Real-time early warning and operation evaluation system of thermal power wet desulphurization device

Abstract

The invention relates to the technical field of thermal power plant environmental protection, in particular to a real-time early warning and operation evaluation system of a thermal power wet desulphurization device, which comprises a real-time early warning subsystem and an operation evaluation subsystem, and the intelligent data collector performs preliminary screening and storage. The intelligent diagnosis module analyzes parameters, judges possible fault types and positions, the prediction model adopts a deep neural network, inputs a diagnosis result and outputs a real-time slurry pH value, density and density difference at a slurry outlet of an absorption tower, the intelligent evaluation module evaluates a final alarm level, and the decision scheme module generates a processing suggestion. The intelligent control execution module converts the data into a control signal, the field alarm module gives out a corresponding alarm, the intelligent query module displays a decision recommendation level on a webpage, the establishment of a prediction model comprises data sampling, model establishment, verification application and update optimization, and the prediction precision is remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of environmental protection in thermal power plants, and particularly to a real-time early warning and operation evaluation system for a wet flue gas desulfurization device in thermal power plants. Background Art

[0002] With the increasingly strict national environmental protection requirements, the stable operation of the desulfurization system in thermal power plants has become particularly important. At present, the wet flue gas desulfurization process is widely used in thermal power plants, which has the advantages of high desulfurization efficiency and mature technology. However, in the actual operation process, due to factors such as fluctuations in flue gas flow and SO2 concentration, equipment failures, and deviations in operating parameters, the desulfurization efficiency often decreases or even exceeds the standard emissions.

[0003] The existing monitoring systems for desulfurization devices mainly make simple judgments based on fixed thresholds, and can only issue alarms for obvious abnormal situations, unable to early warn of potential risks. At the same time, such systems often lack the comprehensive evaluation ability of the operating status of desulfurization devices and cannot provide targeted optimization suggestions for operating personnel. In addition, most of the existing systems use simple linear models or rule-based judgment methods, which are difficult to adapt to complex situations such as changes in working conditions and equipment aging, resulting in delayed alarms or frequent false alarms.

[0004] Therefore, it is urgent to develop a system that can real-time early warn and comprehensively evaluate the operating status of wet flue gas desulfurization devices in thermal power plants to improve the operating efficiency and stability of desulfurization devices. Summary of the Invention

[0005] The purpose of the present invention is to provide a real-time early warning and operation evaluation system for a wet flue gas desulfurization device in thermal power plants, which can real-time monitor the operating status of the desulfurization device, early warn of potential failures, and comprehensively evaluate the operating quality, providing decision-making support for operating personnel.

[0006] The present invention proposes a real-time early warning and operation evaluation system for a wet flue gas desulfurization device in thermal power plants, including:

[0007] A real-time early warning subsystem and an operation evaluation subsystem, the real-time early warning subsystem includes:

[0008] A flue gas composition monitoring device for collecting flue gas flow and flue gas SO2 concentration parameters during the operation of the wet flue gas desulfurization device in thermal power plants;

[0009] An intelligent data collector connected to the flue gas composition monitoring device for receiving the parameters collected by the flue gas composition monitoring device and preliminarily screening and storing the parameters;

[0010] An intelligent diagnosis module connected to the intelligent data collector for analyzing the parameters transmitted by the intelligent data collector and judging the possible types and locations of failures;

[0011] A prediction model, connected to the intelligent diagnosis module, is used to take the diagnosis result of the intelligent diagnosis module as the input of the model, and take the real-time slurry pH value, real-time slurry density, and real-time slurry density difference at the outlet position of the absorption tower corresponding to the diagnosis result as the output of the model;

[0012] An intelligent evaluation module, connected to the prediction model, is used to evaluate the final alarm level of the desulfurization device according to the output of the prediction model;

[0013] A decision-making scheme module, connected to the intelligent evaluation module, is used to generate treatment suggestions according to the final alarm level;

[0014] An intelligent control execution module, connected to the decision-making scheme module, is used to receive the treatment suggestions of the decision-making scheme module and convert them into control signals;

[0015] A field alarm module, connected to the intelligent evaluation module, is used to issue corresponding-level alarms on-site according to the final alarm level;

[0016] An intelligent query module, connected to the prediction model, is used to display the corresponding decision recommendation level on the web page;

[0017] Among them, the prediction model is a deep neural network model, and the establishment method of the prediction model includes the following steps:

[0018] S1. Data sampling: Use the flue gas composition monitoring device to collect flue gas flow X and flue gas SO2 concentration Y parameters during the normal operation of the desulfurization device to obtain a sample set;

[0019] S2. Model construction: Divide the sample set in step S1 into a training set and a test set according to a preset ratio, and use the training set to construct a deep neural network model; Input the flue gas flow X in the test set into the deep neural network model, and the output is the predicted flue gas SO2 concentration Y'', compare the predicted flue gas SO2 concentration Y'' with the flue gas SO2 concentration Y in the test set, calculate the mean square error MSE, and optimize the parameters of the deep neural network model according to the value of the MSE to obtain an optimized deep neural network model;

[0020] S3. Model verification and application: Use the sample set collected in step S1 to verify the optimized deep neural network model obtained in step S2. After passing the verification, use the optimized deep neural network model as the prediction model;

[0021] S4. Update and optimization: During the application process, when the number of samples in the sample set in step S1 exceeds the threshold, a new test set is formed using the flue gas flow rate X and flue gas SO2 concentration Y that exceed the threshold. The optimized deep neural network model in step S2 is further optimized using the new training set to obtain a new optimized deep neural network model as the prediction model applied in step S3.

[0022] Preferably, the decision recommendation levels include the highest priority system alarm, the priority system alarm, and the secondary priority system alarm.

[0023] Preferably, when the final alarm level in step S3 reaches any one or more of the highest priority system alarm, the priority system alarm, and the secondary priority system alarm, the system alarms of the corresponding levels are all given on-site alarms. When giving an alarm, the prediction model accumulates the alarm times of the corresponding on-site alarm devices as 1.

[0024] Preferably, when the mean square error MSE of the optimized deep neural network model in step S3 is less than or equal to 0.01, the optimized deep neural network model is obtained.

[0025] Preferably, the intelligent evaluation module evaluates the diagnostic result of the intelligent diagnosis module. When the diagnostic result is an alarm, the prediction model automatically generates final alarms of three levels, namely the highest priority system alarm, the priority system alarm, and the secondary priority system alarm, according to the real-time slurry pH value, real-time slurry density, and real-time slurry density difference at the corresponding position of the diagnostic result. The intelligent control execution module sends fault instructions of three levels to the on-site alarm module according to the final alarms of the three levels. The fault instructions are composed of the highest priority system alarm, the priority system alarm, and the secondary priority system alarm.

[0026] Preferably, the final alarms of the three levels of the highest priority system alarm, the priority system alarm, and the secondary priority system alarm respectively correspond to fault instructions of three levels. Among them, the fault instruction of the highest priority system alarm is that the slurry supply port of the absorption tower fails and the slurry discharge port of the absorption tower fails; the fault instruction of the priority system alarm is that the slurry supply port and the slurry discharge port of the absorption tower fail simultaneously; the fault instruction of the secondary priority system alarm is that the slurry supply port of the absorption tower fails and the slurry discharge port of the absorption tower is normal.

