineering disciplines to enter the international first-class disciplines and cultivate innovative talents. Introduction The construction of international first-class discipline is the foundation of building a world class university. To adhere to the first-class as the goal, be based on mechanical engineering disciplines, raising the course structure system, training innovative talents [1, 2]. This paper puts the university ranking which is selected by global parameters of engineering technology in international recognized research institutions of higher education Quacquarelli Symonds (QS). Through the design of artificial neural network model, numerous complex indicators quantified. To analyze mechanical engineering discipline for the graduate students compared with a university of China and the world first-class mechanical engineering course construction, with the demand of creative talents training. Put forward more accurate and more available as well as more scientific and objective methods on the construction of mechanical engineer disciplines from the view of course structure to develop innovative talents for Chinese universities. Methods Learning Algorithm of BP Neural Network This study uses a three layers BP neural network model with one hidden layer. BP learning algorithm is the core of BP network, which affects the correctness and accuracy of model prediction [3]. Input network vector 1 2 ( , , ) ( 1,2, ) k k k k T n X x x x k m     , k X excepted output is 1 2 ( , , ) k k k k T q Y y y y   , net input of the middle hidden layer is 1 2 ( , , ) k k k k T p S s s s   , output vector is 1 2 ( , , ) k k k k T p B b b b   , output layer net input is 1 2 ( , , ) k k k k T q L l l l   , actual output vector is 1 2 ( , , ) k k k k T q C c c c   . The connection weights between the hidden layer and the input layer and the output layer are recorded { }( 1,2, , 1,2 ) ij W w i n j p      , { }( 1,2, , 1,2 ) ij V v j p j q      .The threshold of each neuron in the hidden layer and the output layer is { }( 1,2, ) j j p      , { }( 1, 2, ) t t q      .Its detail process is given follows [4]: a) Initialization. b) Select a pairs of random samples ( , ) k k X Y . c) Calculate the input of the input layer. d) Calculate net input and output vector of each neuron in hidden layer. e) Calculate the net input and the actual