Method for monitoring and optimizing adjustment of sulfur trioxide emission control

Abstract

The application provides a monitoring and emission control optimization adjustment method for sulfur trioxide, and belongs to the technical field of optimization adjustment, and comprises the following steps: different types of sensors are used to synchronously monitor sulfur trioxide in flue gas of a supercritical boiler in real time, data preprocessing and data fusion processing are performed on the synchronous monitoring results, a correction model established in combination with parameters of the supercritical boiler is used to correct the processed data, historical monitoring data, historical operation parameters of the supercritical boiler and a historical operation state of flue gas treatment equipment are deeply mined, a sulfur trioxide control model is established, coordinated control instructions of the flue gas treatment equipment and desulfurization adjustment parameters in a desulfurization system are determined according to output results, optimization control adjustment is performed, a model is established, and output effective control results of analysis of the current monitoring results are obtained, and optimization and regulation of the equipment are intelligently realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of optimization adjustment, in particular to a monitoring and emission control optimization adjustment method for sulfur trioxide. BACKGROUND

[0002] With the increasingly stringent environmental requirements, the emission control of sulfur trioxide in the flue gas of a supercritical boiler becomes increasingly important. The emission of sulfur trioxide not only causes air pollution, forming acid rain, haze and other problems that harm the environment and human health, but also may cause corrosion and other adverse effects on the boiler and subsequent equipment.

[0003] At present, although there are some monitoring and control methods for sulfur trioxide in the flue gas of a supercritical boiler, there are still many deficiencies in practical application. For example, the traditional monitoring method may have insufficient accuracy and slow response speed, making it difficult to accurately reflect the emission situation in real time; the existing emission control technology often has poor synergy, and cannot fully utilize the advantages of each flue gas treatment facility, resulting in low efficiency of removing sulfur trioxide.

[0004] Therefore, the present application proposes a monitoring and emission control optimization adjustment method for sulfur trioxide. SUMMARY

[0005] The present application provides a monitoring and emission control optimization adjustment method for sulfur trioxide, which ensures the accuracy of the sulfur trioxide concentration based on the synchronous monitoring of three sensors, and further provides a reliable basis for subsequent control through data preprocessing and fusion, outputs effective control results through the establishment of a model and the analysis of the current monitoring results, and intelligently realizes the optimization and regulation of the equipment.

[0006] The present application provides a monitoring and emission control optimization adjustment method for sulfur trioxide, which includes:

[0007] Step 1: Real-time synchronous monitoring of sulfur trioxide in the flue gas of a supercritical boiler using different types of sensors;

[0008] Step 2: Data preprocessing and data fusion processing of the synchronous monitoring results, and correction of the processed data in combination with a correction model established based on the parameters of the supercritical boiler;

[0009] Step 3: Deep mining of historical monitoring data, historical operating parameters of the supercritical boiler and historical operating states of the flue gas treatment equipment, and establishment of a sulfur trioxide control model;

[0010] Step 4: Inputting the correction results into the sulfur trioxide control model, determining the coordinated control instructions of the flue gas treatment equipment and the desulfurization adjustment parameters in the desulfurization system according to the output results, and performing optimization control adjustment.

[0011] Preferably, the historical monitoring data, historical operating parameters of the supercritical boiler and historical operating states of the flue gas treatment device are deeply mined to establish a sulfur trioxide control model, including:

[0012] Obtain historical monitoring data of the supercritical boiler within a historical specified time period, wherein the historical monitoring information includes sulfur trioxide concentration information at each time point within the historical specified time period;

[0013] Divide the sulfur trioxide concentration information into a plurality of sub-information based on a segmented time window, wherein the sub-information includes concentration values of each to-be-measured point involved at a specified time point;

[0014] Match the historical operating parameters of the supercritical boiler and the historical operating states of the flue gas treatment device at each historical time point with the corresponding sub-information to obtain a first input sample;

[0015] Obtain the sulfur trioxide concentration threshold of the supercritical boiler, the standard operating parameters of the supercritical boiler and the theoretical working state of the flue gas treatment device at each historical concentration value, and compare and analyze the first input sample to obtain a first deviation sample;

[0016] Determine the regression coefficients of the first input sample and the first deviation sample at each sample element, and determine the regulation loss level according to the regression coefficients and the element type of the corresponding sample element;

[0017] Obtain the reference level of the corresponding sample element based on all regulation loss levels of the same sample element, and sort the improvement direction of all sample elements in order to obtain a first output sample;

[0018] Based on all first input samples and the first output sample corresponding to each first input sample, a sample set is constructed;

[0019] Train and test the neural network model according to the sample set to obtain a regulation model.

