CN105466885B - Based on the near infrared online measuring method without measuring point temperature-compensating mechanism - Google Patents

Abstract

The invention relates to a near-infrared spectrum analyzer online real-time measurement method based on a non-measuring point temperature compensation mechanism, including designing a multi-temperature level experiment scheme for spectrum collection, performing preprocessing on the collected spectrum with temperature and physical parameters to be measured as targets, and Statistical outlier processing, and then use partial least squares to establish a temperature prediction model, a low-temperature point physical parameter prediction model, and a high-temperature point physical parameter prediction model; then correct and calculate the prediction models at different temperature levels from the low-temperature section or high-temperature section; Finally, an online recursive algorithm is constructed to complete the near-infrared real-time online measurement with the function of temperature compensation without measuring point. The invention includes the compensation effect into the near-infrared modeling process, and forms a recursive algorithm based on the physical property parameters to be measured, so that the physical property measurement at different temperatures can be completed depending on the adaptability of the model itself to temperature. At the same time, the recursive algorithm for physical parameters ensures that the temperature compensation can automatically adapt to the influence of temperature on the near-infrared online measurement value.

Description

Near-infrared online measurement method based on temperature compensation mechanism without measuring point

technical field

The invention relates to an online real-time measurement method of a near-infrared spectrum analyzer without a temperature compensation mechanism at a measuring point, which is suitable for physical parameters affected by ambient temperature, such as fluid viscosity, material density, component concentration, food quality, agricultural product components, and drug active ingredient content , Gasoline oil quality and other online real-time detection.

Background technique

Most of the traditional analysis and detection methods use offline testing technology, and the measurement has a hysteresis. On the one hand, it cannot provide comprehensive and real-time sample information for the production and quality inspection departments. On the other hand, offline measurement cannot achieve the purpose of computer online monitoring and real-time control. . Near-infrared spectroscopy has become a bright spot in online analytical instruments because of its fast analysis speed, low damage to samples, no chemical pollution, almost suitable for analysis of various samples, and multi-component and multi-channel simultaneous determination. In recent years, with the development of chemometrics, optical fiber and computer technology, online near-infrared spectral analysis technology is widely used in agriculture, food, petrochemical, textile, pharmaceutical and other industries, providing a very broad application for production process control At the same time, it also brings considerable economic and social benefits to enterprises.

However, when the near-infrared spectrum analyzer is used for online measurement, the measurement results will be affected by environmental factors. The research shows that for the near-infrared spectrum of a single component, the influence of temperature is more obvious. For complex systems, temperature has a great influence on the optical properties of biological tissues; especially when measuring liquid samples, the increase in temperature will lead to a decrease in the number of stretching vibration hydroxyl groups and an increase in the number of free vibrations, resulting in a shift in the vibration spectrum. The near-infrared spectrum model established at a specific temperature can only be applied to the analysis of sample quality at this temperature, and the online analysis of sample quality at different temperatures is not ideal. This shortcoming greatly limits the real-time online measurement technology of the near-infrared spectrum analyzer. Applications. Therefore, the study of real-time online measurement methods with strong temperature adaptability, high precision, and good robustness has become the key to the effective online application of near-infrared technology.

Contents of the invention

The method proposed by the present invention is aimed at the situation that the temperature change of the measured object cannot be measured or is not measured during the near-infrared real-time measurement, and the change of the temperature itself will have a significant impact on the near-infrared measurement result. An online measurement method with a temperature compensation mechanism that is insensitive to temperature and has a small error is provided.

In order to achieve the above object, the present invention adopts the following technical solutions:

The steps of the present invention are divided into three parts. The first part is the experimental design and spectrum collection of the modeling data; the second part is the preprocessing of the near-infrared spectrum and the establishment of the correction model; the third part is the construction of an online recursive algorithm to complete the near-infrared online measurement.

The experimental equipment for modeling data includes, (1) a sample cell that can adjust the sample temperature (2) a temperature measurer that can display temperature changes (3) a near-infrared spectrum collection instrument (4) an optical sensor that does not significantly affect the sample temperature probe. (5) A computer recording device connected to the near-infrared spectrum collection instrument. The whole set of equipment is shown in Figure 1.

