CN106970035B - Signal processing method for measuring mercury concentration in thermal power plant flue gas based on CVAFS method - Google Patents

Abstract

The invention provides a signal processing method for measuring the mercury concentration in flue gas of a thermal power plant based on a CVAFS method. The method of the present invention includes: acquisition of the output voltage signal of the mercury analyzer; preprocessing of the voltage signal data; segmental polynomial fitting of the voltage curve; establishment of a corresponding state space equation for the fitted polynomial; Model extended Kalman filter processing. The present invention is used for situations where it is difficult to determine the precise mechanism model of the system in the field of on-line monitoring of flue gas mercury concentration in thermal power plants based on the CVAFS method. The present invention can realize the filtering processing of the output signal of the photomultiplier tube, the method is simple and easy to implement, and the filtering effect is good.

Description

Signal processing method for measuring mercury concentration in thermal power plant flue gas based on CVAFS method

Technical field:

The invention relates to a signal processing method for measuring the mercury concentration in flue gas of a thermal power plant based on a CVAFS method, and belongs to the technical field of signal processing.

Background technique:

As a highly toxic trace heavy metal element, mercury has attracted more and more attention at home and abroad for its potential persistent harm in environmental pollution. In 2011, the Ministry of Environmental Protection issued the "Emission Standards for Air Pollutants from Thermal Power Plants" (GB13223-2011), which included mercury in the scope of control, and stipulated that the maximum emission of mercury and its compounds in coal-fired power plants was 0.05 mg/m ³ . Raise this standard to 0.03mg/m ³ . Although my country currently restricts new coal-fired power plants, they are still the main power generation in my country, so the monitoring and control of mercury pollutant emissions from coal-fired power plants is particularly important.

At present, the online continuous monitoring system Hg-CEM is mainly used to monitor the concentration of mercury in flue gas, and in this system, mercury in flue gas is mainly determined by cold vapor atomic absorption spectrometry (CVAAS) and cold vapor atomic fluorescence spectrometry (CVAFS). concentration. Among them, CVAFS is the most commonly used method. It is a specific ground state atom that is excited by absorbing radiation of a specific wavelength. During the de-excitation process, the excited state atom emits fluorescence of a characteristic wavelength in the form of light radiation, which is measured by a detector. The intensity of fluorescence emitted by the atoms enables a trace analysis method for the determination of element concentrations. Compared with CVAAS, CVAFS has the advantages of higher sensitivity, wider linear range, lower detection limit, less spectral interference, and simpler equipment.

At present, there are Tekran and Thermo Scientific, the mainstream mercury online monitoring instruments in foreign countries. The equipment is very expensive, and the related technology is kept secret. However, domestic flue gas mercury on-line monitoring technology research started late and is not mature enough, and the reliability and accuracy need to be improved. In order to manufacture a mercury form/concentration online monitoring instrument that adapts to the operating characteristics of coal-fired power plants in my country, has high measurement accuracy, low price, and has independent intellectual property rights in my country. The invention starts with the output voltage signal of the mercury analyzer, and performs software filtering on the noise-containing voltage signal, so as to ensure that the finally obtained mercury concentration value is accurate and reliable.

The basic principle of the mercury analyzer to measure the mercury concentration in the flue gas is that the mercury cold vapor is excited under the irradiation of the light source, and the excited mercury atoms will emit fluorescence when they return to the ground state. The fluorescence intensity is detected by the photomultiplier tube, and then determined. Measured mercury concentration. Therefore, in order to ensure the accuracy and precision of the measured concentration, it is necessary to analyze the characteristics and noise of the photomultiplier tube and its related circuits.

Guo Congliang et al. established a mathematical model of photon noise, photocathode noise, secondary emission noise, dynode string noise, and overall noise on the basis of analyzing the overall noise of the photomultiplier tube, which is used for the denoising analysis and correlation matching of the photomultiplier tube. The design of the circuit provides the theoretical basis. In response to the above noise, the following steps are currently taken to reduce the noise. First, standardize the design, manufacture and working conditions of the photomultiplier tube, and then design a series of circuits to perform hardware filtering on the voltage signal output by the photomultiplier tube. The final processed signal ideal. Through the analysis of the above method, it does not involve software filtering, and the design of the hardware circuit is relatively complicated and cumbersome. The invention applies the multi-model extended Kalman filtering method to the nonlinear system, and provides a new method and a new thought for the signal processing of the mercury analyzer. Before filtering the signal, the corresponding state space equation must be obtained first, and the commonly used empirical model is Gaussian. By comparing this model with the actual curve, it is found that the error of the model is large, which will inevitably affect the final filtering effect. .

