A wind-powder parameter on-line measuring device and method based on multi-physical field fusion and adaptive decoupling

Abstract

This invention discloses an online measurement device and method for coal dust parameters based on multi-physics field fusion and adaptive decoupling. The device includes: a co-location fusion sensing probe, which integrates a charge sensing module and a microwave resonant sensing module, the microwave resonant sensing module operating at a frequency of 1.0-5.0 GHz; an adaptive decoupling calculation unit, equipped with an adaptive decoupling algorithm, configured to simultaneously receive electrostatic and dielectric characteristic signals, identify coal type characteristic parameters online through a pre-stored dielectric characteristic-coal type parameter mapping relationship, substitute these parameters into the charge signal physical model, and decouple and calculate the estimated coal dust concentration and flow rate values ​​to eliminate the influence of coal type through the online parameter identification method; and a data communication unit for outputting the estimated coal dust concentration and flow rate values. This invention, through online identification of coal type characteristics using microwave dielectric measurement and dynamic correction of the charge model, fundamentally eliminates measurement drift caused by changes in coal type, achieving long-term calibration-free, high-precision online measurement.

Description

Technical Field

[0001] This invention relates to the field of combustion monitoring and optimization control technology for power plant boilers, specifically to a device and method for online measurement of gas-solid two-phase flow parameters in the primary air-coal duct of a boiler. Background Technology

[0002] Combustion optimization control in power plant boilers relies on accurate, real-time measurement of pulverized coal concentration and flow rate within the primary air-coal duct. Currently, the electrostatic charge method (or electrostatic method) based on electrostatic induction is the mainstream measurement technology in this field. Its principle is that pulverized coal particles generate static charges through friction and collision with the duct wall and themselves during pneumatic conveying. By measuring the intensity and fluctuation characteristics of the electrostatic signal on the induction electrode, the concentration and flow rate of the pulverized coal can be inferred.

[0003] However, this method has a fundamental flaw in its measurement mechanism: the charge-generating capacity of pulverized coal particles (i.e., the amount of charge generated per unit mass of pulverized coal) is not a constant physical constant, but rather strongly depends on the physicochemical properties of the coal itself. When the type of coal used in power plants changes, or when the properties of the same type of coal fluctuate due to different sources or washing processes, the charge-generating rate of pulverized coal will change significantly.

[0004] This characteristic leads to the following technical challenges for existing charge-based measurement devices: the functional relationship between the charge signal and the coal powder mass concentration, established through pre-calibration, will systematically "drift" due to changes in coal type. For example, when switching from high-volatile bituminous coal to low-volatile lean coal, even if the actual coal powder concentration remains unchanged, the amplitude of the charge signal may increase by 30%-50%, resulting in severely inaccurate measurement results and failing to provide a reliable input signal for the combustion control system.

[0005] To address the aforementioned issues, existing technologies disclose a periodically calibrated charge measurement device that corrects measurement errors through periodic shutdowns and offline sampling calibration. However, this method has the following drawbacks: first, frequent shutdowns for calibration affect the continuous operation of the unit and increase maintenance costs; second, if the coal type changes again within the calibration period, the error cannot be corrected in a timely manner; and third, it cannot adapt to the complex operating conditions of dynamic blending and continuous changes in coal quality commonly used in power plants.

[0006] Other technologies attempt to establish coal type databases for offline compensation, such as pre-measuring the electrical characteristics of different coal types and storing them in a lookup table. However, the actual sources of coal are complex and their properties are variable, making it difficult for pre-established databases to cover all possible situations. Furthermore, they cannot adapt to the characteristic fluctuations of the same coal type caused by different origins and batches, resulting in limited compensation effectiveness.

[0007] Therefore, developing an online measurement device and method for air-coal parameters that can fundamentally decouple the influence of coal type characteristics, provide real-time adaptive correction measurement models, and achieve long-term calibration-free stable operation has become a key technical challenge that urgently needs to be solved in this field. Summary of the Invention

[0008] The purpose of this invention is to provide an apparatus and method for online measurement of gas-solid two-phase flow parameters in the primary air-coal pipeline of a boiler, in order to solve the problem mentioned in the background art of the lack of a real-time adaptive correction measurement model and a comprehensive solution for online measurement of air-coal parameters that can achieve long-term calibration-free stable operation.