[0027] Preferably, the prediction model generates a decision plan recommendation level according to the diagnostic result of the intelligent diagnosis module corresponding to the final alarm. The decision plan recommendation level is composed of the highest priority plan, the priority plan, and the secondary priority plan; the intelligent query module simultaneously displays the highest priority plan, the priority plan, and the secondary priority plan on the web page and the background operation according to the decision plan recommendation level.

[0028] Preferably, the intelligent query module includes a parameter query unit, a fault warning unit, and a historical query unit. The parameter query unit includes the real-time slurry density of the absorption tower, the real-time slurry pH value of the absorption tower, the real-time slurry density difference of the absorption tower, the real-time absorption slurry circulation pump flow rate, and the real-time alarm count of the on-site alarm device. The fault warning unit includes the highest-priority system alarm, the priority system alarm, the sub-priority system alarm, and a suggested solution. The historical query unit includes a fault history query and an effect history query. The fault history query is divided into the last month, the last three months, and the last six months according to the alarm count of the absorption tower; the fault history query generates corresponding table titles according to the alarm count of the absorption tower; the effect history query is divided into queries for the last month, the last three months, and the last six months.

[0029] Preferably, the decision-making solution module includes: the first absorption tower slurry density grade and the first flue gas inlet SO2 concentration grade. According to the first absorption tower slurry density grade and the first flue gas inlet SO2 concentration grade, the decision-making solution module generates different priority solutions. The first absorption tower slurry density is obtained by using the slurry pH value and slurry density at the absorption tower outlet slurry, and the first flue gas inlet SO2 concentration grade is obtained by using the flue gas inlet gas flow rate and flue gas SO2 concentration.

[0030] Preferably, the operation evaluation subsystem includes a data collector, an evaluation module, a data processor, and a display module. The data collector collects equipment operation data and sends it to the data processor. The equipment operation data includes the flue gas inlet SO2 concentration and the flue gas inlet flow rate. The data processor processes the equipment operation data to obtain an equipment operation data evaluation grade. The equipment operation data evaluation grade is obtained by the data processor. When the equipment operation data evaluation grade is the highest-priority solution, the display module issues an alarm prompt message. When the equipment operation data grade is the highest-priority solution, the display module issues a warning prompt message. When the equipment operation data grade is the priority solution, the display module issues a poor operation prompt message. When the equipment operation data evaluation grade is the sub-priority solution, the display module issues a good prompt message. Among them, for the equipment operation data evaluation grade of the highest-priority solution, the SO2 concentration is greater than or equal to 2000 mg / Nm 3 , and the flue gas inlet flow rate is greater than or equal to 1000000 Nm 3 / h. For the equipment operation data evaluation grade of the priority solution, the SO2 concentration is greater than or equal to 1000 mg / Nm 3 and less than 2000 mg / Nm 3 , and the flue gas inlet flow rate is greater than or equal to 100000 Nm 3 / h and less than 1000000 Nm3 / h, the SO2 concentration of the secondary priority scheme is less than 1000 mg / Nm 3 , and the flue gas inlet flow rate is greater than or equal to 10000 Nm 3 / h and less than 100000 Nm 3 / h.

[0031] The beneficial effects of the present invention include:

[0032] 1. By applying the deep neural network model, the prediction accuracy is significantly improved, and the mean square error is controlled within 0.01, and the prediction accuracy rate is increased by more than 30% compared with the traditional linear model;

[0033] 2. Adopting the model update mechanism triggered by the sample threshold enables the model to adapt to the changes in working conditions and equipment aging, and the model accuracy decay rate is reduced to less than 5% during long-term operation;

[0034] 3. Innovatively establishing a two-way mapping relationship between the diagnostic results and physical parameters, and the abnormal capture rate is increased by more than 40% compared with the traditional one-way prediction;

[0035] 4. Automatically generating a multi-level decision-making scheme based on the combined state of key process parameters, reducing the abnormal handling time by 30% - 50% and increasing the handling success rate by more than 25%;

[0036] 5. Adopting a three-level alarm mechanism to accurately distinguish the severity of faults, reducing the false alarm rate by more than 15% and improving the alarm accuracy;

[0037] 6. The overall system realizes the closed-loop management from data collection to intelligent decision-making, reduces the manual intervention by more than 80%, and significantly improves the system operation efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 It is a schematic diagram of the overall architecture of the system of the present invention;

[0039] Figure 2 It is a connection relationship diagram of the function modules of the real-time early warning subsystem of the present invention;

[0040] Figure 3 It is a schematic diagram of the structure of the deep neural network prediction model of the present invention;

[0041] Figure 4 It is a flow chart for establishing the prediction model of the present invention;

[0042] Figure 5 It is a flow chart for recommending the decision-making scheme of the present invention;

[0043] Figure 6 It is a connection relationship diagram of the function modules of the operation evaluation subsystem of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0044] Please refer to the attached Figure 1-6 , and the specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0045] As Figure 1 shown, the present invention provides a real-time warning and operation evaluation system for a thermal power wet flue gas desulfurization device, which includes a real-time warning subsystem and an operation evaluation subsystem.

[0046] As Figure 2 shown, the real-time warning subsystem includes a flue gas composition monitoring device 1, an intelligent data collector 2, an intelligent diagnosis module 3, a prediction model 4, an intelligent evaluation module 5, a decision-making scheme module 6, an intelligent control execution module 7, a field alarm module 8, and an intelligent query module 9.

[0047] The flue gas composition monitoring device 1 is installed at the inlet and outlet of the absorption tower, and is used to collect the flue gas flow rate and flue gas SO2 concentration parameters during the operation of the thermal power wet flue gas desulfurization device. Preferably, the flue gas composition monitoring device 1 adopts a high-precision on-line analyzer, and the sampling frequency is 5 times per second to ensure the real-time and accuracy of the data. In actual operation, the flue gas flow rate generally fluctuates within the range of 50,000 - 1,500,000 Nm³ / h, while the SO2 concentration varies between 400 - 5,000 mg / Nm³. The accurate collection of these parameters is the basis of the system's warning ability.

[0048] The intelligent data collector 2 is connected to the flue gas composition monitoring device 1, and is used to receive the parameters collected by the flue gas composition monitoring device 1, and perform preliminary screening and storage on the parameters. In an embodiment of the present invention, the intelligent data collector 2 adopts an industrial-grade data acquisition processor, which has anti-interference ability and data caching function, and can save data for at least 24 hours in case of network interruption. In addition, the intelligent data collector 2 can also perform validity verification on the original data, and filter out abnormal values through a set reasonable range, such as abnormal readings where the SO2 concentration suddenly drops to zero or exceeds 10,000 mg / Nm³, to avoid these incorrect data from affecting the system's judgment.

[0049] The intelligent diagnosis module 3 is connected to the intelligent data collector 2, and is used to analyze the parameters transmitted by the intelligent data collector 2, and judge the possible fault types and locations. The intelligent diagnosis module 3 adopts a hybrid diagnosis method based on rules and pattern recognition, and can identify various typical fault modes, including slurry circulation system faults, absorption tower spray faults, demister blockages, etc. For example, when the system detects that the differential pressure of the absorption tower rises rapidly and the slurry flow rate decreases, the diagnosis module will judge that the spray layer is blocked; when it detects that the slurry density continues to increase and the pH value decreases, it will judge that the preparation of limestone slurry is abnormal.

