A desulfurization system absorption tower liquid level measurement system and method based on Gaussian process regression

Abstract

The present invention proposes a desulfurization system absorption tower liquid level measurement system based on Gaussian process regression, wherein a self-learning module is built into the DCS in the form of a PLC, between the density measurement circuit and the liquid level measurement circuit, and includes a data import module, a data storage module, a data cleaning module, a function fitting module, and an output module. The module temporarily stores the density data in the DCS in the storage module after preliminary processing by the import module, replaces the original data after cleaning and function fitting, and finally returns the data to the DCS system and outputs it to the liquid level measurement circuit by the output module, and realizes the function of real-time liquid level correction in a reciprocating manner. The present invention adopts a self-learning method to efficiently and accurately realize density correction, reduce the fluctuation of density measurement, optimize the soft measurement technology of liquid level, and improve the accuracy of measuring liquid level.

Description

Technical Field

[0001] The invention belongs to the technical field of desulfurization, and in particular relates to a desulfurization system absorption tower liquid level measurement system and method based on Gaussian process regression. Background Art

[0002] In conventional wet desulfurization processes, the liquid level and slurry density in the desulfurization absorption tower are important parameters related to the desulfurization efficiency and operational stability of the entire desulfurization system. If the liquid level in the desulfurization tower is too high, the energy consumption of equipment such as the oxidation blower and slurry circulation pump will increase, and the risk of overflow will also increase; if the liquid level is too low, the oxidation effect will deviate and there will be safety risks.

[0003] In the context of flexible power generation, due to the rapid changes in unit load, the desulfurization load changes more dramatically, which makes the water balance fluctuate greatly during the entire wide load operation process, resulting in large fluctuations in the liquid level. Therefore, it is very necessary to accurately measure the liquid level during the operation process to provide a reference for the operation. However, due to reasons such as instrument structure and harsh measurement environment, it is not easy to measure the liquid level accurately. Therefore, it is necessary to correct the measurement of the liquid level.

[0004] Commonly used liquid level measurement devices are mainly divided into the following types according to the sensor type:

[0005] (1) Differential pressure type: The difference between static pressures is measured by a sensor, and the liquid level of the desulfurization tower is calculated by the slurry density.

[0006] like Figure 1 The differential pressure level gauge shown above uses the principle that a liquid column or material accumulation generates pressure on a certain point. When the density ρ (kg / m3) of the measured medium is known, the formula The liquid level measurement problem can be converted into a differential pressure measurement problem.

[0007] If the measured medium is corrosive, an isolation container needs to be installed between the positive and negative pressure chambers of the differential pressure transmitter and the pressure pipe to prevent the corrosive medium from directly contacting the transmitter. Figure 1 If the density of the isolation fluid is ρ1 (ρ1>ρ), the differential pressure Δp (Pa) measured on the differential pressure transmitter is calculated as

[0008] Δp=ρgh+ρ1g(h1-h2)

[0009] Where g is the acceleration due to gravity, 9.8 m / s 2

[0010] Since the differential pressure signal has an additional term ρ1g(h1-h2), when h=0, Δp is not equal to 0, and a negative zero point migration is required to overcome the influence of the fixed differential pressure ρ1g(h1-h2).

[0011] The actual liquid level H(m) in the container is

[0012]

[0013] Where h0 is the height of the liquid level measuring point from the bottom of the container.

[0014] The pressure measuring membrane of the differential pressure level gauge is prone to sediment accumulation, and the differential pressure level gauge with the flange directly fixed on the tank wall cannot be disassembled during operation. The slurry density, a parameter associated with the liquid level measured by the differential pressure level gauge, is inaccurate.

[0015] (2) Ultrasonic type: Ultrasonic pulse waves are emitted to the liquid surface, and the transmission time of the pulse reflected by the liquid surface is converted into the liquid level.

