Device and method for detecting concentration of particulate matters in sintering flue gas of steel plant

Abstract

The present application relates to the technical field of concentration detection, and provides a steel plant sintering flue gas particulate matter concentration detection device and a detection method.The device comprises a different particle size section concentration collection module, a sintering process optimization module, a dust removal optimization module and a detection data visualization module.The present application can obtain particulate matter mass concentration data of different particle size sections such as ultrafine particles, fine particles, medium particles and coarse particles, thereby accurately locating the abnormal source of the sintering process based on the particle size section characteristics, and generating targeted sintering process control strategies and dust remover adjustment strategies.Meanwhile, the detection data visualization module integrates real-time data, the current sintering process state level, the main abnormal source, the sintering process control strategy and the dust remover adjustment strategy and displays them in a graphical interface, greatly reducing the experience dependence of the operation personnel and providing intuitive operation guidance.

Description

Technical Field

[0001] This invention belongs to the field of concentration detection technology, specifically relating to a device and method for detecting particulate matter concentration in sintering flue gas from steel plants. Background Technology

[0002] The sintering flue gas from steel plants has a complex composition, with particulate matter being one of the main air pollutants. Effective monitoring and control of its concentration is crucial for achieving environmental emission standards. Currently, this field mainly relies on traditional particulate matter monitors installed in flues or at the main exhaust outlet. However, these technical solutions have significant limitations and are no longer suitable for the development needs of intelligent sintering production. The shortcomings of existing technologies are mainly reflected in the following three aspects:

[0003] (1) Existing technologies can only provide the total mass concentration of particulate matter, but cannot obtain the concentration data of different particle size ranges such as ultrafine particles, fine particles, medium particles and coarse particles. Since the generation mechanism of particulate matter in different particle size ranges is directly related to the sintering process parameters, the inability to detect the concentration of particulate matter in different particle size ranges makes it impossible to accurately locate the source of abnormality in the sintering process based on the concentration of particulate matter in different particle size ranges, thus losing the opportunity to optimize and control the sintering process from the source;

[0004] (2) Although existing technologies have particle size concentration monitoring points installed before and after key equipment such as electrostatic precipitators and wet dust collectors, the concentration data has not been effectively correlated and analyzed, resulting in the concentration data being disconnected from the application and making it impossible to achieve precise and coordinated control of the dust removal system.

[0005] (3) The visualization interface of existing concentration detection devices is mainly based on numbers and charts, which cannot directly provide clear operating condition assessment and operation guidance. This requires operators to have a high level of experience to interpret it. This mode of only alarming and not diagnosing, only displaying and not guiding, greatly reduces the value of concentration data. Summary of the Invention

[0006] This invention provides a device and method for detecting particulate matter concentration in sintering flue gas from steel plants, thereby solving at least one of the aforementioned technical problems. This invention is achieved through the following technical solution:

[0007] A device and method for detecting particulate matter concentration in sintering flue gas from a steel plant, the device comprising:

[0008] The different particle size range concentration acquisition module is used to collect the mass concentration of particulate matter of different particle size ranges in the flue gas in the main flue, electrostatic precipitator inlet and outlet, and wet dust collector inlet and outlet by several particulate matter concentration acquisition devices installed at the main flue of the sintering flue gas treatment system, the inlet and outlet of the electrostatic precipitator, and the inlet and outlet of the wet dust collector.

[0009] The sintering process optimization module is used to analyze the process status of particulate matter mass concentration data of different particle size ranges collected at the large flue, identify the current sintering process status level and locate the main sources of anomalies, and generate sintering process control strategies.

[0010] The dust removal optimization module is used to calculate the graded dust removal efficiency of each particle size segment under the corresponding dust collector based on the mass concentration of particulate matter in different particle size segments collected at the inlet and outlet of the electrostatic precipitator and the inlet and outlet of the wet dust collector, and to generate the corresponding adjustment strategy for the dust collector.

