CN116008379B - Automatic titration system, method and device based on model fitting and machine learning - Google Patents

Abstract

The invention discloses an automatic system, method and device based on model fitting and machine learning. The automatic system includes an input and output module, a titration module for preparatory processing steps when the system starts working, and data processing for calculating titration curves. module, a model fitting module for processing simple system curves, and a machine learning module for processing complex system curves. Different from the existing methods, the present invention no longer relies on monitoring titration jumps for quantitative analysis but adopts a model fitting method to achieve quantitative analysis, and can achieve quantitative analysis of multiple analyte systems. At the same time, the method of pre-deploying machine learning modules is introduced, and the pre-deployed machine learning modules can be called to achieve quantitative analysis of complex systems. The machine learning module does not learn based on historical data, but on "data enhancement" technology. Big data originates from the program module in Kapok software.

Description

Automatic titration system, method and device based on model fitting and machine learning

Technical field

The invention relates to the technical field of quantitative analysis, in particular to an automatic potentiometric titration method based on model fitting and machine learning.

Background technique

Titration technology is a very classic quantitative analysis technology. The establishment of an automatic titrator is to integrate existing titration technology to facilitate the actual quantitative analysis process. The steps used by current automatic titrators are essentially instrumentation and automation of classic titration steps.

In terms of the determination of the titration end point, the traditional method of judging the position of the titration sudden jump is adopted, and the sudden jump of the electrode potential of the titration process is used to determine the titration end point. For simple systems, this approach is usually feasible. However, for complex systems composed of multiple components, it is difficult for existing automatic titrators to achieve automatic quantitative analysis.

Patents that have appeared in recent years have introduced machine learning in fully automatic titration methods, but it is difficult to implement in practice. For example, using historical data to perform machine learning modeling on an automated titrator is inherently impossible. The reasons are: A. Machine learning requires big data, and it takes a long time to obtain big data in the same feature space from actual measurements; B. Machine learning requires good hardware support (such as RTX3090 graphics card, 32G or more memory). It is basically unfeasible to configure these hardware on a fully automatic titrator; C. Machine learning requires specialized knowledge, and routine analysts are not competent.

Contents of the invention

The purpose of the present invention is to break through existing titration methods. The quantitative analysis of analytes is no longer limited to monitoring titration jumps, but introduces a model fitting method to achieve quantitative analysis of multiple analyte systems. At the same time, a pre-deployed machine learning module is introduced. By calling the pre-deployed machine learning module, quantitative analysis of complex systems can be achieved.

The invention provides an automatic potentiometric titration system based on model fitting and machine learning, including:

Data input and output module, used to input parameter files into the instrument and export titration results to the instrument;

Titration module, used to titrate according to parameter files and record titration data;

The data processing module is used to organize and calculate the titration data and draw it as a titration curve;

The model fitting module is used to calculate the concentration value of the analyte when the titrated solution belongs to a simple system;

The machine learning module is used to calculate the concentration value of the analyte when the solution to be titrated does not belong to a simple system.

For the present invention, for this automatic titrator, when the measurement system contains only one analyte (such as monoacid, dibasic acid, polybasic acid, single ion, multiple ion but forms 1:1 with the titrant) type complex, etc.), this system can be regarded as a simple system.

Further, the parameters in the parameter file include: parameters that control the operation of the titration process hardware, the volume of the solution to be measured, the amount of titrant added each time during the titration, the number of volumes of titrant to be added when the titration is completed, the volume of the titrant to be measured. The number of substances, the concentration of the analyte, and the system complexity identifier. When the analyte is an acid or a base, the acid dissociation constant of the analyte and the titrant, the identifier of each form in the solution, the number of hydrogen ions contained in the original acid, and the number of hydrogen ions that can be dissociated by the current acid form. Number; when the analyte is an ion that can be measured by a selective electrode, the formation constant of the complex formed between the analyte and the titrant, the formation constant of the complex formed by other ligands in the system, the analyte and the titrant, and others The property parameters of the ligand itself.

Further, the data input and output module transmits the parameter file into the instrument in the form of a text file. When the titrator starts working, it reads the parameters in the parameter file and then performs the automatic titration process. After the titration is completed, the measurement data can be output to a PC or mobile device for subsequent data processing.

Further, the data processing module is also used to judge the accuracy of the concentration value of the analyte calculated by the model fitting module. When the accuracy does not reach the preset threshold, the titration process can be continued until the titration is reached. The volume of titrant that needs to be added when the titration is completed as set in the parameter file.