[0009] To achieve the above objectives, the present invention provides the following technical solution: A wind and dust measurement device based on multiphysics field fusion and adaptive decoupling includes: The co-location fusion sensing probe is installed on the measurement section of the primary air-powder pipeline. Its exterior is a wear-resistant sleeve, and its interior integrates a charge sensing module and a microwave resonant sensing module. The microwave resonant sensing module operates at a frequency of 1.0 GHz to 5.0 GHz. An adaptive decoupling calculation unit, connected to the co-location fusion sensing probe signal, has an internally built-in adaptive decoupling algorithm configured to perform the following operations: Synchronously receive electrostatic signals from the charge sensing module. And the dielectric characteristic signal Sd(t) from the microwave resonant sensing module; Based on the dielectric property signal Sd(t), the current coal type characteristic parameter Θ is identified online by querying the pre-stored dielectric property-coal type parameter mapping relationship; A physical model of charge signal Se(t)=Hc(C,V,Θ) is established, which includes coal powder concentration C, flow velocity V and coal type characteristic parameter Θ. Substitute the identified coal type characteristic parameters Θ into the physical model Hc, and use the real-time electrostatic signal Se(t) to decouple and calculate the coal powder concentration estimate C and flow velocity estimate V to eliminate the influence of coal type characteristics through the online parameter identification method. The data communication unit is used to output the estimated coal powder concentration C and the estimated flow rate V.

[0010] Furthermore, in the co-location fusion sensing probe, the sensing electrode of the charge sensing module is a ring structure arranged on the inner wall of the wear-resistant sleeve; the microwave resonant sensing module is a coaxial resonant cavity structure, with its central conductor and the wear-resistant sleeve forming the inner and outer conductors of the resonant cavity, and the central conductor is arranged along the flow direction, with pulverized coal flowing through the annular space between the central conductor and the wear-resistant sleeve.

[0011] Furthermore, the coaxial resonant cavity has a gradually changing microwave impedance matching structure at its inlet and outlet ends; the impedance matching structure is a quarter-wavelength stepped impedance transformer, used to ensure that the voltage standing wave ratio (VSWR) is less than 1.5 within the operating frequency band.

[0012] Furthermore, the pre-stored dielectric property-coal type parameter mapping relationship is obtained by training a machine learning classification model based on a dielectric property database covering a variety of industrial coal samples.

[0013] Furthermore, machine learning classification models include support vector machines.

[0014] Furthermore, the adaptive decoupling calculation unit also includes a signal quality diagnosis module, which is used to calculate the signal-to-noise ratio and stability of the charge signal and microwave signal in real time, and trigger an alarm or output hold when the signal quality is lower than a preset threshold.

[0015] Furthermore, the adaptive decoupling computation unit also includes an adaptive update module, which is configured as follows: Monitoring the stability of microwave coal type identification results and charge signal model residuals; Determine whether the coal type has undergone a continuous and stable change; In response to the determination that the coal type has undergone a continuous and stable change, the baseline coefficients of the model parameters and / or charge signal physical model of the dielectric property-coal type parameter mapping relationship are fine-tuned online using new data.

[0016] Furthermore, the adaptive update module includes: The sliding window consistency verification submodule is used to maintain the first-in-first-out data buffer and calculate the entropy value of the microwave coal type identification result and the root mean square (RMS) of the charge signal model residual within the window in real time. The parameter fine-tuning submodule is used to determine that the coal type has undergone a continuous and stable change when the entropy value of multiple consecutive windows is lower than the first threshold and the RMS value is higher than the second threshold, and to fine-tune the model parameters online using the new data in the current sliding window.

[0017] Furthermore, the adaptive decoupling algorithm also includes robustness processing logic to handle edge cases, including: The high-moisture coal processing logic is configured to introduce the signal quality index (SQI). When the microwave signal amplitude attenuation is abnormal and causes the SQI to decrease, the output filter time constant is automatically increased, and a signal degradation prompt is sent to the host system. The probe dust accumulation compensation logic is configured to automatically trigger a baseline compensation algorithm by monitoring the long-term slow drift trend of the resonant frequency, and subtract the slow drift amount from the current measurement value.

[0018] The wind-dust measurement method based on multi-physics field fusion and adaptive decoupling, applied to the aforementioned wind-dust measurement device, includes the following steps: S1: The electrostatic signal Se(t) and microwave dielectric property signal Sd(t) of pulverized coal in the primary air-coal pipeline are simultaneously acquired by the co-position fusion sensing probe. S2: Based on the microwave dielectric characteristic signal Sd(t), the parameter Θ reflecting the current coal type characteristics is identified in real time by querying the pre-stored dielectric characteristic-coal type parameter mapping relationship; S3: Substitute the coal type characteristic parameter Θ into the preset charge signal physical model Hc(C,V,Θ), and combine it with the real-time acquired electrostatic signal Se(t). Through model inversion and parameter identification, the estimated coal powder concentration is obtained through decoupling. and flow velocity estimates The decoupling process described above compensates online for the influence of changes in coal type characteristic parameter Θ on the charge signal; S4: Output the estimated coal powder concentration. and flow velocity estimates .

[0019] Furthermore, the pre-stored dielectric property-coal type parameter mapping relationship mentioned in step S2 is pre-constructed through the following steps: Various typical coal samples were collected, and their complex permittivity spectra were measured in the 1.0-5.0 GHz frequency band. Extract the dielectric parameters at the characteristic frequency points to construct a multidimensional feature vector; The feature vector is associated with and stored in conjunction with the standard coal type label and industrial analysis parameters of the coal sample to form a dielectric property database. A machine learning classification model is trained based on the database to establish a mapping relationship between dielectric parameters and coal type characteristic parameters.