[0020] Preferably, determining the regulation loss level comprises:

[0021]

[0022] wherein, represents the regulation loss level of the corresponding sample element; represents the measured value of the i-th sample element in the first input sample; represents the theoretical value of the i-th sample element; represents the maximum value in the measured value of the i-th sample element among all first input samples; represents the minimum value in the measurement value of the i-th sample element under all first input samples; represents a preset regulatory loss coefficient of the h-th element type; represents the back-calibration coefficient of the i-th sample element.

[0023] Preferably, the training and testing of the neural network model according to the sample set are performed to obtain the regulatory model, which comprises:

[0024] The division ratio of the sample set is determined, and the sample set is divided based on the division ratio to obtain the training set and the test set;

[0025] The neural network model is trained according to the training set;

[0026] The trained model is tested according to the test set to obtain the regulatory model.

[0027] Preferably, the training of the neural network model according to the training set comprises:

[0028] Each training sample in the training set is received based on the input layer of the neural network;

[0029] Each training sample received by the multiple convolution layers and the multiple input layers of the nonlinear activation function of each convolution layer of the neural network is subjected to dimension extraction to determine the dimension control feature of each training sample;

[0030] Each training sample in the training set is subjected to noise extraction based on the multiple deep convolution layers of the neural network to determine the noise control feature of each training sample;

[0031] The dimension control feature and the noise control feature of each training sample in the training set are respectively subjected to feature reconstruction using the multiple deconvolution layers of the neural network to obtain a noise-free sample;

[0032] The difference value of each training sample and the corresponding noise-free sample is determined, and if all the difference values are less than a preset value, it is determined that the training of the neural network model is completed.

[0033] Preferably, the determination of the difference value of each training sample and the corresponding noise-free sample comprises:

[0034]

[0035]

[0036] wherein, represents the difference value of the u-th training sample and the corresponding noise-free sample; represents the dimension loss value of the u-th training sample; 、 Xi u represents the value of the i th sample element under the u th training sample and the u th noiseless sample respectively; N1 represents the total number of sample elements existing in each training sample; The difference fine-tuning coefficient is represented by 0.01. The expectation of the difference fine-tuning coefficient is represented by The expectation of the difference fine-tuning coefficient is represented by

[0037] Preferably, different types of sensors include: sensors based on controlled condensation method, sensors based on ion chromatography method and sensors based on optical absorption method.

[0038] Different types of sensors are combined and deployed at each monitoring position.

[0039] Preferably, the processed data is corrected in combination with a correction model established based on supercritical boiler parameters, including:

[0040] The supercritical boiler parameters in the operation process of the supercritical boiler are collected, wherein the supercritical boiler parameters include: boiler load, coal type, combustion temperature and excess air coefficient.

[0041] The initial mathematical relationship between the concentration value of sulfur trioxide and the supercritical boiler parameters is established by using the multivariate regression analysis method, and the correction coefficient between the concentration value of sulfur trioxide and each supercritical boiler parameter is determined by using the least square method, as the to-be-supplemented coefficient of the corresponding supercritical boiler parameter in the initial mathematical relationship, to obtain the correction model.

[0042] The concentration value in the processed data is corrected based on the correction model.

[0043] Compared with the prior art, the beneficial effects of the present application are as follows:

[0044] The synchronization monitoring based on the three types of sensors ensures the accuracy of the concentration of sulfur trioxide, and further provides a reliable basis for subsequent control through the preprocessing and fusion of data, outputs effective control results through the establishment of a model and the analysis of the current monitoring results, and intelligently realizes the optimization and regulation of the equipment. BRIEF DESCRIPTION OF DRAWINGS

[0045] In order to more clearly illustrate the technical solutions in the present application or prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor.