[0016] During the measurement, the ultrasonic transducer (probe) emits a high-frequency pulse sound wave. After the sound wave is reflected back from the liquid surface, the reflected echo is received by the transducer and converted into an electrical signal through a piezoelectric crystal or magnetostrictive device. The distance from the sensor to the surface of the measured liquid is calculated based on the time between the emission and reception of the sound wave. The distance measurement principle of the ultrasonic level meter used in the experiment is as follows:

[0017]

[0018] In the formula: h is the distance from the ultrasonic sensor to the bottom; L is the distance from the sensor to the liquid surface; C is the ultrasonic sound velocity; H is the actual height of the measured liquid surface; T is the time from the ultrasonic wave being emitted to being received. During the experiment, after each change in the liquid level volume, the liquid level height measured by the ultrasonic level meter can be directly displayed and read directly.

[0019] Ultrasonic level gauges are easily affected by environmental noise. For example, when foam appears in the slurry, or when water vapor is generated in the slurry due to evaporation, the slurry will be raised to produce a false level, which is difficult to distinguish with ultrasonic waves, resulting in inaccurate level measurement.

[0020] (3) Capacitive type. The liquid level is measured by measuring the capacitance between the probe and the container wall.

[0021] Capacitive level gauges measure the height of the liquid by measuring the change in capacitance. Insert the level gauge into a liquid container, with the metal rod of the level gauge as one pole of the capacitor and the container wall as the other pole of the capacitor. The medium between the two electrodes is the liquid and the gas above it. The dielectric constant ε1 of the liquid is different from the dielectric constant ε2 on the liquid surface. If ε1>ε2, the total dielectric constant value between the two electrodes increases with the increase of the liquid level, and thus the capacitance increases; conversely, when the liquid level drops, the ε value decreases, and the capacitance also decreases. Therefore, the height of the liquid level can be measured by the change in capacitance between the two electrodes.

[0022] Capacitive level meters use a contact measurement method, which can easily lead to solid accumulation or even scaling, affecting measurement.

[0023] (4) Float type: A float floating on the liquid or slurry is used to indicate the tank liquid level. Mechanical, electronic or magnetic linkage devices can be used.

[0024] The commonly used density meters for measuring the slurry density of limestone flue gas desulfurization system (FGD) include nuclear radiation type, ultrasonic type, photoelectric vibration type, tuning fork type, etc. Considering the cost, safety and other factors, the most commonly used is the tuning fork density meter. The specific measurement methods are as follows:

[0025] The tuning fork density meter is mainly composed of a tuning fork, a resonance unit, and a vibration pickup unit. It uses a piezoelectric crystal to generate a resonant frequency, and then measures the resonant frequency through another piezoelectric crystal on the head of the tuning fork. When the flowing liquid passes through the tuning fork, due to the influence of the liquid density, the damping of the tuning fork by the measured liquid is different from that in the undamped state, and the change in the measured liquid density is reflected by the change in the resonant frequency.

[0026] The resonant frequency of the tuning fork is known to be:

[0027]

[0028] Where: f′——the changed resonant frequency;

[0029] f——resonant frequency before change;

[0030] m g ——effective mass of the cantilever beam;

[0031] k – a constant related to the fork size.

[0032] The liquid level density can be obtained by calibrating the resonant frequency of the tuning fork before it changes.

[0033] There are also several problems with float-type level gauges; 1) The installation must be in a vertical state to ensure that the float assembly can move freely in the main pipeline. 2) Crystallization on the float surface. After the float crystallizes, the weight increases and it cannot float normally on the liquid surface. 3) The impact of mechanical failure. Corrosion, external force, crystallization on the float surface, etc. can cause mechanical failure and hinder the sliding of the float assembly. 4) Turbulence of the liquid. When the float-type level gauge is measuring, the turbulence of the liquid will cause the float to be unstable and affect the measurement. Instruments such as liquid level and density will measure inaccurately.