Experiment of the present invention and data collection steps are as follows:

Experimental step 1: Confirm the maximum and minimum temperature values of the sample under online conditions. Divide the temperature range into levels. Each temperature level is generally 5 times larger than the resolution of the temperature measuring instrument to achieve effective discrimination accuracy.

Experimental step two: determine the temperature range of the sample under real-time conditions. At the specified standard temperature, obtain the original standard analysis data for all sample physical parameters.

Experimental step 3: Collect spectral data for the same sample at different temperature levels. At the same time record the corresponding sample temperature value. This temperature value is used in the establishment of the temperature correction model.

The steps for modeling temperature as an explicit factor variable are as follows:

Modeling step 1: Preprocessing the spectrum with the temperature model as the target: perform the first derivative or second derivative operation on the original spectrum to generate the first derivative spectrum or the second derivative spectrum. Here, the determination of the derivative order varies with the characteristics of the physical parameters. For high-molecular high-viscosity samples, the second-order derivative is better; for low-viscosity samples, the first-order derivative is better.

Modeling step 2: Perform principal component analysis (PCA) on the derivative spectrum generated above, and remove statistical outliers, so that the principal component mode of the entire derivative spectrum data is within a statistical reliability.

Modeling step three: temperature is used as the predictor variable, and the derivative spectral wavenumber is used as the independent variable. A temperature correction model of the following form is established using the partial least squares algorithm (PLS):

T _c ＝A ₁ x ₁ +A ₂ x ₂ +...A _n x _n

Here, A _i , i=1, 2,...n are regression coefficients, and _xi is the value of the derivative spectrum at wavenumbers i=1, 2,...n.

Modeling step 4: Preprocessing the original spectrum with the target physical parameter model as the target. These pretreatments include one or more superposition operations of the following algorithms: first derivative, second derivative, max-min normalization, base baseline correction, scatter correction, constant bias correction, etc. The determination of the preprocessing algorithm here varies with the physical parameters to be measured.

Modeling step 5: Perform principal component analysis (PCA) on the preprocessed spectrum generated above, and remove statistical outliers, so that the principal component mode of the entire preprocessed spectral data is within a statistical reliability.

Modeling step 6: Select the spectral data set corresponding to the lowest experimental temperature, use the physical parameters to be measured as predictive variables, and preprocess the spectral wavenumber as an independent variable. The correction model of low temperature physical property parameters in the following form is established by partial least squares algorithm (PLS):

P _l ＝C ₁ z ₁ +C ₂ z ₂ +…C _n z _n

Here, C _i , i=1, 2, . . . n are regression coefficients, _zi is the value of the preprocessed spectrum at wavenumbers i=1, 2, . . . n.

Modeling step 7: Select the spectral data set corresponding to the highest experimental temperature, use the physical parameters to be measured as predictive variables, and preprocess the spectral wavenumber as an independent variable. Using the partial least squares algorithm (PLS) to establish the following form of high temperature physical property parameter correction model:

P _h ＝B ₁ y ₁ +B ₂ y ₂ +...B _n y _n

Here, B _i , i=1, 2,...n are regression coefficients, and y _i is the value of the preprocessed spectrum at wavenumbers i=1, 2,...n.

Modeling step 8: Construct the following physical property parameter formulas at any temperature based on the predicted values of the low temperature model:

P _c ＝P _l +{(P _l0 -P _h0 )/(T _l -T _h )}×(T _c -T _l )

Here, P _l0 and P _h0 are the model prediction values of the lowest and highest temperature points of the same sample in the low temperature model and the high temperature model, respectively. T _h and T _l are the predicted values of the temperature model at the highest and lowest temperature points of the experiment, respectively, and P _c is the measured value of physical properties at the temperature T _c .

