On-line detection control method and system for fly ash blackening in thermal power plant

Abstract

The invention relates to the technical field of fire coal control, and discloses a thermal power plant fly ash floating black online detection control method and system, and the method comprises the steps: 1, employing microwave, infrared and laser sensors for cooperative work, detecting carbon dielectric characteristics through microwave dust interference resistance, identifying thermal radiation difference through infrared, and analyzing particle distribution through laser; 2, on the basis of the detection requirements in the step 1, triple protection of air curtain barrier, ultrasonic cleaning and hydrophobic coating is designed, dust condensation is blocked, and it is ensured that the light transmittance of the sensor is larger than 90%; the invention provides a multi-mode sensing fusion-anti-interference current stabilization-closed loop optimization full-link solution for the industrial problem of signal distortion caused by optical window dew formation and dust deposition, airflow disturbance and environmental interference in fly ash floating black online detection. The light transmittance of the optical window is greater than 93% by combining triple protection of air curtain barrier, ultrasonic cavitation stripping and nano hydrophobic coating.

Description

Technical Field

[0001] The present invention relates to the technical field of coal combustion control, in particular to an online detection and control method and system for fly ash blackening in a thermal power plant. Background Art

[0002] Boiler fly ash blackening refers to the appearance of black oil or black precipitate on the surface of fly ash or after contact with water. Blackening can negatively impact the performance of concrete. Fly ash with a high carbon content contains carbon particles that absorb a significant amount of cement hydration products, affecting the concrete's hardening process and strength development. Fly ash with a high loss on ignition can negatively impact the concrete's workability, setting time, and hardening process, further impacting the quality of construction. Blackening of fly ash is combustible, implying reduced furnace efficiency and carbon loss. Severe blackening can also make fly ash sales difficult, leading to direct economic losses.

[0003] There are many possible causes of fly ash blackening, including unburned fuel due to poor combustion organization, carbon blackening caused by localized oxygen deficiency in the furnace due to an irrational combustion flow field, combustion equipment design flaws, and contamination with fuel additives and impurities. Combustion theory research and practical experience indicate that these issues can be addressed to varying degrees through combustion optimization, maintaining near-optimal combustion conditions. This can significantly reduce the majority of fly ash blackening issues, while also potentially lowering the carbon content of fly ash and improving economic efficiency.

[0004] However, in the existing technology, online detection of fly ash black is susceptible to multiple environmental interferences such as condensation and dust accumulation on the optical window, airflow disturbance, temperature drift and dust humidity changes, resulting in unstable signals, frequent false alarms and missed alarms, and a significant increase in maintenance costs and shutdown calibration frequency.

[0005] Therefore, those skilled in the art provide a method and system for online detection and control of fly ash blackening in a thermal power plant to solve the above-mentioned problems. Summary of the Invention

[0006] In view of the deficiencies in the prior art, the present invention provides an online detection and control method and system for fly ash blackening in a thermal power plant to solve the problems raised in the above-mentioned background technology.

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions: an online detection and control method and system for fly ash blackening in a thermal power plant, comprising the following steps:

[0008] Step 1: Using microwave, infrared, and laser sensors to work together, microwaves detect carbon dielectric properties against dust interference, infrared identifies thermal radiation differences, and lasers analyze particle distribution. Cross-validation provides a reliable data foundation.

[0009] Step 2: Based on the detection requirements of step 1, a triple protection system consisting of an air curtain barrier, ultrasonic cleaning, and a hydrophobic coating is designed to block dust condensation and ensure that the sensor transmittance is greater than 90%.

[0010] Step 3: Combined with the clean environment in step 2, temperature, humidity, and dust sensors are deployed. The multi-sensor data from step 1 is integrated with environmental parameters in real time through the LSTM model to dynamically correct the signal, reducing the error from ±15% to ±3%.

[0011] Step 4: Install a porous ceramic flow stabilizer upstream of the detection area in step 1 to suppress airflow disturbances to ±5% fluctuations. Combined with adaptive negative pressure control, this ensures sampling uniformity and improves the accuracy of the compensation model in step 3.

[0012] Step 5: Based on the system stability of steps 1-4, the sensor parameters are automatically calibrated every 4 hours; when an abnormality occurs, multi-sensor cross-validation is triggered, the false alarm rate is less than 0.5%, and the calibration cycle is extended by 3 times;

[0013] Step 6: Integrate the calibration data from step 5 and link it to the power plant control system in real time to dynamically adjust the air-coal ratio and pulverized coal temperature, stabilizing the carbon content at 3% to 5%, and reducing operation and maintenance costs by 50%.

[0014] Preferably, the step 1 includes:

[0015] Step 1.1: Based on the dielectric properties of the unburned carbon and ash in fly ash, a dielectric model with a shape factor correction term is introduced:

[0016]

[0017] ∈ eff : mixed dielectric constant;

[0018] ∈ c ,∈ a : Dielectric constant of carbon particles and ash;

[0019] f c : carbon volume fraction;

[0020] κ: shape factor;

[0021] Step 1.2: Based on the strong absorption characteristics of unburned carbon in the mid-infrared band, a nonlinear relationship between emissivity and carbon content is established:

[0022]

[0023] ε(λ,C): surface emissivity at wavelength λ and carbon content C;

[0024] ε0(λ): base emissivity of pure ash in the λ band;

[0025] α: absorption coefficient;

[0026] C0: characteristic carbon content threshold;

[0027] Step 1.3 uses the multi-angle scattered light intensity data I(θi) to solve the particle size distribution n(d) through Tikhonov regularization:

[0028]

[0029] V(θ i ,d,m): Mie scattering kernel function;

[0030] γ: regularization parameter;

[0031] d min , d max : Particle size range;

[0032] Step 1.4 defines the basic probability assignment (BPA) function of the three sensors for the proposition “high black float concentration”:

[0033]

[0034] m sensor(.) : the confidence level assigned by each sensor to the proposition;

[0035] A,B,C,D: a set of propositions.

