Self-adaptive blending combustion method for thermal power generating unit based on coal quality parameters

Abstract

The invention discloses a thermal power generating unit self-adaptive blending combustion method based on coal quality parameters. The thermal power generating unit self-adaptive blending combustion method comprises the following steps that S1, the coal quality parameters are obtained in real time; s2, digitizing the detected coal quality parameters, and inputting the digitized coal quality parameters into a system database; updating the digital twin model in real time based on the latest data; s3, dynamic characteristic analysis is carried out through a digital twinborn model, and the response of coal combustion is predicted; s4, generating a three-dimensional control matrix, and generating an optimal blending combustion proportion; performing dynamic compensation correction according to the real-time operation data; s5, issuing the optimized blending combustion instruction to a combustion system for execution; s6, the emission index and the combustion efficiency after combustion are monitored in real time; s7, carrying out deviation analysis on the collected operation data and an expected target; and S8, carrying out machine learning training on the actual blending combustion data and the theoretical model deviation. According to the method, the blending combustion proportion is dynamically optimized by generating the three-dimensional control matrix; the combustion boundary is predicted based on a transfer learning engine, rolling horizon optimization responds to deviation within 3 minutes, and it is ensured that pollutants do not exceed the limit.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of coal blending combustion, and particularly relates to a self-adaptive blending combustion method for thermal power generating units based on coal quality parameters. BACKGROUND

[0002] At present, the multi-coal blending combustion technology is generally used in thermal power generating units to reduce fuel cost and adapt to coal source fluctuation, but the off-line detection is mainly relied on to fail to respond to coal quality change in real time, so that the blending combustion ratio is out of touch with actual demand. In addition, the existing blending combustion scheme is mostly formulated based on static experience model, the optimization target is single, and the real-time dynamic adjustment capability is lacked, in addition, the traditional method is difficult to accurately control pollutant emission while guaranteeing the safety of the boiler, and frequent manual intervention is needed. SUMMARY

[0003] Therefore, the present application provides a self-adaptive blending combustion method for thermal power generating units based on coal quality parameters to solve the problems in the prior art.

[0004] In order to achieve the above purpose, the present application provides the following technical scheme:

[0005] A self-adaptive blending combustion method for thermal power generating units based on coal quality parameters, comprising the following steps:

[0006] S1. Laser-induced breakdown spectroscopy (LIBS) technology and near-infrared spectroscopy (NIR) technology are used to detect raw coal to obtain coal quality parameters in real time;

[0007] S2. The detected coal quality parameters are digitized and input into a system database; based on the latest data, a digital twin model is updated in real time;

[0008] S3. Dynamic characteristic analysis is performed through the digital twin model to predict the response of coal quality combustion; and a safety constraint boundary is calculated;

[0009] S4. Within the safety boundary, an improved NSGA-II multi-objective optimization algorithm is used to generate a three-dimensional control matrix containing coal feeder speed setting, coal mill output distribution and air-coal ratio adjustment, and the optimal blending combustion ratio is generated; dynamic compensation correction is performed according to real-time operation data to adjust the blending combustion strategy;

[0010] S5. The optimized blending combustion instruction is issued to the combustion system for execution to control the equipment and realize precise blending combustion;

[0011] S6. The emission indicators and combustion efficiency after combustion are monitored in real time, and actual operation data are collected;

[0012] S7. Deviation analysis of the collected operation data from the expected target; when the key parameters of operation deviate from the preset threshold value, trigger the rolling horizon optimization to make strategy correction; if the detection data is in the normal range, only update the historical database and maintain the current operation state;

[0013] S8. Machine learning training of the deviation between actual blending data and theoretical model; generate blending economy analysis report regularly.

[0014] Further: in S2, the system database based on coal quality data is needed to establish a pyrolysis characteristic index and NOx generation characteristic correlation model.

[0015] Further: in S4, the introduced coal type dynamic matching rules include: high-sulfur coal and low-sulfur coal are matched according to sulfur balance, high-ash melting point coal and low-ash melting point coal are matched according to ash sticking temperature characteristics, and inferior coal blending ratio is dynamically limited.

