Quality detection system for fuel of thermal power plant

Abstract

The invention provides a thermal power plant fuel quality detection system, and relates to the technical field of detection, and the system comprises a multi-modal acquisition module which is used for acquiring multi-modal parameters of thermal power plant fuel; the data fusion processing module is used for preprocessing the multi-modal parameters of the thermal power plant fuel and generating a multi-dimensional feature vector from the preprocessed multi-modal parameters; the evaluation module is used for evaluating the multi-dimensional feature vector through a multi-modal fusion evaluation model; the intelligent early warning module comprises a first early warning unit which carries out first-level early warning when a certain modal parameter exceeds a preset threshold value; the second early warning unit is used for performing secondary early warning when the cosine similarity between the multi-dimensional feature vector and the historical inferior fuel feature is greater than or equal to a preset value; and the third early warning unit is used for carrying out third-level early warning by predicting a multi-modal parameter trend when the predicted quality score is lower than a preset value. The multi-modal acquisition module covers various quality related parameters of the fuel; compared with traditional single parameter detection, the fuel quality description dimension is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of fuel quality detection, and in particular to a quality detection system for fuel of a thermal power plant. BACKGROUND

[0002] In the production process of thermal power generation, fuel quality is one of the core factors affecting power generation efficiency, equipment operation and maintenance cost, and environmental protection index compliance. The traditional fuel quality detection method of a thermal power plant has problems such as multi-dimensional detection missing, rough data processing, and lagging evaluation and early warning, which cannot meet the needs of fine management of a smart power plant. The specific manifestations are as follows: I. Multi-modal parameter detection blank Traditional detection focuses on a single physical or chemical parameter (such as only detecting calorific value and sulfur content), ignoring the influence of fuel particle size distribution, water content spatial heterogeneity, and surface morphology on combustion efficiency and equipment wear.

[0003] II. Insufficient data processing and feature mining After collecting multi-source parameters, the temporal and spatial synchronization, abnormal values, and noise are not accurately preprocessed. III. Relies on manual experience or single-modal shallow model, and cannot fuse multi-dimensional features for accurate quality grading. SUMMARY

[0004] The present application provides a quality detection system for fuel of a thermal power plant to solve at least one of the technical problems in the background art.

[0005] To solve the above technical problems, the present application discloses a quality detection system for fuel of a thermal power plant, comprising: a multi-modal acquisition module for acquiring multi-modal parameters of fuel of a thermal power plant; a data fusion processing module for preprocessing the multi-modal parameters of the fuel of the thermal power plant and generating a multi-dimensional feature vector from the preprocessed multi-modal parameters; an evaluation module for evaluating the multi-dimensional feature vector through a multi-modal fusion evaluation model; an intelligent early warning module, comprising: a first early warning unit for performing a first-level early warning when a certain modal parameter exceeds a preset threshold; a second early warning unit for performing a second-level early warning when the cosine similarity between the multi-dimensional feature vector and the historical poor-quality fuel features is greater than or equal to a preset value; and a third early warning unit for performing a third-level early warning when the predicted quality score is lower than a preset value by predicting the trend of the multi-modal parameters.

[0006] Preferably, the parameters include: chemical composition data obtained by a near-infrared spectrum analyzer; moisture content data obtained by a microwave moisture meter; Particle size distribution data, obtained by a laser particle size analyzer; Element composition data, obtained by an X-ray fluorescence spectrometer.

[0007] Preferably, the preprocessing includes: Temporal and spatial synchronization calibration: based on the GPS time module, the multi-modal parameters are timestamped and aligned, and the spatial position deviation of the sensor is compensated by the Kalman filtering algorithm; Eliminate outliers: use improved Relyda criterion combined with local outlier factor algorithm to identify and eliminate abnormal data points with deviation exceeding 3σ; High-frequency noise parameter filtering and denoising: use adaptive wavelet packet threshold denoising + morphological filtering fusion algorithm for filtering and denoising.

[0008] Preferably, it also includes: Adaptive calibration module: automatically collect standard fuel samples for cross-validation every week; When the detection value deviates from the standard value by more than the preset range, trigger the sensor self-calibration program; the calibration algorithm uses the weighted least squares method, and the weight factor is dynamically adjusted according to the historical reliability of the sensor.

[0009] Preferably, it also includes: Intelligent decision-making module: based on the quality evaluation results, generate the optimal coal blending scheme through reinforcement learning algorithm; The output parameters include the mixing ratio of different coal types and the adjustment suggestion of the coal mill output.

