基于小波变换和人工神经网络方法的煤热转化预测模型研究

【作者】 崔阳；

【导师】 徐龙；

【作者基本信息】 西北大学 ， 化学工程， 2011， 硕士

【摘要】 基于我国总体能源资源消费结构特点及能源安全、环保要求,煤炭高效清洁转化利用已成为必然的发展趋势。但是煤炭转化利用常常是在高温、高压等苛刻的条件下操作,同时由于煤炭的复杂性导致煤炭清洁利用过程中影响因素繁多,这些都增加了研究过程的难度和实验工作量,大大增加了实验成本,并且得到的实验数据往往含有“噪音”,谱图时常出现重叠峰。为了解决这些问题,本文采用连续小波变换处理实验数据,采用改进BP神经网络方法建立预测模型,为煤转化过程的工艺设计与优化,催化剂的筛选等提供重要的依据与参考。本文建立的第一个模型是基于改进BP神经网络的煤加氢热解过程预测模型,采用小波变换方法提取失重速率-温度图谱中的第一温峰和峰个数信息,模型选取碳氢比、灰分和挥发分作为网络输入,失重率、第一温峰和峰个数作为输出。预测结果为：多输出预测模型校验样本失重率和第一温峰预测平均相对误差为3.35%和0.54%,单输出预测模型的平均相对误差为2.26%和0.30%,失重速率峰个数预测模型无论训练样本还是校验样本预测值和实验值也都完全吻合。同时,与多输出预测模型相比,单输出预测模型训练速度快,信息学习全面,预测误差小,泛化能力强。基于建立好的各个单输入预测模型,发现碳氢比、灰分和挥发分对煤加氢热解都有着重要的影响,其中C/H的影响最大,这与我们的实际煤热解理论相符合。第二个模型是基于改进BP神经网络的煤二氧化碳催化气化反应预测模型,输入选择碳氢比、灰分、挥发分、催化剂种类和催化剂含量5个影响因素,输出考察气化率、气化初始温度和最大气化速率所对应的温度。预测结果为：多输出预测模型校验样本气化率、气化初始温度和最大气化速率所对应的温度预测平均相对误差是2.68%、3.64%、1.99%,单输出预测模型的平均相对误差分别为1.70%、0.58%、0.55%,显著小于归回公式的预测误差。同时,单输出预测模型收敛快速,对这3个考察参数的预测效果明显优于多输出预测模型,预测精度高,泛化能力强。基于建立好的各个单输入预测模型,发现催化剂种类对煤二氧化碳催化气化反应有着极大地影响,因此筛选合适的催化剂对于提高煤气化反应效率来说是非常有效而且重要的工作。本文建立的预测模型对煤气化反应有着重要的指导作用,同时能够避免大量的重复实验过程。更多还原

【Abstract】 According to the overall consumption structure of energy resources, safe and environmental requirements, clean coal conversion has become an inevitable trend. However, coal conversion is in the high temperature and high pressure. The reaction conditions are very harsh. Moreover, the complexity of coal leads to the result that there are many influence factors in coal conversion process. Those increase the research difficulty, experimental workload as well as cost. Meanwhile, the experimental data often contain noise and spectrum peaks overlap. Therefore, to solve these problems, we use continuous wavelet transform to process experimental data and the improved BP neural network to establish prediction model of coal conversion process. That will provide an important basis and reference for coal conversion process design, optimization, catalyst screening and so on.In this paper, the first established model is prediction model of coal pyrolysis based on the improved BP neural network. We use the continuous wavelet transform to extract informations of the first temperature peak position and peak numbers from the weight loss rate-temperature figures. We select carbon and hydrogen ratio, ash, and volatile as inputs, weight loss, the first temperature peak position and peak numbers as outputs. The results are: in the multi-outputs prediction model, average relative errors of weight loss and the first temperature peak position for test samples are 3.35% and 0.54% respectively. In the single output prediction model, average relative errors are 2.26% and 0.30% respectively. And in the peak numbers prediction model, calculated and measured values are also same. Moreover, compared with the multi-outputs prediction model, single output prediction model has fast training speed, comprehensive study ability, small prediction error, strong generalization ability. Based on the established single output prediction model, it is found that carbon and hydrogen ratio, ash, and volatile all have important effects on the coal pyrolysis process, in which carbon and hydrogen ratio is greatest influence factor. That is consistent with actual theory of coal pyrolysis.The second prediction model is prediction model of coal catalytic gasification based on the improved BP neural network. We select carbon and hydrogen ratio, ash, volatile, type and concentration of catalyst as inputs, gasification efficiency, the initial temperature of gasification and the temperature of maximum gasification rate as outputs. The results are:in the multi-outputs prediction model, average relative errors of gasification efficiency, the initial temperature of gasification and the temperature of maximum gasification rate are 2.68%,3.64% and 1.99% respectively, and, in the single output prediction model, average relative errors are 1.70%,0.58% and 0.55% respectively, which are much smaller than those predicted by regression equation. Moreover, in the single output prediction model, the convergence is fast, prediction results for three output parameters are superior to those of multi-outputs prediction model, prediction precision is high, and generalization ability is strong. Based on the established single output prediction model, it is found that the type of catalyst has a great influence on coal catalytic gasification reaction. Therefore, screening appropriate catalyst is very effectve and important work to improve the efficiency of coal gasification. In this paper, the established prediction models have an important guiding role in coal gasification process and avoid a large number of the repeated experiments.更多还原