[0036] Preferably, the step 2 includes:

[0037] Step 2.1: To quantify the protective effect of the air curtain on the optical window, a modified air film cooling efficiency model is introduced:

[0038] Define the minimum energy density threshold required for ultrasonic cleaning of dust deposits:

[0039]

[0040] f: ultrasonic frequency;

[0041] Q: oscillator amplitude;

[0042] ρ dust : Dust bulk density;

[0043] δ: thickness of dust layer;

[0044] σ adhesion : Dust-window adhesion strength;

[0045] σ dust : Shear strength of dust itself;

[0046] Step 2.2: A contact angle aging equation is proposed to address the degradation of hydrophobicity under high-speed airflow:

[0047]

[0048] θ0: initial contact angle;

[0049] θ min : limiting contact angle;

[0050] U air : air curtain velocity;

[0051] L: wear coefficient;

[0052] t: running time;

[0053] Step 2.3: Integrate the effects of the three mechanisms on transmittance;

[0054]

[0055] z: light transmittance (target > 90%);

[0056] w: basic pollution rate;

[0057] η,E\E crit ,θ(t): comes from the three sub-models mentioned above.

[0058] Preferably, the step 3 includes:

[0059] Step 3.1 Inject physical constraints of environmental parameters into the standard LSTM forget gate:

[0060]

[0061] Physical consistency loss term;

[0062] φ i : Intermediate feature vectors output by microwave, infrared, and laser sensors;

[0063] λ i :Physical constraint weight;

[0064] Step 3.2 defines the dynamic attention weights between sensor data and environmental parameters:

[0065]

[0066] Characteristics of the i-th sensor at time t;

[0067] Environmental parameters of category j;

[0068] W a :Learnable projection matrix;

[0069] Step 3.3 adds a residual path between the sensor's original signal and the hidden state inside the LSTM cell:

[0070]

[0071] β i : sensor residual gain;

[0072] The residual projection matrix of sensor i;

[0073] Step 3.4 designs a loss function for the fast and slow dual-scale characteristics of fly ash concentration changes:

[0074]

[0075] mean absolute error;

[0076] ||·|| TV : total variation penalty term;

[0077] J1=0.3, J2=0.7: fast / slow scale penalty weights;

[0078] T long :Long period window.

[0079] Preferably, the step 4 includes:

[0080] Step 4.1: Based on the Forchheimer modified equation, the pore structure shape factor is introduced:

[0081] s: permeability (1e-105e-10m' determined by pore size distribution);

[0082] x: inertial drag coefficient;

[0083] n: pore curvature shape factor; |

[0084] g mean : average pore size;

[0085] Step 4.2 Design a fractional sliding surface to suppress sudden disturbances in the airflow:

[0086] s=D O e+λ1e+λ2∫edt(0<α<1),

[0087] Control law:

[0088]

[0089] e=U real -U set: Flow rate deviation (U s et=1.0m / s);

[0090] O=0.7, I=0.5: fractional-order damping term to enhance anti-interference performance;

[0091] Disturbance observer estimate.

[0092] Preferably, the step 4.2 includes:

[0093] Step 4.2.1 Combine the flow stability and sampling efficiency goals to construct a dynamic weight loss function:

[0094] Step 4.2.2 Dynamic Weight Strategy:

[0095]

[0096] C opt =200g / m' optimal dust concentration;

[0097] t i,last : The last time the i-th target exceeded the limit.

[0098] Preferably, the step 5 includes:

[0099] Step 5.1 Dynamically assign sensor weights based on Bayesian posterior probability, prioritize correcting low-confidence sensor parameters, and verify the validity of calibration results through a sliding window T test to avoid miscalibration:

[0100] Define the real-time confidence B of sensor i i :

[0101]

[0102] P(D i |S true ):Sensor i is in the real state S t Likelihood function under rue;

[0103] P prior (S true ): Prior probability;

[0104] T: time decay factor;

[0105] When B i <0.7 triggers calibration, weight w i =B i / ∑B j Used for multi-sensor data fusion;

[0106] Step 5.2: Construct the sensor state transition matrix:

[0107]

[0108] Among them, status 0: normal; status 1: suspected abnormality; status 2: confirmed failure.

[0109] Preferably, the step 5.2 includes:

[0110] Step 5.2.1: Establish a time decay model for the sensor gain parameters, predict the drift trend and pre-compensate. The calibration period is adaptively adjusted according to the drift rate. The time evolution model of the sensor gain is:

[0111]

[0112] m i :Dust concentration cumulative drift coefficient;

[0113] T thermal : Thermal cycle;

[0114] l: thermal expansion effect amplitude;

[0115] Pre-compensation: Calculate ΔG in real time during the calibration cycle i =G i (t)-G i,meas And reversely correct the output signal;

[0116] Step 5.2.2 Periodic update rule based on drift rate:

[0117]

[0118] T base =4h,

[0119] ΔG avg : Average gain deviation of the last five calibrations;

[0120] δ th =1%: maximum allowed deviation threshold;

[0121] Constraint: T n ew∈[4h,72h].