Furthermore, the model fitting module is used to calculate the concentration value of the analyte when the liquid to be titrated belongs to a simple system. The model fitting module specifically includes data input, data processing, model fitting process, and data output. The model fitting process includes fitting the calculated titration curve with the measured titration curve; the algorithm of the model fitting process is a sequential number theory optimization method, and the sequential number theory optimization method includes: using the test substance Based on the possible concentration values, construct a parameter space containing the concentration value and evenly arrange the grid points, and then gradually reduce the parameter space according to the optimal point of this parameter space, and finally make the parameter space small enough, so that the optimal point approaches the true concentration value.

Furthermore, the machine learning module also uses data enhancement technology to construct big data, and uses dilated convolution technology to build a deep learning network for quantitative model construction. The algorithm included is: completely using one-dimensional convolution transformation layer to construct a deep neural network In the framework, only one-dimensional convolution transformation is performed in each layer to extract information. Inflated convolution layers with different amplification coefficients are used to expand the receptive field to obtain information on different sections of the titration curve. ResNet technology is used to prevent the gradient of the deep network from disappearing. The convolutional layers use Leaky ReLU activation function, and the final output layer uses non-activated 1/n weight summation.

Furthermore, in the step of applying data enhancement technology to construct big data, and using a deep learning architecture including dilated convolution layers to build a quantitative model, the source of the big data includes: using Kapok software to construct a large curve data set.

On the other hand, the present invention also provides an automatic potentiometric titration method based on model fitting and machine learning, including:

Read the edited parameter file;

Read the parameter file and perform titration initialization;

Determine the system type of the solution being measured based on the system complexity identifier in the parameter file, perform titration and record the titration data;

When the solution to be measured is a simple system, the model fitting module is used to process the relevant parameters and calculate the concentration value of the substance to be measured;

When the solution being measured is a complex system, the machine learning module is used to process the relevant parameters and calculate the concentration value of the substance to be measured.

On the other hand, the present invention also provides a fully automatic titration device, which is characterized in that it includes a central processing unit and a titration device for running the at least one program to perform the automatic potential based on model fitting and machine learning. Titration method.

The beneficial effects of the present invention are: when measuring the concentration value of the substance to be measured, the model fitting module is used to perform fitting, and titration end point determination is no longer required, thereby ensuring accuracy in terms of accuracy; using machine learning technical means, it is possible to Quantitative analysis of complex systems.

Description of the drawings

Figure 1 is a schematic diagram of the working flow of the titrator according to the embodiment of the present invention;

Figure 2 is a schematic diagram of the titrator hardware module according to the embodiment of the present invention.

Detailed ways

The present application will be further described below in conjunction with the accompanying drawings and specific embodiments. The described embodiments should not be regarded as limitations of this application. All other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of this application.

In the following description, reference is made to "some embodiments" which describe a subset of all possible embodiments, but it is understood that "some embodiments" may be the same subset or a different subset of all possible embodiments, and Can be combined with each other without conflict.

Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the technical field to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of the present application and are not intended to limit the present application.

The present invention no longer relies on monitoring titration jumps for quantitative analysis but adopts a model fitting method to achieve quantitative analysis, and can achieve quantitative analysis of multiple analyte systems. At the same time, the method of pre-deploying machine learning modules is introduced, and the pre-deployed machine learning modules can be called to achieve quantitative analysis of complex systems. These machine learning modules do not learn based on historical data, but on big data generated by "data enhancement" technology. Big data originates from the Kapok software module.

The scheme described in the embodiment of the present invention can obtain the following beneficial effects: A. Only the titration curve is measured and fitted through the model fitting module, without determining the titration end point, thus ensuring greater accuracy; B. For systems with multiple components to be tested, the present invention can calculate the titration curve based on the Kapok software module, and then fit the measured titration curve and the calculated titration curve based on the global optimization method, thereby estimating the titration curve of each component to be tested. Concentration value; C. The pre-deployed machine learning module no longer relies on historical data, but first calculates a large data set of the titration curve through the Kapok software module, and uses a special deep neural network architecture on the server for learning modeling. This generates a quantitative analysis model that can be used in actual systems. This process is often called "data augmentation", which can solve quantitative analysis problems of complex systems that cannot be solved by model fitting.

The embodiment of the present invention provides an automatic potentiometric titration system based on model fitting and machine learning, including a data input and output module, a titration module, a model fitting module, and a machine learning module.