[0011] The detection data visualization module integrates and renders the real-time data collected by the concentration acquisition module for different particle size ranges, the current sintering process status level and main sources of anomalies and sintering process control strategies analyzed by the sintering process optimization module, and the graded dust removal efficiency calculated by the dust removal optimization module and the generated dust collector adjustment strategies, and displays them in the form of a graphical interface.

[0012] Preferably, the sintering flue gas treatment system includes a sintering machine, a large flue, an electrostatic precipitator, a wet desulfurization device, a wet dust collector, an SCR denitrification tower, and a chimney connected in sequence.

[0013] Preferably, the particulate matter concentration collector includes:

[0014] A multi-wavelength laser emitting unit is used to emit lasers of at least two different wavelengths into the flue gas under test;

[0015] The multi-angle scattered light acquisition unit includes multiple photoelectric sensors arranged at different scattering angles, which are used to synchronously acquire the scattered light signals generated after the particulate matter is irradiated by lasers of different wavelengths.

[0016] The signal processing unit is used to condition and convert the scattered light signal from analog to digital before inputting it into the microprocessor, which then outputs the mass concentration of particles in different particle size ranges.

[0017] Preferably, the sintering process optimization module includes:

[0018] The curve construction submodule is used to construct time curves of particulate matter concentration in different particle size segments of flue gas in the large flue based on the mass concentration of particulate matter in different particle size segments of flue gas.

[0019] The feature matrix construction submodule is used to extract the multidimensional feature values ​​corresponding to each detection cycle of the particulate matter concentration time curve of different particle size segments in the large flue, and generate the feature vector corresponding to each detection cycle of the particulate matter concentration time curve of different particle size segments. Based on the feature vector corresponding to each detection cycle of the particulate matter concentration time curve of different particle size segments, a process analysis comparison matrix of different particle size segments is constructed.

[0020] The sintering process comparison submodule is used to perform multi-feature fusion and weighted analysis on the process analysis comparison matrix of each particle size range, generate a comprehensive process feature matrix, and compare the comprehensive process feature matrix with the pre-stored benchmark process matrix in multiple dimensions. Based on the comparison results, the current sintering process status level is identified and the main sources of anomalies are located.

[0021] The process optimization and control submodule generates targeted sintering process control strategies based on the current sintering process status level and the identified main sources of anomalies through a multi-level rule base.

[0022] Preferably, the sintering process comparison submodule includes:

[0023] The feature extraction and encoding unit is used to extract four types of features from the process analysis comparison matrix of each particle size segment: statistical features, structural features, dynamic features, and distribution features. The four types of features corresponding to each particle size segment are then concatenated into a feature vector corresponding to each particle size segment.

[0024] The attention weight learning unit is used to input the feature vector corresponding to each particle size segment into the trained attention network, and output the attention weight values ​​of ultrafine particle size segment, fine particle size segment, medium particle size segment and coarse particle size segment through the trained attention network.

[0025] The dynamic matrix fusion unit is used to standardize the process analysis comparison matrix of each particle size segment, and to perform weighted summation on the standardized process analysis comparison matrix of each particle size segment based on the attention weight values ​​of ultrafine particles, fine particles, medium particles, and coarse particles to generate a comprehensive process feature matrix.

[0026] The multi-granularity similarity calculation unit is used to calculate the element-level similarity, feature-level similarity, and structural-level similarity between the integrated process feature matrix and the pre-stored benchmark process matrix, and to calculate the overall similarity based on the element-level similarity, feature-level similarity, and structural-level similarity.

[0027] The process status identification unit is used to determine the current sintering process status level based on the overall similarity. The sintering process status level includes five levels: excellent, good, attention, abnormal, and serious abnormal. If the current sintering process status level is attention, abnormal, or serious abnormal, the particle size segment with the largest attention weight value in each particle size segment is located as the main source of abnormality.