[0046] Figure 1 It is a flow chart of the method for monitoring and emission control optimization adjustment of sulfur trioxide in the embodiments of the present application. DETAILED DESCRIPTION

[0047] In order to make the objects, technical solutions and advantages of the present application clearer, the technical solutions in the present application will be described clearly and completely below in conjunction with the drawings in the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the protection scope of the present application.

[0048] The present application provides a monitoring and emission control optimization adjustment method for sulfur trioxide, as shown in the formula (I), comprising: Figure 1

[0049] Step 1: Real-time synchronous monitoring of sulfur trioxide in the flue gas of the supercritical boiler by using different types of sensors;

[0050] Step 2: Data preprocessing and data fusion processing are performed on the synchronous monitoring results, and the processed data is corrected in combination with a correction model established based on the parameters of the supercritical boiler;

[0051] Step 3: Deep mining is performed on the historical monitoring data, historical operation parameters of the supercritical boiler and historical operation state of the flue gas treatment equipment, and a sulfur trioxide control model is established;

[0052] Step 4: The correction result is input into the sulfur trioxide control model, and the collaborative control instruction of the flue gas treatment equipment and the desulfurization adjustment parameter in the desulfurization system are determined according to the output result for optimization control adjustment.

[0053] In this embodiment, three sensors are used to synchronously measure the flue gas in the boiler, and finally the concentration is averaged, that is, the concentration value monitored at the corresponding position is obtained, and the sensors are: sensors under control condensation method, ion chromatography method and optical absorption method.

[0054] In this embodiment, in the data preprocessing stage, 3σ principle is used to identify and eliminate abnormal data, and linear interpolation method is used to fill in the missing data; data fusion uses weighted average fusion algorithm, and weights are allocated according to the historical measurement accuracy of each sensor (such as control condensation method sensor weight 0.4, ion chromatography method sensor weight 0.3, optical absorption method sensor weight 0.3).

[0055] The correction model is pre-constructed.

[0056] In this embodiment, the boiler operation parameters (such as load, temperature, pressure), equipment operation state data (such as desulfurization tower pH value, dust removal efficiency, denitration ammonia escape rate) are taken as input, and the equipment control parameters (such as desulfurizer addition amount, dust removal electric field voltage, denitration ammonia injection amount) are taken as output, ​

[0057] In this embodiment, the corrected sulfur trioxide concentration data is input into the sulfur trioxide control model, and the model outputs device cooperative control instructions and desulfurization adjustment parameters. The control instructions are transmitted to the PLC control system through the Modbus protocol to realize the operation of devices such as air preheater temperature adjustment, desulfurization tower slurry circulating pump flow control, dust removal system pulse valve injection frequency adjustment, denitration system ammonia injection grid opening adjustment, and real-time optimization of sulfur trioxide emission control.

[0058] The beneficial effects of the above technical solution are: based on the synchronous monitoring of the three sensors to ensure the accuracy of the sulfur trioxide concentration, and further through the preprocessing and fusion of the data to provide a reliable basis for subsequent control, through the establishment of the model and the analysis of the current monitoring results to output effective control results, and intelligently realize the optimization and regulation of the device.

[0059] The present application provides a kind of for sulfur trioxide monitoring and emission control optimization adjustment method, to historical monitoring data, supercritical boiler historical operating parameter and flue gas treatment equipment historical operating state are deeply mined, establish sulfur trioxide control model, including:

[0060] Obtain the historical monitoring data of supercritical boiler in historical specified time period, wherein the historical monitoring information includes the sulfur trioxide concentration information of each time point in the historical specified time period;

[0061] Sulfur trioxide concentration information is divided into multiple sub-information based on segmented time window, wherein the sub-information includes the concentration value of each to-be-measured point involved at specified time point;

[0062] The historical operating parameters of supercritical boiler and the historical operating state of flue gas treatment equipment at each historical time point are matched with the corresponding sub-information, to obtain a first input sample;

[0063] Obtain the sulfur trioxide concentration threshold of the supercritical boiler, the standard operating parameters of the supercritical boiler and the theoretical working state of the flue gas treatment equipment at each historical concentration value, and compare and analyze the first input sample to obtain a first deviation sample;

[0064] Determine the regression coefficient of the first input sample and the first deviation sample at each sample element, and determine the regulation loss level according to the regression coefficient and the element type of the corresponding sample element;

[0065] Based on all regulation loss levels of the same sample element, the reference level of the corresponding sample element is obtained, and the improvement direction of all sample elements is sorted in order to obtain a first output sample;

[0066] Based on all the first input samples and the first output samples corresponding to each first input sample, a sample set is constructed;

[0067] The neural network model is trained and tested according to the sample set to obtain a regulation model.