[0050] The prediction model 4 is connected to the intelligent diagnosis module 3, and is used to take the diagnosis result of the intelligent diagnosis module 3 as the input of the model, and take the real-time slurry pH value, real-time slurry density, and real-time slurry density difference at the outlet of the absorber corresponding to the diagnosis result as the output of the model. Preferably, the prediction model 4 is a deep neural network model, such as Figure 3 As shown, it includes an input layer, multiple hidden layers, and an output layer. In the thermal power wet flue gas desulfurization system, the parameters at the outlet of the absorber reflect the operating status of the desulfurization system. For example, the pH value usually needs to be maintained between 5.0 and 6.0 to ensure sufficient SO2 absorption and good gypsum crystallization, while the slurry density should be kept within the range of 1.05 - 1.15 g / cm³ to balance the desulfurization efficiency and pumping energy consumption.

[0051] The intelligent evaluation module 5 is connected to the prediction model 4, and is used to evaluate the final alarm level of the desulfurization device according to the output of the prediction model 4. In an embodiment of the present invention, the intelligent evaluation module 5 adopts a fuzzy logic evaluation method, comprehensively considering the parameter deviation degree, change trend, and duration to improve the accuracy of the evaluation. For example, when the slurry pH value drops rapidly and has been lower than 5.0 for more than 15 minutes, the system will evaluate it as a high-priority alarm; while when the pH value drops slowly but is still within the safe range, it may be evaluated as a low-priority early warning.

[0052] The decision-making plan module 6 is connected to the intelligent evaluation module 5, and is used to generate treatment suggestions according to the final alarm level. The decision-making plan module 6 is based on historical treatment experience and an expert knowledge base, and provides customized treatment plans for different fault situations. For example, for a slurry circulation pump failure, the system will first recommend checking the current and vibration values of the pump, and then provide operation guidance for pump switching or emergency startup of the standby pump; for the situation of insufficient limestone slurry concentration, it will recommend adjusting the slurry addition ratio and stirring intensity.

[0053] The intelligent control execution module 7 is connected to the decision-making plan module 6, and is used to receive the treatment suggestions of the decision-making plan module 6 and convert them into control signals. Preferably, the intelligent control execution module 7 adopts a step-by-step execution strategy, first implementing adjustment measures with lower risks, and then deciding whether to implement more radical measures after monitoring the effects to ensure the safety of system adjustment. For example, when dealing with the problem of excessive differential pressure of the demister, the system will first fine-tune the flushing water pressure and frequency, and then decide whether to increase the flushing intensity or time after observing the effects.

[0054] The on-site alarm module 8 is connected to the intelligent evaluation module 5 and is used to issue corresponding-level alarms on-site according to the final alarm level. The on-site alarm module 8 includes an audible and visual alarm and a remote notification function, and can remind the duty personnel in various ways. According to the severity of the fault, the system will trigger alarm signals of different levels: for emergency faults, a red flashing light and continuous alarm sound are used; for general faults, a yellow flashing light and intermittent alarm sound are used; for minor anomalies, only a prompt message is displayed on the control panel, and the audible and visual alarm is not activated.

[0055] The intelligent query module 9 is connected to the prediction model 4 and is used to display the corresponding decision recommendation level on the web page. In an embodiment of the present invention, the intelligent query module 9 provides a Web interface and a mobile application interface, supporting remote query and monitoring. The operating personnel can understand the system status in real time through the intelligent query module, view historical fault records and treatment effects, and can also obtain optimization suggestions for the current working conditions. For example, how to adjust the slurry circulation rate during low-load operation to reduce energy consumption while ensuring the desulfurization efficiency.

[0056] The prediction model 4 is a deep neural network model, and its establishment method is as Figure 4 shown and includes the following steps:

[0057] S1. Data sampling: Use the flue gas composition monitoring device 1 to collect the flue gas flow rate X and flue gas SO2 concentration Y parameters during the normal operation of the desulfurization device to obtain a sample set. Preferably, the sample set contains at least 30 days of operation data, covering the operation states under different load conditions, and the total number of samples is not less than 100,000. In the desulfurization system of a thermal power plant, data sampling needs to consider seasonal factors and unit load changes. For example, the flue gas volume and SO2 concentration are usually significantly higher in the peak heating period in winter than in summer. Data sampling should cover these typical working conditions to ensure the applicability of the model.

[0058] S2. Model construction: Divide the sample set in step S1 into a training set and a test set according to a ratio of 7:3, and use the training set to construct a deep neural network model; input the flue gas flow rate X in the test set into the deep neural network model, and the output is the predicted flue gas SO2 concentration Y'', compare the predicted flue gas SO2 concentration Y'' with the flue gas SO2 concentration Y in the test set, calculate the mean square error MSE, and optimize the parameters of the deep neural network model according to the value of MSE to obtain an optimized deep neural network model.

[0059] In one embodiment of the present invention, the deep neural network model adopts a multi-layer perceptron structure, including 4 hidden layers, with 128, 64, 32, and 16 neurons in each layer respectively. The ReLU function is used as the activation function, and the linear activation function is used for the output layer. The model is optimized using the Adam optimizer, with the initial learning rate set to 0.001 and adjusted dynamically during the training process. The model is trained using batch gradient descent, with a batch size of 64 and 200 training epochs. The mean squared error (MSE) calculation formula is as follows:

[0060] ,

[0061] where, is the number of samples in the test set, is the actual SO2 concentration value (unit: mg / Nm³) of the th sample, is the predicted SO2 concentration value (unit: mg / Nm³) of the th sample. This formula calculates the sum of the squares of the differences between the actual SO2 concentration and the predicted values, and then divides by the number of samples to obtain the mean squared error value, which reflects the prediction accuracy of the model. The smaller the value, the more accurate the prediction.

[0062] In the application of the desulfurization system, special attention should be paid to the sudden changes in SO2 concentration during model training, such as the rapid changes in SO2 concentration during the start-up and shutdown of the unit or the emission fluctuations caused by changes in coal composition. To improve the adaptability of the model to such situations, the training set should include sufficient samples under extreme working conditions, and the weights of these samples can be appropriately increased. Practice has shown that through this targeted training, the prediction accuracy of the model during severe fluctuations in flue gas emissions can be increased from 85% in standard training to over 95%.

[0063] S3. Model verification and application: Use the sample set collected in step S1 to verify the optimized deep neural network model obtained in step S2. After passing the verification, use the optimized deep neural network model as the prediction model. Preferably, 10-fold cross-validation is used for verification to ensure that the model has good performance on different data sets. During the actual verification process, in addition to paying attention to the overall prediction accuracy, it is also necessary to specifically evaluate the performance of the model under different working conditions, such as the prediction accuracy in key scenarios such as high load, low load, and coal switching, to ensure that the model can provide reliable predictions under various conditions.