[0010] Preferably, the evaluation module includes: Temporal and spatial perception feature encoding unit: Map the preprocessed multi-dimensional feature vector to a unified dimensional space, introduce the timestamp and spatial position of fuel collection, and generate temporal and spatial features through sinusoidal position encoding; Adjust the importance of each modal feature by calculating the temporal and spatial attention weight through multilayer perception machine; Deep cross-modal interaction unit: use multi-head cross-modal attention mechanism to calculate the correlation matrix between modalities, and integrate the attention output and the original feature through cross-modal gating unit; Quality multi-dimensional evaluation unit: Quality category prediction subunit: output quality classification results to provide prediction quality score basis for three-level early warning; Continuous score regression subunit: output 0-1 continuous quality score, which is used for cosine similarity calculation of secondary warning in cooperation with historical inferior fuel feature library.

[0011] Preferably, it also includes: Environmental humidity detection module: for detecting environmental humidity; a conveying speed detection module, configured to detect a conveying speed of the fuel; a storage module, configured to store historical detection data of the multi-modal acquisition module; an adjustment module, electrically connected with the ambient humidity detection module, the conveying speed detection module, and the storage module, and configured to adjust a sampling frequency of the multi-modal acquisition module based on the storage module, the conveying speed detection module, and the storage module.

[0012] Preferably, the adjustment module comprises: a first acquisition unit, configured to acquire acquisition data of the ambient humidity detection module and the conveying speed detection module; a second acquisition unit, configured to acquire latest historical detection data of the multi-modal acquisition module for a plurality of times of the current coal material; a first calculation unit, configured to calculate a current sampling frequency of the multi-modal acquisition module based on the first acquisition unit and the second acquisition unit; a first control unit, configured to control the sampling frequency of the multi-modal acquisition module in a preset time period to be the sampling frequency calculated by the first calculation unit.

[0013] Preferably, the current sampling frequency of the multi-modal acquisition module is calculated based on the following formula: = min (f1, f2, f3, f4) ; wherein, min is a minimum value; f0 is a reference sampling frequency of the current coal material; σ2 is a current window quality parameter variance; σ2ref is a variance of the fuel quality parameter in a reference time window; v is a current detection value of the conveying speed detection module; vmax is a rated maximum speed of the fuel conveying equipment of the thermal power plant; a is a correction coefficient of the fuel particle size state, and when the fuel is coal powder, a is 1, and when the fuel is coal block, a is 0; h is a current detection value of the ambient humidity detection module; fmax is a maximum allowed sampling frequency of the multi-modal acquisition module; hmax is a historical maximum humidity on site; p is a data proportion of the latest historical detection data of the multi-modal acquisition module that exceeds a corresponding preset allowed range; 、 、 、 respectively sampling frequency adjustment weight one, sampling frequency adjustment weight two, sampling frequency adjustment weight three, and sampling frequency adjustment weight four.

[0014] Preferably, a short-term time window for predicting the multi-modal parameter trend is calculated based on the following formula: ; wherein, is a current short-term time window for predicting the multi-modal parameter trend; is a preset minimum short-term time window for predicting the multi-modal parameter trend for the current fuel; is a preset maximum short-term time window for predicting the multi-modal parameter trend for the current fuel; is a variance of the current fuel quality parameter; is a variance of the reference fuel quality parameter within a certain time length; The long-term time window for predicting the multi-modal parameter trend is calculated based on the following formula: ; wherein, is a current long-term time window for predicting the multi-modal parameter trend; is a preset minimum long-term time window for predicting the multi-modal parameter trend; is a preset maximum long-term time window for predicting the multi-modal parameter trend for the current fuel; is a trend degree of the current fuel quality parameter, i.e. a quality parameter change rate within a second time length; is a reference quality parameter change rate within a third time length for the current fuel.

[0015] Compared with the prior art, the present application has the following beneficial effects: 1. The multi-modal acquisition module covers parameters such as chemical composition (near-infrared spectrum), moisture (microwave moisture meter), particle size (laser particle size instrument), and elemental composition (X-ray fluorescence spectrum); Compared with traditional single parameter detection, the fuel quality characterization dimension is improved (from "single dimension of calorific value / sulfur content" to "multi-dimension of physics + chemistry + morphology"); In the coal blending optimization scenario, due to the synergistic matching of particle size and moisture, the energy consumption of the coal mill is reduced (avoiding idling and over-pulverization caused by excessive particle size), and the stability of boiler combustion is improved.