Similarly, the following physical property parameter formulas at any temperature can be constructed based on the predicted values of the high temperature model:

P _c ＝P _h -{(P _l0 -P _h0 )/(T _l -T _h )}×(T _h -T _c )

Modeling Step Nine: Obtain a new spectral data set online, and use the following method to form a recursive correction algorithm:

(1) Take the result obtained in the above step 6 as the current value P(k)

(2) Calculate the next measurement: P _r (k+1)=P(k)+K[L(k-1)-P(k-1)]

(3) Assign the current corrected predicted value P _r (k) to the measured value P(k-1) at the previous moment, repeat the above steps, and perform recursive operations.

Here P _r (k) is the current correction value of the near-infrared physical property measurement with temperature compensation, P(k-1) is the measured value of the near-infrared physical property without correction in the previous step, and L(k-1) is the value used in the last calculation The actual physical property parameter value, K is the correction factor or digital filter.

In the above ninth modeling step, the correction factor or the low-order filter can be a more general statistical judgment and logical judgment, or a combination of the two.

In the above-mentioned modeling step nine, in each step of calculation, the physical parameter correction model used can be regenerated from the updated spectral data. The entire calculation algorithm constitutes a recursive form.

The invented method includes the temperature compensation effect into the near-infrared modeling process, and forms a recursive algorithm based on the physical parameters to be measured. Therefore, when near-infrared is used for real-time online measurement, the physical property measurement at different temperatures can be completed relying on the adaptability of the model itself to temperature, without direct temperature measurement information and related calculations. At the same time, the recursive algorithm for physical parameters ensures that the temperature compensation can automatically adapt to the influence of temperature on the near-infrared online measurement value.

Description of drawings

Figure 1 Temperature compensation experimental device without measuring point

Fig.2 The local spectrum of the second derivative of a polymer material

Figure 3 The principal element pattern generated by the second derivative spectrum

Figure 4 Temperature prediction model of polymer

Fig.5 The first derivative preprocessing local spectrum of a high molecular weight polymer

Fig. 6 The main element pattern diagram of the pretreatment spectrum of a polymer

Figure 7 Prediction model of low temperature point of polymer viscosity

Figure 8 Modeling wavenumbers used in the prediction model for the low temperature point of polymer viscosity

Figure 9 Prediction model of high temperature point of polymer viscosity

Figure 10 The modeling wavenumber used in the prediction model of high temperature point of polymer viscosity

Figure 11 Block diagram of online implementation

Figure 12 Comparison of measured and model predicted values of temperature

Figure 13 The effect of temperature compensation on real-time measurement of the viscosity of a polymer

detailed description

The following takes the viscosity measurement of a polymer compound as an example to illustrate the specific implementation method. This example does not constitute a limitation on the scope of the method of the present invention.

The block diagram of the implementation steps is shown in Figure 11.

Step 1: Collect samples under different online conditions, and ensure that the physical parameters of the samples to be measured can cover the range required for measurement. The total number of samples is 40-60.

Step 2: Using the laboratory equipment shown in Figure 1, collect the near-infrared spectra of each sample at five different temperature levels of 24°C, 35°C, 50°C, 60°C, and 70°C, and record the experimental temperature at the same time.

Step 3: Perform preprocessing and principal component analysis on the collected spectra. Different preprocessing methods are performed on the spectra and compared to determine the final suitable preprocessing method. In the example, the second derivative processing is performed on a high-molecular high-viscosity sample. The processing effect is shown in Figure 2. The second-order derivative preprocessing of the original spectrum eliminates the up-and-down drift of the spectrum caused by the aging of the near-infrared light source, the contact between the online sample and the probe, or the vibration of the probe, while maintaining the spectral peak and shape changes caused by temperature changes. The principal element pattern generated by the second-order derivative spectrum is shown in Figure 3, and one of the singular points is eliminated, so that the principal element pattern of the entire spectral data is within the statistical reliability.