[0068] Preferably, determining the regulation loss level comprises:

[0069]

[0070] wherein, represents the regulation loss level of the corresponding sample element; represents the measured value of the i-th sample element in the first input sample; represents the theoretical value of the i-th sample element; represents the maximum value of the i-th sample element with respect to the measured value under all first input samples; represents the minimum value of the i-th sample element with respect to the measured value under all first input samples; represents the preset regulation loss coefficient of the h-th element type; represents the back-calibration coefficient of the i-th sample element.

[0071] In this embodiment, the period from January 1, 2024 to January 31, 2024 is selected as the historical specified time period, and the sulfur trioxide concentration information at every 5-minute time point in this period is successfully obtained, totaling 8784 data records. The sulfur trioxide concentration shows fluctuating changes in different time periods, with the lowest concentration being 8 mg / m³ and the highest concentration reaching 18 mg / m³, providing a rich data basis for subsequent analysis.

[0072] The continuous sulfur trioxide concentration information is divided into multiple sub-information segments according to a certain time interval (i.e., a segmented time window). Each sub-information contains the sulfur trioxide concentration values of each test point (such as sensor installation points at different positions of the flue) at the specified time point. This division facilitates subsequent fine-grained analysis and processing of the data and mining of concentration change characteristics in different time periods. With 1 hour as the segmented time window, the sulfur trioxide concentration information obtained in the above experiment is divided into 448 sub-information. Through analysis, it is found that the concentration values of each test point in different sub-information differ, and the difference in sulfur trioxide concentration between the upstream and downstream of the flue in some sub-information is large, reflecting the diffusion and reaction of flue gas in the flue, and providing effective data units for studying the concentration distribution law.

[0073] The historical operation parameters of the supercritical boiler (such as boiler load, combustion temperature, steam pressure, etc.) and the historical operation state of the flue gas treatment equipment (such as the pH value of the desulfurization tower, the electric field strength of the dust removal equipment, the ammonia injection amount of the denitration system, etc.) are associated and matched with the sulfur trioxide concentration sub-information at the corresponding time point. Ensure that each sub-information corresponds to the same time point under the complete boiler operation condition and equipment state data, and combine to form a first input sample as the basis data set for subsequent analysis.

[0074] In the matching process, each sub-information is successfully associated with the corresponding boiler operation parameters and equipment state data, forming 448 first input samples. For example, in a certain sub-information, the sulfur trioxide concentration is 12 mg / m³, the corresponding boiler load is 500 MW, the combustion temperature is 1300°C, the desulfurization tower pH value is 5.8, and the dust removal equipment electric field strength is 35 kV, etc. These data collectively constitute a complete first input sample, clearly presenting the running condition of the system at that time point.

[0075] Determine the sulfur trioxide concentration environmental protection emission standard or process requirement threshold of the supercritical boiler, and the standard operation parameter range of the boiler and the ideal working state parameter of the flue gas treatment equipment under different sulfur trioxide concentrations. Then compare the actual data in the first input sample with these standard or theoretical data, calculate the deviation value of each data item, and obtain the first deviation sample to reflect the difference between the actual operation and the ideal state.

[0076] After comparison and analysis, it is found that among the 448 first input samples, 87 samples have sulfur trioxide concentration exceeding the threshold value, part of the boiler operation parameters deviating from the standard range, and the flue gas treatment equipment running state not reaching the theoretical best state. For example, in a certain sample, the sulfur trioxide concentration is 16 mg / m³ (threshold value is 15 mg / m³), the boiler load exceeds the standard range by 10%, and the desulfurization tower pH value is 0.3 lower than the theoretical value. Record these deviation data to form the first deviation sample, which directly shows the problem points in the operation.