[0016] The multi-modal acquisition module supports multi-type detection of coal powder, coal blocks, and biomass fuels, and the parameter acquisition range is adapted to different fuel forms; Under complex working conditions (such as high moisture in the rainy season and frozen blocks in winter), multi-modal parameters are coordinated, and the accuracy of abnormal identification is improved (frozen blocks and cohesive coal are jointly judged by particle size distribution and moisture content).

[0017] 2. The data fusion processing module uses GPS time service + Kalman filtering to realize time synchronization of multi-modal parameters; after time synchronization, the error of causal relationship analysis of calorific value and sulfur content is reduced (calibrating the time difference of belt conveying to avoid misjudgment of fuels with high calorific value but excessive sulfur content); After spatial calibration, the accuracy of the fuel quality thermodynamic map in the stacking area is improved, guiding the precise material taking of the stacker-reclaimer and improving the utilization rate of high-quality fuel (avoiding the "mixing of high-quality and low-quality areas" due to spatial deviation).

[0018] Outlier rejection: improved Relyda criterion + local outlier factor (LOF), improved recognition rate; Noise filtering: adaptive wavelet packet threshold denoising + morphological filtering, improved signal-to-noise ratio (SNR); The misjudgment rate of quality assessment caused by abnormal value interference is reduced; After filtering high-frequency noise (such as particle size fluctuations caused by belt vibration), the standard deviation of particle size detection is reduced, providing stable parameters for coal blending.

[0019] 3. The evaluation module uses spatiotemporal perception coding + multi-head cross-modal attention to fuse physical, chemical, and morphological characteristics; the quality classification accuracy is improved, and the missed detection rate of low-quality fuel is reduced; Continuous quality scoring compared with laboratory calorimeter, error reduction, providing accurate input for intelligent early warning.

[0020] 4. Multi-level early warning: from "passive response" to "active prevention": First-level early warning (single parameter over-limit): linkage of belt conveyor speed reduction (to avoid rapid entry of abnormal fuel into the furnace); Second-level early warning (historical similar low-quality): cosine similarity threshold, linkage of fuel isolation door (5s for low-quality fuel cutting); Third-level early warning (trend prediction deterioration): based on LSTM to predict future 30min quality, score <0.5, linkage of boiler load reduction (10min early warning); Equipment failure early warning lead time is improved, coal mill liner wear early warning is changed from "after failure alarm" to "2h before failure warning", maintenance cost is reduced by 25%.

[0021] 5. Technology association: adaptive calibration module automatically collects standard fuel samples every week, using weighted least squares method for calibration; detection errors caused by sensor drift are controlled after automatic calibration; manual calibration frequency is reduced from "once a week" to "once a month", reducing operation and maintenance labor cost.

[0022] The intelligent decision-making module generates a coal blending scheme based on reinforcement learning (PPO algorithm); After optimizing the coal blending scheme, the standard coal consumption of power generation is reduced (due to the coordinated matching of calorific value, sulfur content, and ash content); The response speed of the coal mill output adjustment is improved, and the output adjustment time is shortened from 5min to 2min when the load changes (such as from 500MW to 600MW), enhancing the climbing ability of the unit. BRIEF DESCRIPTION OF DRAWINGS

[0023] In order to more clearly illustrate the technical solutions in the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings described below are only some embodiments of the present application, and the ordinary skilled in the art can obtain other drawings according to these drawings without any creative effort.

[0024] Figure 1 It is a schematic diagram of a fuel quality detection system for a thermal power plant. DETAILED DESCRIPTION

[0025] In order to make the objects, technical solutions and advantages of the present application clearer, the following will describe the technical solutions in the present application clearly and completely with reference to the drawings in the present application. Obviously, the described embodiments are only some embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by the ordinary skilled in the art without any creative effort belong to the protection scope of the present application.