[0064] S4. Update and optimization: During the application process, when the number of samples in the sample set in step S1 exceeds the threshold, a new test set is formed using the flue gas flow rate X and flue gas SO2 concentration Y that exceed the threshold, and the optimized deep neural network model described in step S2 is further optimized using the new training set to obtain a new optimized deep neural network model as the prediction model for application in step S3.

[0065] In one embodiment of the present invention, the threshold is set to 20% of the number of original training samples, that is, when the number of newly added samples reaches 20% of the original training samples, the model update is triggered. In addition, a time threshold of 30 days can be set to ensure that even if the sample growth is slow, the model can be updated regularly to adapt to the changes in equipment performance. This dynamic update mechanism is particularly important in the desulfurization system of thermal power plants because factors such as equipment aging, catalyst deactivation, and seasonal changes will cause changes in system performance. The regularly updated model can capture these changes in a timely manner and maintain prediction accuracy. For example, after a thermal power plant put this system into use, it was found that the prediction deviation of the model increased after equipment cleaning and maintenance. By setting a forced update mechanism after maintenance, the model can quickly adapt to the new state after the equipment performance is restored, and the prediction accuracy rate has recovered from 85% after maintenance to over 95%.

[0066] As Figure 5 shown, the decision recommendation levels of the present invention include the highest priority system alarm, the priority system alarm, and the secondary priority system alarm. These three levels correspond to different degrees of fault severity and provide a gradient warning mechanism.

[0067] Preferably, the highest priority system alarm is used to indicate a serious fault that may cause an emergency shutdown of the system and requires immediate handling; the priority system alarm is used to indicate a fault that will affect the normal operation of the system but will not cause a shutdown in a short time and requires prompt handling; the secondary priority system alarm is used to indicate minor abnormalities or performance degradation and requires handling during regular maintenance.

[0068] In the desulfurization system of thermal power plants, typical highest priority faults include the complete stop of the slurry circulation pump, the serious corrosion of equipment caused by the pH value being lower than 4.5, the blockage of pipelines caused by the slurry density exceeding 1.25 g / cm³, etc.; priority faults include the failure of a single slurry circulation pump, the pH value deviating from the set value by more than ±0.8, the differential pressure of the demister exceeding the standard, etc.; secondary priority faults include slurry density fluctuations, the decline of gypsum dehydration efficiency, and a slight increase in energy consumption. This hierarchical alarm mechanism enables operators to reasonably allocate resources, prioritize key issues, and improve the overall reliability of the system.

[0069] When the final alarm level in step S3 reaches any one or more of the highest priority system alarm, the priority system alarm, and the secondary priority system alarm, the system alarms of the corresponding levels will all give on-site alarms. When alarming, the prediction model 4 will accumulate the alarm times of the corresponding on-site alarm device as 1.

[0070] Preferably, the system records and analyzes the alarm frequencies under different time periods and operating conditions, constructs an alarm history database by accumulating the number of alarms, and provides data support for future fault mode recognition and preventive maintenance. In an embodiment of the present invention, the system sets the alarm silence period to 5 minutes to avoid repeated alarms for the same fault within a short time and relieve the burden on the operator.

[0071] In practical applications, the statistics and analysis of the number of alarms have important value. For example, by analyzing the alarm frequency distribution related to the slurry circulation pump in a desulfurization system within one month, it is found that the alarm frequency is the highest during the period from 8:00 to 10:00 every day. Further investigation reveals that this is exactly the stage of the daily load increase in the power plant, and it is a temporary abnormality caused by the insufficiently rapid response of the slurry circulation system. Based on this discovery, the operation personnel adjusted the start-stop strategy of the circulation pump, increased the circulation volume in advance, and reduced the number of alarms during the same period by 82%, significantly improving the system stability.

[0072] When the mean square error MSE of the optimized deep neural network model in step S3 is less than or equal to 0.01, the optimized deep neural network model is obtained. This accuracy requirement ensures the high reliability of the model prediction.

[0073] In an embodiment of the present invention, in addition to the mean square error, multiple indicators are also used to comprehensively evaluate the model performance, including the mean absolute error (MAE) less than 5 mg / Nm³, the coefficient of determination (R²) greater than 0.95, etc. The model is considered to be optimized successfully only when all performance index requirements are met simultaneously. This strict standard ensures the stability and reliability of the prediction model under various operating conditions.

[0074] The calculation formula for the mean absolute error (MAE) is:

[0075] ,

[0076] where is the number of samples in the test set, is the actual SO2 concentration value (unit: mg / Nm³) of the th sample, is the predicted SO2 concentration value (unit: mg / Nm³) of the th sample, represents the absolute value of the difference between the predicted value and the actual value. MAE directly reflects the average deviation degree between the predicted value and the actual value, and the unit is the same as the original data, which is convenient for engineering personnel to intuitively understand.

[0077] The calculation formula for the coefficient of determination (R²) is:

[0078] ,

[0079] where is the number of samples in the test set, is the actual SO2 concentration value of the is the predicted SO2 concentration value of the is the average value of the actual SO2 concentration values, calculated as . The R² value ranges from 0 to 1. The closer it is to 1, the more variance in the data the model explains and the stronger the predictive ability.

[0080] In the application of the desulfurization system in thermal power plants, the model accuracy directly affects the reliability of early warning. For example, in the desulfurization system of a 300MW unit, when the emission concentration is close to the limit value (200mg / Nm³), an error control within 5mg / Nm³ means that the system can accurately give an early warning before actual over - standard, avoiding environmental protection penalties; when the normal emission is in the range of 50 - 100mg / Nm³, this accuracy is also sufficient to detect early abnormal fluctuations and timely discover the decay of equipment performance. Practice has proved that when the model accuracy MSE ≤ 0.01, the system can predict the downward trend of desulfurization efficiency 15 - 30 minutes in advance, leaving enough intervention time for operators.

[0081] The intelligent evaluation module 5 evaluates the diagnostic results of the intelligent diagnosis module 3. When the diagnostic result is an alarm, the prediction model 4 automatically generates final alarms at three levels: the highest - priority system alarm, the priority system alarm, and the secondary - priority system alarm, based on the real - time slurry pH value, real - time slurry density, and real - time slurry density difference at the corresponding position of the diagnostic result. The intelligent control execution module 7 sends three - level fault instructions composed of the highest - priority system alarm, the priority system alarm, and the secondary - priority system alarm to the on - site alarm module 8 according to the three - level final alarms.

[0082] Preferably, the following criteria are adopted in the evaluation process: When the real - time slurry pH value deviates from the set value by more than ±1.0 and the duration exceeds 30 minutes, the alarm evaluation is triggered; when the real - time slurry density deviates from the normal range (1.05 - 1.15g / cm³) by more than ±0.05g / cm³ and the duration exceeds 15 minutes, the alarm evaluation is triggered; when the real - time slurry density difference (the density difference between the inlet and the outlet) exceeds 0.1g / cm³ and the duration exceeds 10 minutes, the alarm evaluation is triggered.