Contents of the invention

The purpose of the present invention is to address the above existing problems, provide a signal processing method based on the CVAFS method for measuring the mercury concentration in the flue gas of a thermal power plant, and use a multi-model Kalman filter algorithm to perform filtering processing, thereby ensuring that the final mercury concentration value is accurate reliable.

The above-mentioned purpose is achieved through the following technical solutions:

A signal processing method based on the CVAFS method for measuring the mercury concentration in the flue gas of a thermal power plant. The method includes: acquisition of the output voltage signal of the mercury analyzer; preprocessing of the voltage signal data; segmental polynomial fitting of the voltage curve; for the fitted polynomial Establish the corresponding state space equation; perform multi-model extended Kalman filter processing on noisy signals.

Further, the output voltage signal of the mercury analyzer is obtained from the output voltage signal after hardware filtering of the photomultiplier tube used to detect the fluorescence intensity generated by the trace amount of mercury.

Further, the voltage signal data preprocessing includes the following steps:

S11. Randomly find a peak from the obtained several peaks, and obtain a voltage sample value under the premise of satisfying the Shannon sampling theorem;

S12. Draw a corresponding voltage curve according to the obtained voltage sample value.

Further, the segmental polynomial fitting of the voltage curve includes the following steps:

S21. According to the characteristics of the voltage curve, the voltage curve is segmented, the demarcation point is the peak, and the peak is divided into a rising peak and a falling peak;

S22. Carry out polynomial fitting for the rising peak and the falling peak respectively. The fitting order is from 1 to 6. By comprehensively considering the complexity of the fitting curve function and the fitting accuracy, an appropriate order is determined, wherein the polynomial The fitting method is realized by the polyfit function in MATLAB, and the root mean square error RMSE and the mean absolute error MAE are used to test the overall fitting effect. The expressions of RMSE and MAE are as follows:

Among them, N is the number of samples, yi is the voltage value after the i-th fitting, is the true value of the i-th sample voltage.

Further, the establishment of the system state space equation includes the following steps:

S31. Determine the state quantity, state matrix, observation quantity and observation matrix in the state space equation according to the fitted polynomial;

S32. Determine the characteristics of the noise in the voltage signal by analyzing the mercury analyzer, so as to add such noise to the voltage signal.

Further, the multi-model extended Kalman filter is specifically aimed at the two established state space equations, using two extended Kalman filter algorithms with the same principle, and checking the overall The filtering effect, the expressions of RMSE and MAE are as follows:

Among them, N is the number of samples, and Zi is the voltage value after the i-th filtering process.

Beneficial effect:

Compared with the prior art, the present invention has the following beneficial effects:

1. Applicable to systems whose mechanism models are difficult to determine;

2. The effect of piecewise polynomial fitting is not only higher than the overall polynomial fitting, but also higher than the existing empirical model;

3. The piecewise polynomial fitting is not only simple, but also derivatives of any order can be obtained, which is highly compatible with the extended Kalman filter, and there will be no problem that the Jacobian matrix does not exist when obtaining the observation matrix;

4. Compared with inertial filtering, the multi-model extended Kalman is applied to the signal processing of mercury analyzer, and its filtering effect is better.

5. Compared with the conventional hardware filtering method, the software filtering method designed by the present invention is simpler, more convenient and easier to implement under the condition that certain filtering requirements are met.

Description of drawings

Figure 1 comes from the processed wave peak diagram generated when monitoring the mercury content in the flue gas of a coal-fired power plant given in the TekranMercury DataSystem in the Tekran 2600Mercury Analysis System for a certain period of time;

Fig. 2 is the simulation curve of voltage peak of the present invention in MATLAB;

Fig. 3 is the fitting effect contrast figure of segmental polynomial of the present invention and integral polynomial, empirical model;

Fig. 4 is the flowchart of multi-model Kalman filtering of the present invention;

Fig. 5 is the filtering effect diagram of the multi-model extended Kalman filter of the present invention to the noise voltage signal;

FIG. 6 is a diagram showing the filtering effect of the inertial filter of the present invention on a noisy voltage signal.