[0016] Further: the process of dynamic characteristic analysis is as follows:

[0017] (1) The received digitized data is input into the data fusion preprocessing module, and after cleaning, denoising, correction and standardization processing, high-quality data is obtained;

[0018] (2) The processed high-quality data is input into the multi-physical field coupling model for simulation calculation of thermodynamics, fluid mechanics and chemical reaction kinetics;

[0019] (3) The simulation results are output in the form of combustion characteristics and transmitted to the transfer learning engine;

[0020] (4) The transfer learning engine analyzes and processes the combustion characteristics to generate a combustion characteristic matrix;

[0021] (5) The combustion characteristic matrix is sent to the dynamic safety boundary calculation module on one hand to calculate the safety operation boundary of the unit in real time and output safety constraint parameters; on the other hand, the combustion characteristic matrix is sent to the real-time working condition prediction module to predict the future working condition of the unit and give optimization suggestion parameters;

[0022] (6) The safety constraint parameters and optimization suggestion parameters are transmitted to the control system of the unit.

[0023] Further: the multi-physical field coupling model includes a thermodynamics simulation module, a fluid mechanics simulation module, and a chemical reaction kinetics module;

[0024] The thermodynamic simulation module simulates heat transfer phenomena in the combustion process; the fluid dynamics simulation module simulates the flow characteristics of the gas in the combustion chamber and predicts possible flow instability phenomena in the combustion process; and the chemical reaction kinetics module simulates the chemical reaction process of the coal powder combustion, and can calculate the combustion rate, combustion product composition and pollutant generation under different coal types and different working conditions.

[0025] Further, the transfer learning engine takes the output of the multi-physical field coupling model, i.e., the combustion feature information; the transfer learning engine analyzes and processes the combustion features by using the transfer learning method, and establishes a combustion feature matrix.

[0026] Further, in the S1, laser-induced breakdown spectroscopy (LIBS) and near-infrared spectroscopy (NIR) are used to detect the raw coal, generate spectral data, and the process of processing the spectral data is as follows:

[0027] 1) Original spectrum acquisition: the laser-induced breakdown spectroscopy (LIBS) technology focuses high-energy laser pulses on the surface of the coal sample to generate high-temperature plasma, and the plasma emits light of specific wavelengths during the cooling process. These lights are captured by a spectrometer to form a LIBS spectrum; the LIBS spectrum contains the characteristic emission lines of elements in the coal; the NIR technology measures the absorption, reflection or scattering characteristics of the coal sample to the near-infrared light by using a near-infrared spectrometer to obtain a NIR spectrum; the NIR spectrum reflects the vibration and rotation information of the organic components in the coal;

[0028] 2) Wavelet denoising: select a wavelet basis function and a decomposition level to decompose the original spectral signal into approximate coefficients and detail coefficients of different scales; noise is usually concentrated in the high-frequency detail coefficients, and the noise components are removed by threshold processing or soft and hard threshold methods, and then the signal is reconstructed;

[0029] 3) Feature peak extraction: extract feature peaks from the wavelet denoised spectral data; for the LIBS spectrum, the feature peaks correspond to the characteristic emission lines of elements in the coal; for the NIR spectrum, the feature peaks are related to the vibration and rotation modes of the organic components in the coal;

[0030] 4) Convolutional neural network processing: input the extracted feature peak data into the convolutional neural network (CNN) for deep analysis, and the convolutional neural network (CNN) learns the complex relationship between the feature peaks and the coal quality characteristics;

[0031] 5) Element composition analysis: determine the element composition in the coal by combining the feature emission line information in the LIBS spectrum through the processing of the convolutional neural network (CNN).

[0032] Further, in the S7, the quadratic programming problem of the rolling horizon optimization (RTO) is calculated as:

[0033] ;

[0034] Constraint: A u (t)≤b, u min ≤u(t)≤u max ;

[0035] where, is the state estimation value of the system at time t; x * (t) is the reference trajectory; Q, R are state and control weight matrices respectively; A, b are linear inequality constraint matrices; u is the control variable; u * (t) is the reference control input.

[0036] The present application has the following advantages: the present application generates a three-dimensional control matrix of the coal feeder speed, the coal mill output and the air-coal ratio, dynamically optimizes the blending ratio, and improves the economy and environmental protection; the migration learning engine is used for predicting the combustion boundary, and the rolling horizon optimization (RTO) responds to the deviation within 3 minutes, so that the boiler efficiency is greater than or equal to 92%, and the pollutants do not exceed the limit.