Figure 1 is the work flow in the embodiment of the present invention. The specific work flow is as follows:

S1. Edit parameter file;

S2. Carry out titration initialization work;

S3. Perform titration according to the parameter file, record the titration data, and calculate the titration curve;

S4. When the measured solution is a simple system, use the model fitting module to process the relevant parameters;

S401. Calculate the concentration value of the substance to be tested and determine whether the concentration meets the accuracy requirements;

S402. If the concentration reaches the accuracy requirement, end the titration process, otherwise continue the titration until the dripping volume threshold is reached;

S5. When the solution being measured is a complex system, the machine learning module is used to process the relevant parameters, calculate the concentration value of the substance to be measured, and end the titration process.

Step S1 In the embodiment of the present invention, any text editor that supports UTF-8 can be used for parameter editing. The parameter text can be edited in advance on the PC side and the mobile device side. The parameters involved in the parameter file include: parameters that control the hardware operation of the titration process, the volume of the solution to be measured, the amount of titrant added each time during titration, the volume of titrant that needs to be added when the titration is completed, the number of objects to be measured, The concentration of the analyte and the system complexity identifier. When the analyte is an acid or a base, the acid dissociation constant of the analyte and the titrant, the identifier of each form in the solution, the number of hydrogen ions contained in the original acid, and the number of dissociable hydrogen ions of the current form. ; When the analyte is an ion that can be measured by a selective electrode, the formation constant of the complex formed between the analyte and the titrant, the formation constant of the complex formed by other ligands in the system, the analyte and the titrant, and other ligands The attribute parameters of the body itself. The prepared parameter file is stored in the form of a text file. When the titrator starts working, it reads the parameters in the parameter file and then performs the automatic titration process.

In particular, based on the acid-base proton theory, acids and bases are conjugates, and the acid dissociation constant can be used uniformly to describe their behavior. Therefore, only "acid dissociation constant" is used here to uniformly describe its characteristic parameters.

Step S2 In the embodiment of the present invention, the processing work of titration initialization includes a series of preparation work before the formal titration work, such as electrode calibration, titration glass pipe flushing, etc. These preparation work belong to the routine operations of titration analysis. The following describes the process and details related to pH electrode calibration: pH electrode calibration is a standard process for all titrators. The basic method is to use a pH electrode to measure three acid-base buffer solutions. According to the difference between the measured value and the standard value, the electronic circuit is The module adjusts the circuit status of the instrument so that the displayed value of the titrator is equal to the standard value.

Steps S3 and S4 in the embodiment of the present invention include specific operations in the model fitting module: including data input, data processing, model fitting process, data output and other processes in the module. The data input of the model fitting module includes the titration curve of the system to be tested and the titration curve calculated based on the parameters of the test object in the parameter file. Calculated titration curves can be calculated using modules in the Kapok software package. The measured titration curve is measured through the circuit system of the automatic titrator. The concentration of the analyte is estimated through the sequential number theory optimization method. The core method is to construct a grid point set according to the concentration parameter space. Each point is a combination of a set of concentration values of the analyte to calculate the titration curve and measurement. The sum of the squared residuals of the titration curve subtraction is the objective function, and the objective function is minimized to obtain the concentration value of the test substance. Kapok software is a software package that can be used to calculate titrations of complex acid-base systems and titration curves of complex coordination systems. This software package can theoretically calculate titration curves of any complex acid-base titration systems and EDTA-based coordination titration systems. Titration curve.

In the embodiment of the present invention, the basis for calculating the concentration value of the analyte, the calculation formula, the meaning of each variable in the calculation formula, etc.; the basis for the calculation process of the concentration value of the analyte is as follows:

For an acid-base titration system, when multiple titrant mixed solutions are used to titrate multiple analyte mixed solutions, the titration equation is:

Here, the subscript t represents the titrant; the subscript s represents the substance to be measured; Δ=[H ⁺ ]-[OH ^- ]; the calculation formula of F is as follows

Here, the acid dissociation constant K _a,0 ≡0; p represents the maximum number of protons that can be dissociated from the acid; q represents the number of protons that can be dissociated in the current state of the acid. Note: Acids and bases are conjugates under the framework of acid-base proton theory. Here we use acid to express acids and bases.

In the embodiment of the present invention, the Kapok software module can solve the titration curve of the acid-base titration system composed of equation (1) and equation (2).

For an ion titration system, assume that titrant R is used to titrate m ions to be measured M _j (j=1,,2,...,m.). The coordination reaction is as follows (for convenience, the ion charge labels are ignored below)

Here, β is the cumulative formation constant; Is the standard concentration, unit 1mol/L. The titration equation is

here, is the added volume of titrant;/> is the initial volume of the metal ion solution; α is the side reaction coefficient of the corresponding type in brackets.