[0020] Furthermore, the characteristic frequency points include 2.45 GHz and 3.5 GHz; the dielectric parameters include the real part of the dielectric, the imaginary part of the dielectric, the loss tangent, and the curve fitting coefficients of the dielectric spectrum across the entire frequency band.

[0021] Furthermore, step S3 is followed by an adaptive update step S5: The entropy value of the microwave coal type identification result and the root mean square (RMS) of the charge signal model residual are calculated in real time using a sliding window. When the entropy value of multiple consecutive windows is lower than the first threshold and the RMS value is higher than the second threshold, it is determined that the coal type has undergone a continuous and stable change. Using the new data within the current sliding window, the baseline coefficients of the model parameters and / or the physical model of the charge signal mapping relationship of the dielectric property-coal type parameters are fine-tuned online.

[0022] Furthermore, online fine-tuning employs the gradient descent algorithm to update the model parameters with a preset small learning rate.

[0023] Furthermore, a signal preprocessing step is included after step S1 and before step S2: Perform Butterworth bandpass filtering, power frequency notch filtering, and sliding median filtering on the charge signal; The microwave signal is smoothed by moving average and temperature compensation is performed using the probe's built-in temperature sensor.

[0024] Compared with the prior art, the beneficial effects of the present invention are: (1) Fundamentally solve the coal type drift problem and achieve long-term calibration-free operation. By introducing microwave dielectric measurement to identify the coal type characteristic parameter Θ online and substituting it into the charge signal physical model for dynamic decoupling, the signal contribution of coal quality change and concentration / flow rate change is distinguished from the measurement principle, thereby fundamentally eliminating the measurement drift caused by coal type change; (2) High measurement accuracy and fast dynamic response. The co-position integrated sensing design ensures strict spatiotemporal synchronization between charge and microwave signal, eliminating material distribution unevenness and signal asynchrony errors caused by sensor spatial separation; (3) Strong engineering adaptability and high operational reliability. The device can directly replace the existing charge method probe and is easy to install; it has a built-in sliding window consistency verification and parameter fine-tuning mechanism to realize model self-learning and self-maintenance; for edge working conditions such as high moisture coal, probe ash accumulation, and communication failure, it sets up multiple robust processing logics such as signal quality index monitoring, baseline drift compensation, and fault safety output, which significantly improves the system reliability under complex working conditions; (4) Provide auxiliary coal quality information to support intelligent combustion optimization. The microwave resonant signal reflects the dielectric properties of pulverized coal in real time and can be used as an indirect indicator of coal quality changes, providing decision-making reference for coal blending and combustion adjustment, which helps to improve the combustion economy of boilers. Attached Figure Description

[0025] Figure 1 This is a diagram showing the overall architecture of the wind and dust measurement device based on multi-physics field fusion and adaptive decoupling of the present invention. Figure 2 This is a flowchart of the wind and dust measurement method based on multi-physics field fusion and adaptive decoupling of the present invention; Figure 3 This is a flowchart of the adaptive decoupling algorithm. Detailed Implementation

[0026] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] Please see Figure 1 This invention provides a technical solution: an online wind and dust parameter measurement device based on multi-physics field fusion and adaptive decoupling, comprising: Co-location fusion sensing probe: Installed on the measurement section of the primary air-powder duct, it adopts an integrated mechanical and electrical structure. Its exterior is a wear-resistant alloy steel sleeve, and its interior integrates two conformally designed sensor modules that sense the same flow field micro-element. Charge sensing module: Used to sense broadband electrostatic signals generated by the collision and friction of pulverized coal particles. Its sensing electrodes are one or more concentric ring-shaped metal electrodes, embedded and insulated and fixed to the inner wall of the wear-resistant sleeve to ensure direct contact with the pulverized coal flow.

[0028] Microwave resonant sensing module: Used to excite and sense the equivalent complex permittivity of the coal-air mixture within the pipeline. This module employs a coaxial or cylindrical resonant cavity structure, with its central conductor (inner conductor) and a wear-resistant sleeve serving as the outer conductor forming the cavity boundary, through which the coal powder flows. Crucially, the operating frequency of the microwave resonant sensing module is selected within the 1.0 GHz to 5.0 GHz frequency band.

[0029] Adaptive decoupling computation unit: Connected to the co-located fusion sensing probe via a shielded cable, its core is an embedded microprocessor (such as an ARM Cortex-A series or a high-performance DSP). This unit contains an adaptive decoupling algorithm and is configured to perform the following operations: Synchronously acquire and preprocess raw electrostatic signals from the charge sensing module. The resonant frequency offset Δf and the quality factor change ΔQ from the microwave resonant sensing module (together characterizing the dielectric properties) .

