Studies On Simultaneous Determination Of Multi-component By Chemiluminescence And Spectrophotometry With ANN

Chemometrics, a basic theory and methodology of chemical measure and determination, has taken up an important position and has become an important tool and means in chemistry subject. In the field of analytical chemistry, multivariate resolution and calibration was one of hotspots that analytical chemists pay attention to. In the various methods of chemometrics, artificial neural network(ANN) and wavelet analysis were widely studied and applied. ANN was a system that used engineering and technology means to simulates the structure and function of human brain neuron network. It was made up of large numbers of simple and non-linear processing units. These units were connected each other in complicated, agile and alterable mode. ANN could discretionarily carry out non-linear mapping. ANN could do side-by-side processing and had a strong learning ability by itself. Therefore, ANN has a wide applied foreground in information engineering and control engineering.ANN could be divided into feed forward network and feedback network according to the characteristic of information transfer. The feed forward network with error back propagation learning algorithmic was called BP network. Because this network had a strong ability ofnon-linear mapping, simple network structure, stable working state and simpleness of needed hardwares, BP network was of widely application. Since BP network presented by Ruraelhart 1985, the theory of neural network was developed rapidly. A non-linear 3 layers network with Sigmoid neuron could approach any continuous function at any precision with introduce of hidden node. The main idea of BP network algorithmic was that it divided the learning process into two stages. One was positive direction transfer of information; the other was reverse propagating of error. The input information was calculated step by step at each hidden layer network with Sigmoid neuron could approach any continuous function at any precision with introduce of hidden node. The main idea of BP network algorithmic was that it divided the learning process into two stages. One was positive direction transfer of information; the other was reverse propagating of error. The input information was calculated step by step at each hidden layer, and the output of one layer acted on the input of next layer in the process of positive direction transfer. If without obtaining desired outputs of output layer, the transferring direction of output layer' s error was turned back to modify the weight of each layer till obtaining desired goal.Although having the advantage of distributing storage, side by side processing of information and self-learning ability, BP network had some insufficiency such as network constringency speed being slow, the training time being long, generalization performance being feeblish and so on. So that some improved BP algorithm were put forward by many scholars including adding momentum, reasonably adjusting step length, altering transfer function and the form of error function and so on.Chemiluminescence analysis is an analysis method based on chemiluminescent phenomenon. Without disturbance of foreign excitated light during ray radiating that was produced by chemical or biologicalchemical reaction, chemiluminescent analysis had characteristics of high sensitivity, broad linear range, simple instruments and so on. It had been used in the fields of environmental chemistry, clinic test, medicament analysis, industrial analysis etc. At the same time, with the development of modern science and technology, a lot of advanced researches provided new technique and means for chemiluminescent analysis. Therefore, many new luminescent methods such as nanometer luminescence, luminescent imaging, luminescent live body analysis promoted the development and application of chmiluminescent analysis.In this paper, ANN was combined with liquid chemiluminescent analysis to establish the feasibility of ANN-liquid chemiluminescent analysis method.This paper summarized application of ANN on simultaneous determination of multi-components in analytical chemistry. ANN was used in almost all modern analytical methods, for example, ultraviolet-visible spectrophometry, infrared and near infrared spectrometry, fluorescence spectrometry, atomic spectrometry, electrochemical method, sensor, catalytic kinetic method, chromatographic analysis, mass spectroscopy, nuclear magnetic resonance spectroscopy, X-ray spectrometry analysis, etc. As we known, there isn' t report that ANN was used in chemiluminescent analysis.It was important to determination of As (III) > As(V) in environment water samples. The main methods to determine arsenic were HPLC-ICP-MS, GC-MS, HPLC-ASS, LC-ASS, IC-ICP-MS and neutron activation analysis. Liquid chemiluminescent analysis was used to determine total arsenic in sample. The present paper investigated a method to determine As(III)and