基于卷积神经网络算法判定稀土总量滴定终点技术研究

由于稀土产业发展讯速,市场需求越来越大,应用范围越来越广,需建立一个长期有效且适用于测定各类稀土的方法。针对目前传统稀土总量检测方法效率低、准确度低、滴定终点差异大等问题,难以满足在线检测的需要。本文提出一种基于卷积神经网络的稀土总量浓度在线分析仪。卷积神经网络算法通过摄像头记录样品在滴定过程中的溶液颜色变化,对溶液进行图像特征提取和学习,从而有效、准确地实现化学反应过程中溶液颜色的自动化识别,配合步进电机和注射泵等部件实现自动滴定过程。图像识别本质上是对图像信息进行特征提取,而卷积神经网络(CNN)有着传统识别方法不具备的优点,比如能够自行训练、识别速度更快、所需特征更少等。本设备将自动滴定与神经网络相结合,实现了滴定流程的自动化和样品前处理、滴定、终点判定等过程的一体化,且设备内可同时进行五个样品滴定试验,提高了滴定效率。     

强弱酸混合体系被分步直接准确滴定的判据

以一元强酸和一元弱酸组成的混合体系为对象,并以指示剂确定终点,推导能否实现混酸分步直接准确滴定的判据。研究发现:只有待滴定酸的浓度≥0.544 mol·L-1且弱酸的p Ka处于5.6–8.3之间时,才有可能在滴定误差不超过0.1%的要求下实现同浓度混合液中一元强酸和弱酸的分步直接准确滴定,这两个条件会随着滴定准确度要求的变化而变化。     

改进的人工神经网络在传感器阵列信号处理中的应用

本文提出传感器阵列信号处理的人工神经网络模型，以Cu^ 2/Ca^ 2系统为研究对象．尝试了神经网络方法的效果。其最大相对误差不超过5．％，最大相对预测误差不超过2．4％，结果表明，该方法性能良好，在各种传感器阵列的信号处理方面有广泛的应用前景。     

主成分分析法-BP神经网络算法用于电位滴定法测定有机酸

采用电位滴定法测定混合液中3种以上的多元弱酸,主成分分析法处理了电位滴定的曲线数据,用BP神经网络进行有机酸预测计算。建立了滴定时滴定剂和混合液中有机酸浓度之间关系的模型,实现了不经分离直接测定溶液中几种小分子多元弱酸。结果表明:该神经网络计算出有机酸的预测值与实际值相对误差不大于5.0%。     

冷空气蒸馏分离酸滴定法测定复合肥中铵态氮

提出了用冷空气蒸馏分离酸滴定法在多种形态氮共存时测定复合肥料中铵态氮含量的方法。取适量复合肥料样品于蒸馏瓶中,与550g.L-1碳酸钾溶液60mL反应,溶液总体积以不宜超过110mL。通过空气在流量为3L.min-1和不加热条件下蒸馏8h,生成的氨馏出并先后吸收于相互连接的3个吸收瓶中,每一吸收瓶中预置硼酸吸收溶液50mL,用0.05mol.L-1盐酸标准溶液滴定3溶液至指示剂呈暗红色,记录所耗盐酸标准溶液的总量。将吸收瓶重新按装在蒸馏系统中继续蒸馏1h后再进行滴定,以所耗盐酸标准溶液量不超过0.03mL认为测定完成。根据滴定所耗盐酸标准溶液的总量计算样品中铵态氮的含量。根据3个样品各5次测定结果计算,得到方法的回收率在99.7%～100.1%之间,相对标准偏差在0.12%～0.55%之间。     

热动力学的滴定量热法研究2: 考虑热滞后的一级 反应精确热动力学

建立了滴定热量计滴定期一级反应热动力学的精确数学模型,并在此基础上探讨了热滞后对此模型的影响。经过改进的模型大大提高了利用滴定热量计获取动力学数据的准确度,且扩大了滴定热量计研究的一级反应体系的范围,使其适用反应的半衰期可以小至十几秒。本文采用此模型,研究了去离子水溶剂中乙酸乙酯皂化反应的热动力学,实验结果验证了该模型研究一级反应热动力学的适用性。     

热动力学的滴定量热法研究2: 考虑热滞后的一级 反应精确热动力学

建立了滴定热量计滴定期一级反应热动力学的精确数学模型,并在此基础上探讨了热滞后对此模型的影响。经过改进的模型大大提高了利用滴定热量计获取动力学数据的准确度,且扩大了滴定热量计研究的一级反应体系的范围,使其适用反应的半衰期可以小至十几秒。本文采用此模型,研究了去离子水溶剂中乙酸乙酯皂化反应的热动力学,实验结果验证了该模型研究一级反应热动力学的适用性。     

神经网络法用于多元混酸同时测定

利用多层神经网络误差反向传播算法处理酸碱电位滴定数据，求得出多元混合酸各组分的浓度，优化了神经网络的结构和参数，测定了三组分有机酸混合样品，结果良好，平均相对偏差ＲＳＤ≤４％。     