[0034] In coal-fired power plants, considering factors such as cost and accuracy, differential pressure level gauges with high cost performance are generally selected. Differential pressure level gauges use soft measurement to convert the difficult-to-measure liquid level into the easy-to-measure slurry density. However, due to instrument reasons, the measured value of density does not reflect the actual operating level. Therefore, it is necessary to make corresponding correction measurements on measurement data such as density to optimize the soft measurement technology of liquid level. Summary of the invention

[0035] In order to solve the above problems, the present invention provides a desulfurization system absorption tower liquid level measurement method based on self-learning, which uses a self-learning algorithm to solve the fluctuation problem of density measurement and improve the liquid level measurement accuracy.

[0036] On the one hand, the present invention proposes a desulfurization system absorption tower liquid level measurement system based on Gaussian process regression, including a density measurement circuit, a liquid level measurement circuit, and a density data correction system between the two, wherein the liquid level measurement circuit includes a differential pressure liquid level gauge; characterized in that: the density data correction system places a Gaussian process regression algorithm in a distributed control system in the form of a programmable logic controller to correct the density of the slurry measured by the density measurement circuit, and the density data correction system includes a data import module, a data storage module, a data cleaning module, a function fitting module, and an output module; the density data in the distributed control system is temporarily stored in the data storage module after preliminary processing by the data import module, and the data cleaning module and the function fitting module replace the original data after cleaning and function fitting the data stored in the data storage module to achieve data correction; the data import module receives a loop instruction issued by the output module to achieve reciprocating density data correction; the final corrected density data is returned to the distributed control system by the output module and output to the liquid level measurement circuit to achieve accurate measurement of the liquid level.

[0037] Furthermore, the data import module receives the original data of each parameter in the distributed control system, and makes the data of each parameter evenly distributed in time based on an interpolation algorithm.

[0038] Furthermore, the data storage module includes data storage, data classification and data clearing functions, and is used to temporarily store the data output from the import module and the function fitting module, and export the corrected data therein to the output module.

[0039] Furthermore, the data storage module includes a data storage area, which is divided into a historical data area and a real-time data area; the historical data area stores the accumulated qualified data and corresponding time series of each stage, and the real-time data area stores the time, value, logical signal and other information of the newly imported data.

[0040] Furthermore, the data cleaning module receives data samples from the real-time data area and divides the data into qualified data and unqualified data that needs to be corrected; and the function fitting module is used to process and output the unqualified data.

[0041] Furthermore, the function fitting module uses the qualified data input from the storage module as a prediction model training set, simulates the variation curve of slurry density over a period of time based on the Gaussian process regression algorithm, predicts the density change in the time window where the unqualified data is located, and outputs the corrected data to replace the unqualified data in the storage module.

[0042] Furthermore, when the storage module receives the correction data from the function fitting module, it retrieves the correction logic of the real-time data. When the correction logics of all real-time data are high-order signals, it gives a signal to the output module to start the output module. After the output module is started, the storage module issues a cleanup instruction to copy the real-time data to the output module and output it to the distributed control system and the liquid level measurement circuit. After the copying is completed, the cleanup function of the storage module is started to clear the real-time data area and clean up some historical data, waiting for the input of the next data sample set.

[0043] Furthermore, the data classification function sets correction logic and quality logic for each real-time data tuple. When the storage module receives the density data from the import module, the logic signals of the correction logic and the quality logic are set to low-order signals, and the density data is placed in the real-time data area; the correction logic is used to distinguish whether the density data has been processed by the algorithm, and the quality logic is used to judge the quality of the density data when it is imported from the distributed control system, so as to determine whether the density data can be used as a training set sample for the function fitting module.

[0044] Furthermore, when the correction logic signal is a low-order signal, the density data cannot be output to the distributed control system; when the data storage module receives a high-order instruction from the data cleaning module or the function fitting module, the corresponding correction logic signal is set to a high-order signal, and the density data can be output to the distributed control system.

[0045] Furthermore, only when the data cleaning module issues a high-order instruction can the quality logic be converted into a high-order signal; after receiving the high-order instruction, the corresponding data will be copied from the real-time data area to the historical data area in sequence.