[0037] Other features and advantages of the present application will be set forth in the following description, and in part will become apparent to those skilled in the art, and will be learned from practice of the present application. BRIEF DESCRIPTION OF DRAWINGS

[0038] In order to more intuitively illustrate the prior art and the present application, exemplary drawings are given below. It should be understood that the specific shapes, structures shown in the drawings should not be regarded as limiting conditions in the implementation of the present application; for example, based on the technical concept disclosed in the present application and the exemplary drawings, those skilled in the art can easily make routine adjustments or further optimization to the increase / decrease / attribute division of certain units (components), specific shapes, positional relationship, connection mode, size ratio relationship, etc.

[0039] Figure 1 An implementation flowchart of a coal quality parameter-based self-adaptive blending method for a thermal power generating unit is provided.

[0040] Figure 2 An implementation architecture diagram of dynamic characteristic analysis in the coal quality parameter-based self-adaptive blending method for a thermal power generating unit is provided.

[0041] Figure 3 A processing flowchart of spectral data in the coal quality parameter-based self-adaptive blending method for a thermal power generating unit is provided. DETAILED DESCRIPTION

[0042] The following embodiments of the present application are described by way of specific examples, and other advantages and effects of the present application can be easily understood by those skilled in the art from the description. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. It should be understood that these embodiments are only for further illustration of the present application, and cannot be understood as a limitation on the scope of protection of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0043] Please refer to Figure 1 A coal quality parameter-based self-adaptive blending combustion method for thermal power generating units, comprising the following steps:

[0044] S1. Raw coal entering and detection: Before raw coal enters the thermal power generating unit, coal quality parameters (such as calorific value, moisture content, ash content, sulfur content, etc.) are obtained in real time by a dual-mode online detection technology (laser-induced breakdown spectroscopy LIBS and near-infrared spectroscopy NIR).

[0045] S2. Data digitization and digital twin model updating: digitize the detected coal quality parameters and input them into the system database; based on the latest data, update the digital twin model in real time;

[0046] In addition, a pyrolysis characteristic index (such as ignition index D and stable combustion index W) and NOx generation characteristic correlation model need to be established based on the coal quality database.

[0047] The calculation formula of the pyrolysis characteristic index is:

[0048] ;

[0049] Among them, V daf is the dry ash-free volatile matter (%); A d is the dry ash content (%); FC d is the fixed carbon content (%); and D is the ignition characteristic index (the larger the value, the easier the ignition).

[0050] The NOx generation prediction model is as follows:

[0051] ;

[0052] Among them, T max is the peak temperature of the furnace (K); E a is the activation energy (J / mol); R is the gas constant (8.314 J / (mol·K)); [O2] / [N2] is the oxygen-nitrogen ratio (molar concentration ratio); τ is the gas residence time (s); and k is the coal-related coefficient (fitted by historical data).

[0053] The constraint system of the multi-objective optimization model is constructed: boiler efficiency ≥ 92%, main reheat steam temperature deviation < 10℃, furnace outlet flue gas temperature < 1350℃, pollutant emission ≤ ultra-low limit value, and pulverizing system safety margin > 15%.

[0054] The objective function of the multi-objective optimization model is:

[0055] ;

[0056] The constraint conditions are:

[0057] ;

[0058] wherein x i is the blending ratio of the i-th coal feeder (decision variable); C coal is the coal price vector (yuan / ton); E j is the emission intensity of pollutant j (g / kWh); w j is the pollutant weight coefficient (set according to environmental protection policy); η boiler is the boiler efficiency (calculated through the heat balance equation).

[0059] S3. Dynamic characteristics and safety analysis: dynamic characteristics analysis is performed through the digital twin model to predict the response of coal combustion (such as combustion rate, heat release curve); at the same time, the safety constraint boundary (such as temperature, pressure, emission limit value) is calculated to ensure operation within the safety range.

[0060] S4. Economic optimization and dynamic correction: within the safety boundary, an improved NSGA-II multi-objective optimization algorithm is used to take economic efficiency (coal consumption cost) and environmental protection (converted pollutant discharge) as the objective function, to generate a three-dimensional control matrix containing coal feeder speed setting, coal mill output distribution, and air-coal ratio adjustment, and to generate the optimal blending ratio; dynamic compensation correction is performed according to real-time operation data (such as load fluctuation, coal quality change) to adjust the blending strategy.

[0061] Among them, the coal type dynamic matching rules that need to be introduced include: high-sulfur coal and low-sulfur coal are matched according to sulfur balance, high-ash fusion point coal and low-ash fusion point coal are matched according to ash adhesion temperature characteristics, and the blending ratio of inferior coal is dynamically limited (single unit < 30%).