[0030] Establish a dual-physics field coupled measurement model: For charge signals:

[0031] in, It is a nonlinear function that describes the electrostatic signal generated by the combined action of pulverized coal mass concentration C, flow velocity V, and coal type characteristic parameter vector Θ (such as equivalent dielectric constant, loss tangent, etc.). For measuring noise.

[0032] For microwave signals:

[0033] The function F establishes a mapping relationship from the microwave resonance parameters (Δf, ΔQ) to the coal type characteristic parameter vector Θ. This relationship can be pre-established using a basic dielectric experimental database of pulverized coal samples.

[0034] Within each sampling period, the algorithm utilizes real-time measurements Solving for function F yields estimated values ​​of the characteristic parameters for the current coal type. .

[0035] Will Substituting into the charge model Hc, the model now only has two unknown variables: C and V. Combined with the charge signal... The time-domain or frequency-domain characteristics (such as the principle of cross-correlation velocity measurement) are used to solve for the coal powder mass concentration in real time through an online parameter identification algorithm based on recursive least squares (RLS) or a pre-trained lightweight convolutional neural network (CNN) model. and flow rate The optimal estimate. This decoupling process essentially compensates online for the drift of the charge model parameter Θ caused by changes in coal type.

[0036] Data communication unit: Integrated within the computing unit, used to transmit the decoupled high-precision coal powder concentration signal. Flow velocity signal It is uploaded to the power plant's distributed control system (DCS) or monitoring system via 4-20mA analog output, Modbus RTU / TCP, or OPCUA protocol.

[0037] Meanwhile, please see Figure 2 Furthermore, a method for online measurement of wind and dust parameters based on multi-physics field fusion and adaptive decoupling is provided, applied to the aforementioned device, comprising the following steps: S1: Synchronous Co-location Sensing: By using a co-location fusion sensing probe installed on the primary air-coal duct, electrostatic induction signals of the coal powder flowing through it are collected synchronously and at the same spatial point. The dielectric response signal Sd(t) of microwave resonance.

[0038] S2: Online coal type characteristic identification: The adaptive decoupled calculation unit identifies coal type characteristics based on real-time data. The estimated values ​​of the current coal type's characteristic parameters can be obtained by querying the pre-stored dielectric property-coal type parameter mapping table or by online calculation. .

[0039] S3: Dynamic Model Decoupling: [This is followed by a series of seemingly unrelated phrases and sentences, likely due to O Substitute into the physical model of charge signal By utilizing the charge signal sequence Se(t) within the current and historical time windows, the effects of concentration C and flow rate V, which were originally coupled in Hc, are separated through recursive parameter identification or neural network forward computation.

[0040] S4: Output robust results: Calculate and output the decoupled pulverized coal mass concentration value after eliminating the influence of coal type characteristic changes in real time. and flow rate value .

[0041] S5: Adaptive Verification and Update: Continuously monitors the short-term fluctuations and long-term trends of the output results. If a continuous and stable change in coal characteristics is detected, the mapping relationship F or model parameters are updated slightly using new data to track the slow changes in coal quality.

[0042] The following is a detailed explanation using the measurement of air and coal parameters in the primary air duct of a 660MW supercritical unit as an example: 1. Equipment Installation At the outlet of the A coal mill of a 660MW supercritical unit in a power plant, a vertically rising primary air-coal duct was selected as the installation location. This duct section meets the flow field stability requirements of 10D upstream and 5D downstream (D is the duct diameter), and the inner diameter of the duct is 500mm.

[0043] A hole is drilled in the pipe wall and a standard flange is welded on. The co-position fusion sensing probe described in this invention is inserted into the pipe, ensuring that the central axis of the probe coincides with the central axis of the pipe, and is then sealed and fixed using the flange. The probe is inserted to a depth of 250 mm to ensure that the sensing area is located in the fully developed flow field region at the center of the pipe.

[0044] The probe is connected to an adaptive decoupling computing unit installed in a nearby junction box via a dedicated double-shielded cable. The computing unit uses an ARM Cortex-A8 processor with a 1GHz clock speed, 512MB DDR3 memory, and runs an embedded Linux operating system. The computing unit is powered from the power plant's 220VAC, converted to 24VDC via an isolated power module. Its output signals include: 4-20mA analog output (corresponding to concentrations of 0-1.5kg / kg and flow rates of 0-40m / s), a Modbus RTURS-485 interface, and a Modbus TCP Ethernet interface, all connected to the coal mill's DCS control card.

[0045] 2. Specific structure of the probe

[0046] The specific structure of the co-location fusion sensing probe used in this embodiment is as follows: The probe is externally fitted with a wear-resistant alloy steel sleeve (material: 40CrNiMo, surface carburized), with an outer diameter of 50mm and a wall thickness of 5mm.