Detailed ways

The present invention will be further illustrated below in conjunction with specific embodiments, and it should be understood that the following specific embodiments are only used to illustrate the present invention and are not intended to limit the scope of the present invention.

The present invention will be further described below by combining the embodiments and the accompanying drawings.

A signal processing method based on the CVAFS method for measuring the mercury concentration in the flue gas of a thermal power plant, including the acquisition of the output voltage signal of the mercury analyzer; the preprocessing of the voltage signal data; the segmental polynomial fitting of the voltage curve; the establishment of the polynomial for the fitted Corresponding state space equations; multi-model extended Kalman filter processing for noisy signals.

Example 1:

The voltage signal of the mercury analyzer comes from the wave peak after the hardware filtering process produced by the mercury analyzer when monitoring the mercury content in the flue gas of a coal-fired power plant for a certain period of time given in the TekranMercury Data System in the Tekran 2600Mercury Analysis System, as shown in Figure 1 .

Further, the voltage signal data preprocessing includes the following steps:

S11, randomly find a peak from the obtained several peaks, the present invention takes the second peak, and obtains 23 groups of voltage values under the premise of satisfying the Shannon sampling theorem;

S12. Import the obtained 23 groups of voltage values into MATLAB, and draw the voltage peak curve as shown in FIG. 3 .

Further, the segmental polynomial fitting of the voltage curve includes the following steps:

S21. According to the characteristics of the voltage curve, the voltage curve is segmented, and the separation point is the peak, so that the peak is divided into a rising peak and a falling peak;

S22. Carry out polynomial fitting for the rising peak and the falling peak respectively, the fitting order is from 1 to 6, and the appropriate order is determined by comprehensively considering both the complexity of the fitting curve function and the fitting accuracy. The polynomial fitting method is realized by the polyfit function in MATLAB. The order of the fitting polynomials of the rising peak and the falling peak determined by the present invention is 4th order, and the fitting polynomials of the rising peak and the falling peak are respectively ya and yb, and the specific expressions are:

ya＝a1x4 ^{+ a2x3} +a3x2 ⁺ a4x+a5

yb＝b1x4 ^{+ b2x3} +b3x2 ⁺ b4x+b5

The fitting effect diagram of the piecewise polynomial is shown in Figure 4, and the root mean square error RMSE and mean absolute error MAE of the fitting of the piecewise polynomial and the overall polynomial and the empirical model are shown in Table 1.

Table 1 Comparison of fitting effect error

Fit Metrics Segment fit overall fit empirical model RMSE 0.7430 1.4347 8.2549 MAE 0.5447 1.0253 4.7585

It can be seen from Table 1 that compared with the overall fitting and empirical model, the RMSE and MAE of segmental fitting are the smallest, that is, the fitting effect is the best.

Further, the establishment of the system state space equation includes the following steps:

S31. Determine the state quantity, state matrix, observation quantity and observation matrix in the state space equation according to the fitted polynomial. The state quantities are respectively Xa=[a1,a2,a3,a4,a5] ^T , Xb=[b1,b2,b3,b4,b5] ^T ; the state matrix Φ is respectively:

The observation quantities are respectively Z _a ＝a1x ⁴ +a2x ³ +a3x ² +a4x+a5, Z _b ＝b1x ⁴ +b2x ³ +b3x ² +b4x+b5; the observation matrices are:

S32. By analyzing the mercury analyzer, determine the characteristics of the noise in the voltage signal, determine system noise and observation noise, and add such noise to the voltage signal. For the observation noise, according to the mathematical model of photomultiplier tube related noise established by Guo Congliang et al., the noise existing in the voltage signal of the mercury analyzer can be approximated as Gaussian white noise. What the present invention adds in MATLAB simulation is Gaussian white noise with a variance of 5. And because once the fluorescence is detected, its state has been determined, and the theoretical model will not change with time, so the system noise is 0 here.