[0062] S5. Instruction issuance and execution: the optimized blending instruction is issued to the combustion system for execution to control the coal feeder, air volume, and other equipment, and precise blending is realized.

[0063] S6. Emission and efficiency monitoring: real-time monitoring of emission indicators (such as NOx, SO2 concentration) and combustion efficiency after combustion, and collection of actual operation data.

[0064] S7. Deviation analysis and feedback correction: deviation analysis of monitoring data and expected target;

[0065] When the key parameters of the detection data deviate from the preset threshold, trigger the rolling horizon optimization (RTO) to correct the strategy, the response time is less than 3 minutes;

[0066] The quadratic programming problem of rolling horizon optimization (RTO) is calculated as

[0067] ;

[0068] The constraint condition is: A u (t)≤b, u min ≤u(t)≤u max ;

[0069] Where, is the system state estimation value (such as furnace temperature, oxygen content) x * (t) is the reference trajectory (generated by MPC prediction); Q, R are state and control weight matrices; A, b are linear inequality constraint matrices (corresponding to equipment safety limits); u is the control variable (here it refers to the control instruction directly acting on the unit), u(t) is the actual control input at the current time (such as the coal feeder speed set value); u * (t) is the reference control input (such as the optimization result of the last time or the manual set value).

[0070] If the detection data is within the normal range, only update the historical database and maintain the current running state.

[0071] S8. Iterative update of knowledge base: machine learning training of the deviation between actual blending data and theoretical model, updating the coal quality-load-environmental protection characteristic database, accumulating typical working condition case library, forming the mapping relationship table of "working condition characteristics-optimal coal blending-control parameters"; generating blending economy analysis report regularly to guide the optimization of procurement decision.

[0072] The process of dynamic characteristic analysis is as follows:

[0073] (1) The received digitized data is input into the data fusion preprocessing module, and after cleaning, denoising, correction and standardization processing, high-quality data is obtained;

[0074] (2) These processed high-quality data are then sent to the multi-physical field coupling model for simulation calculation of thermodynamics, fluid mechanics and chemical reaction kinetics;

[0075] (3) The simulation results are output in the form of combustion characteristics and transmitted to the transfer learning engine;

[0076] (4) The migration learning engine analyzes and processes these combustion features to generate a combustion feature matrix;

[0077] (5) The combustion feature matrix is sent to the dynamic safety boundary calculation module on one hand, which is used to calculate the safety operation boundary of the unit in real time and output safety constraint parameters; on the other hand, the combustion feature matrix is also sent to the real-time working condition prediction module, which is used to predict the future working condition of the unit and give optimization suggestion parameters;

[0078] (6) Both the safety constraint parameters and the optimization suggestion parameters are transmitted to the control system of the unit to realize self-adaptive optimization control of the combustion process of the thermal power unit.

[0079] Among them, the specific content of data fusion preprocessing is as follows:

[0080] Data fusion preprocessing is the starting step of the whole dynamic feature analysis, which is responsible for receiving raw data from various sensors of the thermal power unit (including laser-induced breakdown spectroscopy LIBS probe), including coal quality parameters, unit operation state parameters and environmental parameters.

[0081] The data fusion preprocessing module will clean, denoise, correct and standardize these raw data; cleaning is to remove outliers and error data in the data; denoising is to eliminate random noise interference in the data; correction is to correct possible system errors of the sensor; standardization is to unify different dimension and order of magnitude data to the same scale, which is convenient for subsequent analysis and processing.

[0082] The data quality after preprocessing is significantly improved, providing a reliable data basis for the subsequent multi-physical field coupling model, ensuring the accuracy and stability of the analysis results.

[0083] The multi-physical field coupling model is the core model of the dynamic analysis layer, including the thermodynamic simulation module, the fluid mechanics simulation module and the chemical reaction dynamics module; the multi-physical field coupling model considers multiple physical fields involved in the combustion process of the thermal power unit, including thermodynamic field, fluid mechanics field and chemical reaction dynamics field.

[0084] Among them, the thermodynamic simulation module is used to simulate the heat transfer phenomenon in the combustion process, such as the temperature distribution in the furnace, the heat enthalpy change of the flue gas, etc.; it can calculate the thermal efficiency of the unit under different working conditions, providing the basis for optimizing combustion from the thermodynamic aspect.

