【作者】 曹晓哲；

【导师】 刘建忠； 程军；

【作者基本信息】 浙江大学 ， 工程热物理， 2010， 硕士

【摘要】 本文主要目的为提高水煤浆成浆性能和建立高精度的水煤浆配煤成浆性能预测模型,应用回归模型和神经网络模型等预测模型对水煤浆的制备与开发进行了一系列的基础性研究。首先,通过实验研究了煤种理化特性参数对煤成浆性的影响,考察了煤种的Mad、Aad和Oad等几种因素与煤种的成浆性之间的关系。由实验结果可以发现煤种的成浆性是由多种因素决定的，而且几种因素之间的关系又较为复杂,单独的用一种影响因素来分析一个煤种的成浆性是不科学的。由于各种单煤的性质不同,混合后的不同煤种在成浆过程中相互影响,相互制约,因此配煤成浆性不可能是各组分煤种特性的简单叠加,而是呈现出非常复杂的非线性特征,而日益兴起的神经网络技术正是解决配煤非线性问题的有效方法。通过分析煤种的各性质与成浆性能的相关性,选择了十种因素进行了回归分析的预测,来与神经网络模型进行对比。通过对十因子、九因子、五因子、四因子和三因子的线性和非线性分析,预测结果最好的是五因子的线性回归模型,预测结果的误差为1.69%。共进行了十因子、九因子、五因子、四因子和三因子的组合的煤种的成浆性能影响因素的神经网络预测分析。通过对比每种输入因子数的最佳模型参数和误差,发现其中五因子神经网络预测模型的结果最好,其预测结果的误差到了0.49%的水平,大大低于五因子线性回归模型,同时每种输入因子数的神经网络预测模型结果都比相对应的回归方程的结果要好。更多还原

【Abstract】 The main purpose of this paper is to improve the performance of coal-water slurry and establish high-precision blending CWS performance prediction models, applicate regression models and neural network models and other forecasting models to research on the coal-water slurry preparation and development.First of all, through the experiment effects of the physical and chemical characteristics on coal slurry parameters have been researched, while coals Mad, Aad, and Oad several factors was investigated in coal slurry parameters. From the experimental results we can find that coal-forming slurry is determined by a variety of factors, and also the relationship between several factors is more complex, and a separate analysis of the impact factors of a coal-water slurry-is not scientific.A variety of single coal are different, mixed into a slurry of different coals in the process of mutual influence and·constraints, blending into a slurry of coal characteristics of each component can not be a simple sum, but it shows a very complex non-linear characteristics, while the increasing emergence of neural network technology to solve nonlinear problems is an effective way.By analyzing correlation between various natures of coal and coal slurry ability, 10 kinds of factors are selected for regression analysis prediction which is compared with the neural network model. Through the ten factors, nine factors, five factors, four factors and three factor linear and nonlinear analysis, the five factors is the best linear regression model, the error is 1.69%.The ten factors, nine factors, five factors, four factors and three factor the neural network models are analyzed. By comparing the number of optimal model parameters and errors, we fine a five-factor neural network prediction model is the best result, the error is 0.49%, while the number of each type of input factors of neural network prediction model results are better than the corresponding results of the regression equation.更多还原