[0077] The back adjustment coefficient is used to measure the cost or difficulty of adjusting the data elements in the first input sample back to the ideal state after deviating from the standard or theoretical value. According to the nature of different data elements (such as sulfur trioxide concentration, boiler load, etc.), combined with the size of the back adjustment coefficient, the sample elements are divided into different control loss levels, such as low loss level, medium loss level and high loss level, in order to subsequently formulate targeted control strategies.

[0078] In the comparison of the first input sample and the first deviation sample, the calibration coefficient of each sample element is determined. For example, for the sulfur trioxide concentration element, if the calibration coefficient is 0.8 (the larger the value, the more difficult the adjustment), according to its element type and the set grading standard, it is divided into a high loss level; for some device operating state parameters, the calibration coefficient is 0.3, which is divided into a low loss level. Finally, all sample elements are classified into different control loss levels, providing a quantitative basis for optimization and adjustment.

[0079] For the same type of sample element (such as all sulfur trioxide concentration elements), the reference level of the sample element is calculated by comprehensively considering the control loss level of each first input sample, to reflect the overall control difficulty level of the element. Then, the improvement priority of all sample elements is sorted according to the reference level, to determine which elements to adjust first to more efficiently improve system performance. The sorting results and related information are arranged to form a first output sample. The reference level of the sulfur trioxide concentration element is high, the reference level of the boiler load element is medium, and the reference level of the desulfurization tower pH value element is low. According to the reference level, the order of the improvement direction is determined as follows: the sulfur trioxide concentration related parameters are adjusted first, followed by the boiler load, and finally the desulfurization tower pH value and other parameters.

[0080] Each sample pair (first input sample and first output sample) represents an actual operating state and its corresponding optimization direction, and the sample set covers various working conditions and optimization strategies in the historical operation, providing sufficient data support for the training of the subsequent neural network model. The sample set includes optimization schemes for high load and high concentration sulfur trioxide emission conditions, as well as adjustment strategies for low load and low concentration emission conditions, providing rich data resources for the model to learn complex operating rules and optimization logic.

[0081] The neural network is, for example, a BP neural network or an LSTM neural network.

[0082] The beneficial effects of the above technical solution are: the deviation value of each data item is calculated to obtain the first deviation sample, which is used to reflect the difference between the actual operating condition and the ideal state, and all sample elements are classified into different control loss levels, providing a quantitative basis for optimization and adjustment. The rich samples provide rich data resources for the model to learn complex operating rules and optimization logic.

[0083] The present application provides a method for monitoring and optimizing adjustment of sulfur trioxide emission control, which trains and tests a neural network model according to the sample set to obtain a control model, including:

[0084] The division ratio of the sample set is determined, and the sample set is divided based on the division ratio to obtain a training set and a test set;

[0085] Train the neural network model according to the training set;

[0086] Test the trained model according to the test set to obtain a regulation model.

[0087] The above technical scheme has the beneficial effects that the proportion of the sample set is divided to effectively ensure the training and testing of the model.

[0088] The application provides a method for monitoring and optimizing adjustment of sulfur trioxide emission control, the neural network model is trained according to the training set, and the method comprises the following steps:

[0089] Each training sample in the training set is received based on an input layer of the neural network;

[0090] Each training sample received by a plurality of convolution layers and a plurality of input layers of a nonlinear activation function of each convolution layer based on the neural network is subjected to dimension extraction to determine the dimension control features of each training sample;

[0091] Each training sample in the training set is subjected to noise extraction based on a plurality of deep convolution layers of the neural network to determine the noise control features of each training sample;

[0092] The dimension control features and the noise control features of each training sample in the training set are subjected to feature reconstruction by using a plurality of deconvolution layers of the neural network respectively to obtain a noiseless sample;

[0093] The difference values of each training sample and the corresponding noiseless sample are determined, and if all the difference values are less than a preset value, it is determined that the training of the neural network model is completed.

[0094] Preferably, the difference values of each training sample and the corresponding noiseless sample are determined, and the determination comprises the following steps:

[0095]

[0096]

[0097] wherein, represents the difference value of the u-th training sample and the corresponding noiseless sample; represents the dimension loss value of the u-th training sample; 、 respectively represent the value of the i-th sample element under the u-th training sample and the u-th noiseless sample; N1 represents the total number of sample elements existing in each training sample; represents a difference fine-tuning coefficient, and the value is 0.01; represents the expectation.