[0028] Preferably, the multi-granularity similarity calculation unit includes:

[0029] The element-level similarity calculation subunit calculates the element-level similarity between the integrated process feature matrix and the pre-stored benchmark process matrix based on the element values ​​of the integrated process feature matrix and the pre-stored benchmark process matrix. ;

[0030] The feature-level similarity calculation subunit performs dimensionality reduction on the integrated process feature matrix and the pre-stored benchmark process matrix based on principal component analysis (PCA). This generates the integrated process principal component eigenvectors and the benchmark process principal component eigenvectors corresponding to the integrated process feature matrix and the pre-stored benchmark process matrix, respectively. The cosine similarity between the integrated process principal component eigenvectors and the benchmark process principal component eigenvectors is then calculated and used as the feature-level similarity between the integrated process feature matrix and the pre-stored benchmark process matrix. ;

[0031] The structural similarity calculation subunit treats both the integrated process feature matrix and the pre-stored benchmark process matrix as weighted adjacency matrices. It constructs weighted graphs corresponding to the integrated process feature matrix and the pre-stored benchmark process matrix, obtaining the integrated process feature weighted graph and the benchmark process weighted graph. It then calculates the graph edit distance D between the integrated process feature weighted graph and the benchmark process weighted graph, and calculates the structural similarity between the integrated process feature matrix and the pre-stored benchmark process matrix based on the graph edit distance D. ;

[0032] The overall similarity calculation subunit calculates the overall similarity based on element-level similarity, feature-level similarity, and structural-level similarity and their corresponding weight coefficients. .

[0033] Preferably, the element-level similarity calculation subunit calculates the element-level similarity between the integrated process feature matrix and the pre-stored benchmark process matrix based on the element values ​​of the integrated process feature matrix and the pre-stored benchmark process matrix using the following formula:

[0034] ;in, The element-level similarity between the comprehensive process feature matrix and the pre-stored benchmark process matrix is ​​given by m and n, respectively, where the comprehensive process feature matrix and the pre-stored benchmark process matrix have the same number of rows and columns. The value of the element in the j-th row and k-th column of the comprehensive process feature matrix. The value of the element in the j-th row and k-th column of the pre-stored baseline process matrix.

[0035] Preferably, the structural similarity calculation subunit calculates the structural similarity between the integrated process feature matrix and the pre-stored benchmark process matrix based on the graph edit distance D. The formula is:

[0036] ;in, To edit the distance in the image, This represents the theoretical upper limit of the graph editing distance.

[0037] The overall similarity calculation subunit calculates the overall similarity based on element-level similarity, feature-level similarity, and structural-level similarity, along with their corresponding weight coefficients. The formula is:

[0038] ;in, , and These are the weight coefficients corresponding to element-level similarity, feature-level similarity, and structural-level similarity, respectively.

[0039] Preferably, the dust removal optimization module includes:

[0040] The dust removal efficiency calculation submodule is used to calculate the graded dust removal efficiency of each particle size segment under the corresponding dust collector based on the mass concentration of particulate matter in different particle size segments collected at the inlet and outlet of the electrostatic precipitator and the inlet and outlet of the wet dust collector.

[0041] The dynamic optimization submodule is used to generate adjustment strategies for electrostatic precipitators and wet dust collectors based on the graded dust removal efficiency of particles of different sizes under the corresponding dust collectors.

[0042] This invention provides a method for detecting particulate matter concentration in sintering flue gas from a steel plant, which is performed using any one of the particulate matter concentration detection devices for sintering flue gas from steel plants in Examples 1-9, and includes:

[0043] S1. By installing several particulate matter concentration collectors at the main flue of the sintering flue gas treatment system, the inlet and outlet of the electrostatic precipitator, and the inlet and outlet of the wet scrubber, the mass concentration of particulate matter of different particle sizes in the flue gas in the main flue, the inlet and outlet of the electrostatic precipitator, and the inlet and outlet of the wet scrubber is collected.