[0083] In an embodiment of the present invention, the evaluation logic considers the combined effect of the parameter deviation degree and the duration, and uses a fuzzy comprehensive evaluation model to calculate the risk level. The model is based on the following formula:

[0084]

[0085] Where, is the risk level score, with a value range of 0 - 100. The larger the value, the higher the risk; is the number of parameters considered. In this example , which are the pH value, slurry density, and density difference respectively; is the weight coefficient of the th parameter, satisfying , reflecting the relative importance of each parameter to system safety; is the current deviation value of the th parameter, calculated as the difference between the current value and the median of the normal range; is the th parameter deviation duration (unit: minutes); is the th parameter risk function, mapping the parameter deviation and duration to a risk value.

[0086] The risk function adopts a non - linear mapping to reflect the combined effect of parameter abnormality degree and duration:

[0087] ,

[0088] where is the absolute value of the parameter deviation; is the maximum allowable deviation value of this parameter; is the time decay coefficient to control the influence degree of the time factor; is the exponential function, which gradually approaches zero as the duration increases, making the risk score increase with time.

[0089] In the desulfurization system of thermal power plants, the weight and threshold settings of each parameter directly affect the evaluation accuracy. Based on operation experience, the pH value is usually given the highest weight (0.5) because it directly affects the equipment corrosion risk; the slurry density is the second (0.3), affecting the desulfurization efficiency and energy consumption; the density difference has the lowest weight (0.2), mainly used to judge the equipment scaling condition. The system automatically judges the alarm level according to the calculated risk score: Risk > 80 is the highest - priority alarm, 60 ≤ Risk ≤ 80 is the priority alarm, and 40 ≤ Risk < 60 is the secondary - priority alarm.

[0090] For example, during a certain operation, the slurry pH value drops to 4.6 (deviation - 0.9, set value 5.5) and lasts for 25 minutes; at the same time, the slurry density rises to 1.19 g / cm³ (deviation + 0.09, median of the normal range 1.10) and lasts for 18 minutes; the density difference is 0.07 g / cm³ and lasts for 15 minutes. Substitute into the risk assessment model for calculation:

[0091] ,

[0092] The calculated result shows Risk = 68.3. The system determines it as a priority alarm, reminding the operator to handle it as soon as possible, but no emergency intervention is required. This refined risk assessment mechanism enables the system to accurately distinguish the severity of different abnormal situations, avoiding over-alarm or under-alarm.

[0093] The final alarms of the three levels of the highest-priority system alarm, priority system alarm, and sub-priority system alarm respectively correspond to three levels of fault instructions. Among them, the fault instruction for the highest-priority system alarm is that the slurry supply port of the absorption tower fails and the slurry discharge port of the absorption tower fails; the fault instruction for the priority system alarm is that the slurry supply port and the slurry discharge port of the absorption tower fail simultaneously; the fault instruction for the sub-priority system alarm is that the slurry supply port of the absorption tower fails and the slurry discharge port of the absorption tower is normal.

[0094] In an embodiment of the present invention, the system establishes a fault mode library through historical data analysis and uses fault feature vectors for fault identification and classification. The fault feature vector is defined as:

[0095] ,

[0096] Where, is the fault feature vector; to are the current values of key parameters; to are the change rates of these parameters within a specified time window, calculated as is the time window width, usually set to 10 minutes. For the thermal power desulfurization system, typical key parameters include slurry flow rate, slurry density, pH value, pump current, demister differential pressure, etc.

[0097] The system uses the cosine similarity algorithm to calculate the similarity between the current feature vector and the known patterns in the fault mode library:

[0098] ,

[0099] Where, is the current fault feature vector; is the feature vector of the th known fault pattern in the fault mode library; and are respectively and 's th element; represents the dot product of the two vectors; and represent the Euclidean norms (vector lengths) of the two vectors. The cosine similarity value ranges from -1 to 1, and the value closer to 1 indicates that the two fault patterns are more similar.

[0100] When the system detects an anomaly, it calculates the similarity between the current feature vector and each known fault in the fault library. The fault type with the highest similarity that exceeds the threshold (usually set to 0.85) is identified as the current fault. For example, the typical characteristics of a pulp supply port blockage are a sudden drop in pulp flow rate (exceeding 20% / minute), fluctuations in the current of the circulation pump, and changes in the pressure difference inside the tower; while a fault in the pulp discharge port is manifested as a slow increase in pulp density (about 0.01 g / cm³ / hour), an obvious downward trend in pH value, etc.

[0101] In an application case of a desulfurization system of a 600MW unit, the system successfully identified a typical pulp supply port fault: the pulp flow rate decreased from 12000 m³ / h to 9500 m³ / h within 2 minutes (a decrease of about 21%), the fluctuation range of the current of the circulation pump increased to ±15A, and the pressure difference inside the tower increased by 200 Pa. The similarity between the calculated feature vector and the pulp supply port blockage mode reached 0.92, far higher than other fault modes, and the system accurately issued a secondary priority alarm. The operator cleaned the pulp supply port in a timely manner according to the system's suggestion, avoiding the expansion of the fault.

[0102] The prediction model 4 generates the recommended level of the decision-making plan according to the diagnostic results of the intelligent diagnosis module 3 corresponding to the final alarm. The recommended level of the decision-making plan consists of the highest priority plan, the priority plan, and the secondary priority plan; the intelligent query module 9 simultaneously displays the highest priority plan, the priority plan, and the secondary priority plan on the web page and the background operation according to the recommended level of the decision-making plan.

[0103] In an embodiment of the present invention, the generation of the decision-making plan adopts a method combining case-based reasoning and rule-based reasoning. The system generates the decision-making plan through the following steps:

[0104] First, calculate the similarity between the current fault and each case in the historical case library. The similarity calculation uses the weighted Euclidean distance:

[0105] ,

[0106] where, is the feature vector of the current fault; is the feature vector of the th case in the historical case library; is the feature dimension; is the th weight of the feature, reflecting the importance of this feature for case similarity judgment; and are respectively and 's th feature value. The smaller the distance, the more similar the cases.

[0107] Then, select the k nearest cases (usually k = 3) from the historical case library, and extract the handling methods of these cases as candidate solutions. Next, the system uses a rule inference engine to adjust and optimize the candidate solutions. The rule inference is based on the characteristics of the current working conditions and the equipment status, and corrects the handling steps and parameter settings. Finally, the system classifies the optimized solutions into three levels: the highest priority, the priority, and the secondary priority according to the expected effects and execution risks of the solutions.

[0108] In the desulfurization system of a thermal power plant, the content of the decision-making solution usually includes five parts: fault description, possible causes, recommended handling steps, expected effects, and precautions. For example, for a typical fault of low slurry pH value, the example of the decision-making solution generated by the system is as follows:

[0109] The highest priority solution (applicable to PH < 4.5):

[0110] Fault description: The pH value of the slurry in the absorption tower is severely low, and the current value is 4.3;

[0111] Possible causes: Failure of the limestone slurry preparation system or problems with the quality of limestone;

[0112] Recommended handling steps:

[0113] 1. Immediately increase the supplementary amount of limestone slurry to 150 m³ / h;

[0114] 2. Check the operating status of the limestone crushing system to ensure that the fineness meets the requirements;

[0115] 3. Take samples from the slurry preparation system for analysis and detect the purity of limestone;

[0116] Expected effects: The pH value should rise above 5.0 within 30 minutes to avoid equipment corrosion;

[0117] Precautions: Closely monitor the change of slurry density during the process of increasing limestone slurry to prevent excessive concentration caused by overfeeding;

[0118] Priority solution (applicable to 4.5 ≤ PH < 5.0):

[0119] Recommended handling steps:

[0120] 1. Appropriately increase the supplementary amount of limestone slurry to 120 m³ / h;

[0121] 2. Check the accuracy of the limestone metering system;

[0122] 3. Confirm the inventory of the limestone silo;

[0123] Secondary priority solution (applicable to slightly low PH but still within the safe range):

[0124] Suggested handling steps:

[0125] 1. Adjust the limestone feed rate and increase it by 10%;

[0126] 2. Pay special attention to the slurry preparation system during the next inspection;

[0127] This multi-level scheme recommendation mechanism provides operators with gradient processing options, which can be flexibly selected according to the severity of the fault and on-site conditions, significantly improving the fault handling efficiency and system reliability.