[0098] In this embodiment, the input layer is the interface through which the neural network interacts with external data. Its role is to convert each training sample in the training set (in the context of supercritical boiler sulfur trioxide emission control, the training sample is a sample pair composed of the previously constructed first input sample and first output sample) into a numerical form that the neural network can process and pass it on to the subsequent network layers. The number of neurons in the input layer is usually consistent with the number of features of the training sample. For example, if a training sample contains 10 features such as boiler load, sulfur trioxide concentration, desulfurization tower pH value, etc., the input layer has 10 neurons, each neuron corresponding to a feature value. In the training experiment of the supercritical boiler sulfur trioxide emission control model, the training set contains 358 sample pairs, and the input layer successfully receives and converts all training samples. Taking one of the samples as an example, the input layer accurately passes the feature values such as boiler load 520 MW, sulfur trioxide concentration 14 mg / m³, desulfurization tower pH value 5.6, etc. to the subsequent network layers.

[0099] After being processed by 3 convolutional layers and ReLU activation functions, the dimension control features are successfully extracted from the training samples. For example, a training sample originally containing multiple complex features is processed to obtain dimension control features such as boiler load trend features and sulfur trioxide concentration fluctuation pattern features. Through visual analysis, it is found that these features can clearly show the internal structure of the data in different dimensions, and are more representative than the original data, providing more effective information for subsequent model learning.

[0100] Using 5 deep convolutional layers to process the training samples, noise control features are successfully extracted. Analysis shows that the noise in some training samples mainly comes from abnormal data fluctuations caused by sensor measurement errors and interference during data transmission. For example, in some samples, the sulfur trioxide concentration has a short-term abnormal jump, and the deep convolutional layer identifies this abnormal change as noise and forms the corresponding noise control feature.

[0101] After 2 deconvolutional layers are used to reconstruct the dimension control features and noise control features, noise-free samples are obtained. Comparing the original training samples with the noise-free samples, it is found that the data in the noise-free samples is smoother and more reasonable, and the abnormal fluctuations and noise interference in the original samples are removed. For example, in a sample, the original sulfur trioxide concentration data has a sharp fluctuation caused by noise, and after processing by the deconvolutional layer, the concentration data trend is more consistent with the actual operation rules, providing higher quality data for model training.

[0102] During the training process, the mean square error is set as the measurement method of the difference value, and the preset value is 0.01. With the training, the mean square error of the training sample and the noise-free sample is calculated. When the training reaches the 45th cycle, the mean square error of all training samples is less than 0.01, which meets the preset condition, and it is determined that the neural network model training is completed.

[0103] The preset value is a threshold value preset in the model training process, which is used as a standard for judging whether the model training is completed, and is reasonably set according to the actual application scene and requirements.

[0104] The beneficial effects of the above technical solutions are: the noise-free sample can more accurately reflect the true characteristics and laws of the data than the original sample, which helps to improve the model training effect. Through the verification of the test set, the model also shows good prediction performance on the test set, proving that the trained model has good generalization ability. By comparing the difference value of the training sample and the noise-free sample with the preset value, it is determined whether the neural network model reaches sufficient training precision and performance. When the condition is met, it is determined that the model training is completed and can be put into use.

[0105] The application provides a monitoring and emission control optimization adjustment method for sulfur trioxide, different types of sensors include: sensors based on controlled condensation method, sensors based on ion chromatography method and sensors based on optical absorption method.

[0106] Different types of sensors are combined and deployed at each monitoring position.

[0107] The beneficial effects of the above technical solutions are: by synchronously deploying 3 types of sensors, the accuracy of measurement can be ensured.

[0108] The application provides a monitoring and emission control optimization adjustment method for sulfur trioxide, and a correction model established in combination with supercritical boiler parameters is used to correct the processed data, including:

[0109] The supercritical boiler parameters in the operation process of the supercritical boiler are collected, wherein the supercritical boiler parameters include: boiler load, coal type, combustion temperature and excess air coefficient.

[0110] A multivariate regression analysis method is used to establish an initial mathematical relationship between the concentration value of sulfur trioxide and the supercritical boiler parameters, and a least square method is used to determine the correction coefficient between the concentration value of sulfur trioxide and each supercritical boiler parameter, as the to-be-supplemented coefficient of the corresponding supercritical boiler parameter in the initial mathematical relationship, to obtain the correction model.