[0026] Embodiment 1, the present application provides a fuel quality detection system for a thermal power plant, as shown in the figure, comprising: Figure 1 Multi-modal acquisition module: for acquiring multi-modal parameters of the fuel of the thermal power plant; Data fusion processing module: for pre-processing the multi-modal parameters of the fuel of the thermal power plant, and generating a multi-dimensional feature vector from the pre-processed multi-modal parameters; Evaluation module: for evaluating the multi-dimensional feature vector through a multi-modal fusion evaluation model; Intelligent early warning module, comprising: First early warning unit: for performing a first-level early warning when a certain modal parameter exceeds a preset threshold; second early warning unit: for performing a second-level early warning when the cosine similarity between the multi-dimensional feature vector and the historical poor-quality fuel features is greater than or equal to a preset value; third early warning unit: for performing a third-level early warning by predicting the trend of the multi-modal parameters when the predicted quality score is lower than a preset value.

[0027] Preferably, the parameters comprise: Chemical composition data, obtained by a near-infrared spectrum analyzer; Moisture content data, obtained by a microwave moisture meter; Particle size distribution data, obtained by a laser particle size analyzer; Element composition data, obtained by an X-ray fluorescence spectrometer.

[0028] Preferably, the pre-processing comprises: ​Space-time synchronization calibration: based on the GPS timing module, the multi-modal parameters are time-stamped aligned, and the sensor spatial position deviation is compensated by Kalman filtering algorithm; Outlier rejection: using improved Rida criterion combined with local outlier factor algorithm, identifying and rejecting abnormal data points with deviation exceeding 3σ; High-frequency noise parameter filtering and denoising: using adaptive wavelet packet threshold denoising + morphological filtering fusion algorithm for filtering and denoising.

[0029] Preferably, it also includes: Adaptive calibration module: automatically collect standard fuel samples for cross-validation every week; When the detection value deviates from the standard value by more than the preset range, trigger the sensor self-calibration program; the calibration algorithm uses weighted least squares method, and the weight factor is dynamically adjusted according to the historical reliability of the sensor.

[0030] Preferably, it also includes: Intelligent decision-making module: based on the quality evaluation results, generate the optimal coal blending scheme through reinforcement learning algorithm; The output parameters include the mixing ratio of different coal types and the adjustment suggestion of the coal mill output.

[0031] Preferably, the evaluation module includes: Space-time perception feature encoding unit: Map the preprocessed multi-dimensional feature vector to a unified dimensional space, introduce the time stamp and spatial position of fuel collection, and generate space-time features through sinusoidal position encoding; Adjust the importance of each modal feature by calculating the space-time attention weight through multi-layer perception machine; Deep cross-modal interaction unit: use multi-head cross-modal attention mechanism to calculate the correlation matrix between modalities, and integrate the attention output and the original features through cross-modal gating unit; Quality multi-dimensional evaluation unit: Quality category prediction sub-unit: output quality classification results to provide prediction quality score basis for three-level early warning; Continuous score regression sub-unit: output 0-1 continuous quality score, which is used for cosine similarity calculation for secondary warning in cooperation with historical inferior fuel feature library.

[0032] The beneficial effects of the above technical solutions are: 1. Multi-modal acquisition module covers chemical composition (near-infrared spectrum), moisture (microwave moisture meter), particle size (laser particle size analyzer), elemental composition (X-ray fluorescence spectrum) and other parameters; Compared with traditional single parameter detection, the fuel quality characterization dimension is improved (from "single dimension of calorific value / sulfur content" to "multi-dimension of physics + chemistry + morphology"); In the coal blending optimization scenario, due to the matching of particle size and moisture, the energy consumption of the coal mill is reduced (avoiding idling and over-grinding caused by excessive particle size), and the stability of boiler combustion is improved.

[0033] The multi-modal acquisition module supports detection of multiple types of fuels such as coal powder, coal blocks, and biomass fuels, and the parameter acquisition range is adapted to different fuel forms. Under complex working conditions (such as high moisture in the rainy season and frozen blocks in winter), multi-modal parameters are coordinated, and the accuracy of abnormality identification is improved (frozen blocks and cohesive coal are judged by combining particle size distribution and moisture content).

[0034] 2, The data fusion processing module uses GPS timing + Kalman filtering to realize time synchronization of multi-modal parameters; after time synchronization, the error of causal relationship analysis of heat value and sulfur content is reduced (calibrating the time difference of belt conveying to avoid misjudgment of high heat value but high sulfur content fuel); belt conveying to avoid misjudgment of high heat value but high sulfur content fuel); After spatial calibration, the fuel quality thermal map precision of the stacking area is improved, guiding the precise taking of the stacker-reclaimer, and the utilization rate of high-quality fuel is improved (avoiding the mixing of high-quality and low-quality areas due to spatial deviation).

