摘要
二维数据表广泛应用于科学与工程计算,当出现数据缺失时,由于数据间存在的非线性关系,采用传统方法在两列数据中插值将产生较大的误差。结合甘油水溶液粘度的预测,提出使用人工神经网络的插值方法,对神经网络模型、参数以及训练样本集的选择进行了实验与优化。实验表明,利用BP神经网络预测25°C甘油水溶液粘度时,网络收敛速度快,预测精度高,优于传统的最邻近插值和双线性插值,能够满足一般科学与工程计算的需要。提出的插值预测方法对类似条件的二维数据表具有普遍指导意义。
Two - dimensional data tables are widely applied in science and engineering, in which some data are absent. Since non - linear relationship exists in the data, the traditional interpolation methods will produce greater error when predicting value between two row data. In this paper, with the viscosity of aqueous glycerol solution predicted, an interpolation method based on artificial neural network is presented. After lots of experiments, the neural network model, parameters and training samples set are selected and optimized. The experiment shows that BP neural network has higher convergence speed and is better to predict 25℃ viscosity of aqueous glycerol solution than traditional methods, such as nearest neighbor interpolation and bilinear interpolation. It is able to meet the need of general science and engineering calculation. The interpolation method based BP neural network presented in this paper.can conduct interpolation and prediction of two - dimensional data tables under similar conditions with universal significance.
出处
《计算机仿真》
CSCD
北大核心
2009年第2期193-195,268,共4页
Computer Simulation
基金
国家自然科学基金项目资助(30770591)
关键词
人工神经网络
非线性插值
预测
Artificial neural network
Non - linear interpolation
Prediction