[0044] S2. Perform process status analysis on the particulate matter mass concentration data of different particle size ranges collected at the main flue, identify the current sintering process status level and locate the main sources of anomalies, and generate sintering process control strategies.

[0045] S3. Based on the mass concentration of particulate matter of different particle size ranges collected at the inlet and outlet of the electrostatic precipitator and the inlet and outlet of the wet dust collector, calculate the graded dust removal efficiency of particulate matter of each particle size range under the corresponding dust collector, and generate the corresponding dust collector adjustment strategy.

[0046] S4. Integrate and render the real-time data collected in step S1, the current sintering process status level, main sources of abnormalities and sintering process control strategies analyzed in step S2, and the graded dust removal efficiency and generated dust collector adjustment strategies calculated in step S3, and display them in the form of a graphical interface.

[0047] Compared with the prior art, the present invention has the following beneficial effects:

[0048] This invention can acquire particulate matter mass concentration data for different particle size ranges, including ultrafine, fine, medium, and coarse particles. Based on particle size characteristics, it can accurately pinpoint the sources of sintering process anomalies and generate targeted sintering process control strategies and dust collector adjustment strategies. Simultaneously, the data visualization module integrates real-time data, the current sintering process status level, major anomaly sources, sintering process control strategies, and dust collector adjustment strategies, displaying them in a graphical interface. This significantly reduces the reliance on operator experience and provides intuitive operational guidance. Attached Figure Description

[0049] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0050] Figure 1 This is a schematic diagram of the component structure of the particulate matter concentration detection device for sintering flue gas in a steel plant according to the present invention.

[0051] Figure 2 This is a circuit diagram of the particulate matter concentration collector in this invention;

[0052] Figure 3 This is a weighted graph corresponding to the comprehensive process feature matrix in this invention;

[0053] Figure 4 This is a weighted graph corresponding to the baseline process matrix in this invention. Detailed Implementation

[0054] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0055] Example 1: This embodiment of the invention provides a device for detecting particulate matter concentration in sintering flue gas from a steel plant, such as... Figure 1-4 As shown, the device includes:

[0056] The different particle size range concentration acquisition module is used to collect the mass concentration of particulate matter of different particle size ranges in the flue gas in the main flue, electrostatic precipitator inlet and outlet, and wet dust collector inlet and outlet by several particulate matter concentration acquisition devices installed at the main flue of the sintering flue gas treatment system, the inlet and outlet of the electrostatic precipitator, and the inlet and outlet of the wet dust collector.

[0057] The sintering process optimization module is used to analyze the process status of particulate matter mass concentration data of different particle size ranges collected at the large flue, identify the current sintering process status level and locate the main sources of anomalies, and generate sintering process control strategies.

[0058] The dust removal optimization module is used to calculate the graded dust removal efficiency of each particle size segment under the corresponding dust collector based on the mass concentration of particulate matter in different particle size segments collected at the inlet and outlet of the electrostatic precipitator and the inlet and outlet of the wet dust collector, and to generate the corresponding adjustment strategy for the dust collector.

[0059] The detection data visualization module integrates and renders the real-time data collected by the concentration acquisition module for different particle size ranges, the current sintering process status level and main sources of anomalies and sintering process control strategies analyzed by the sintering process optimization module, and the graded dust removal efficiency calculated by the dust removal optimization module and the generated dust collector adjustment strategies, and displays them in the form of a graphical interface.

[0060] In this embodiment, the different particle size ranges include ultrafine particle range, fine particle range, medium particle range and coarse particle range.

[0061] In this embodiment, the process status analysis includes constructing particulate matter concentration-time curves for different particle size ranges, extracting features from the particulate matter concentration-time curves for different particle size ranges to construct a comprehensive process feature matrix, and identifying the current sintering process status level and the main abnormal particle size range by comparing the comprehensive process feature matrix with the pre-stored benchmark process matrix in multiple dimensions.