[0085] The fluid mechanics simulation module is used to simulate the flow characteristics of gases in the combustion chamber, including velocity field, pressure field and turbulence characteristics, etc.; it can predict possible flow instability phenomena in the combustion process, such as flame deflection, uneven combustion, etc., and provide guidance for optimizing the design and operating parameters of the combustor from the fluid mechanics aspect.

[0086] The chemical reaction kinetics module is used to simulate the combustion chemical reaction process of coal powder, including coal pyrolysis, gasification, oxidation and other reaction steps; it can calculate the combustion rate, combustion product composition and pollutant generation under different coal types and different working conditions, providing chemical kinetics basis for optimizing combustion ratio and reducing pollutant emissions.

[0087] The multi-physics coupling model organically combines the three modules, fully considering the interaction and coupling relationship between them; for example, the change of temperature field will affect the chemical reaction rate, and the flow of gas will affect the heat transfer; through the simulation of multi-physics coupling, the model can more truly reflect the complexity of the combustion process of thermal power units, and provide more accurate and comprehensive combustion feature information for the subsequent transfer learning engine.

[0088] The transfer learning engine takes the output of the multi-physics coupling model, i.e. the combustion feature information; the transfer learning engine uses the method of transfer learning to analyze and process these combustion features.

[0089] Transfer learning is a machine learning method, and the core idea of transfer learning is to transfer the knowledge learned from one task or field to another related task or field. In the scene of self-adaptive blending of thermal power units, the transfer learning engine can use the knowledge learned from historical data or data of similar units to speed up the learning and adaptation process of new units or new coal types.

[0090] The transfer learning engine extracts key features from the combustion features output by the multi-physics coupling model and uses these features to construct a combustion feature matrix.

[0091] The combustion feature matrix contains various information reflecting the state and performance of the combustion process, such as combustion stability, combustion efficiency, pollutant emission level, etc.

[0092] The dynamic safety boundary calculation module calculates the dynamic safety boundary of the unit operation based on the combustion feature matrix output by the transfer learning engine.

[0093] The operation of thermal power units is limited by various safety constraints, such as main steam temperature, pressure, furnace temperature, pollutant emission concentration, etc.; these safety constraints constitute the safety operation boundary of the unit.

[0094] The dynamic safety boundary calculation module dynamically adjusts and optimizes these safety boundaries according to the current combustion feature matrix; for example, when the coal quality changes, the module will recalculate the safety operation boundary of the unit under the current working condition according to the new coal quality parameters and combustion features, to ensure that the unit always operates within a safe and reliable range.

[0095] Finally, the dynamic security boundary calculation module outputs a series of security constraint parameters, which will be passed to the control system of the unit for real-time monitoring and adjustment of the unit's operating state, ensuring the safety and stability of the unit.

[0096] The real-time working condition prediction module also takes the combustion feature matrix output by the transfer learning engine as input. The real-time working condition prediction module predicts the real-time working condition of the thermal power unit and gives optimization suggestion parameters.

[0097] The real-time working condition prediction module uses historical data and the current combustion feature matrix to establish a prediction model of the unit's operating condition. It can predict the unit's operating state in the future, such as load trend, main parameter trend, etc.

[0098] Based on the prediction results, the real-time working condition prediction module gives optimization suggestion parameters, such as coal supply, air supply, and induced air supply. The optimization suggestion parameters aim to help operators adjust the unit's operating parameters in advance, so that the unit can better adapt to future working condition changes and improve the unit's operating efficiency and stability.

[0099] The optimization suggestion parameters are also passed to the control system of the unit as a reference for automatic control, realizing adaptive optimization operation of the unit.

[0100] Referring to Figure 2 , Figure 3 , the detailed process of processing the spectral data is as follows:

[0101] 1) Obtain the original spectrum

[0102] LIBS technology focuses high-energy laser pulses on the surface of the coal sample to generate high-temperature plasma. During the cooling process of the plasma, specific wavelength light is emitted, which is captured by the spectrometer to form a LIBS spectrum. The LIBS spectrum contains the characteristic emission lines of elements in the coal and can be used for elemental composition analysis.

[0103] NIR technology uses a near-infrared spectrometer to irradiate the coal sample and measure its absorption, reflection, or scattering characteristics of near-infrared light to obtain an NIR spectrum. The NIR spectrum reflects the vibration and rotation information of organic components in the coal, such as carbon, hydrogen, and oxygen, and can be used for rapid analysis of coal quality characteristics such as moisture, volatile matter, and fixed carbon.