[0046] On the other hand, the present invention proposes a method for measuring the liquid level of an absorption tower of a desulfurization system based on Gaussian process regression, which performs measurement according to the above-mentioned system for measuring the liquid level of an absorption tower of a desulfurization system based on Gaussian process regression.

[0047] The beneficial effects of the present invention are:

[0048] The absorption tower liquid level measurement system and method of the present invention use time tuples as standardized storage units, and the tuples use time as the main attribute, which can achieve rapid docking between different modules and reduce the number of data indexing times. The density data correction system uses the storage module as the hub, and the data set only needs to be accumulated in the storage module. And the interaction between modules is mainly based on logical signals, which reduces the transfer of parameter sets between different modules, not only reduces the storage space, but also improves the calculation efficiency. In the process of data preprocessing and fitting, the data processing process refers to the system operation mechanism. For example, when cleaning the data, not only the abnormal values ​​in the conventional sense (such as blank values, missing values, etc.) are eliminated, but also the data fluctuating due to load and pH reasons are eliminated, further improving the accuracy of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 This is the measurement principle diagram of differential pressure level meter;

[0050] Figure 2 It is the curve of slurry density change and distribution in the distributed control system;

[0051] Figure 3 This is the principle of the slurry density correction method in the system of the present invention;

[0052] Figure 4 It is the data storage method and interaction method in the system of the present invention;

[0053] Figure 5 It is a comparison chart between the slurry density data corrected by the system of the present invention and the original data. DETAILED DESCRIPTION

[0054] The embodiments are described in detail below in conjunction with the accompanying drawings.

[0055] The differential pressure level gauge uses a sensor to measure the difference between static pressures and converts the desulfurization tower liquid level height through the slurry density. Figure 2 Shown is the curve of slurry density change and distribution in the distributed control system (DCS).

[0056] The slurry density fluctuates violently in this range. However, by analyzing the parameters such as slurry pH and slurry circulation pump power, the desulfurization condition has not changed significantly. This indicates that the density measurement is inaccurate, which may be caused by wear and blockage of the density meter. At the same time, the uneven distribution of slurry in the absorption tower will also cause density measurement deviation, which in turn affects the liquid level measurement.

[0057] To solve the problem of drastic fluctuations in density data, the present invention adopts Figure 3The steps shown in the figure are used to correct the density data. It embeds the self-learning algorithm into the DCS in the form of a programmable logic controller (PLC), between the density measurement circuit and the liquid level measurement circuit, and mainly includes five modules: data import module, data storage module, data cleaning module, function fitting module, and output module.

[0058] The overall process of liquid level correction measurement is:

[0059] The density data in the DCS is processed by the import module and temporarily stored in the storage module. These data are replaced by the original data after cleaning and function fitting with the storage module as the hub. The final corrected density data is returned to the DCS system by the output module and output to the liquid level measurement circuit, and the function of real-time liquid level correction is realized in a reciprocating cycle.

[0060] The correction system is connected to the DCS database, which enables the functions of fast data reading and real-time correction.

[0061] The specific functions of each module are as follows.

[0062] Data import module:

[0063] The data import module is used to receive the original data in the DCS and make each parameter evenly distributed according to time based on the interpolation algorithm. The imported parameters mainly include slurry density, slurry pH, unit load, etc. and the time corresponding to each parameter, among which slurry pH and unit load are used for data cleaning.

[0064] According to the on-site conditions, the time series of each parameter in the original data are not matched, so it is necessary to normalize the time series of each parameter and establish a parameter index according to the time series to ensure that each parameter can correspond one to one and at the same time enhance the efficiency of data calling.

[0065] The import module mainly uses linear interpolation to process discrete points. For known adjacent points (x0, y0) and (x1, y1), for x0<x<x1, we have:

[0066]

[0067] After interpolation, it is ensured that the time series of each parameter grows evenly.

[0068] Only when the import module receives the loop instruction issued by the output module, the import module will start to work and import the interpolated parameters and corresponding time series from the DCS into the storage module.