[0128] The intelligent query module 9 includes a parameter query unit, a fault warning unit, and a historical query unit. The parameter query unit includes the real-time slurry density in the absorption tower, the real-time slurry pH value in the absorption tower, the real-time slurry density difference in the absorption tower, the real-time flow rate of the absorption slurry circulation pump, and the real-time alarm count of the on-site alarm device. The fault warning unit includes the highest-priority system alarm, the priority system alarm, the secondary-priority system alarm, and the recommended solutions. The historical query unit includes the fault history query and the effect history query. The fault history query is divided into the last month, the last three months, and the last six months according to the alarm count in the absorption tower; the fault history query generates corresponding table titles according to the alarm count in the absorption tower; the effect history query is divided into the last month, the last three months, and the last six months for query.

[0129] Preferably, the parameter query unit adopts a real-time refresh mechanism with a refresh frequency of 5 seconds / time to ensure the timeliness of the displayed data. The system adopts parameter status visualization technology to intuitively display parameter changes through color coding and trend charts. The parameter status evaluation adopts a hemisphere model based on fuzzy intervals:

[0130] ,

[0131] where is the status score of parameter , with a value range of 0 - 1, 1 indicating completely normal, and 0 indicating completely abnormal; is the normal range of the parameter; is the boundary of the normal range or ; is the critical abnormal value, and exceeding this value is regarded as completely abnormal. For example, for the slurry pH value, the normal range is [5.0, 6.0], and the critical abnormal values can be set as the lower limit of 4.0 and the upper limit of 7.0. When PH = 5.5, S(p) = 1, and the state is completely normal; when PH = 4.5, S(p) = 0.5, in a semi-abnormal state; when PH = 3.8, S(p) = 0, and the state is completely abnormal.

[0132] The system visually displays the status score through color mapping: completely normal (S(p)=1) is shown as green, semi-abnormal (0<S(p)<1) is shown as yellow and the brightness is adjusted according to the value of S(p), and completely abnormal (S(p)=0) is shown as red. In addition, the system also displays the trend arrow of the parameter. An upward arrow is shown for an upward trend, a downward arrow is shown for a downward trend, and a horizontal line is shown for stability.

[0133] In the historical query unit, the system provides multi-dimensional analysis functions and supports combined queries according to conditions such as time, fault type, and equipment components. Historical data is visually displayed in the form of charts, including the statistical chart of fault frequency, the distribution chart of fault types, the evaluation chart of treatment effects, etc. For example, the operator can query the distribution of faults related to the circulating pump within a certain month, and the system will generate charts including the proportion of fault types, the comparison of treatment times of different teams, and the treatment effect score, etc., to help analyze the fault pattern and optimize the maintenance strategy.

[0134] In an application case of a certain power plant, through historical query and analysis, it is found that the abnormal pH value caused by the quality problem of limestone frequently occurs at the beginning of each quarter. Further investigation confirms that it is caused by the seasonal change of suppliers. Based on this discovery, the power plant adjusted the limestone procurement strategy, increased the frequency of supplier qualification review and in-plant inspection, resulting in a 65% decrease in the related failure rate and a significant improvement in system stability.

[0135] The decision-making scheme module 6 includes: the density grade of the first absorption tower slurry and the SO2 concentration grade of the first flue gas inlet. According to the density grade of the first absorption tower slurry and the SO2 concentration grade of the first flue gas inlet, the decision-making scheme module 6 generates different priority schemes. The density of the first absorption tower slurry is obtained by using the slurry pH value and slurry density at the slurry outlet of the absorption tower, and the SO2 concentration grade of the first flue gas inlet is obtained by using the flue gas flow rate and flue gas SO2 concentration at the flue gas inlet.

[0136] In an embodiment of the present invention, the density grade of the absorption tower slurry is divided into 5 levels. The grade classification uses the fuzzy clustering algorithm to determine the optimal demarcation point based on historical operation data. The specific classification of the slurry density grade is as follows:

[0137] High density (>1.15g / cm³): indicating that the gypsum saturation is too high, which is easy to cause scaling;

[0138] Medium-high density (1.10 - 1.15g / cm³): the upper limit range of normal operation;

[0139] Normal density (1.05 - 1.10g / cm³): the best operation interval;

[0140] Medium-low density (1.00 - 1.05g / cm³): the lower limit range of normal operation;

[0141] Low density (<1.00 g / cm³): Indicates insufficient slurry concentration, affecting desulfurization efficiency;

[0142] The determination of the slurry density grade also considers the synergistic effect of the pH value, and the two are combined to calculate the comprehensive grade. The comprehensive grade calculation uses the weighted average method:

[0143] ,

[0144] Among them, is the comprehensive grade of the slurry; is the density grade, with a value range of 1 - 5 (corresponding to low density to high density); is the pH value grade, with a value range of 1 - 5 (corresponding to too low, slightly low, normal, slightly high, too high); and are the weight coefficients, satisfying + = 1. According to operation experience, usually = 0.7, = 0.3.

[0145] The SO2 concentration grade at the flue gas inlet is also divided into 5 levels:

[0146] Ultra-high concentration (>4000 mg / Nm³): Extreme working condition;

[0147] High concentration (2000 - 4000 mg / Nm³): High load working condition;

[0148] Medium concentration (1000 - 2000 mg / Nm³): Normal working condition;

[0149] Low concentration (500 - 1000 mg / Nm³): Low load working condition;

[0150] Very low concentration (<500 mg / Nm³): Start-up and shutdown transition working condition;

[0151] The determination of the SO2 concentration grade considers the flue gas flow rate factor and uses the SO2 load index for comprehensive evaluation:

[0152] ,

[0153] Among them, is the SO2 load index, indicating the SO2 treatment load relative to the design state; is the inlet SO2 concentration (unit: mg / Nm 3 ); is the actual flue gas flow rate (unit: Nm³ / h); is the design flue gas flow rate (unit: Nm³ / h). The system determines the final SO2 concentration grade according to the value.

[0154] Based on the combination of the comprehensive slurry grade and the SO2 concentration grade, the decision-making solution module 6 forms a 5×5 solution matrix, corresponding to 25 sets of operating conditions and their treatment solutions. For example, when the slurry is of low density and the SO2 is of high concentration, the system will identify it as a state of severely insufficient desulfurization capacity, generate the highest-priority solution, and recommend immediately increasing the concentration of limestone slurry and adjusting the flow rate of the circulation pump; when the slurry is of medium-high density and the SO2 is of medium concentration, the system judges it as a slight scaling risk, generates a priority solution, and recommends appropriately reducing the slurry density and increasing the slurry discharge period.