[0035] Outlier rejection: improved Relyda criterion + local outlier factor (LOF), recognition rate improved; Noise filtering: adaptive wavelet packet threshold denoising + morphological filtering, signal-to-noise ratio (SNR) improved; The misjudgment rate of quality evaluation caused by abnormal value interference is reduced; After filtering high-frequency noise (such as particle size fluctuations caused by belt vibration), the standard deviation of particle size detection is reduced, providing stable parameters for coal blending and blending.

[0036] 3, The evaluation module uses spatio-temporal perception coding + multi-head cross-modal attention to fuse physical, chemical, and morphological features; the quality classification accuracy is improved, and the missed detection rate of low-quality fuel is reduced; Continuous quality scoring compared with laboratory calorimeter, error reduced, providing accurate input for intelligent early warning.

[0037] 4. Multi-level early warning: from "passive response" to "active prevention": First-level early warning (single parameter exceeding limit): linkage of belt conveyor speed reduction (to avoid rapid feeding of abnormal fuel into the furnace); Second-level early warning (historical poor quality similar): cosine similarity threshold, linkage of fuel isolation door (5s for cutting off low-quality fuel); Third-level early warning (trend prediction deterioration): based on LSTM to predict future 30min quality, score <0.5, linkage of boiler load reduction (10min early warning); The early warning time of equipment failure is improved, the mill liner wear early warning is changed from "after failure alarm" to "2h before failure warning", and the maintenance cost is reduced by 25%.

[0038] 5. Technology association: The adaptive calibration module automatically collects standard fuel samples every week, and uses weighted least squares method for calibration; the detection error caused by sensor drift is controlled after automatic calibration; the frequency of manual calibration is reduced from "once a week" to "once a month", and the operation and maintenance labor cost is reduced.

[0039] The intelligent decision module generates a coal blending scheme based on reinforcement learning (PPO algorithm); After optimization of the coal blending scheme, the coal consumption of power generation is reduced (due to the synergistic matching of calorific value, sulfur content and ash content); The response speed of the coal mill output adjustment is improved, and when the load changes (such as from 500MW to 600MW), the output adjustment time is shortened from 5min to 2min, and the climbing ability of the unit is enhanced.

[0040] In embodiment 2, on the basis of embodiment 1, further comprising: An environmental humidity detection module for detecting environmental humidity; A conveying speed detection module for detecting the conveying speed of the fuel; A storage module for storing historical detection data of the multi-modal acquisition module; An adjustment module electrically connected with the environmental humidity detection module, the conveying speed detection module and the storage module, and configured to adjust the sampling frequency of the multi-modal acquisition module based on the storage module, the conveying speed detection module and the storage module.

[0041] Preferably, the adjustment module comprises: A first acquisition unit configured to acquire the acquisition data of the environmental humidity detection module and the conveying speed detection module; A second acquisition unit configured to acquire the latest historical detection data of the multi-modal acquisition module for the current coal material for several times; A first calculation unit configured to calculate the current sampling frequency of the multi-modal acquisition module based on the first acquisition unit and the second acquisition unit; A first control unit configured to control the sampling frequency of the multi-modal acquisition module in the current preset time period to be the sampling frequency calculated by the first calculation unit.

[0042] Preferably, the current sampling frequency of the multi-modal acquisition module is calculated based on the following formula: ; Wherein, min is the minimum value; is the reference sampling frequency of the current coal material; is the current window quality parameter variance; is the variance of the fuel quality parameter in the reference time window; ​is the current detection value of the conveying speed detection module; is the rated maximum speed of the fuel conveying equipment of the thermal power plant; is the correction coefficient of the fuel particle size state, which is 1 when the fuel is coal powder and 0 when the fuel is coal block; is the current detection value of the ambient humidity detection module; is the maximum allowed sampling frequency of the multi-modal acquisition module; is the maximum historical humidity of the thermal power plant site; is the data proportion of the historical detection data of the multi-modal acquisition module in the latest several times that exceeds the corresponding preset allowed range; 、 、 、 respectively sampling frequency adjustment weight one, sampling frequency adjustment weight two, sampling frequency adjustment weight three, sampling frequency adjustment weight four (all take values greater than 0 and less than 1, and 1 after the four).