[0104] 2) Wavelet denoising

[0105] Selecting appropriate wavelet basis function and decomposition level, the original spectral signal is decomposed into approximate coefficients and detail coefficients of different scales; noise is usually concentrated in high-frequency detail coefficients, and noise components are removed by threshold processing or soft and hard threshold method, and then the signal is reconstructed to realize denoising.

[0106] The noise in the spectral data is removed, the data quality is improved, and accurate data basis is provided for subsequent feature peak extraction and element composition analysis.

[0107] 3) Feature peak extraction

[0108] Feature peaks are extracted from the spectral data after wavelet denoising. For LIBS spectrum, the feature peaks correspond to the characteristic emission lines of elements in coal; for NIR spectrum, the feature peaks are related to the vibration and rotation modes of organic components in coal.

[0109] Feature peaks contain the composition information of coal, which is the key basis for subsequent element composition analysis and coal quality evaluation.

[0110] 4) Convolutional neural network processing

[0111] In spectral data processing, convolutional neural network CNN can be used for feature extraction, classification and regression tasks; the extracted feature peak data is input into CNN for deep analysis, and CNN can learn the complex relationship between feature peaks and coal quality characteristics, improve the accuracy of element composition analysis and coal quality evaluation.

[0112] 5) Element composition analysis

[0113] Through the processing of CNN, combined with the characteristic emission line information in LIBS spectrum, the element composition (such as carbon, hydrogen, oxygen, nitrogen, sulfur, etc.) in coal can be determined; these element compositions are important indicators for evaluating coal quality and combustion characteristics; element composition analysis provides basic data for combustion characteristic evaluation of coal, which helps to optimize combustion process and improve combustion efficiency.

[0114] Through the above process, LIBS and NIR dual-mode online detection technology combined with wavelet denoising, feature peak extraction and convolutional neural network realizes accurate processing and analysis of spectral data of coal, and finally obtains the combustion characteristic matrix of coal, which provides strong support for clean and efficient utilization of coal.

[0115] The above only describes the preferred embodiments of the present application, and does not limit the present application, any modification, equivalent replacement and improvement within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. An adaptive co-firing method for thermal power units based on coal quality parameters, characterized in that, Includes the following steps: S1. Laser-induced breakdown spectroscopy (LIBS) and near-infrared spectroscopy (NIR) are used to detect raw coal and obtain coal quality parameters in real time; S2. Digitize the detected coal quality parameters and input them into the system database; The digital twin model is updated in real time based on the latest data; S3. Dynamic characteristic analysis is performed using a digital twin model to predict the response of coal combustion; at the same time, safety constraint boundaries are calculated. S4. Within the safety boundary, the improved NSGA-II multi-objective optimization algorithm is adopted, with economy and environmental protection as objective functions, to generate a three-dimensional control matrix including coal feeder speed setting, coal mill output distribution, and air-coal ratio adjustment, and to generate the optimal blending ratio. Dynamic compensation and correction are performed based on real-time operational data to adjust the blending strategy. S5. Send the optimized blending command to the combustion system for execution, control the equipment, and achieve precise blending; S6. Real-time monitoring of post-combustion emissions and combustion efficiency, and collection of actual operating data; S7. Perform deviation analysis between the collected operational data and the expected target; when the key parameters of the operation deviate from the preset threshold, trigger rolling time-domain optimization to correct the strategy; if the detected data is within the normal range, only update the historical database and maintain the current operating state; S8. Use machine learning to train the model to compare the deviations between the actual calcination data and the theoretical model; Regularly generate economic analysis reports on co-firing.

2. The adaptive co-firing method for thermal power units based on coal quality parameters according to claim 1, characterized in that, In step S2, a correlation model between the pyrolysis characteristic index and NOx generation characteristics needs to be established based on the system database of coal quality data.

3. The adaptive co-firing method for thermal power units based on coal quality parameters according to claim 1, characterized in that, In S4, the dynamic coal matching rules that need to be introduced include: high-sulfur coal and low-sulfur coal are matched according to sulfur balance ratio, high-ash fusion point coal and low-ash fusion point coal are matched according to ash viscosity temperature characteristics, and the proportion of low-quality coal blending is dynamically limited.