[0122] Preferably, the step 6 includes:

[0123] Step 6.1 Based on the black smoke concentration data calibrated in step 5, the ratio of primary air volume to coal feed rate is adjusted through feedforward-feedback composite control;

[0124] Step 6.1.1 Dynamically set the pulverizer outlet temperature based on the volatile matter content of the pulverized coal to suppress the formation of unburned carbon;

[0125] Step 6.1.2 Dynamic feedforward compensation term:

[0126]

[0127] Feedback PID correction term:

[0128]

[0129] Total air-to-coal ratio command:

[0130] R total =R ff +R fb ,

[0131] u=0.15、p=0.05: coal quality response coefficient, τ s =25s: system inertia time constant, γ=

[0132] 0.3: nonlinear damping factor, suppressing differential oscillations when large deviations occur;

[0133] Step 6.2: Construct the Pareto frontier of combustion efficiency (η), NOx emissions (E), and power generation load (P), and find the optimal balance point between carbon content (C) and operation and maintenance cost (Cost);

[0134] Step 6.2.1 Update the target weights through online rolling optimization to adapt to the peak load demand of the power grid:

[0135] Step 6.2.2 Pareto objective function:

[0136]

[0137] Dynamic weight update rules:

[0138]

[0139] Q = 1.5: cost penalty curvature, κ = 200 mg / MJ: environmental protection constraint threshold;

[0140] Among them, the improved NSGA-II algorithm is used, the population size is 50, and the crossover probability is 0.8

[0141] Step 6.3: Design a time-delay observer to compensate for the transmission delay from the coal mill to the burner, ensuring that the control period is ≤ 30 seconds.

[0142] Step 6.3.1 Transmission Delay Estimation Model:

[0143]

[0144] Predictive-corrective control law:

[0145]

[0146] K obs =0.05: observer gain, L pipe : Length of pipeline from coal mill to burner.

[0147] Preferably, an online detection and control system for fly ash blackening in a thermal power plant uses the online detection and control method for fly ash blackening in a thermal power plant according to any one of claims 1 to 9, comprising:

[0148] The data monitoring module uses microwave, infrared, and laser sensors to work together. Microwaves detect carbon dielectric properties against dust interference, infrared identifies thermal radiation differences, and lasers analyze particle distribution. Cross-validation provides a reliable data foundation.

[0149] The data fusion module, based on the detection requirements of step 1, designs a triple protection system of air curtain barrier + ultrasonic cleaning + hydrophobic coating to block dust condensation and ensure that the sensor transmittance is greater than 90%;

[0150] The dynamic correction module combines the clean environment of step 2 with temperature, humidity, and dust sensors. Using the LSTM model, the multi-sensor data from step 1 is integrated with environmental parameters in real time to dynamically correct the signal, reducing the error from ±15% to ±3%.

[0151] The sampling suppression module installs a porous ceramic flow stabilizer upstream of the detection area in step 1 to suppress airflow disturbances to ±5% fluctuations. This, combined with adaptive negative pressure control, ensures sampling uniformity and improves the accuracy of the compensation model in step 3.

[0152] The calibration module automatically calibrates sensor parameters every four hours based on steps 1-4 to stabilize the system. It triggers multi-sensor cross-validation when an anomaly occurs, with a false alarm rate of less than 0.5% and a calibration cycle extended by three times.

[0153] The integration module integrates the calibration data from step 5, links with the power plant control system in real time, dynamically adjusts the air-coal ratio and pulverized coal temperature, stabilizes the carbon content at 3% to 5%, and reduces operation and maintenance costs by 50%.

[0154] The present invention provides a method and system for online detection and control of fly ash blackening in thermal power plants. It has the following beneficial effects:

[0155] 1. This invention addresses the industry problem of signal distortion caused by condensation and dust accumulation on optical windows, airflow disturbances and environmental interference in online detection of fly ash blackening, and proposes a full-link solution of multimodal sensing fusion, anti-disturbance and flow stabilization, and closed-loop optimization: through microwave / infrared / laser cross-physical field cross-validation, combined with triple protection of air curtain barrier + ultrasonic cavitation stripping + nano-hydrophobic coating, the optical window transmittance can reach >93%.

[0156] 2. This invention suppresses airflow disturbance to ±5% through innovative gradient porous ceramic flow stabilizers and fractional-order sliding mode control, and couples the physical constraint LSTM model to compress the detection error from ±15% to ±3%.

[0157] 3. The present invention extends the calibration period from 8 hours to 72 hours through the Bayesian-HMM dual-trigger calibration and gain drift pre-compensation model; ultimately, the decoupling control of the air-coal ratio and the Pareto optimization of the pulverized coal temperature are linked to achieve 30-second dynamic adjustment, stabilize the carbon content at 3%-5%, increase the power generation per ton of coal by 2.1%, reduce annual CO2 emissions by 5,200 tons / unit, and reduce operation and maintenance costs by 52%. BRIEF DESCRIPTION OF THE DRAWINGS

[0158] Figure 1 is a flow chart of the present invention;

[0159] Figure 2 It is a system diagram of the present invention. DETAILED DESCRIPTION

[0160] To help those skilled in the art understand the present invention, the following will provide a clear and complete description of the technical solutions in the embodiments of the present invention, in conjunction with the accompanying drawings. Obviously, the described embodiments are only partial embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0161] The present invention is described in detail below with reference to the accompanying drawings:

[0162] Example 1:

[0163] Please see the attached Figure 1 The embodiment of the present invention provides an online detection and control method for fly ash blackening in a thermal power plant, comprising the following steps:

[0164] Step 1: Using microwave, infrared, and laser sensors to work together, microwaves detect carbon dielectric properties against dust interference, infrared identifies thermal radiation differences, and lasers analyze particle distribution. Cross-validation provides a reliable data foundation.