[0043] Preferably, the short-term time window for predicting the multi-modal parameter trend is calculated based on the following formula: ; wherein, is the current short-term time window for predicting the multi-modal parameter trend; is the preset minimum short-term time window (taking a value of 1-60S) for predicting the multi-modal parameter trend of the current fuel; is the preset maximum short-term time window (taking a value of 30-300S) for predicting the multi-modal parameter trend of the current fuel; is the variance of the current fuel quality parameter; is the variance of the reference fuel quality parameter within a certain time length; The long-term time window for predicting the multi-modal parameter trend is calculated based on the following formula: ; wherein, is the current long-term time window for predicting the multi-modal parameter trend; is the preset minimum long-term time window (taking a value of 300-1800S) for predicting the multi-modal parameter trend; is the preset maximum long-term time window (taking a value of 1800-3000S) for predicting the multi-modal parameter trend of the current fuel; is the trend degree of the current fuel quality parameter, i.e. the quality parameter change rate within the second time length; is the reference quality parameter change rate within the third time length of the current fuel.

[0044] The above technical solution has the following beneficial effects: The scheme realizes intelligent sampling control through "multi-parameter perception, historical data fusion, and dynamic adjustment of sampling frequency". The core is to dynamically adapt the sampling frequency using environmental humidity, conveying speed, historical data, and other factors, solving the problem of "redundancy / insufficiency coexistence" of traditional fixed sampling.

[0045] The short-term window ensures accurate real-time data, and the long-term window ensures reliable trend analysis. The combination of the two ensures that the system can not only perceive the present in detail but also clearly understand the long-term trend in fuel quality detection. The detection data quality and analysis depth are greatly improved, providing a comprehensive and accurate data foundation for real-time monitoring to long-term optimization of power plant fuel management.

[0046] Enhance system robustness: In the face of complex and variable fuel characteristics (changes in humidity, particle size, etc.), conveying conditions (speed fluctuations, equipment start-stop), and environmental interference (high humidity in rainy weather, open-air dust, etc.), the dual-window dynamic adjustment mechanism allows the system to adapt flexibly, enhancing the system's robustness in various complex scenarios, and ensuring stable output of high-quality detection results, ensuring stable operation of power plant fuel quality detection and subsequent production links.

[0047] Finally, it should be noted that: the above examples are only used to illustrate the technical solutions of the present application, and not to limit it; although the present application has been described in detail with reference to the foregoing examples, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing examples, or make equivalent replacement for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A quality inspection system for fuel in thermal power plants, characterized in that, include: Multimodal acquisition module: used to acquire multimodal parameters of fuel in thermal power plants; Data fusion processing module: preprocesses the multimodal parameters of the thermal power plant fuel and generates a multidimensional feature vector from the preprocessed multimodal parameters; Evaluation module: Evaluates multi-dimensional feature vectors using a multi-modal fusion evaluation model; Intelligent early warning module: includes: First warning unit: When a certain modal parameter exceeds a preset threshold, a first-level warning is issued; Second warning unit: When the cosine similarity between the multidimensional feature vector and the historical inferior fuel features is greater than or equal to a preset value, a second-level warning is issued; Third warning unit: By predicting the trend of multimodal parameters, when the predicted quality score is lower than a preset value, a third-level warning is issued.

2. The quality detection system for thermal power plant fuel according to claim 1, characterized in that, The parameters include: Chemical composition data were obtained using a near-infrared spectroscopy analyzer. Moisture content data were obtained using a microwave moisture meter; Particle size distribution data were acquired using a laser particle size analyzer. Elemental composition data were obtained using an X-ray fluorescence spectrometer.

3. The quality detection system for thermal power plant fuel according to claim 1, characterized in that, The preprocessing includes: Spatiotemporal synchronization calibration: Based on the GPS timing module, timestamps are aligned for multimodal parameters, and the spatial position deviation of the sensor is compensated by the Kalman filter algorithm; Outlier removal: An improved Laida criterion combined with a local outlier factor algorithm is used to identify and remove outlier data points with a deviation exceeding 3σ. High-frequency noise parameter filtering and denoising: An adaptive wavelet packet threshold denoising + morphological filtering fusion algorithm is used for filtering and denoising.

4. The quality detection system for thermal power plant fuel according to claim 1, characterized in that, Also includes: Adaptive calibration module: Automatically collects standard fuel samples weekly for cross-validation; When the detected value deviates from the standard value by more than a preset range, the sensor self-calibration program is triggered. The calibration algorithm uses weighted least squares, and the weighting factors are dynamically adjusted based on the sensor's historical reliability.