[0047] Charge sensing module: Two concentric ring-shaped stainless steel electrodes, 8mm wide and 15mm apart, are embedded in the inner wall of a wear-resistant sleeve. A polytetrafluoroethylene (PTFE) insulating layer is filled between the electrodes and the sleeve to ensure electrical isolation. Electrode leads are led out through sealed ceramic terminals to the signal conditioning circuit box at the probe's tail.

[0048] Microwave resonant sensing module: Employs a coaxial resonant cavity structure, with an 8mm diameter copper inner rod as the central conductor, running through the entire probe along its axis. A wear-resistant sleeve serves as the outer conductor of the resonant cavity. The annular space between the central conductor and the wear-resistant sleeve is the pulverized coal flow region; as the pulverized coal flows through this region, it alters the equivalent dielectric constant of the cavity.

[0049] A quarter-wavelength stepped impedance transformer is installed at the entrance and exit ends of the resonant cavity as a microwave impedance matching structure. This structure consists of two coaxial line segments with different characteristic impedances, and their specific dimensions were determined through electromagnetic simulation optimization: the first segment has a characteristic impedance of 75Ω and a length of 12.5mm; the second segment has a characteristic impedance of 50Ω and a length of 12.5mm. Actual measurements show that this structure maintains a voltage standing wave ratio (VSWR) of <1.5 within the operating frequency band of 1.0-5.0GHz, ensuring efficient microwave energy transmission.

[0050] The probe integrates a PT100 temperature sensor to monitor the probe temperature in real time, providing a basis for temperature compensation of the microwave signal.

[0051] 3. Construction of Dielectric Property Database

[0052] Before the device is put into operation, a dielectric property database is pre-built. The specific steps are as follows: (1) Sample collection: Fifty representative coal samples were collected from the cooperative power plant and surrounding coal mines, covering the entire range of lignite, bituminous coal, lean coal, and anthracite, as well as variants with different ash contents (Aad between 10% and 40%) and different volatile matter contents (Vdaf between 8% and 48%). The amount of each coal sample collected was no less than 5 kg; (2) Sample preparation: All coal samples were dried and ground to the fineness of coal powder commonly used in power plants (R90=20±2%), and their complete industrial analysis data (moisture Mad, ash Aad, volatile matter Vdaf, fixed carbon FCad) and elemental analysis data were recorded; (3) Dielectric parameter measurement: Using a Keysight PNA series vector network analyzer equipped with a coaxial probe fixture, the complex dielectric constant of each coal powder sample was measured in the 1.0-5.0 GHz frequency band under constant temperature and humidity conditions (temperature 25±1℃, humidity 45±5%RH). Each coal sample was measured three times at different apparent densities (0.4-0.7 g / cm³), and the average value was taken as the dielectric property data of the coal sample. (4) Feature extraction: For the measurement data of each coal sample, the following feature parameters are extracted to form a multidimensional feature vector: Real part of dielectric at 2.45 GHz Imaginary part of dielectric Loss tangent tanδ; Real part of dielectric at 3.5 GHz Imaginary part of dielectric Loss tangent tanδ; Curve fitting coefficients for the full-band dielectric spectrum (using the Debye model). (5) Database construction: Associate the above multidimensional feature vectors with the standard coal type labels of the coal sample (such as "high volatile bituminous coal-HVBit", "low volatile lean coal-LVAnth", etc.) and key industrial analysis parameters (Vdaf range, FCdaf range) and store them in an embedded SQLite database; (6) Mapping Model Training: Based on the above database, a multi-classification model was trained using the Support Vector Machine (SVM) algorithm. The model input consisted of real-time measured microwave resonance parameters (after preprocessing and feature extraction), and the output was the most likely coal type classification label. The model outputs estimates of volatile matter and fixed carbon, along with their probabilities, through a regression model. The SVM model, after training, is stored in the Flash memory of the adaptive decoupled computing unit.

[0053] 4. Initialization Calibration

[0054] When the unit is operating stably at 100% load (660MW) and burning the designed coal type (bituminous coal A, Vdaf=28%), the unit initialization calibration is performed: (1) Reference value acquisition: The reference value of coal powder concentration under the current working condition, C0=0.48kg / kg, is obtained by taking samples at the measurement section simultaneously through the offline isokinetic sampling method; the wind speed value corrected by the standard Venturi flow meter is read by the DCS and used as the flow velocity reference V0=24.5m / s.

[0055] (2) Signal acquisition: Input C0 and V0 on the human-machine interface to start the "basic calibration" program. The device automatically acquires and records the stable charge signal Se0 and microwave resonance parameters (Δf0, ΔQ0) within 60 seconds, and takes the average value as the calibration reference value.

[0056] (3) Model initialization: Based on (C0,V0,Se0) and (Δf0,ΔQ0), and combined with the built-in dielectric database, the algorithm calculates and stores the initial parameters of the charge model Hc and the mapping function F. Specifically, it determines the charge generation coefficient k0, the exponent α, and β in the charge model Hc (after fitting α=0.95, β=1.08), and establishes the association between the current coal type label Type_A and the dielectric feature vector. The entire calibration process takes about 15 minutes.