标准溶液制备中几个问题的探讨

标准溶液是化学实验室用于分析工作中的标准试剂,标准溶液浓度不准确会导致分析结果不可靠,无法对产品质量做出正确判断。标准滴定溶液是专用于滴定分析法的标准溶液,在制备标准滴定溶液过程中,每步操作正确与否直接影响到分析结果的准确性。     

人工神经网络方法预测气相色谱保留值

本文运用一典型的人工神经网络模型－“反向传播“模型的改进形式，研究了诱导效应指数Ｉ，摩尔折射度Ｒｏ，疏水亲脂参数ＩｇＰ，以及分子联通性指数与气象色谱保留行为的关系，实现了对色谱保留植的预测。神经网络预测模型的最大相对误差不超过８．７％。结果表明，该方法性能良好，可望成为色谱保留值预测的有效手段。     

传感器阵列信号处理的人工神经网络方法

本文提出传感器阵列信号处理的人工神经网络方法,并以K~+/Ca~(2+)/NO_3~-/Cl~-阵列系统为对象,尝试了该方法的效果。结果表明,其拟合最大相对误差不超过7.4％,预测最大相对误差不超过6.9％。可见,其性能良好,可望成为各种传感器阵列信号处理的有用工具。     

络合体系中的相对滴定法及其应用于铝的测定

本文提出了络合体系中的相对滴定分析法，它以滴定到一定的状态时，滴定剂体积与被测物的基之间的线性关系为定量基础，用仪器分析的定量方法进行定量，适应于所有类型的络合体系，即使对滴定突跃很小或根本无突跃的体系也同样适用。用氟离子滴定铝离子的实验结果表明本方法的精密度很高。一般相对误差小于0．1％。即使对于铝离子的量小到10-7mol／L时，在最佳实验条件下的相对误差也不大于0．2％。该方法既具有滴定分析法的高精密度，又具有仪器分析法可以测量微量物质的优点，并且设备和方法简单。     

Rapid new methods for the determination of nitrogen in fertilisers and other compounds : Automatic analysis

The determination of ammonia nitrogen, or the sum of nitrate and ammonia nitrogen in fertilisers as described in a previous article, has been automated. In routine work one operator can analyse 50 samples (including grinding, weighing, distillation and titration) within one hour; the total time of analysis per sample does not exceed 3.5 min.     

Statistical Analysis of Differential Potentiometric Precipitation Titration Data for a Mixture of Three Heterovalent Ions Using Linear Characteristics

A computer-based method is proposed for data processing in the successive potentiometric precipitation titration of a three-component mixture of differently charged ions. It is based on a piecewise transformation of the titration curve into linear regression characteristics, the parameters of which determine the successive equivalence points and the solubility products of the precipitates formed in the course of titration to the accuracy of least-squares estimates. To implement the calculation algorithm, a DIFTITR2 program written in Turbo Pascal was developed. The method was tested on the titration of an artificial solution containing a mixture of Cd(II), Hg(II), and Te(IV) with diethyldithiocarbamate, using a chalcogenide Cd-selective or a glassy-carbon indicator electrode, and was used for the analysis of the main components in a semiconductor material, namely, a solid solution of mercury and cadmium tellurides. For concentrations of test ions in the model solution from 7.5 × 10^–3 to 3 × 10^–4 M, the RSD value varies from 0.01 to 0.3; the determination error does not exceed 6%. The RSD value for the semiconductor alloy lies in the range of from 0.002 to 0.01.     

Titrimetric Microdetermination of chloral hydrate by amplification reactions

A simple titrimetric method for estimation of 0.1–5 mg of chloral hydrate is presented. It depends on oxidation of an alkaline solution of chloral hydrate with a chloroform solution of iodine, removal of excess of iodine, oxidation of the resulting iodide with bromine, and iodometric titration of the iodate produced, giving 6-fold amplification. Alternatively, the iodide formed is oxidized with periodate, masking of the excess of periodate with molybdate at pH 3, and iodometric titration of the iodate, giving 24-fold amplification. The coefficient of variation does not usually exceed 2%, for above 1 mg of chloral hydrate but increases to 3.8% at the 0.1-mg level.     

Determination of carbonate alkalinity and apparent dissociation constants in a multiprotolytic system

A method for analysing the carbonate system in a multicomponent solultion is presented, which does not need knowledge of the total composition of the system. It is based on two titrations with acid, starting at the same pH, one of the original solution and the other after removal of carbonate species as carbon dioxide and restoration of the pH to the value for the original solution by addition of carbonate-free base. The differential titration curve, obtained by subtracting one titration curve from the other, is associated with the carbonate system. A procedure is proposed for calculating from the differential titration curve the apparent first and second dissociation constants of carbonic acid, total CO(2) and the carbonate alkalinity at the original pH of the solution.     

Artificial neural networks applied to potentiometric titration of multi-component polybasic acid mixtures

A three-layer artificial neural network model with back-propagation of error is used to treat potentiometric acid-base titration data for estimating the concentrations of individual components in polybasic weak acid mixtures. The network's architecture and parameters were optimized and an empirical rule for dynamically adjusting the learning rate is put forward to improve the network's performance. Satisfactory prediction results were obtained for three-component samples containing maleic acid, propandioic acid and succinic acid with an average relative error of 4.5%.