[0069] The import module arranges the irregular features according to the DCS raw data and uses an interpolation algorithm to make the data set more normalized, which helps in the fast and normalized storage of data.

[0070] Data storage module:

[0071] The data storage module is equivalent to a data transfer station, including three functions: data storage, data classification and data clearing. This module is used to temporarily store the data output from the import module and the function fitting module, and export the corrected data to the output module.

[0072] The data storage function is to ensure the efficiency of data import and export by building a small ordered database. In the storage module, data will be stored in the form of tuples, and all data are associated with time, which is convenient for the system to index the data. When the data in the corresponding time window is needed, a series of data can be called out by calling the time series. At the same time, the data storage area is divided into a historical data area and a real-time data area. The historical data area stores the qualified data and corresponding time series accumulated at each stage, and is usually directly combined with the fitting module. The real-time data area stores the time, density, logical signal and other information of the newly imported data. Through partitioned storage, the number of indexes can be reduced and work efficiency can be improved.

[0073] In order to prevent the storage module from being overloaded and slow in operation, the storage module usually cannot store too much data. Therefore, a data clearing function is added to the module to clean up some historical data and real-time data area data to ensure the module's operating efficiency. At the same time, it can reduce the possibility of overfitting in function simulation. The instruction to clean up the data will be given by the output module.

[0074] The data classification function is to distinguish whether the density data has been processed by the algorithm and the quality of the original data, so that other modules can quickly retrieve the required data from the data sample. Figure 4 As shown, it is mainly realized by adding two logic variables to each real-time data tuple. The two logic variables are called correction logic and quality logic. When the storage module receives the density data from the import module, both logic signals are set to low signals, and the data is placed in the real-time data area.

[0075] The correction logic is used to distinguish whether the data has been cleaned, fitted, etc., and to determine whether the data can be output to the DCS. When the correction logic signal is a low-order signal, it means that it has not been processed and cannot be output to the DCS; when the data storage module receives a high-order instruction from the data cleaning module or the function fitting module, the corresponding correction logic signal is set to a high-order signal, indicating that it has been processed and can be output to the DCS.

[0076] The quality logic is used to judge the quality of the density data when it is imported from the DCS, and to determine whether the density data can be used as a training set sample for the function fitting module. Only when the cleaning module issues a high-order instruction can the quality logic be converted into a high-order signal. After receiving the high-order instruction, the corresponding data will be copied from the real-time data area to the historical data area in sequence.

[0077] The data storage module is the hub of the overall correction system and interacts with other modules. A small database is built in the storage module. Other modules only need to transmit the processing results to the storage module in the form of logical signals. There is no need to transfer a large amount of data information to different modules, which greatly reduces the storage space and improves the computing efficiency.

[0078] The addition of data tuples and logic signals further improves the processing efficiency of the storage module. The direct docking of modules only requires the comparison of time nodes and the replacement of data units.

[0079] Data cleaning module:

[0080] The data cleaning module is used to distinguish data quality and is the main preprocessing link of the function fitting step. The data cleaning module receives data samples from the real-time data area and divides them into qualified data and unqualified data that need to be corrected. According to the mechanism, it cleans density data under unstable load conditions, abnormal pH conditions, and abnormal values ​​such as blank values, missing values, and over-limit values.

[0081] Among them, the steady-state extraction adopts the sliding window method. For a data set with a continuous time series, the window size and standard deviation threshold R are set as needed, the window is slid along the time series, and the standard deviation std of the data in the window is obtained to meet the conditions:

[0082] std<|R×mean|

[0083] Indicates that the data segment in the window is in a stable state. Wherein, mean represents the average value of the data in the window.

[0084] For the density value cleaning of abnormal pH, the range cleaning method is adopted. The upper and lower limits of pH value are set according to the unit operation parameters, and the density data within the range is retained.

[0085] The data judged as qualified outputs high-order instructions to the corresponding data in the data storage module, changes its two logic signals so that it can be output to the DCS. At the same time, it is determined that it can be used as a training set for function fitting, and unqualified data is eliminated. Only the time series of these data needs to be output to the function fitting module as a test set for function fitting.