[0057] 5. Implementation details of the adaptive decoupling algorithm

[0058] The algorithm firmware is written in C language and has a running cycle of 10ms. See [link / details] for the specific process. Figure 3 : (1) Signal acquisition and preprocessing module Charge signal preprocessing: The charge signal is acquired every 10ms and then subjected to the following steps in sequence: 100Hz-10kHz fourth-order Butterworth bandpass filtering to remove low-frequency disturbances and high-frequency noise; 50Hz and its harmonics notch filtering to eliminate power frequency interference; and sliding median filtering (window length 5 points) to remove impulse noise.

[0059] Microwave signal preprocessing: The resonant frequency f and quality factor Q are acquired every 100ms, and the following steps are performed sequentially: A 5-point moving average smoothing method is used to eliminate random fluctuations; temperature compensation: the PT100 temperature value is read, and real-time compensation is performed according to the pre-stored temperature-frequency drift coefficient lookup table (the compensation coefficient has been experimentally calibrated to -0.2MHz / ℃).

[0060] (2) Coal type characteristic identification module

[0061] Coal type identification is performed once every 1 second (i.e., every 100 sampling cycles): Calculate the average value (Δf) of the microwave resonance parameters within that second. avg ,ΔQ avg ); Extract feature values ​​(dielectric parameters at 2.45 GHz and 3.5 GHz); The input is fed into a pre-trained SVM classifier, which outputs the current coal type label. (e.g., "Type_A", "Type_B", etc.) and their corresponding confidence levels P; Simultaneously, a continuous coal type characteristic vector is output through a regression model. (Including estimates of volatile matter, ash content, etc.)

[0062] (3) Dynamic decoupling calculation module

[0063] The charge model Hc is simplified as follows:

[0064] in, The power generation coefficients are related to the characteristics of coal type, and α and β are constants determined by basic calibration (α=0.95, β=1.08 in this embodiment).

[0065] The algorithm performs the following calculations every 10ms: Flow velocity estimation: The peak time difference Δt is calculated using the cross-correlation function of the signals from two adjacent charged electrodes. Combined with the electrode spacing L=15mm, the instantaneous flow velocity estimate is obtained. ; Rough concentration calculation: V raw and the current coal type power generation rate coefficient k est (Initial value: calibration value k0) Substitute into the model to back-calculate the rough concentration value. ; Online RLS identification: Employs a recursive least squares algorithm, comparing the actual measured Se with the model's predicted value. The error is the driving force for online iterative updates of k. est The output includes estimates of V and the optimal concentration estimate C. hat The forgetting factor of the RLS algorithm is set to λ=0.98 to ensure both rapid tracking of coal type changes and sufficient stability. Confidence assessment: Calculate the model fit residuals If e > 0.15, then lower the confidence flag of the current output result.

[0066] (4) Output and communication module

[0067] For C hat and V hat Perform a moving average filter (time constant 1 second, i.e., the average of 100 points), then: Outputs a 4-20mA analog signal (C) via a 16-bit D / A converter. hat Corresponding to 0-1.5kg / kg, V hat (corresponding to 0-40m / s) Upload digital data via RS-485 interface using Modbus RTU protocol; It uploads via the ModbusTCP protocol through an Ethernet interface, and internal diagnostic information can be accessed simultaneously.

[0068] 6. Adaptive update mechanism

[0069] The algorithm incorporates a sliding window consistency check and parameter update mechanism: (1) Sliding window maintenance: The system maintains a first-in-first-out data buffer with a length of M=600, corresponding to 10 minutes of historical data. Real-time recording within the window: Coal type identification results per second And confidence level P.

[0070] RMS value of charge model residuals per second.

[0071] (2) Update condition determination: Calculate the statistics within the window every 10 seconds: Entropy value of coal type identification results , where p i H represents the frequency of occurrence of various coal types. When H < 0.3, it indicates that the coal type is stable.

[0072] Residual RMS mean R avg When R avg When the value is greater than 0.1 (i.e., the model residual exceeds 10%), it indicates that the existing model deviates significantly from reality.

[0073] When six consecutive windows (i.e., N=6, corresponding to 1 hour) simultaneously satisfy H<0.3 and R avg When the value is greater than 0.1, it is determined that the coal type has undergone a stable and continuous change, triggering a parameter update.

[0074] (3) Parameter fine-tuning update: After the update is triggered, the system enters "learning mode": Using the new data within the current sliding window, perform one round of gradient descent fine-tuning on the decision boundary auxiliary parameters of the SVM classifier with a learning rate η=0.01. The reference charge rate coefficient k0 in the charge model is adjusted synchronously. After the update is completed, the system automatically exits the learning mode and resumes online measurement.