5. A quality detection system for thermal power plant fuel according to claim 1, characterized in that, Also includes: Intelligent decision-making module: Based on the quality assessment results, it generates the optimal coal blending scheme through reinforcement learning algorithms; The output parameters include the mixing ratio of different coal types and suggestions for adjusting the output of the coal mill.

6. The quality detection system for thermal power plant fuel according to claim 1, characterized in that, The evaluation module includes: Spatiotemporal awareness feature encoding unit: The preprocessed multidimensional feature vectors are mapped to a unified dimensional space, and the timestamp and spatial location of fuel collection are introduced. Spatiotemporal features are generated through sinusoidal position encoding. The importance of each modality feature is adjusted by calculating the spatiotemporal attention weights using a multilayer perceptron. Deep cross-modal interaction unit: Employs a multi-head cross-modal attention mechanism to calculate the correlation matrix between modalities and integrates the attention output with the original features through a cross-modal gating unit; Multidimensional Quality Assessment Unit: Quality Category Prediction Subunit: Outputs quality classification results, providing a basis for predictive quality scoring for Level 3 early warning; Continuous scoring regression subunit: Outputs a continuous quality score of 0 to 1, which, in conjunction with the historical poor-quality fuel feature library, is used for cosine similarity calculation for secondary early warning.

7. The quality detection system for thermal power plant fuel according to claim 1, characterized in that, Also includes: Ambient humidity detection module: used to detect ambient humidity; Conveying speed detection module: used to detect the fuel conveying speed; Storage module: Used to store historical detection data from the multimodal acquisition module; Adjustment module: The adjustment module is electrically connected to the ambient humidity detection module, the conveying speed detection module, and the storage module respectively. The adjustment module is used to adjust the sampling frequency of the multimodal acquisition module based on the storage module, the conveying speed detection module, and the storage module.

8. A quality detection system for thermal power plant fuel according to claim 7, characterized in that, The adjustment module includes: First acquisition unit: used to acquire data collected by the ambient humidity detection module and the conveying speed detection module; The second acquisition unit is used to acquire the latest historical detection data from the multimodal acquisition module for the current type of coal. First calculation unit: used to calculate the current sampling frequency of the multimodal acquisition module based on the first acquisition unit and the second acquisition unit; First control unit: used to control the sampling frequency of the current preset time period of the multimodal acquisition module to be the sampling frequency calculated by the first calculation unit.

9. A quality detection system for thermal power plant fuel according to claim 8, characterized in that, The current sampling frequency of the multimodal acquisition module Calculated based on the following formula: ； Where min is the minimum value; This is the current benchmark sampling frequency for this type of coal. The variance of the current window quality parameter; This represents the variance of fuel quality parameters over a reference time window. The current detection value of the conveyor speed detection module; This refers to the rated maximum speed of fuel conveying equipment in thermal power plants. This is a correction factor for the particle size of the fuel. For example, it takes a value of 1 when it is pulverized coal and a value of 0 when it is lumpy coal. This is the current detection value from the ambient humidity detection module; This is the maximum allowable sampling frequency for the multimodal acquisition module; This represents the highest humidity level ever recorded at a thermal power plant site. The percentage of historical detection data from the latest multimodal acquisition modules that exceeds the corresponding preset allowable range; , , , The sampling frequency adjustment weights are set as follows: weight 1, weight 2, weight 3, and weight 4.

10. A quality inspection system for thermal power plant fuel according to claim 2 or 8, characterized in that, The short-term time window for predicting multimodal parameter trends is calculated based on the following formula: ； in, For predicting the current short-term time window of multimodal parameter trends; The preset minimum short-term time window for predicting the trend of multimodal parameters for the current fuel; The preset maximum short-term time window for the current fuel to predict the trend of multimodal parameters; This represents the variance of the current fuel quality parameters; The variance of the reference fuel quality parameters over a certain period of time; The long-term time window for predicting multimodal parameter trends is calculated based on the following formula: ； in, For predicting the current long-term time window of multimodal parameter trends; A preset minimum long-term time window for predicting the trend of multimodal parameters; The preset maximum long-term time window for the current fuel to predict the trend of multimodal parameters; The trend of the current fuel quality parameters, i.e., the rate of change of the quality parameters within the second time period; This represents the rate of change of the reference mass parameter over the third time period for the current fuel type.

Images