Step 4: Establish a near-infrared prediction model for the sample temperature. This model will take the temperature value of the sample directly from the spectrum. Figure 4 is an example of a temperature model, using the modeling bands of 7397-6880cm ^-1 and 5299-4558cm ^-1 . Figure 12 is a comparison of the measured temperature and the model predicted value. It can be seen from the figure that the correlation between the temperature model predicted value and the measured value is 0.99, and the model accuracy ^R2 is 0.98.

Step 5: Carry out first-order preprocessing and principal component analysis (PCA) on the original spectrum with the target physical parameter mode as the target, and remove statistical outliers, so that the principal component mode of the entire preprocessed spectral data is in a statistically credible within degrees. Fig. 5 is a partial spectrum of a polymer first derivative pretreatment. Fig. 6 is a PCA model diagram of a pretreatment spectrum of a polymer.

Step 6: Establish near-infrared prediction models for low temperature and high temperature points respectively. Figure 7 and Figure 9 are the low temperature and high temperature model results respectively. Figures 8 and 10 below are examples of the modeled spectral wavenumber ranges used. Select the waveband range 8900-4497cm ^-1 shown in Figure 8 to model to obtain the low-temperature model Figure 7, and select the waveband range 8955-4497cm ^-1 shown in Figure 10 to model to obtain the high-temperature model Figure 9. It can be seen from Figure ⁷ that the correlation between the predicted value of the low temperature model and the measured value is 0.991, and the model precision R2 is 0.98. It can be seen from Figure 9 that the correlation between the predicted value of the high temperature model and the measured value is 0.988, and the model precision R2 is ^0.9772 .

Step 7: Note that the established low-temperature near-infrared physical parameter model is more accurate in the low-temperature section. However, the high-temperature near-infrared physical parameter model is more accurate in the high-temperature section. The correction calculation can be performed from different directions of the low temperature section or the high temperature section.

The formulas of the physical property parameters at any temperature based on the prediction value of the low temperature model from the low temperature section are as follows:

P _c ＝P _l +{(P _l0 -P _h0 )/(T _l -T _h )}×(T _c -T _l )

Here, P _l0 and P _h0 are the model prediction values of the lowest and highest temperature points of the same sample in the low temperature model and the high temperature model, respectively. T _h and T _l are the predicted values of the temperature model at the highest and lowest temperature points of the experiment, respectively, and P _c is the measured value of physical properties at the temperature T _c .

Step 8: Obtain 10 new spectral data sets online, and obtain the corresponding laboratory raw data at the same time.

Step 9: Calculate the error E(k)=L(k)-P(k) for the past 10 samples and form an error time series

E(k-1), E(k-2),...E(k-10).

Step 10: Perform a low-pass dynamic filtering operation on the above error time series to obtain a one-step forecast value, denoted as B.

Step 11: Calculate Viscosity Corrected Measurements: P _r =P+B

Here P is the current measured value of near-infrared physical properties with temperature compensation.

Step 12: assign the current correction value P _r (k) to the measured value P(k-1) at the previous moment, and perform recursive operation.

Repeat steps 8-12 above.

The legend shown in Figure 13 is an example of the effect of temperature compensation for real-time measurement of polymer viscosity. The temperature range of the sample is about 20-70 degrees Celsius, and the required measurement value is the viscosity value of the sample at 50 degrees Celsius. The fixed temperature model is based on a physical property measurement model established at a temperature of 50 degrees, and its measured values are highly sensitive to temperature. The result obtained by the method of the present invention is insensitive to temperature changes. Furthermore, because the recursive algorithm is adopted, the measurement as a whole is better in line with the real analytical value of the sample.