[0165] Step 2: Based on the detection requirements of step 1, a triple protection system consisting of an air curtain barrier, ultrasonic cleaning, and a hydrophobic coating is designed to block dust condensation and ensure that the sensor transmittance is greater than 90%.

[0166] Step 3: Combined with the clean environment in step 2, temperature, humidity, and dust sensors are deployed. The multi-sensor data from step 1 is integrated with environmental parameters in real time through the LSTM model to dynamically correct the signal, reducing the error from ±15% to ±3%.

[0167] Step 4: Install a porous ceramic flow stabilizer upstream of the detection area in step 1 to suppress airflow disturbances to ±5% fluctuations. Combined with adaptive negative pressure control, this ensures sampling uniformity and improves the accuracy of the compensation model in step 3.

[0168] Step 5: Based on the system stability of steps 1-4, the sensor parameters are automatically calibrated every 4 hours; when an abnormality occurs, multi-sensor cross-validation is triggered, the false alarm rate is less than 0.5%, and the calibration cycle is extended by 3 times;

[0169] Step 6: Integrate the calibration data from step 5 and link it to the power plant control system in real time to dynamically adjust the air-coal ratio and pulverized coal temperature, stabilizing the carbon content at 3% to 5%, and reducing operation and maintenance costs by 50%.

[0170] Preferably, the step 1 includes:

[0171] Step 1.1: Based on the dielectric properties of the unburned carbon and ash in fly ash, a dielectric model with a shape factor correction term is introduced:

[0172]

[0173] ∈ eff : mixed dielectric constant;

[0174] ∈ c ,∈ a : Dielectric constant of carbon particles and ash;

[0175] f c : carbon volume fraction;

[0176] κ: shape factor);

[0177] Step 1.2: Based on the strong absorption characteristics of unburned carbon in the mid-infrared band, a nonlinear relationship between emissivity and carbon content is established:

[0178]

[0179] ε(λ,C): surface emissivity at wavelength λ and carbon content C;

[0180] ε0(λ): base emissivity of pure ash in the λ band;

[0181] α: absorption coefficient;

[0182] C0: characteristic carbon content threshold;

[0183] Step 1.3 uses the multi-angle scattered light intensity data I(θi) to solve the particle size distribution n(d) through Tikhonov regularization:

[0184]

[0185] V(θi ,d,m): Mie scattering kernel function;

[0186] γ: regularization parameter;

[0187] d min , d max : Particle size range;

[0188] Step 1.4 defines the basic probability assignment (BPA) function of the three sensors for the proposition “high black float concentration”:

[0189]

[0190] m sensor(.) : the confidence level assigned by each sensor to the proposition;

[0191] A, B, C, D: set of propositions;

[0192] Specifically, the shape factor corrected dielectric model accurately characterizes irregular unburned carbon particles, and is combined with the mid-infrared nonlinear absorption model to improve the dynamic detection range of carbon content. Tikhonov regularization and anti-noise optimization are used to achieve high-resolution inversion of the entire particle size range of 0.1 to 200 μm. Multi-sensor BPA evidence fusion is used to achieve a >99% confidence level in abnormality detection, forming a dielectric-infrared-particle size multi-physics field cross-validation system, which reduces the error in floating black concentration detection from ±15% to ±3%. In actual measurements of 600MW units, the boiler efficiency is increased by 1.8%, saving 8 million yuan in coal costs annually, overcoming the problem of online detection signal distortion under high dust conditions, and filling the technical gap in the field of clean coal combustion.

[0193] Preferably, the step 2 includes:

[0194] Step 2.1: To quantify the protective effect of the air curtain on the optical window, a modified air film cooling efficiency model is introduced:

[0195] Define the minimum energy density threshold required for ultrasonic cleaning of dust deposits:

[0196]

[0197] f: ultrasonic frequency;

[0198] Q: oscillator amplitude;

[0199] ρ dust : dust bulk density);

[0200] δ: thickness of dust layer;

[0201] σ adhesion : Dust-window adhesion strength;

[0202] σ dust: Shear strength of dust itself;

[0203] Step 2.2: A contact angle aging equation is proposed to address the degradation of hydrophobicity under high-speed airflow:

[0204]

[0205] θ0: initial contact angle;

[0206] θ min : limiting contact angle);

[0207] U air : air curtain velocity;

[0208] L: wear coefficient;

[0209] t: running time;

[0210] Step 2.3: Integrate the effects of the three mechanisms on transmittance;

[0211]

[0212] z: light transmittance;

[0213] w: basic pollution rate;

[0214] η,E\E crit ,θ(t): from the three sub-models mentioned above;

[0215] Specifically: The air film cooling efficiency model is used to accurately match the dust adhesion strength, the ultrasonic cavitation threshold is combined to optimize the cleaning energy density, and a **hydrophobic coating contact angle aging equation is established to predict the performance attenuation under high-speed airflow scouring, forming a triple dynamic protection mechanism of air curtain isolation-ultrasonic stripping-hydrophobic anti-fouling. The optical window transmittance is stabilized at >93%, the dust removal efficiency is increased by 80%, the maintenance cycle is extended to 3 months, and the annual maintenance cost is reduced by 70%, effectively solving the problem of sensor condensation pollution and performance degradation in high humidity and high dust environments.

