大豆秸秆纤维素和半纤维素含量近红外检测模型研究与建立
发布时间：2018-07-21 13:03
【摘要】：我国的经济目前正处于快速发展的阶段中,且近几年我国对于能源的要求和需求也在不断地提高。2015年,我国提出要合理有效地利用已发现和正在开发的新能源,并着重地要求实施改善能源体系利用结构,充分使用可再生能源替代原有石化能源的指导方针。生物质能源是可再生能源中重要的一部分,在将来有可能变成最具有工业价值的能源原料之一。如何结合我国现有情况,开发并合理地利用生物质能源来缓解能源短缺问题,也就变得至关重要。生物质经化学、物理以及生物化学等手段转化为可利用的生物燃料时,其转化过程中所产生的经济效益会直接受转化产出率所影响。但由于受制于繁琐的检验工艺和传统化学技术的限制,无法满足实验或者生产人员对生物质成分配比进行快速检测,因此很多生产过程都缺少明确配比的指导,对产出率也不是非常明确。因此本文将以此为着手点,采用脱豆之后的大豆秸秆作为研究对象,利用近红外光谱结合大豆秸秆中的纤维素和半纤维素含量进行图谱拟合,以求探索快速检测实现的可能性。主要进行了如下研究工作:(1)在黑龙江省内各个地区,收集了不同品种的大豆秸秆173株,并对其所含有的纤维素和半纤维素含量进行了定标及其原始光谱的采集,通过对所采集的样本使用正态直方分析统计,以保证所选的样本都具有一定的代表性。(2)采用导数和平滑预处理方法,对原始光谱做了简单的去噪处理,使用了基于X-Y残差和杠杆值的3D视图分析法剔除样本中的异常,经过分析发现,校正模型的精度也有了大幅度的提升。(3)针对纤维素和半纤维素去噪后的光谱,进行筛选最优的特征波段,选用35和50间隔的间隔偏最小二乘(IPLS)、评估次数为50和100次的遗传算法(GA)、21至51窗口大小的移动窗口最小二乘法(MWPLS)和随机噪声变量100至1000的无信息变量消除法(UVE)进行选择。并对所使用的波段特征选取方法的参数进行多种尝试,然后建立了各特征波段的校正模型。最后得出,各个选取方法之间验证结果相差很大,但相对于全谱的校正模型,结果提升都很明显。(4)建立半纤维素和纤维素各自的预测模型,建立PLS回归模型,同时也利用BP神经网络的非线性拟合优势完成BP验证模型的建立,并且对预测结果进行比较。半纤维素和纤维素在两种预测模型下也分别看到了优劣,半纤维素预测的过程中,我们可以看到PLSR的预测能力远远的强于BP神经网络,而纤维素的预测,结果却是截然相反。综上所述,本文研究的近红外模型对大豆秸秆中纤维素和半纤维素的快速检测,具有一定的可行性,也解决了过.去检测方法中遇到的一些实质性问题,以期运用该技术的日益发展和完善为日后生物燃料生产过程中的快速检测提供新方法。
[Abstract]:Our country's economy is at the stage of rapid development. In recent years, our country's demand and demand for energy have also been increasing in.2015 years. Our country proposes to use the new energy that has been found and being developed effectively and effectively, and emphasizes the improvement of the structure of the energy system and the full use of renewable energy instead of the original. There is a guideline for petrochemical energy. Biomass energy is an important part of renewable energy and may become one of the most valuable energy sources in the future. It is also important to develop and rationally utilize biomass energy to alleviate the problem of energy shortage in the future. The economic benefits produced in the transformation process will be directly affected by the conversion output rate when the means of Biochemistry and biochemistry are converted into available biofuels. However, due to the restriction of tedious test technology and traditional chemical technology, it is impossible to meet the rapid detection of the ratio of biomass components by the experiment or the producer. Many of the production processes are lack of clear ratio guidance, and the output rate is not very clear. Therefore, this paper will take this as the starting point, using soybean straw after pea as the research object, using near infrared spectroscopy combined with the content of cellulose and hemicellulose in soybean straw to carry out atlas fitting, in order to explore the rapid detection and realization. The main research work is as follows: (1) 173 strains of different varieties of soybean straw were collected in various regions of Heilongjiang Province, and the content of cellulose and hemicellulose contained in the samples was calibrated and the original spectrum was collected. The samples were analyzed by direct normal analysis and statistics to ensure the selected samples. All of them have certain representativeness. (2) using the derivative and smoothing preprocessing method, the original spectrum is simply de-noised, and the 3D view analysis method based on X-Y residual and lever value is used to eliminate the anomaly in the sample. After analysis, the accuracy of the correction model has also been greatly improved. (3) De-noising for cellulose and hemicellulose. After the spectrum, the optimal feature band is selected, 35 and 50 spaced interval partial least squares (IPLS), the 50 and 100 times genetic algorithm (GA), the 21 to 51 window size moving window least squares (MWPLS) and the random noise variable 100 to 1000 free variable elimination method (UVE) are selected. The parameters of the feature selection method are tried, and then the correction model of each feature band is established. Finally, the results of each selection method are very different, but the results are very obvious compared with the full spectrum correction model. (4) to establish the prediction model of hemicellulose and fibrin and to establish PLS regression model, and at the same time Using the nonlinear fitting advantage of BP neural network, the BP validation model is established and the prediction results are compared. Hemicellulose and cellulose are also seen under the two prediction models. In the process of hemicellulose prediction, we can see that the prediction ability of PLSR is much stronger than that of the BP neural network, and the cellulose is predisposed. In summary, the near-infrared model studied in this paper has a certain feasibility for rapid detection of cellulose and hemicellulose in soybean straw. It also solved some substantive problems encountered in the detection method, in order to use the technology to develop and improve the production process of biofuel in the future. A new method of rapid detection is provided.