[0075] 7. Robust processing logic

[0076] For edge cases, the system has the following robust processing logic built in: (1) High-moisture coal processing: When the moisture content of the pulverized coal is too high, the microwave signal amplitude will attenuate abnormally. The system calculates the signal quality index SQI in real time: SQI = (current signal amplitude / reference signal amplitude). When SQI < 0.7: The output filter time constant is automatically increased from 1 second to 5 seconds to smooth the results. Send a "signal degradation" alarm to the host system (Modbus register 0x1001 is set to 1). Pause the adaptive update mechanism to avoid mislearning. (2) Compensation for slight dust accumulation on the probe: Dust accumulation can cause a slow drift in the microwave resonant frequency f0. The system addresses this by: Monitor the 24-hour long-term trend of f0 and calculate the drift rate df / dt. When df / dt exceeds 0.5 MHz / day but is less than 2 MHz / day, it is determined to be ash accumulation rather than a change in coal type. Automatic baseline compensation is triggered by subtracting the slow drift from the current measurement value Δf, i.e.:

[0077] (3) Handling communication or transient faults: Set a watchdog timer to automatically restart the algorithm if it does not respond for 100ms. When a continuous abnormal signal loss is detected (such as 10 consecutive ADC acquisition failures), the system enters fault-safe mode, maintains the average value of the last valid output, and reports a communication fault to the host computer (Modbus register 0x1002 is set to 1). 8. Industrial verification test To comprehensively evaluate the robustness, accuracy, and engineering practical value of the device of this invention, a 30-day continuous industrial test was conducted on the 660MW unit. The test covered a variety of complex operating conditions and was systematically compared with the traditional charge method and offline isokinetic sampling method.

[0078] (1) Rapid switching test of multiple coal types

[0079] Operating conditions: Four typical coal types are burned sequentially within 4 hours: Bituminous coal A: Vdaf=28%, Aad=15% Lean coal B: Vdaf=12%, Aad=25% Lignite C:Vdaf=45%, Aad=10% High ash content coal D: Vdaf=22%, Aad=35% The unit load remained stable at 500MW, and the coal mill feed rate command remained unchanged at 45.0t / h.

[0080] Results: The traditional charge method output concentration values ​​fluctuated wildly, with maximum relative deviations of +58.3% (lean coal B) and -36.7% (lignite C). In contrast, the output concentration of the device of this invention was highly consistent with the coal feed command, with a maximum dynamic deviation of <±6.5% and a mean absolute percentage error (MAPE) of 3.2%. The correlation coefficient R² between the offline sampling and testing points (a total of 12) and the output values ​​of this device was 0.983.

[0081] The specific data is shown in the table below:

[0082] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A wind powder measuring device based on multi-physical field fusion and adaptive decoupling, characterized in that, include: The co-position fusion sensing probe is installed on the measurement section of the primary air-powder pipeline. Its outer part is a wear-resistant sleeve, and its interior integrates a charge sensing module and a microwave resonant sensing module. The microwave resonant sensing module operates at a frequency of 1.0 GHz to 5.0 GHz; An adaptive decoupling calculation unit, connected to the co-location fusion sensing probe signal, has an internally built-in adaptive decoupling algorithm configured to perform the following operations: Synchronously receive electrostatic signals from the charge sensing module. and dielectric property signals from the microwave resonant sensing module ; Based on the dielectric property signal By querying the pre-stored mapping relationship between dielectric properties and coal type parameters, the characteristic parameter Θ of the current coal type can be identified online; Establish a physical model of charge signal that includes pulverized coal concentration C, flow velocity V, and coal type characteristic parameter Θ. ; Substitute the identified coal type characteristic parameter Θ into the physical model. And using real-time electrostatic signals By using an online parameter identification method, a coal powder concentration estimate that eliminates the influence of coal type characteristics is calculated through decoupling. and flow velocity estimates ; The data communication unit is used to output the estimated coal powder concentration. and flow velocity estimates .

2. The apparatus according to claim 1, characterized in that, In the co-position fusion sensing probe, the sensing electrode of the charge sensing module is a ring structure and is arranged on the inner wall of the wear-resistant sleeve; the microwave resonant sensing module is a coaxial resonant cavity structure, and its central conductor and the wear-resistant sleeve constitute the inner and outer conductors of the resonant cavity, and the central conductor is arranged along the flow direction, and the coal powder flows through the annular space between the central conductor and the wear-resistant sleeve.

3. The apparatus according to claim 2, characterized in that, The coaxial resonant cavity has a gradually changing microwave impedance matching structure at its inlet and outlet ends; the impedance matching structure is a quarter-wavelength stepped impedance transformer, used to ensure that the voltage standing wave ratio (VSWR) is less than 1.5 within the operating frequency band.

4. The apparatus according to claim 1, characterized in that, The pre-stored dielectric property-coal type parameter mapping relationship is obtained by training a machine learning classification model based on a dielectric property database covering a variety of industrial coal samples.