4. The adaptive co-firing method for thermal power units based on coal quality parameters according to claim 1, characterized in that, The process of dynamic characteristic analysis is as follows: (1) The received digitized data is input into the data fusion preprocessing module. After cleaning, noise reduction, correction and standardization, high-quality data is obtained. (2) Input the processed high-quality data into the multiphysics coupling model to perform simulation calculations of thermodynamics, fluid mechanics and chemical reaction kinetics; (3) The simulation results are output in the form of combustion features and passed to the transfer learning engine; (4) The transfer learning engine analyzes and processes the combustion features to generate a combustion feature matrix; (5) The combustion feature matrix is ​​sent to the dynamic safety boundary calculation module to calculate the safe operating boundary of the unit in real time and output the safety constraint parameters; on the other hand, the combustion feature matrix is ​​sent to the real-time operating condition prediction module to predict the future operating conditions of the unit and provide optimization suggestion parameters. (6) Safety constraint parameters and optimization suggestion parameters are transmitted to the unit's control system.

5. The adaptive co-firing method for thermal power units based on coal quality parameters according to claim 4, characterized in that, The multiphysics coupling model includes a thermodynamic simulation module, a fluid dynamics simulation module, and a chemical reaction kinetics module; Among them, the thermodynamic simulation module simulates the heat transfer phenomenon during combustion and can calculate the thermal efficiency of the unit under different operating conditions; the fluid dynamics simulation module simulates the flow characteristics of gas in the combustion chamber and predicts the flow instability that may occur during combustion; the chemical reaction kinetics module simulates the chemical reaction process of pulverized coal combustion and can calculate the combustion rate, composition of combustion products and amount of pollutants generated under different coal types and operating conditions.

6. The adaptive co-firing method for thermal power units based on coal quality parameters according to claim 4, characterized in that, The transfer learning engine receives the output of the multiphysics coupling model, namely combustion feature information; The transfer learning engine uses transfer learning methods to analyze and process combustion features and build a combustion feature matrix.

7. The adaptive co-firing method for thermal power units based on coal quality parameters according to claim 1, characterized in that, In step S1, laser-induced breakdown spectroscopy (LIBS) and near-infrared spectroscopy (NIR) are used to detect the raw coal, generating spectral data. The process of processing the spectral data is as follows: 1) Acquisition of raw spectrum: Laser-induced breakdown spectroscopy (LIBS) technology uses high-energy laser pulses focused onto the surface of a coal sample to generate high-temperature plasma. During the cooling process of the plasma, light of a specific wavelength is emitted, which is captured by a spectrometer to form a LIBS spectrum. LIBS spectroscopy contains the characteristic emission lines of elements in coal; NIR technology uses a near-infrared spectrometer to irradiate a coal sample and measures the absorption, reflection, or scattering characteristics of the coal sample to near-infrared light to obtain the NIR spectrum; NIR spectroscopy reflects the vibrational and rotational information of organic components in coal; 2) Wavelet denoising: Select the wavelet basis function and the number of decomposition levels to decompose the original spectral signal into approximation coefficients and detail coefficients at different scales; noise is usually concentrated in the high-frequency detail coefficients, and the noise components are removed by thresholding or soft and hard thresholding methods, and then the signal is reconstructed. 3) Feature peak extraction: Feature peaks are extracted from the wavelet-denoised spectral data. For LIBS spectra, the feature peaks correspond to the characteristic emission lines of elements in coal; for NIR spectra, the feature peaks are related to the vibration and rotation modes of organic components in coal. 4) Convolutional Neural Network Processing: The extracted feature peak data is input into a convolutional neural network (CNN) for deep analysis. The CNN learns the complex relationship between the feature peaks and coal quality characteristics. 5) Elemental composition analysis: The elemental composition of coal is determined by processing with a convolutional neural network (CNN) and combining it with the characteristic emission line information in the LIBS spectrum.

8. The adaptive co-firing method for thermal power units based on coal quality parameters according to claim 1, characterized in that, In S7, the quadratic programming problem of Rolling Time-Domain Optimization (RTO) is calculated as follows: ； The constraint is: A u (t)≤b,u min ≤u(t)≤u max ; in, x is the estimated state of the system at time t; * (t) represents the reference trajectory; Q and R are the state and control weight matrices, respectively; A and b are the linear inequality constraint matrices; u is the control variable; u(t) is the actual control input at the current moment; u * (t) is the reference control input.

Images