【学位授予单位】：东北农业大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：S216.2
，
本文编号：2135611
[Abstract]:Our country's economy is at the stage of rapid development. In recent years, our country's demand and demand for energy have also been increasing in.2015 years. Our country proposes to use the new energy that has been found and being developed effectively and effectively, and emphasizes the improvement of the structure of the energy system and the full use of renewable energy instead of the original. There is a guideline for petrochemical energy. Biomass energy is an important part of renewable energy and may become one of the most valuable energy sources in the future. It is also important to develop and rationally utilize biomass energy to alleviate the problem of energy shortage in the future. The economic benefits produced in the transformation process will be directly affected by the conversion output rate when the means of Biochemistry and biochemistry are converted into available biofuels. However, due to the restriction of tedious test technology and traditional chemical technology, it is impossible to meet the rapid detection of the ratio of biomass components by the experiment or the producer. Many of the production processes are lack of clear ratio guidance, and the output rate is not very clear. Therefore, this paper will take this as the starting point, using soybean straw after pea as the research object, using near infrared spectroscopy combined with the content of cellulose and hemicellulose in soybean straw to carry out atlas fitting, in order to explore the rapid detection and realization. The main research work is as follows: (1) 173 strains of different varieties of soybean straw were collected in various regions of Heilongjiang Province, and the content of cellulose and hemicellulose contained in the samples was calibrated and the original spectrum was collected. The samples were analyzed by direct normal analysis and statistics to ensure the selected samples. All of them have certain representativeness. (2) using the derivative and smoothing preprocessing method, the original spectrum is simply de-noised, and the 3D view analysis method based on X-Y residual and lever value is used to eliminate the anomaly in the sample. After analysis, the accuracy of the correction model has also been greatly improved. (3) De-noising for cellulose and hemicellulose. After the spectrum, the optimal feature band is selected, 35 and 50 spaced interval partial least squares (IPLS), the 50 and 100 times genetic algorithm (GA), the 21 to 51 window size moving window least squares (MWPLS) and the random noise variable 100 to 1000 free variable elimination method (UVE) are selected. The parameters of the feature selection method are tried, and then the correction model of each feature band is established. Finally, the results of each selection method are very different, but the results are very obvious compared with the full spectrum correction model. (4) to establish the prediction model of hemicellulose and fibrin and to establish PLS regression model, and at the same time Using the nonlinear fitting advantage of BP neural network, the BP validation model is established and the prediction results are compared. Hemicellulose and cellulose are also seen under the two prediction models. In the process of hemicellulose prediction, we can see that the prediction ability of PLSR is much stronger than that of the BP neural network, and the cellulose is predisposed. In summary, the near-infrared model studied in this paper has a certain feasibility for rapid detection of cellulose and hemicellulose in soybean straw. It also solved some substantive problems encountered in the detection method, in order to use the technology to develop and improve the production process of biofuel in the future. A new method of rapid detection is provided.
【学位授予单位】：东北农业大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：S216.2
，
本文编号：2135611
本文链接：https://www.wllwen.com/jingjilunwen/jiliangjingjilunwen/2135611.html