5. The apparatus according to claim 4, characterized in that, The machine learning classification model includes a support vector machine.

6. The apparatus according to claim 1, characterized in that, The adaptive decoupling calculation unit also includes a signal quality diagnosis module, which is used to calculate the signal-to-noise ratio and stability of the charge signal and microwave signal in real time, and to trigger an alarm or output hold when the signal quality is lower than a preset threshold.

7. The apparatus according to claim 1, characterized in that, The adaptive decoupling calculation unit further includes an adaptive update module, which is configured as follows: Monitoring the stability of microwave coal type identification results and charge signal model residuals; Determine whether the coal type has undergone a continuous and stable change; In response to the determination that the coal type has undergone a continuous and stable change, the baseline coefficients of the model parameters and / or charge signal physical model of the dielectric property-coal type parameter mapping relationship are fine-tuned online using new data.

8. The apparatus according to claim 7, characterized in that, The adaptive update module includes: The sliding window consistency verification submodule is used to maintain the first-in-first-out data buffer and calculate the entropy value of the microwave coal type identification result and the root mean square (RMS) of the charge signal model residual within the window in real time. The parameter fine-tuning submodule is used to determine that the coal type has undergone a continuous and stable change when the entropy value of multiple consecutive windows is lower than the first threshold and the RMS value is higher than the second threshold, and to fine-tune the model parameters online using the new data in the current sliding window.

9. The apparatus according to claim 1, characterized in that, The adaptive decoupling algorithm also includes robust processing logic for handling edge cases, including: The high-moisture coal processing logic is configured to introduce the signal quality index (SQI). When the microwave signal amplitude attenuation is abnormal and causes the SQI to decrease, the output filter time constant is automatically increased, and a signal degradation prompt is sent to the host system. The probe dust accumulation compensation logic is configured to automatically trigger a baseline compensation algorithm by monitoring the long-term slow drift trend of the resonant frequency, and subtract the slow drift amount from the current measurement value.

10. A method for online measurement of wind and dust parameters based on multiphysics fusion and adaptive decoupling, applied to the apparatus of any one of claims 1 to 9, characterized in that, Includes the following steps: S1: Simultaneously acquire the electrostatic signal of pulverized coal in the primary air-coal pipeline using a co-location fusion sensing probe. and microwave dielectric property signals ; S2: Based on the microwave dielectric property signal By querying the pre-stored mapping relationship between dielectric properties and coal type parameters, the parameter Θ reflecting the current coal type characteristics can be identified in real time. S3: Substitute the coal type characteristic parameter Θ into the preset charge signal physical model H c (C,V,Θ), and combined with real-time acquired electrostatic signals By decoupling model inversion and parameter identification, the estimated coal powder concentration is obtained. and flow velocity estimates The decoupling process described above compensates online for the influence of changes in coal type characteristic parameter Θ on the charge signal; S4: Output the estimated coal powder concentration. and flow velocity estimates .

11. The method according to claim 10, characterized in that, The pre-stored dielectric property-coal type parameter mapping relationship mentioned in step S2 is pre-constructed through the following steps: Various typical coal samples were collected, and their complex permittivity spectra were measured in the 1.0-5.0 GHz frequency band. Extract the dielectric parameters at the characteristic frequency points to construct a multidimensional feature vector; The feature vector is associated with and stored in conjunction with the standard coal type label and industrial analysis parameters of the coal sample to form a dielectric property database. A machine learning classification model is trained based on the database to establish a mapping relationship between dielectric parameters and coal type characteristic parameters.

12. The method according to claim 11, characterized in that, The characteristic frequency points include 2.45 GHz and 3.5 GHz; the dielectric parameters include the real part of the dielectric, the imaginary part of the dielectric, the loss tangent, and the curve fitting coefficients of the dielectric spectrum across the entire frequency band.

13. The method according to claim 10, characterized in that, Step S3 is followed by an adaptive update step S5: The entropy value of the microwave coal type identification result and the root mean square (RMS) of the charge signal model residual are calculated in real time using a sliding window. When the entropy value of multiple consecutive windows is lower than the first threshold and the RMS value is higher than the second threshold, it is determined that the coal type has undergone a continuous and stable change. Using the new data within the current sliding window, the baseline coefficients of the model parameters and / or the physical model of the charge signal mapping relationship of the dielectric property-coal type parameters are fine-tuned online.

14. The method according to claim 13, characterized in that, The online fine-tuning uses the gradient descent algorithm to update the model parameters with a preset small learning rate.

15. The method according to claim 10, characterized in that, The process following step S1 and preceding step S2 includes a signal preprocessing step: Perform Butterworth bandpass filtering, power frequency notch filtering, and sliding median filtering on the charge signal; The microwave signal is smoothed by moving average and temperature compensation is performed using the probe's built-in temperature sensor.

Images