[0216] Preferably, the step 3 includes:

[0217] Step 3.1: Inject physical constraints of environmental parameters (temperature T, humidity H, dust concentration D) into the standard LSTM forget gate:

[0218]

[0219] Physical consistency loss term;

[0220] φ i : Intermediate feature vectors output by microwave, infrared, and laser sensors;

[0221] λ i:Physical constraint weight;

[0222] Step 3.2 defines the dynamic attention weights between sensor data and environmental parameters:

[0223]

[0224] Characteristics of the i-th sensor at time t;

[0225] Environmental parameters of category j;

[0226] W a :Learnable projection matrix;

[0227] Step 3.3 adds a residual path between the sensor's original signal and the hidden state inside the LSTM cell:

[0228]

[0229] β i : sensor residual gain;

[0230] The residual projection matrix of sensor i;

[0231] Step 3.4 designs a loss function for the fast and slow dual-scale characteristics of fly ash concentration changes:

[0232]

[0233] mean absolute error;

[0234] ||·|| TV : total variation penalty term;

[0235] J1=0.3, J2=0.7: fast / slow scale penalty weights;

[0236] T long :Long period window.

[0237] Preferably, the step 4 includes:

[0238] Step 4.1: Based on the Forchheimer modified equation, the pore structure shape factor is introduced:

[0239]

[0240] s: permeability;

[0241] x: Inertial drag coefficient:

[0242] n: pore curvature factor;

[0243] g mean : average pore size;

[0244] Step 4.2 Design a fractional sliding surface to suppress sudden disturbances in the airflow:

[0245] s=D O e+λ1e+λ2∫edt(0<α<1),

[0246] Control law:

[0247]

[0248] e=U real -U set : Flow rate deviation;

[0249] O=0.7, I=0.5: fractional-order damping term to enhance anti-interference performance;

[0250] disturbance observer estimate;

[0251] Specifically, by embedding temperature, humidity, and dust concentration physical constraints within the LSTM forget gate and combining a dual-scale loss function with the sensor residual path, the model prediction error was reduced from ±15% to ±3%. A gradient porous ceramic was designed based on the Forchheimer equation, coupled with fractional-order sliding mode control to suppress airflow disturbances to ±5% fluctuations and improve sampling uniformity by 40%. Through the dual-driven fusion of physics and data, the model response time is ≤ 30 seconds, the transmittance decay rate is reduced by 80%, and the annual maintenance cost is reduced by 52%, overcoming the challenges of detection signal instability and dynamic compensation lag under high dust and strong disturbance conditions.

[0252] Preferably, the step 4.2 includes:

[0253] Step 4.2.1 Combine the flow stability and sampling efficiency goals to construct a dynamic weight loss function:

[0254]

[0255] Step 4.2.2 Dynamic Weight Strategy:

[0256]

[0257] C opt =200g / m' optimal dust concentration;

[0258] t i,last : The last time the i-th target exceeded the limit;

[0259] Specifically, the weight distribution is constrained by the optimal concentration threshold. When dust exceeds the standard, priority is given to stabilizing the flow to suppress signal distortion, and when dust is normal, emphasis is placed on sampling speed. Under the high-disturbance operating conditions of a 1000MW unit, sampling uniformity is improved by 40%, dust limit-exceeding events are reduced by 80%, and annual maintenance costs are reduced by 45%, breaking through the performance bottleneck of traditional fixed-weight strategies under multi-objective conflicts.

[0260] Preferably, the step 5 includes:

[0261] Step 5.1: Dynamically assign sensor weights based on Bayesian posterior probability, prioritize correcting low-confidence sensor parameters, and verify the validity of calibration results using a sliding window T-test (window length = 10 cycles) to avoid miscalibration.

[0262] Define the real-time confidence B of sensor i i :

[0263]

[0264] P(D i |S true ):Sensor i is in the real state S t Likelihood function under rue;

[0265] P prior (S true ): Prior probability;

[0266] T: time decay factor;

[0267] When B i <0.7 triggers calibration, weight w i =B i / ∑B j Used for multi-sensor data fusion;

[0268] Step 5.2: Construct the sensor state transition matrix:

[0269]

[0270] Among them, status 0: normal; status 1: suspected abnormality; status 2: confirmed failure;

[0271] Specifically, sensor weights are dynamically allocated based on posterior probabilities, and calibration is triggered when the real-time confidence level B is less than 0.7. The sliding window T test is combined to avoid misoperation, and the calibration accuracy is improved to 99%. A three-state transfer matrix is constructed, and an automatic isolation mechanism for failed sensors is integrated, reducing the system false alarm rate from 5% to 0.8%. In the high dust environment of a 600MW unit, the sensor data drift rate is reduced by 90%, the maintenance frequency is extended from once a week to once a quarter, and the annual operation and maintenance cost is reduced by 60%, breaking through the technical bottleneck of achieving both multi-sensor cross-interference and long-term reliability.