[0155] In the application case of a 1000MW unit, the system detected that the slurry density was 1.16 g / cm³ (slightly higher than the lower limit of high density), the pH value was 5.2 (normal), and the comprehensive grade was calculated as 4.3; at the same time, the SO2 concentration was 3500 mg / Nm³, the flue gas flow rate was 2.8 million Nm³ / h, and the SO2 load index was 3.2 (high load). The system generated a priority solution, recommending that, while maintaining the slurry pH value, the slurry density be controlled at about 1.12 g / cm³, and at the same time increasing the gypsum discharge to prevent scale formation. After the operating personnel adopted the recommendation, the system ran smoothly, avoiding the possible risk of equipment scaling and shutdown.

[0156] As Figure 6 shown, the operation evaluation subsystem includes a data collector 11, an evaluation module 12, a data processor 13, and a display module 14. The data collector 11 collects the equipment operation data and sends it to the data processor 13. The equipment operation data includes the SO2 concentration at the flue gas inlet and the flue gas inlet flow rate. The data processor 13 processes the equipment operation data to obtain the evaluation grade of the equipment operation data. The evaluation grade of the equipment operation data is obtained by the data processor 13. When the evaluation grade of the equipment operation data is the highest-priority solution, the display module 14 issues an alarm prompt message. When the equipment operation data grade is the highest-priority solution, the display module 14 issues a warning prompt message. When the equipment operation data grade is the priority solution, the display module 14 issues a poor operation prompt message. When the evaluation grade of the equipment operation data is the sub-priority solution, the display module 14 issues a good prompt message. Among them, for the evaluation grade of the equipment operation data of the highest-priority solution, the SO2 concentration is greater than or equal to 2000 mg / Nm 3 , and the flue gas inlet flow rate is greater than or equal to 1000000 Nm 3 / h. For the evaluation grade of the equipment operation data of the priority solution, the SO2 concentration is greater than or equal to 1000 mg / Nm 3 and less than 2000 mg / Nm 3 , and the flue gas inlet flow rate is greater than or equal to 100000 Nm 3 / h and less than 1000000 Nm 3 / h, the SO2 concentration of the secondary priority scheme is less than 1000 mg / Nm 3 , the flue gas inlet flow rate is greater than or equal to 10000 Nm 3 / h and less than 100000 Nm 3 / h.

[0157] In an embodiment of the present invention, the operation evaluation subsystem adopts a multi-index comprehensive evaluation method to construct an operation quality evaluation model for the desulfurization system. The evaluation index system includes three categories: technical indexes, economic indexes, and environmental protection indexes. The specific indexes include desulfurization efficiency, energy consumption ratio, limestone consumption rate, equipment reliability, etc.

[0158] The comprehensive evaluation uses the analytic hierarchy process (AHP) to determine the weights of each index, and then uses weighted summation to calculate the comprehensive score:

[0159] ,

[0160] Among them, Score is the comprehensive score, and the value range is 0-100. The higher the value, the better the operation quality; is the number of indexes; is the weight of the th index, satisfying is the th actual value of the index; is the index scoring function, which maps the actual index value to a score of 0-100.

[0161] The index scoring function adopts different forms according to different index types:

[0162] 1. For indexes of the "higher the better" type (such as desulfurization efficiency):

[0163] ,

[0164] 2. For indexes of the "lower the better" type (such as energy consumption ratio):

[0165] ,

[0166] 3. For indexes of the "closest to a certain value is the best" type (such as pH value):

[0167] ,

[0168] Among them, and are respectively the lower limit and upper limit of the index; is the optimal value.

[0169] The system determines the operation quality level according to the comprehensive score: excellent (90 - 100 points), good (80 - 90 points), average (70 - 80 points), poor (60 - 70 points), very poor (below 60 points). At the same time, the system also conducts short-term trend prediction based on the current working condition parameters, estimates the system performance within 2 - 4 hours, and provides a basis for operation adjustment.

[0170] In a specific application case, when the desulfurization system of a 660MW unit operates under high load conditions (SO2 concentration 3500mg / Nm³, flue gas flow 3.1 million Nm³ / h), the system evaluates various indicators in real time: desulfurization efficiency 96.2% (score 92, weight 0.3), energy consumption ratio of slurry circulation pump 0.28kWh / kg-SO2 (score 85, weight 0.2), limestone consumption rate 1.05kg / kg-SO2 (score 88, weight 0.15), system pressure drop 850Pa (score 78, weight 0.15), equipment reliability index 0.92 (score 90, weight 0.2). The calculated comprehensive score is 87.7 points, and it is determined to be in the good level.

[0171] Based on historical data and current trends, the system predicts that after continuing the current operation mode for 4 hours, the system pressure drop will increase to 920Pa, the energy consumption ratio will rise to 0.31kWh / kg-SO2, and the comprehensive score may drop to 82 points. Therefore, the system generates optimization suggestions, suggesting appropriate increase in slurry discharge to control the slurry density and prevent excessive system pressure drop and increased energy consumption. After the operation personnel adopt the suggestions for adjustment, the actual system pressure drop is controlled at 880Pa after 4 hours, and the energy consumption ratio remains at 0.29kWh / kg-SO2, maintaining the high-efficiency operation state of the system.

[0172] The above are only the preferred embodiments of the present invention and are not used to limit the present invention. The patent protection scope of the present invention is subject to the claims. Changes, modifications, additions, or substitutions made by those skilled in the art within the substantial scope of the present invention all belong to the protection scope of the present invention.

