基于BP神经网络的教学质量评价

基于MATLAB设计神经网络评价流程，以3层BP神经网络进行教学质量评价，其中隐含层的传递函数为tansig，输出层的传递函数为purelin．为保证网络的泛化能力，在训练样本的选择上，选择比较有代表性的数据，打乱顺序，对数据进行归一化预处理；训练函数采用Levenberg—Marquardt算法，选择学习速率为0．01，网络训练误差为le^-5，训练次数1000次，步长选择为10．成功训练神经网络后，对给出的4组数据进行了预测，画出预测拟合及误差分析图形．结果表明：在设计BP神经网络模型的基础上，采用合理的样本数据，神经网络就会通过训练样本，学习到样本的有关环境本身的内在规律性，对输入给出正确的反应，以实现教师的教学质量评价．

Evaluation of Teaching Quality of Based on BP Neural Network

The method of evaluation of teaching quality based on the BP neural network aims at many advantages BP neural network has, including self-studying, adaptation, and associative memory, etc. The evaluation is carried on through the MATLAB neural network evaluation design process formed by 3 layer BP neural network. Among them, one hidden layer transfers function tansig, the other layer transfers function purelin. To ensure the network＇s generalization ability, we choose more representative selection of data, disrupt the order of the data and handle the data normatively. The training function adopts Levenberg-Marquardt algorithm which parameters learning rate is 0.01, the network training error is le-5 with 1 000 training times and 10 steps. After the neural network is trained successfully, four sets of data are forecasted, prediction fitting and error analysis are drawn into graphics. Based on the design of the 13P neural network model, the network, with reasonable sample of data and the samples＂ help, would learn the intrinsic regularity of the samples＂ environment selves and reacts on the input response correctly. Thus, the evaluation of teaching quality works at last.