[0272] Preferably, the step 5.2 includes:

[0273] Step 5.2.1: Establish a time decay model for the sensor gain parameters, predict the drift trend and pre-compensate. The calibration period is adaptively adjusted according to the drift rate. The time evolution model of the sensor gain is:

[0274]

[0275] m i :Dust concentration cumulative drift coefficient:

[0276] T thermal : Thermal cycle;

[0277] l: thermal expansion effect amplitude;

[0278] Pre-compensation: Calculate ΔG in real time during the calibration cycle i =G i (t)-G i,meas And reversely correct the output signal;

[0279] Step 5.2.2 Periodic update rule based on drift rate:

[0280]

[0281] T base =4h,

[0282] ΔG avg : Average gain deviation of the last five calibrations;

[0283] δ th =1%: maximum allowed deviation threshold;

[0284] Constraint: T n ew∈[4h,72h];

[0285] Specifically, a gain time evolution model is established, and the pre-compensation algorithm corrects the output signal in real time, reducing the drift error by 90%; the calibration period is dynamically updated based on the drift rate, and calibration is forced when the average gain deviation Δ≥0.1 to avoid excessive maintenance; under the high-temperature vibration conditions of coal-fired boilers, the sensor calibration period is extended from once a day to once a month, reducing maintenance costs by 70%, and the signal stability meets the requirements of industrial-grade high-precision detection, overcoming the stubborn problem of sensor time drift caused by temperature cycling and mechanical stress.

[0286] Preferably, the step 6 includes:

[0287] Step 6.1 Based on the black smoke concentration data calibrated in step 5, the ratio of primary air volume to coal feed rate is adjusted through feedforward-feedback composite control;

[0288] Step 6.1.1 Dynamically set the pulverizer outlet temperature (60-90°C) based on the volatile matter content of the pulverized coal to suppress the formation of unburned carbon;

[0289] Step 6.1.2 Dynamic feedforward compensation term:

[0290]

[0291] Feedback PID correction term:

[0292]

[0293] Total air-to-coal ratio command:

[0294] R total =R ff +R fb ,

[0295] u=0.15、p=0.05: coal quality response coefficient,

[0296] τ s =25s: system inertia time constant,

[0297] γ=0.3: nonlinear damping factor, suppressing the oscillation of the differential term when the deviation is large;

[0298] Step 6.2: Construct the Pareto frontier of combustion efficiency (η), NOx emissions (E), and power generation load (P), and find the optimal balance point between carbon content (C) and operation and maintenance cost (Cost);

[0299] Step 6.2.1 Update the target weights through online rolling optimization to adapt to the peak load demand of the power grid:

[0300] Step 6.2.2 Pareto objective function:

[0301]

[0302] Dynamic weight update rules:

[0303]

[0304] Q=1.5: cost penalty curvature;

[0305] κ = 200 mg / MJ: environmental protection constraint threshold;

[0306] Among them, the improved NSGA-II algorithm is used, the population size is 50, and the crossover probability is 0.8

[0307] Step 6.3: Design a time-delay observer to compensate for the transmission delay from the coal mill to the burner, ensuring that the control period is ≤ 30 seconds.

[0308] Based on step 6.3, include:

[0309] Step 6.3.1 Transmission Delay Estimation Model:

[0310]

[0311] Predictive-corrective control law:

[0312]

[0313] K obs =0.05: observer gain;

[0314] L pipe : Length of pipeline from coal mill to burner;

[0315] Specifically, the pulverizer outlet temperature is dynamically set based on volatile matter, combined with feedforward compensation and feedback PID composite control, to reduce the generation of unburned carbon by 40%; a three-dimensional frontier surface of combustion efficiency, NOx emissions, and power generation load is constructed, and the improved NSGA-II algorithm is used to solve the optimal balance point between carbon content and operation and maintenance costs in real time, increasing the peak-shaving response speed by 60%; the control period is predicted and corrected to ≤30 seconds through the transmission delay observer, eliminating the impact of coal powder transportation lag; in the 1000MW ultra-supercritical unit, the boiler efficiency is increased by 1.5%, NOx emissions are reduced by 25%, and the annual coal cost is saved by 12 million yuan, solving the industry problem of the difficulty in coordinated optimization of combustion stability, environmental protection compliance and peak-shaving flexibility.

[0316] Example 2:

[0317] Please see the attached Figure 2 An embodiment of the present invention provides an online detection and control system for fly ash blackening in a thermal power plant, comprising the following steps:

[0318] The data monitoring module uses microwave, infrared, and laser sensors to work together. Microwaves detect carbon dielectric properties against dust interference, infrared identifies thermal radiation differences, and lasers analyze particle distribution. Cross-validation provides a reliable data foundation.

[0319] The data fusion module, based on the detection requirements of step 1, designs a triple protection system of air curtain barrier + ultrasonic cleaning + hydrophobic coating to block dust condensation and ensure that the sensor transmittance is greater than 90%;

[0320] The dynamic correction module combines the clean environment of step 2 with temperature, humidity, and dust sensors. Using the LSTM model, the multi-sensor data from step 1 is integrated with environmental parameters in real time to dynamically correct the signal, reducing the error from ±15% to ±3%.

[0321] The sampling suppression module installs a porous ceramic flow stabilizer upstream of the detection area in step 1 to suppress airflow disturbances to ±5% fluctuations. This, combined with adaptive negative pressure control, ensures sampling uniformity and improves the accuracy of the compensation model in step 3.

[0322] The calibration module automatically calibrates sensor parameters every four hours based on steps 1-4 to stabilize the system. It triggers multi-sensor cross-validation when an anomaly occurs, with a false alarm rate of less than 0.5% and a calibration cycle extended by three times.