[0086] The data cleaning module is a further preprocessing of the original data. Different from the definition of abnormal values ​​in traditional algorithms, the cleaning module also determines the fluctuating data caused by load and pH changes as unqualified data. The module combines physical mechanisms with intelligent algorithms, and the retained excellent data can better reflect the density changes in the system.

[0087] Function fitting module:

[0088] The function fitting module is mainly used to process and output unqualified data. Qualified data is input from the storage module as the prediction model training set. Based on the regression algorithm, the density change curve over a period of time is simulated, and the density change in the time window where the unqualified data is located is predicted. Then, the unqualified data in the storage module is replaced by the output.

[0089] The fitting function of the present invention mainly adopts Gaussian process regression in the regression algorithm. The basic idea is as follows: for the fitting function f(x) that conforms to the Gaussian distribution, there is:

[0090] f(x)~N(μ(x), k(x, x′))

[0091] Among them, μ(x) represents the mean function, and k(x, x′) represents the covariance function. The covariance function will take the form of a kernel function, and the specific expression is as follows:

[0092]

[0093] where δ ij is the Dirac function, θ0 and θ1 are hyperparameters. The maximum likelihood estimation expression of hyperparameters is:

[0094]

[0095] The predicted value is finally obtained by solving the hyperparameters θ0 and θ1.

[0096] by Figure 2 Taking the data sample shown as an example, through the analysis of the system mechanism, it can be seen that the desulfurization conditions have not changed significantly during this period. Therefore, a Gaussian process regression model with time as the variable and slurry density as the predicted quantity can be constructed to correct the density data.

[0097] like Figure 5 As shown in the figure, the broken line segment represents the original data before correction, and the smooth curve represents the slurry density curve after data screening and regression algorithm fitting. After correction by Gaussian process regression, the floating range of density data is reduced, and the quality of density data is greatly improved.

[0098] The density curve is self-learned according to the Gaussian process regression method, which reflects the density change law of the current desulfurization system.

[0099] The fitting module is the core of the entire correction system and takes the longest time to calculate. By storing the quality logic signal in the tuple, qualified data can be quickly retrieved as a training set. This includes qualified density data series and corresponding time series. These qualified data include historical data and real-time data. When the storage module receives the signal from the cleaning module, it only needs to fill in the good data in the real-time data in order, without retrieving historical data, thus reducing the number of indexes and storage space.

[0100] Output Module:

[0101] The functions of the output module include data export and providing loop instructions. When the storage module receives the correction data from the function fitting module, it will retrieve the correction logic of the real-time data. When the correction logic of all real-time data is a high signal, it will give a signal to the output module to start the function of the output module.

[0102] After the output module is started, it will send a cleanup instruction to the storage module, copy the real-time data to the output module, and output it to the DCS and liquid level measurement circuit to realize the data export function, and then combine the liquid level pressure difference to realize the liquid level measurement. After the copy is completed, the cleanup function of the storage module is started to clear the real-time data area and clean up some historical data, waiting for the input of the next data sample set.

[0103] After the data export is completed, a loop instruction will be issued to the import module, and the import module will export the next set of data samples from the DCS in chronological order into the real-time data area of ​​the storage module. Among them, the frequency of the loop instruction provided by the output module and the size of the data set exported by the import module need to be manually set according to the module operation speed, on-site requirements, etc.

[0104] In order to facilitate maintenance and adjustment of the correction system, a termination command is added to the branch of the loop instruction, which is manually controlled by the operator to start and stop the correction system. When the system is working in a loop, the command does not work. When the system needs to be stopped, the termination command will shield the loop instruction and terminate the loop fitting process.