[0111] The concentration value in the processed data is corrected based on the correction model.

[0112] In this embodiment, real-time data is collected by sensors, instruments and coal quality analysis equipment installed at various parts of the boiler. For example, the boiler load can be calculated by monitoring parameters such as steam flow and pressure; the coal type is determined by coal quality analysis to determine its sulfur content, volatile matter and other characteristics; the combustion temperature is measured by temperature sensors such as thermocouples inside the furnace; and the excess air coefficient is calculated based on flue gas composition analysis data (such as oxygen content). These parameters will serve as basic variables for establishing mathematical relationships subsequently.

[0113] In a 72-hour continuous operation monitoring experiment of a 600 MW supercritical boiler, complete supercritical boiler parameter data was successfully collected. Among them, the boiler load fluctuates between 400-600 MW, involving 3 different coal types (low-sulfur coal, medium-sulfur coal, and high-sulfur coal), the combustion temperature ranges from 1200-1400℃, and the excess air coefficient varies between 1.1-1.3. These data provide a rich sample for subsequent analysis.

[0114] The initial mathematical relationship is: ye = b0 + b1f1 + b2f2 + b3f3 + b4f4 + p0, where ye represents the concentration value, f1, f2, f3, and f4 represent the boiler load, coal type, combustion temperature, and excess air coefficient, respectively, b0 is the constant term; b1 to b4 are the correction coefficients to be determined, and p0 is the random error term. The correction model obtained is: ye = -5.2 + 0.012f1 + 2.5f2 + 0.003f3 - 1.8f4 + 0.01.

[0115] The sulfur trioxide concentration monitoring data is corrected. Taking one time point as an example, the original monitoring concentration is 12 mg / m³, the boiler load is 550 MW, the coal type is high-sulfur coal (code 3), the combustion temperature is 1350℃, and the excess air coefficient is 1.2. Substituting into the correction model, the corrected concentration value is 13.8 mg / m³. Compared with the laboratory offline analysis data, the relative error of the corrected data and the actual concentration is reduced from 15% to 5%, significantly improving the accuracy of the sulfur trioxide concentration monitoring data.

[0116] The beneficial effects of the above technical solution are: by constructing a mathematical relationship, the relationship between the supercritical boiler parameters and the sulfur trioxide concentration value can be quantitatively described, the original concentration monitoring value is adjusted, the measurement error caused by the change of the boiler operating parameters is eliminated, and the corrected data that more accurately reflects the actual sulfur trioxide concentration is obtained.

[0117] It should be pointed out finally that the above embodiments are only used to illustrate the technical solutions of the present application, but not to limit the same; and although the present application has been described in detail with reference to the foregoing embodiments, it should be appreciated by those skilled in the art that the technical solutions recorded in the foregoing embodiments can be modified, or some technical features thereof can be replaced equivalently; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for monitoring and emission control optimization adjustment of sulfur trioxide, characterized in that, The method comprises the following steps: Step 1: Real-time synchronous monitoring of sulfur trioxide in the flue gas of a supercritical boiler by using different types of sensors; Step 2: Data preprocessing and data fusion processing of the synchronous monitoring results, and correction of the processed data in combination with a correction model established based on the parameters of the supercritical boiler; Step 3: Deep mining of historical monitoring data, historical operating parameters of the supercritical boiler and historical operating states of the flue gas treatment equipment, and establishment of a sulfur trioxide control model; Step 4: Inputting the correction results into the sulfur trioxide control model, determining the coordinated control instructions of the flue gas treatment equipment and the desulfurization adjustment parameters in the desulfurization system according to the output results, and performing optimization control adjustment; The deep mining of the historical monitoring data, the historical operating parameters of the supercritical boiler and the historical operating states of the flue gas treatment equipment, and the establishment of the sulfur trioxide control model comprise: Obtaining historical monitoring data of the supercritical boiler within a historical specified time period, wherein the historical monitoring data comprises sulfur trioxide concentration information at each time point within the historical specified time period; Dividing the sulfur trioxide concentration information into a plurality of sub-information based on a segmented time window, wherein the sub-information comprises the concentration value of each to-be-measured point at a specified time point; Matching the historical operating parameters of the supercritical boiler and the historical operating states of the flue gas treatment equipment at each historical time point with the corresponding sub-information to obtain a first input sample; Obtaining the sulfur trioxide concentration threshold of the supercritical boiler, the standard operating parameters of the supercritical boiler and the theoretical working state of the flue gas treatment equipment at each historical concentration value, and comparing and analyzing the first input sample to obtain a first deviation sample; Determining the regression coefficients of the first input sample and the first deviation sample at each sample element, and determining the regulation and control loss level according to the regression coefficients and the element type of the corresponding sample element; Obtaining the reference level of the corresponding sample element based on all regulation and control loss levels of the same sample element, and sequentially sorting the improvement direction of all sample elements to obtain a first output sample; Based on all the first input samples and the first output samples corresponding to each first input sample, a sample set is constructed; Training and testing a neural network model according to the sample set to obtain a regulation and control model.