Currently, the most commonly used ionic titrants are aminocarboxylic titrants, such as EDTA. Since EDTA forms a 1:1 complex with metal ions, in this case, the cumulative formation constant is the usual formation constant. The Kapok software module can solve the titration curve of the titration system in which EDTA is the titrant. Other coordination titration types are less commonly used and are not currently integrated into the Kapok software module, but can be developed according to equation (4).

In the embodiment of the present invention, the basis for judging accuracy is to obtain the global optimal point through the sequential number theory optimization method, through global distribution of points, and then narrowing the search space. The minimum search space size of the concentration value can be preset according to the accuracy requirements of the quantitative analysis. When the search space of the sequential number theory optimization method is smaller than the preset search space of the concentration value, the required accuracy is achieved.

Step S5 In the embodiment of the present invention, the machine learning module uses Kapok software to construct big data based on "data enhancement" technology, and uses the "dilated convolution" deep learning model to build quantitative models.

Figure 2 is a brief schematic diagram of the hardware part of the embodiment of the present invention. The hardware part of the embodiment of the present invention is introduced below with reference to Figure 2:

central processing unit

The central processing unit of the titrator in the example of the present invention includes a CPU and related auxiliary circuits, and is paired with a Linux operating system (such as Ubuntu) to realize the calling of other modules by writing corresponding programs.

Data input and output module

The data input and output module in the example of the present invention is responsible for the input of parameter files and the output of measurement data. This module contains wireless and wired communication modules to connect with instrumentation equipment. The wireless communication module is mainly composed of ESP32/ESP8266 modules commonly used in industry. It can perform data communication with the instrument by connecting to wifi on the mobile phone or computer. The wired communication module is mainly composed of the RS485 module commonly used in industry, and is connected to a PC through wiring to realize data transmission.

Machine learning module

The machine learning module in the example of the present invention is a pre-deployment module, and the related quantitative analysis models involved are trained through the server (equipped with an RTX 3090 graphics card) in the early stage, and then deployed to the titrator in the form of files. During the actual titration process, the machine learning prediction program calls up the corresponding machine learning module to use the titration curve obtained by titration to calculate the concentration value of the substance to be measured.

Data processing module

The data processing module in the example of the present invention will be responsible for processing data. This module mainly includes various data processing programs written in C and C++ programming languages. This module can calculate the titration curve based on the parameter file. It can predict the concentration parameters of the analyte by combining the calculated titration curve and the titration curve obtained by titration through the optimization algorithm. It can use the pre-deployed machine learning module and the titration curve obtained by titration to predict the analyte. Concentration parameters are predicted.

memory module

The memory module in the example of the present invention includes an SD card. The operating system, programs, models, data, etc. are all stored in the memory module and are uniformly scheduled by the operating system and the central processing unit. The central processing unit and the SD card module are connected via the serial peripheral interface SPI protocol. Based on C and C++ programming languages, functions related to SD card identification, reading and writing can be written to realize the storage and reading of SD card data. The communication method between the central processing unit and the SD card is the FAT16/32 protocol. Through this protocol, the central processing unit can save data files in real time.

display module

The display module in the example of the present invention will be used to display the titration status of the instrument. The display module is composed of a liquid crystal control chip and a liquid crystal screen. The liquid crystal screen will be connected to the liquid crystal control chip through an RGB interface. The interface with the chip will be connected through 8080/SPI and other methods. The specific connection method should be selected based on chip resources and actual conditions. This module can be used to display instrument titration data or titration curves.

Titration module

The titration module in the example of the present invention consists of a peristaltic pump module and an electrode module. The peristaltic pump module controls the stepper motor and the driver. The driver can adjust the signal to regulate the motor speed, and also has the function of protecting the motor. High-precision flow transmission control can be achieved to control the amount of titrant added during the titration process. The electrode module is used to measure changes in electromotive force caused by changes in ion concentration in the solution system to be measured. The module consists of an electrode and an electrode signal conditioning module. Optional electrode modules include pH electrodes and ion selective electrodes. The electrode is used to measure the electrode signal of the object to be measured in the solution, and the electrode signal modulation module will amplify the signal. Its working principle: Use a signal-taking circuit to extract the electrode data, and pass the signal through a proportional amplification circuit. The ADC digital-to-analog conversion module samples, quantizes, and encodes the data. The sampling frequency is greater than 2 times Shannon's law, thereby ensuring that the signal No distortion. Quantization can approximate a limited number of amplitude values to the original continuously changing amplitude value, and change the continuous amplitude of the analog signal into a limited number of discrete values with certain intervals. Coding follows certain rules. Represent the quantized value as a binary number. The final data is transmitted to the chip, and the electrode potential value is accurately calculated based on the calibration signal reference of the electrode at the beginning of the titration.