As(V) simultaneously through altering the acidity of hydride generation reaction. Iodine was used for absorbent of hydride, luminol chemiluminescent reaction with iodine was used for indicating reaction,flow injection technology was used for determination method of chemiluminescent signal, the chemiluminescent signals were used for input of ANN, the concentration of As (HI) and As (V) were used for object output. The reaction conditions of hydride generation, the absorption solution, the parameters of flow injection, the structure and the training parameters of ANN were optimized in this paper. The present method was used to determine As (III) and As(V) in synthetic samples simultaneously. The relative standard deviations of forecast results for selected synthetic sample were 3. 7% for As (III), 3. 1% for As(V) (Bayesian regularization BP network, n=5) respectively. The average relative errors of forecast results for synthetic samples were 3. 1% for As (III) ,2.4% for As(V)( Bayesian regularization BP network).The determination of iron and antimony is very important in industrial analysis, food analysis, environment analysis, farming analysis, etc. Based on the catalytic action of iron (II) and antimony (III) to the reaction between Cr(VI) with I", this paper studied influence of the temperature to kinetic status of catalytic reaction. It was found that when Fe( II ) and Sb(III) exist simultaneously the average chemiluminescence intensity was higher than the total chemiluminescence intensity, which equaled the chemiluminescence intensity of the Fe(II) catalytic system plus the Sb(III) catalytic system within the detection time. Therefore, the present system was nonlinear. The total chemiluminescence intensity could not be obtained by simply adding the chemiluminescence intensity of the Fe(II ) catalytic system to the chemiluminescence intensity of the Sb (III) catalytic system. On the basis of this phenomenon, a new method of simultaneous determination iron and antimony was proposed by combining ANN with chemiluminescence analysis, because ANN offered a distinct advantage of dealing with the non-linear system. The present paper studied the appropriate conditions ofcatalytic reaction, selected the construction of BP network, optimized the parameters of network training by processing chemi luminescent signal using gradient descent with momentum BP network and Bayesian regularization BP network. The present method could primely forecast the concentration of Fe(II) and Sb(IH) in synthetic samples, the relative standard deviation of Fe(II) and Sb(III) for selected samples was 3.2% and 4. 5%( Bayesian regularization BP network, n=5), the average relative error of the forecast results was 2. 1% for iron and 3.8% for antimony(Bayesian regularization BP network).On the basis of catalytic action of Fe( II ) and Sb(III), used iodine-starch color-producing reaction for indicator reaction, a new kinetic spectrophotmetry method to determine simultaneously Fe(II) and Sb(III) with ANN was established with the reaction system in a high original temperature, injecting it into flow cell, stopping flow and determining the absorbance at different time. The absorbance intensity of catalytic reaction was used for the input of ANN and the concentration of Fe(II) and Sb (III) in reaction system was used for the object output of ANN. The determination conditions, the network structure and training parameters were optimized respectively. The relative standard deviations of forecast results were 3.1% of Fe(II), 2.5% of Sb(III) respectively (Bayesian regularization BP network, n=10) for selected synthetic sample. The average relative errors of forecast results were 4. 8% forof Fe(II), 5. 6% for Sb (III) respectively (Bayesian regularization BP network).A new hydride generation-spectrophotometry method of determination As (III) and As(V) simultaneously with ANN was investigated. The hydride generation was done under different acidity. The silver nitrate is absorption reagent in this method. This present method was applied successfully to determine As (III) and As( V) simultaneously in watersamples. The relative standard deviations of forecast results for selected synthetic sample were 2.6% of As(HI), 5.0% of As(V) respectively (Bayesian regularization BP network, n=5). The average relative errors of forecast results for synthetic samples were 4. 1% for As (III) and 3.8% for As(V) respectively (Bayesian regularization BP network).ANN technique with liquid chemiluminescent analysis and spectrophotometry were studied in the present paper. A new ANN-chemiluminescence analysis approach was searched and a new research field for liquid chemiluminescent analysis was developed. All researches in this paper weren't reported by other people.All researches of the present paper were financially supported by the National Key Technologies R&D Program of China (No. 2002AA636290) and Trackling Key problem of Science and Technology Program of Sichuan Province (No. 02zx028~045).