[0105] This embodiment is only a preferred specific implementation of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A desulfurization system absorption tower liquid level measurement system based on Gaussian process regression, comprising a density measurement circuit, a liquid level measurement circuit, and a density data correction system between the two, wherein the liquid level measurement circuit comprises a differential pressure liquid level gauge; characterized in that: The density data correction system places the Gaussian process regression algorithm in the distributed control system in the form of a programmable logic controller to correct the density of the slurry measured by the density measurement circuit. The density data correction system includes a data import module, a data storage module, a data cleaning module, a function fitting module, and an output module; the density data in the distributed control system is temporarily stored in the data storage module after preliminary processing by the data import module; the data storage module includes a data storage area, which is divided into a historical data area and a real-time data area; the historical data area stores and accumulates qualified data and corresponding time series at each stage, and the real-time data area stores the time, value, and logic signal information of the newly imported data; the data cleaning module and the function fitting module clean and fit the data stored in the data storage module respectively Replace the original data to achieve data correction; the data cleaning module receives the data samples of the real-time data area and divides them into qualified data and unqualified data that need to be corrected; the function fitting module is used to process the unqualified data and output it; the function fitting module uses the qualified data input from the storage module as the prediction model training set, simulates the change curve of the slurry density over a period of time based on the Gaussian process regression algorithm, and predicts the density change in the time window where the unqualified data is located, and outputs the corrected data to replace the unqualified data in the storage module; the data import module receives the loop instruction issued by the output module to realize the cyclic density data correction; the final corrected density data is returned to the distributed control system by the output module and output to the liquid level measurement circuit to achieve accurate measurement of the liquid level.

2. The desulfurization system absorption tower liquid level measurement system based on Gaussian process regression according to claim 1, characterized in that: The data import module receives the original data of each parameter in the distributed control system, and makes the data of each parameter evenly distributed in time based on an interpolation algorithm.

3. The desulfurization system absorption tower liquid level measurement system based on Gaussian process regression according to claim 1, characterized in that: The data storage module includes data storage, data classification and data clearing functions, and is used to temporarily store the data output from the import module and the function fitting module, and export the corrected data therein to the output module.

4. The desulfurization system absorption tower liquid level measurement system based on Gaussian process regression according to claim 1, characterized in that: When the storage module receives the correction data from the function fitting module, it retrieves the correction logic of the real-time data. When the correction logics of all real-time data are high-order signals, it gives a signal to the output module to start the output module. After the output module is started, the storage module issues a cleanup instruction to copy the real-time data to the output module and output it to the distributed control system and the liquid level measurement circuit. After the copying is completed, the cleanup function of the storage module is started to clear the real-time data area and clean up some historical data, waiting for the input of the next data sample set.

5. The desulfurization system absorption tower liquid level measurement system based on Gaussian process regression according to claim 1, characterized in that: The data classification function sets correction logic and quality logic for each real-time data tuple. When the storage module receives the density data from the import module, the logic signals of the correction logic and the quality logic are set to low-order signals, and the density data is placed in the real-time data area. The correction logic is used to distinguish whether the density data has been processed by the algorithm, and the quality logic is used to judge the quality of the density data when it is imported from the distributed control system, so as to determine whether the density data can be used as a training set sample for the function fitting module.

6. A desulfurization system absorption tower liquid level measurement system based on Gaussian process regression according to claim 5, characterized in that: When the correction logic signal is a low-order signal, the density data cannot be output to the distributed control system; when the data storage module receives a high-order instruction from the data cleaning module or the function fitting module, the corresponding correction logic signal is set to a high-order signal, and the density data can be output to the distributed control system.

7. A desulfurization system absorption tower liquid level measurement system based on Gaussian process regression according to claim 6, characterized in that: Only when the data cleaning module issues a high-order instruction can the quality logic be converted into a high-order signal; after receiving the high-order instruction, the corresponding data will be copied from the real-time data area to the historical data area in sequence.

8. A method for measuring the liquid level of an absorption tower in a desulfurization system based on Gaussian process regression, characterized in that: The measurement is performed according to a desulfurization system absorption tower liquid level measurement system based on Gaussian process regression as described in claims 1-7.

Images