2. The method for monitoring and emission control optimization adjustment of sulfur trioxide according to claim 1, characterized in that, Determining the regulation and control loss level comprises: wherein, represents a regulation loss level corresponding to a sample element; represents a measured value of an i-th sample element in the first input sample; represents a theoretical value of the i-th sample element; represents a maximum value of the i-th sample element with respect to the measured value among all first input samples; represents a minimum value of the i-th sample element with respect to the measured value among all first input samples; represents a preset regulation loss coefficient of an h-th element type; represents a back-calibration coefficient of the i-th sample element.

3. The method for monitoring and emission control optimization adjustment of sulfur trioxide according to claim 2, characterized in that, Training and testing a neural network model according to the sample set to obtain a regulation and control model comprises: Determining the division ratio of the sample set, dividing the sample set based on the division ratio to obtain a training set and a test set; Training a neural network model according to the training set; Testing the trained model according to the test set to obtain a regulation and control model.

4. The method for monitoring and emission control optimization adjustment of sulfur trioxide according to claim 3, characterized in that, Training a neural network model according to the training set comprises: Receiving each training sample in the training set based on the input layer of the neural network; Performing dimension extraction on each training sample received by the multiple convolution layers and the multiple input layers of the nonlinear activation function of each convolution layer of the neural network to determine the dimension control features of each training sample; The neural network-based multiple deep convolutional layers extract noise from each training sample in the training set to determine noise control features of each training sample; The multiple deconvolutional layers of the neural network are used to respectively reconstruct the dimensional control features and the noise control features of each training sample in the training set to obtain noise-free samples; Differences between each training sample and the corresponding noise-free sample are determined, and if all the differences are less than a preset value, it is determined that the neural network model is trained.

5. The method for monitoring and emission control optimization adjustment of sulfur trioxide according to claim 3, wherein, Determining the difference between each training sample and the corresponding noise-free sample includes: wherein, represents the difference value of the u-th training sample and the corresponding noise-free sample; represents the dimension loss value of the u-th training sample; , respectively represent the value of the i-th sample element under the u-th training sample and the u-th noise-free sample; N1 represents the total number of sample elements existing in each training sample; represents the difference fine-tuning coefficient, and the value is 0.01; represents the expectation of .

6. The method for monitoring and emission control optimization adjustment of sulfur trioxide according to claim 1, wherein, Different types of sensors include sensors based on controlled condensation method, sensors based on ion chromatography method, and sensors based on optical absorption method; Different types of sensors are combined and deployed at each monitoring location.

7. The method for monitoring and emission control optimization adjustment of sulfur trioxide according to claim 1, wherein, The processed data is corrected by combining a correction model established based on the supercritical boiler parameters, including: Collecting supercritical boiler parameters during the operation of the supercritical boiler, wherein the supercritical boiler parameters include boiler load, coal type, combustion temperature, and excess air coefficient; Using a multiple regression analysis method, an initial mathematical relationship between the concentration value of sulfur trioxide and the supercritical boiler parameters is established, and a least squares method is used to determine the correction coefficient between the concentration value of sulfur trioxide and each supercritical boiler parameter, as the to-be-supplemented coefficient of the corresponding supercritical boiler parameter in the initial mathematical relationship, to obtain the correction model; Based on the correction model, the concentration value in the processed data is corrected.

Images