In the embodiment of the present invention, the titrator further includes a power supply module. The power supply module of the embodiment of the present invention is the same as the power supply module of a commonly used titrator on the market.

Claims (5)

1. An automatic potentiometric titration system based on model fitting and machine learning, comprising:

the data input and output module is used for importing the parameter file and exporting titration result data;

the titration module is used for titrating according to the parameter file and recording titration data;

the data processing module is used for sorting and calculating the titration data and drawing the titration data into a titration curve;

the model fitting module is used for calculating the concentration value of the object to be detected when the solution of the object to be detected belongs to a simple system; the simple system is a measurement system which only comprises one object to be measured;

the machine learning module is used for calculating the concentration value of the object to be detected when the solution of the object to be detected does not belong to a simple system;

the model fitting module specifically comprises data input, data processing, a model fitting process and data output; the model fitting process comprises fitting the calculated titration curve and the measured titration curve, wherein the algorithm of the model fitting process is a sequential number theory optimization method, and the sequential number theory optimization method comprises the following steps: constructing a parameter space containing the concentration value by utilizing the concentration possible value of the object to be detected, uniformly arranging lattice points, sequentially reducing the parameter space according to the optimal point of the parameter space, and finally enabling the parameter space to be small enough so as to enable the optimal point to approach to a real concentration value;

the model fitting process of the model fitting module specifically comprises the following steps: constructing a lattice point set according to the parameter space, wherein each point is a concentration value of an object to be measured, taking the sum of squares of residual errors subtracted by a calculated titration curve and a measured titration curve as an objective function, and realizing the minimization of the objective function, thereby obtaining an estimated value of the concentration value of the object to be measured;

the machine learning module specifically includes: constructing big data by using a data enhancement technology, and constructing a quantitative model by means of a deep learning framework comprising an expansion convolution layer, wherein the specific algorithm comprises the following steps: constructing a deep neural network frame by completely adopting one-dimensional convolution transformation layers, and carrying out one-dimensional convolution transformation in each layer; expanding the receptive field by using expansion convolution layers with different amplification coefficients so as to obtain information of different sections of the titration curve; the ResNet technology is adopted to prevent gradient disappearance of the deep network, and each convolution layer activates a function by using a leak ReLU; the output layer of the framework uses inactive 1/n weight summation.

2. An automatic potentiometric titration system based on model fitting and machine learning according to claim 1, wherein the titration experimental parameters include: controlling the hardware operation parameters of the titration process, the volume of the solution to be measured, the volume of the titrant added each time during titration, the volume number of the titrant to be added when the titration is completed, the number of the objects to be measured, the concentration of the objects to be measured and the system complexity identifier; when the object to be detected is acid or alkali, the acid dissociation constants of the object to be detected and the titrant, identifiers of all types of bodies in the solution, the number of hydrogen ions contained in original acid and the number of hydrogen ions dissociable in the current acid type; when the analyte is an ion that can be measured by the selective electrode, the formation constant of the complex formed by the analyte and the titrant, the formation constant of the complex formed by other ligands in the system, the analyte and the titrant, and the attribute parameters of the other ligands themselves.

3. An automatic potentiometric titration system based on model fitting and machine learning according to claim 1, in which the automatic potentiometric titration system stores the parameter file in the form of a text file, and when the titrator is started, the parameters in the parameter file are read to control the subsequent titration process.

4. The automatic potentiometric titration system based on model fitting and machine learning according to claim 1, wherein the data processing module is further configured to fit the calculated titration curve with the measured titration curve during the titration process, determine the accuracy of the concentration value of the analyte calculated by the model fitting module, and continue to drop the titrant until the volume of titrant to be added reaches the volume of titrant to be added when the accuracy does not reach the preset threshold, where the volume of titrant to be added reaches the completion of the titration set by the titration parameter file.

5. An automatic potentiometric titration method based on model fitting and machine learning, applied to an automatic potentiometric titration system according to any one of claims 1-4, comprising:

reading the edited parameter file;

reading a parameter file, and performing titration initialization work;

judging the type of a system of the measured solution according to the system complexity identifier in the parameter file, titrating and recording titration data;

when the measured solution is a simple system, a model fitting module is adopted to process related parameters, and the concentration value of the object to be measured is obtained through calculation;

when the measured solution does not belong to a simple system, the machine learning module is adopted to process related parameters, and the concentration value of the object to be measured is calculated.