[0323] The integration module integrates the calibration data from step 5, links with the power plant control system in real time, dynamically adjusts the air-coal ratio and pulverized coal temperature, stabilizes the carbon content at 3% to 5%, and reduces operation and maintenance costs by 50%.

[0324] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. An online detection and control method for fly ash blackening in a thermal power plant, characterized in that: include: Step 1: Using microwave, infrared, and laser sensors to work together, microwaves detect carbon dielectric properties against dust interference, infrared identifies thermal radiation differences, and lasers analyze particle distribution. Cross-validation provides a reliable data foundation. Step 2: Based on the detection requirements of step 1, a triple protection system consisting of an air curtain barrier, ultrasonic cleaning, and a hydrophobic coating is designed to block dust condensation and ensure that the sensor transmittance is greater than 90%. Step 3: Combined with the clean environment in step 2, temperature, humidity, and dust sensors are deployed. The multi-sensor data from step 1 is integrated with environmental parameters in real time through the LSTM model to dynamically correct the signal, reducing the error from ±15% to ±3%. Step 4: Install a porous ceramic flow stabilizer upstream of the detection area in step 1 to suppress airflow disturbances to ±5% fluctuations. Combined with adaptive negative pressure control, this ensures sampling uniformity and improves the accuracy of the compensation model in step 3. Step 5: Based on the system stability of steps 1-4, the sensor parameters are automatically calibrated every 4 hours; when an abnormality occurs, multi-sensor cross-validation is triggered, the false alarm rate is less than 0.5%, and the calibration cycle is extended by 3 times; Step 6: Integrate the calibration data from step 5 and link it to the power plant control system in real time to dynamically adjust the air-coal ratio and pulverized coal temperature, stabilizing the carbon content at 3% to 5%, and reducing operation and maintenance costs by 50%.

2. The on-line detection and control method for fly ash blackening in a thermal power plant according to claim 1, characterized in that: The step 1 includes: Step 1.1: Based on the dielectric properties of the unburned carbon and ash in fly ash, a dielectric model with a shape factor correction term is introduced: ∈ eff : mixed dielectric constant; ∈ c ,∈ a : Dielectric constant of carbon particles and ash; f c : carbon volume fraction; κ: shape factor; (where ∈ eff represents the mixed dielectric constant, ∈ c ,∈ a represents the dielectric constant of carbon particles and ash, f c represents the carbon volume fraction; κ represents the shape factor); Step 1.2: Based on the strong absorption characteristics of unburned carbon in the mid-infrared band, a nonlinear relationship between emissivity and carbon content is established: (ε(λ, C): surface emissivity at wavelength λ and carbon content C, ε0(λ): baseline emissivity of pure ash in the λ band, α: absorption coefficient; C0: characteristic carbon content threshold); Step 1.3 Use the multi-angle scattered light intensity data I(θ i ), and the particle size distribution n(d) is solved by Tikhonov regularization: (V(θ i ,d,m):Mie scattering kernel function,γ:regularization parameter,d min , d max : Particle size range) Step 1.4 defines the basic probability distribution function of the three sensors for the proposition "high floating black concentration": (m sensor(.) : Confidence distribution of each sensor to the proposition, A, B, C, D: proposition set).

3. The on-line detection and control method for fly ash blackening in a thermal power plant according to claim 1, characterized in that: The step 2 includes: Step 2.1: To quantify the protective effect of the air curtain on the optical window, a modified air film cooling efficiency model is introduced: Define the minimum energy density threshold required for ultrasonic cleaning of dust deposits: (f: ultrasonic frequency, Q: vibrator amplitude, ρ dust : Dust accumulation density, δ: Dust layer thickness, σ adhesion : Dust-window adhesion strength, σ dust : shear strength of dust itself); Step 2.2: A contact angle aging equation is proposed to address the degradation of hydrophobicity under high-speed airflow: (θ0: initial contact angle, θ min : Limiting contact angle, U air : air curtain flow rate, L: wear coefficient, t: running time); Step 2.3: Integrate the effects of the three mechanisms on transmittance; (z: transmittance, w: basic pollution rate, η, E\E crit ,θ(t): from the three sub-models mentioned above).

4. The method for online detection and control of fly ash blackening in a thermal power plant according to claim 1, characterized in that: The step 3 includes: Step 3.1 Inject physical constraints of environmental parameters into the standard LSTM forget gate: Physical consistency loss term; (φ i : Intermediate eigenvector output by microwave, infrared, and laser sensors, λ i :Physical constraint weight) Step 3.2 defines the dynamic attention weights between sensor data and environmental parameters: ( The characteristics of the i-th sensor at time t, Environmental parameter of type j, W a : learnable projection matrix); Step 3.3 adds a residual path between the sensor's original signal and the hidden state inside the LSTM Cell: (β i : sensor residual gain, Residual projection matrix of sensor i); Step 3.4 designs a loss function for the fast and slow dual-scale characteristics of fly ash concentration changes: mean absolute error; (||·|| TV : Total variation penalty term, J1=0.3, J2=0.7: Fast / slow scale penalty weight, T long : long period window).

5. The on-line detection and control method for fly ash blackening in a thermal power plant according to claim 1, characterized in that: The step 4 includes: Step 4.1: Based on the Forchheimer modified equation, the pore structure shape factor is introduced: (s: permeability; x: inertial resistance coefficient, n: pore curvature factor, g mean : average pore diameter); Step 4.2 Design a fractional sliding surface to suppress sudden disturbances in the airflow: s=D O e+λ1e+λ2∫edt(0<α<1), Control law: (e=U real -U set : Flow rate deviation, O = 0.7, I = 0.5, fractional damping term, enhanced anti-interference performance, disturbance observer estimate).