Claims

1. A real-time warning and operation evaluation system for a thermal power wet flue gas desulfurization device, characterized in that The system includes a real-time warning subsystem and an operation evaluation subsystem. The real-time warning subsystem includes: A flue gas component monitoring device, which is used to collect the flue gas flow rate and flue gas SO2 concentration parameters during the operation of the thermal power wet flue gas desulfurization device; An intelligent data collector, which is connected to the flue gas component monitoring device, and is used to receive the parameters collected by the flue gas component monitoring device, and perform preliminary screening and storage on the parameters; An intelligent diagnosis module, which is connected to the intelligent data collector, and is used to analyze the parameters transmitted by the intelligent data collector, and judge the possible fault types and positions; A prediction model, which is connected to the intelligent diagnosis module, and is used to take the diagnosis result of the intelligent diagnosis module as the input of the model, and take the real-time slurry pH value, real-time slurry density and real-time slurry density difference at the slurry outlet position of the absorption tower corresponding to the diagnosis result as the output of the model; An intelligent evaluation module, which is connected to the prediction model, and is used to evaluate the final alarm level of the desulfurization device according to the output of the prediction model; A decision-making scheme module, which is connected to the intelligent evaluation module, and is used to generate treatment suggestions according to the final alarm level; An intelligent control execution module, which is connected to the decision-making scheme module, and is used to receive the treatment suggestions of the decision-making scheme module and convert them into control signals; A field alarm module, which is connected to the intelligent evaluation module, and is used to issue corresponding-level alarms on site according to the final alarm level; An intelligent query module, which is connected to the prediction model, and is used to display the corresponding decision recommendation level on the web page; Among them, the prediction model is a deep neural network model, and the establishment method of the prediction model includes the following steps: S1. Data sampling: Use the flue gas component monitoring device to collect the flue gas flow rate X and flue gas SO2 concentration Y parameters from the normal operation of the desulfurization device to obtain a sample set; S2. Model construction: Divide the sample set in step S1 into a training set and a test set according to a preset ratio, and use the training set to construct a deep neural network model; input the flue gas flow rate X in the test set into the deep neural network model, and the output is the predicted flue gas SO2 concentration Y'', compare the predicted flue gas SO2 concentration Y'' with the flue gas SO2 concentration Y in the test set, calculate the mean square error MSE, and optimize the parameters of the deep neural network model according to the value of the MSE to obtain an optimized deep neural network model; S3. Model verification and application: Use the sample set collected in step S1 to verify the optimized deep neural network model obtained in step S2. After passing the verification, use the optimized deep neural network model as the prediction model; S4. Update and optimization: During the application process, when the number of samples in the sample set in step S1 exceeds the threshold, use the flue gas flow rate X and flue gas SO2 concentration Y that exceed the threshold to form a new test set, and use the new training set to further optimize the optimized deep neural network model in step S2 to obtain a new optimized deep neural network model as the prediction model applied in step S3.

2. The real-time warning and operation evaluation system of the thermal power wet flue gas desulfurization device according to claim 1, characterized in that, The decision recommendation levels include the highest-priority system alarm, the priority system alarm, and the sub-priority system alarm.

3. The real-time early warning and operation evaluation system for the thermal power wet desulfurization device according to claim 2, wherein, When the final alarm level in step S3 reaches any one or more of the highest-priority system alarm, the priority system alarm, and the sub-priority system alarm, on-site alarms are triggered for the corresponding-level system alarms. When the alarm is triggered, the prediction model increments the alarm count of the corresponding on-site alarm device by 1.

4. The real-time warning and operation evaluation system of the thermal power wet flue gas desulfurization device according to claim 1, characterized in that, When the mean squared error (MSE) of the optimized deep neural network model in step S3 is less than or equal to 0.01, the optimized deep neural network model is obtained.

5. The real-time warning and operation evaluation system for the thermal power wet flue gas desulfurization device according to claim 1, characterized in that, The intelligent evaluation module evaluates the diagnostic results of the intelligent diagnosis module. When the diagnostic result is an alarm, the prediction model automatically generates final alarms at three levels, namely the highest-priority system alarm, the priority system alarm, and the sub-priority system alarm, based on the real-time slurry pH value, real-time slurry density, and real-time slurry density difference at the corresponding position of the diagnostic result. The intelligent control execution module sends fault instructions at three levels to the on-site alarm module according to the final alarms at the three levels. The fault instructions consist of the highest-priority system alarm, the priority system alarm, and the sub-priority system alarm.

6. The real-time early warning and operation evaluation system for the thermal power wet flue gas desulfurization device according to claim 5, characterized in that, The final alarms at the three levels of the highest-priority system alarm, the priority system alarm, and the sub-priority system alarm respectively correspond to fault instructions at three levels. Among them, the fault instruction for the highest-priority system alarm is that the slurry supply port of the absorption tower fails and the slurry discharge port of the absorption tower fails; the fault instruction for the priority system alarm is that the slurry supply port and the slurry discharge port of the absorption tower fail simultaneously; the fault instruction for the sub-priority system alarm is that the slurry supply port of the absorption tower fails and the slurry discharge port of the absorption tower is normal.

7. The real-time early warning and operation evaluation system for the thermal power wet flue gas desulfurization device according to claim 1, characterized in that, The prediction model generates a decision scheme recommendation level according to the diagnostic results of the intelligent diagnosis module corresponding to the final alarm. The decision scheme recommendation level consists of the highest-priority scheme, the priority scheme, and the sub-priority scheme; the intelligent query module simultaneously displays the highest-priority scheme, the priority scheme, and the sub-priority scheme on the web page and the background operation according to the decision scheme recommendation level.

8. The real-time early warning and operation evaluation system for the thermal power wet flue gas desulfurization device according to claim 1, characterized in that, The intelligent query module includes a parameter query unit, a fault warning unit, and a history query unit. The parameter query unit includes the real-time slurry density of the absorption tower, the real-time slurry pH value of the absorption tower, the real-time slurry density difference of the absorption tower, the real-time flow rate of the absorption slurry circulation pump, and the real-time alarm count of the on-site alarm device. The fault warning unit includes the highest-priority system alarm, the priority system alarm, the sub-priority system alarm, and a recommended solution. The history query unit includes a fault history query and an effect history query. The fault history query is divided into the last month, the last three months, and the last six months according to the alarm count of the absorption tower; The fault history query generates corresponding table titles according to the alarm count of the absorption tower; the effect history query is divided into queries for the last month, the last three months, and the last six months.

9. The real-time warning and operation evaluation system of the thermal power wet desulfurization device according to claim 1, characterized in that, The decision-making scheme module includes: the first absorption tower slurry density level and the first flue gas inlet SO2 concentration level. According to the first absorption tower slurry density level and the first flue gas inlet SO2 concentration level, the decision-making scheme module generates different priority schemes. The first absorption tower slurry density is obtained by using the slurry pH value and slurry density at the absorption tower slurry outlet, and the first flue gas inlet SO2 concentration level is obtained by using the flue gas inlet gas flow rate and flue gas SO2 concentration.

10. The real-time warning and operation evaluation system for the thermal power wet flue gas desulfurization device according to claim 1, wherein, The operation evaluation subsystem includes a data collector, an evaluation module, a data processor, and a display module. The data collector collects equipment operation data and sends it to the data processor. The equipment operation data includes the SO2 concentration at the flue gas inlet and the flue gas inlet flow rate. The data processor processes the equipment operation data to obtain the evaluation grade of the equipment operation data. The evaluation grade of the equipment operation data is obtained by the data processor. When the evaluation grade of the equipment operation data is the highest priority scheme, the display module issues an alarm prompt message. When the equipment operation data level is the highest priority scheme, the display module issues a warning prompt message. When the equipment operation data level is the priority scheme, the display module issues a poor operation prompt message. When the evaluation grade of the equipment operation data is the sub-priority scheme, the display module issues a good prompt message. Among them, for the highest priority scheme, the SO2 concentration of the evaluation grade of the equipment operation data is greater than or equal to 2000mg / Nm 3 , and the flue gas inlet flow rate is greater than or equal to 1000000Nm 3 / h. For the priority scheme, the SO2 concentration of the evaluation grade of the equipment operation data is greater than or equal to 1000mg / Nm 3 and less than 2000mg / Nm 3 , and the flue gas inlet flow rate is greater than or equal to 100000Nm 3 / h and less than 1000000Nm 3 / h. For the sub-priority scheme, the SO2 concentration is less than 1000mg / Nm 3 , and the flue gas inlet flow rate is greater than or equal to 10000Nm 3 / h and less than 100000Nm 3 / h.

Images