Claims (6)

1. a kind of existed based on the near-infrared spectrometers On-line sampling system method without measuring point temperature-compensating mechanism, its feature In comprising the following steps：

Step 1：The physical parameter measured value of testing sample is obtained, and near infrared spectrum is gathered under different temperatures level；

Step 2：The near infrared spectrum gathered in step 1 is carried out using temperature as the pretreatment of target and counted at exceptional value Reason, produce derivative spectrum；

Step 3：Using temperature as predictive variable, derivative spectrum wave number caused by step 2 establishes temperature mould as independent variable Type；

Step 4：The near infrared spectrum gathered in step 1 is carried out abnormal as the pretreatment of target and statistics using physical parameter Value processing, produces pre-processed spectrum；

Step 5：The data group corresponding to minimum experimental temperature is chosen, using physical parameter measured value as predictive variable, step 4 Caused pre-processed spectrum wave number is independent variable, establishes the geophysical parameter prediction model of low temperature point；

Step 6：Choose the data group corresponding to highest experimental temperature, using physical parameter measured value as predictive variable, step 4 Caused pre-processed spectrum wave number is independent variable, establishes the geophysical parameter prediction model of high temperature dot；

Step 7：Calculating is modified to forecast model of the different temperatures under horizontal from low temperature point or high temperature dot, amendment is any At a temperature of geophysical parameter prediction model；

Step 8：New near infrared spectrum collection is obtained online, and physical parameter measured value is updated using online recursive algorithm,

Online recursive algorithm is described in step 8：

Pr (k+1)=Pr (k)+K [L (k-1)-P (k-1)]

Wherein Pr (k+1) is the measurement correction value that subsequent time has temperature-compensating, and Pr (k) is that have temperature-compensating at current time Measurement correction value, P (k-1) is the physical parameter measured value of last moment, and L (k-1) is the actual physical property used in last computation Reference value, K are modifying factor or lower order filter.

It is 2. according to claim 1 based on the near-infrared spectrometers On-line sampling system without measuring point temperature-compensating mechanism Method, it is characterised in that：The foundation of temperature model carries out linear regression using offset minimum binary：

Tc=A1x1+A2x2+ ... Anxn

Herein, Ai, i=1,2 ... n are regression coefficients, and xi is the numerical value at the wave number i=1,2 ... n of derivative spectrum.

It is 3. according to claim 1 based on the near-infrared spectrometers On-line sampling system without measuring point temperature-compensating mechanism Method, it is characterised in that：The foundation of low temperature point geophysical parameter prediction model is entered using partial least squares algorithm in the step 5 The following linear regression of row：

Pl=C1z1+C2z2+ ... Cnzn

Herein, Ci, i=1,2 ... n are regression coefficients, and zi is numerical value of the pre-processed spectrum at wave number i=1,2 ... n.

It is 4. according to claim 1 based on the near-infrared spectrometers On-line sampling system without measuring point temperature-compensating mechanism Method, it is characterised in that：The foundation of the model of high temperature dot geophysical parameter prediction described in step 6 is entered using partial least squares algorithm The following linear regression of row：

Ph=B1y1+B2y2+ ... Bnyn

Herein, Bi, i=1,2 ... n are regression coefficients, and yi is numerical value of the pre-processed spectrum at wave number i=1,2 ... n.

It is 5. according to claim 1 based on the near-infrared spectrometers On-line sampling system without measuring point temperature-compensating mechanism Method, it is characterised in that：Model described in step 7 is from low temperature point correction algorithm：

Pc=Pl+ { (Pl0-Ph0)/(Tl-Th) } × (Tc-Tl)

Pl0 herein, Ph0 are model of the same sample in the minimum warm spot and highest warm spot of low-temperature model and high temperature model respectively Predicted value, Th, Tl are the highest of experiment and the temperature model predicted value of minimum temperature point respectively, and Pc is the physical property under temperature Tc Measured value, Pl are the physical parameter measured values of low temperature point.

It is 6. according to claim 1 based on the near-infrared spectrometers On-line sampling system without measuring point temperature-compensating mechanism Method, it is characterised in that：Model described in step 7 is from high temperature dot correction algorithm：

Pc=Ph- { (Pl0-Ph0)/(Tl-Th) } × (Th-Tc)

Pl0 herein, Ph0 are model of the same sample in the minimum warm spot and highest warm spot of low-temperature model and high temperature model respectively Predicted value, Th, Tl are the highest of experiment and the temperature model predicted value of minimum temperature point respectively, and Pc is the physical property under temperature Tc Measured value, Ph are the physical parameter measured values of high temperature dot.