6. The on-line detection and control method for fly ash blackening in a thermal power plant according to claim 5, characterized in that: The step 4.2 includes: Step 4.2.1 Combine the flow stability and sampling efficiency goals to construct a dynamic weight loss function: Step 4.2.2 Dynamic Weight Strategy: C opt =200g / m'optimal dust concentration; t i,last : The last time the i-th target exceeded the limit.

7. The on-line detection and control method for fly ash blackening in a thermal power plant according to claim 1, characterized in that: The step 5 includes: Step 5.1 Dynamically assign sensor weights based on Bayesian posterior probability, prioritize correcting low-confidence sensor parameters, and verify the validity of calibration results through a sliding window T test to avoid miscalibration: Define the real-time confidence B of sensor i i : P(D i |S true ):Sensor i is in the real state S t Likelihood function under rue; P prior (S true ): Prior probability; T: time decay factor; When B i <0.7 triggers calibration, weight w i =B i / ∑B j Used for multi-sensor data fusion; Step 5.2: Construct the sensor state transition matrix: Among them, status 0: normal; status 1: suspected abnormality; status 2: confirmed failure.

8. The on-line detection and control method for fly ash blackening in a thermal power plant according to claim 7, characterized in that: The step 5.2 includes: Step 5.2.1: Establish a time decay model for the sensor gain parameters, predict the drift trend and pre-compensate. The calibration period is adaptively adjusted according to the drift rate. The time evolution model of the sensor gain is: (m i : Dust concentration cumulative drift coefficient; T thermal : thermal cycle period, l: thermal expansion effect amplitude); Pre-compensation: Calculate ΔG in real time during the calibration cycle i =G i (t)-G i,meas And reversely correct the output signal; Step 5.2.2 Periodic update rule based on drift rate: T base =4h, ΔG avg : Average gain deviation of the last five calibrations; δ th =1%: maximum allowed deviation threshold; Constraint: T n ew∈[4h,72h].

9. The method for online detection and control of fly ash blackening in a thermal power plant according to claim 1, characterized in that: The step 6 includes: Step 6.1 Based on the black smoke concentration data calibrated in step 5, the ratio of primary air volume to coal feed rate is adjusted through feedforward-feedback composite control; Step 6.1.1 Dynamically set the pulverizer outlet temperature based on the volatile matter content of the pulverized coal to suppress the formation of unburned carbon; Step 6.1.2 Dynamic feedforward compensation term: Feedback PID correction term: Total air-to-coal ratio command: R total =R ff +R fb , u=0.15、p=0.05: coal quality response coefficient, τ s =25s: system inertia time constant, γ=0.3: nonlinear damping factor, suppressing differential oscillation when large deviation occurs; Step 6.2: Construct the Pareto frontier of combustion efficiency (η), NOx emissions (E), and power generation load (P), and find the optimal balance point between carbon content (C) and operation and maintenance cost (Cost); Step 6.2.1 Update the target weights through online rolling optimization to adapt to the peak load demand of the power grid: Step 6.2.2 Pareto objective function: Dynamic weight update rules: Q = 1.5: cost penalty curvature, κ = 200 mg / MJ: environmental protection constraint threshold; Among them, the improved NSGA-II algorithm is used, the population size is 50, and the crossover probability is 0.8 Step 6.3: Design a time-delay observer to compensate for the transmission delay from the coal mill to the burner, ensuring that the control period is ≤ 30 seconds. Step 6.3.1 Transmission Delay Estimation Model: Predictive-corrective control law: K obs =0.05: observer gain, L pipe : Length of pipeline from coal mill to burner.

10. An online detection and control system for fly ash blackening in a thermal power plant, using the online detection and control method for fly ash blackening in a thermal power plant according to any one of claims 1 to 9, characterized in that: include: The data monitoring module uses microwave, infrared, and laser sensors to work together. Microwaves detect carbon dielectric properties against dust interference, infrared identifies thermal radiation differences, and lasers analyze particle distribution. Cross-validation provides a reliable data foundation. The data fusion module, based on the detection requirements of step 1, designs a triple protection system of air curtain barrier + ultrasonic cleaning + hydrophobic coating to block dust condensation and ensure that the sensor transmittance is greater than 90%; The dynamic correction module combines the clean environment of step 2 with temperature, humidity, and dust sensors. Using the LSTM model, the multi-sensor data from step 1 is integrated with environmental parameters in real time to dynamically correct the signal, reducing the error from ±15% to ±3%. The sampling suppression module installs a porous ceramic flow stabilizer upstream of the detection area in step 1 to suppress airflow disturbances to ±5% fluctuations. This, combined with adaptive negative pressure control, ensures sampling uniformity and improves the accuracy of the compensation model in step 3. The calibration module automatically calibrates sensor parameters every four hours based on steps 1-4 to stabilize the system. It triggers multi-sensor cross-validation when an anomaly occurs, with a false alarm rate of less than 0.5% and a calibration cycle extended by three times. The integration module integrates the calibration data from step 5, links with the power plant control system in real time, dynamically adjusts the air-coal ratio and pulverized coal temperature, stabilizes the carbon content at 3% to 5%, and reduces operation and maintenance costs by 50%.

Images