Through the above two steps, the two final state space equations obtained are:

(1) The system equation of the rising peak is:

The observation equation is:

Z _a (k)=a1(k)x ⁴ +a2(k)x ³ +a3(k)x ² +a4(k)x+a5(k)+v(k),

(2) The equation of the descending peak system is:

The observation equation is:

Zb(k)= _b1 (k) ^x4 +b2(k) ^x3 +b3(k) ^x2 +b4(k)x+b5(k)+v(k),

Where w is 0, v is Gaussian white noise with a mean of 0 and a variance of 5.

Further, the multi-model extended Kalman filter is specifically aimed at the two established state space equations, using two extended Kalman filter algorithms with the same principle for filtering processing. The specific filtering process is shown in Figure 4, in which N is the number of sampling points. In order to fully reflect the filtering effect, the number of sampling points is 200, of which 50 are rising peaks and 150 are falling peaks. The cycle number of the extended Kalman filter for the rising peak is 6000, and the cycle number of the extended Kalman filter for the falling peak is 5000. The multi-model extended Kalman filter effect diagram is shown in Figure 6. In order to verify the superiority of the algorithm of the present invention, inertial filtering is performed on voltage signals containing the same noise, and the inertial filtering coefficient is 0.5. The filtering effect diagram is shown in FIG. 6 . The root mean square error RMSE and mean absolute error MAE after multi-model extended Kalman filtering and inertial filtering are shown in Table 2.

Table 2 Comparison of filtering effects

Filter Effect Metrics Multi-Model Extended Kalman Filter inertial filtering RMSE 0.7502 1.8308 MAE 0.6056 1.4633

It can be seen from Table 2 that the RMSE and MAE of the multi-model extended Kalman filter based on the piecewise polynomial model are smaller than the inertial filter, indicating that the multi-model extended Kalman filter based on the piecewise polynomial model has a better effect.

Claims (1)

1. a kind of signal processing method based on CVAFS method measurement coal steam-electric plant smoke mercury concentration, it is characterised in that: this method packet It includes: the acquisition of mercury analyzer output voltage signal；The pretreatment of voltage signal data；It is quasi- that voltage curve carries out piecewise polynomial It closes；Corresponding state space equation is established for the multinomial fitted；Multi-model spreading kalman is carried out for signals and associated noises Filtering processing；

The acquisition of the mercury analyzer output voltage signal is from the photoelectricity for detecting fluorescence intensity caused by Trace Hg The hardware filtering of multiplier tube treated output voltage signal；

The voltage signal data pretreatment includes the following steps:

S11, from several wave crests of acquisition, look for a peak at random, under the premise of meeting Shannon's sampling theorem, obtain voltage sample This value；

S12, corresponding voltage curve is drawn out according to the voltage example values of acquisition；

The voltage curve piecewise polynomial fitting includes the following steps:

S21, the characteristic according to voltage curve carry out segment processing to voltage curve, and separation is that peak is divided into rising at peak value Peak and lower decresting；

S22, fitting of a polynomial is carried out respectively for rising peak and lower decresting, the order of fitting is from 1 to 6, by matched curve Both the complexity of function and fitting precision comprehensively consider, and determine appropriate order, wherein polynomial fitting method passes through Polyfit function is realized in MATLAB, examines the whole fitting to imitate with root-mean-square error RMSE and mean absolute error MAE The expression formula of fruit, RMSE and MAE are as follows:

Wherein, N is number of samples, and yi is the voltage value after i-th of fitting,For i-th of sample voltage true value；

The system state space equation of establishing includes the following steps:

S31, the multinomial according to fitting determine quantity of state, state matrix, observed quantity and observation square in state space equation Battle array；

S32, by analyzing mercury analyzer, the characteristic of noise in voltage signal is determined, to add in voltage signal This noise like；

The multi-model Extended Kalman filter is particularly directed to two state space equations established, using two principle phases Same expanded Kalman filtration algorithm, examines whole filtering to imitate by root-mean-square error RMSE and mean absolute error MAE The expression formula of fruit, RMSE and MAE are as follows:

Wherein, N is number of samples, and Zi is the voltage value after i-th of filtering processing.