基于时序建模的教学质量评估系统的神经网络辨识

采用时序分析对教学质量评估体系进行系统分析和建模,引入Box-Cox进行数据的非平稳性预处理,选用L-MBP网络进行系统辨识。通过比较网络对校验样本的预测效果,对隐层节点数和各层间的传递函数进行优化,同时采用重复训练法来提高网络的稳定性和预测精度。算例表明这种辨识方法能对教学质量进行更准确的评估和预测。

Neural Network Evaluation System for Teaching Quality Based on Time Series Modeling

Time series analysis is used to study and modeling the teaching quality evaluation system.The complicated procedures of non-stationary data pre-processing is achieved by Box-Cox transformation.Levenberg-Marquardt BP Artificial Neural Network is introduced to identify the system.The effect of the number of hidden layer neurons and transfer function on the precision of prediction to improve the convergence and the training speed of the network are studied.The structure and parameters of ANN are optimized by comparing the prediction results of samples.And the stability and precision in prediction of the neural network are improved by way of repetitive training.Simulation results show that the proposed method can give more accurate prediction of compared with traditional ANN.