[0062] In this embodiment, the sintering process status level includes five levels: excellent, good, caution, abnormal, and serious abnormal.

[0063] In this embodiment, the main source of anomaly refers to the specific particle size range that contributes the most to the deterioration of the current overall process status.

[0064] In this embodiment, the sintering process control strategy is based on the current sintering process status level and the particle size range of the main abnormal sources, and is generated by querying a pre-stored multi-level rule base to generate adjustment instructions for the sintering machine process parameters.

[0065] The beneficial effects of the above technical solution are as follows: By setting up modules for collecting concentrations of different particle sizes, optimizing sintering processes, optimizing dust removal, and visualizing detection data, comprehensive collection, analysis, and optimization of particulate matter mass concentrations of different particle sizes in the sintering flue gas of steel plants are achieved. Compared with existing technologies that can only provide total mass concentration, this device can acquire particulate matter mass concentration data for different particle sizes, such as ultrafine, fine, medium, and coarse particles. This allows for precise location of sintering process anomalies based on particle size characteristics, and the generation of targeted sintering process control strategies and dust collector adjustment strategies. Simultaneously, the detection data visualization module integrates real-time data, the current sintering process status level, major anomaly sources, sintering process control strategies, and dust collector adjustment strategies, displaying them in a graphical interface. This significantly reduces the reliance on operator experience and provides intuitive operational guidance.

[0066] Example 2: Based on Example 1, the sintering flue gas treatment system includes a sintering machine, a large flue, an electrostatic precipitator, a wet desulfurization device, a wet dust collector, an SCR denitrification tower, and a chimney connected in sequence.

[0067] In this embodiment, the electrostatic precipitator efficiently removes particulate matter from the flue gas, ensuring that the emitted flue gas has a low particulate matter concentration in the subsequent treatment stage. The working principle is that when the dust-laden gas passes through the high-voltage electrostatic field, it is electrically separated. After the dust particles combine with negative ions and become negatively charged, they tend to discharge and deposit on the anode surface.

[0068] In this embodiment, the wet desulfurization equipment mainly uses the limestone / lime-gypsum method or the ammonia method for desulfurization, effectively removing sulfides from the flue gas and reducing environmental pollution. The wet desulfurization equipment is located at the end of the main flue, after the electrostatic precipitator. The reaction temperature of the desulfurization process is below the dew point, so the desulfurized flue gas needs to be reheated before being discharged. Due to the gas-liquid reaction, its desulfurization reaction speed is fast, its efficiency is high, and the utilization rate of desulfurization additives is high.

[0069] In this embodiment, the wet scrubber is a new type of dust removal equipment used to treat dust and microparticles. It is mainly used to remove harmful substances such as dust, acid mist, water droplets, aerosols, odors, and PM2.5 from humid gases, and is an ideal device for controlling atmospheric dust pollution.

[0070] In this embodiment, the SCR denitrification tower utilizes a metal catalyst within a certain temperature range by injecting gaseous... A reducing agent that reduces nitrogen oxides in flue gas to harmless forms. Water.

[0071] The beneficial effects of the above technical solution are as follows: By clearly defining the sintering flue gas treatment system as comprising a sintering machine, a large flue, an electrostatic precipitator, a wet desulfurization device, a wet scrubber, an SCR denitrification tower, and a chimney connected in sequence, efficient multi-stage removal of particulate matter is ensured in the treatment process. The electrostatic precipitator performs initial and efficient capture of particulate matter, the wet desulfurization device effectively removes sulfides, the wet scrubber further treats fine particles and harmful substances, and the SCR denitrification tower reduces nitrogen oxides. The entire system works in synergy, not only improving the overall removal efficiency of particulate matter but also addressing the needs of desulfurization and denitrification, thereby achieving comprehensive treatment and compliant emissions of sintering flue gas, and enhancing the system's reliability and environmental friendliness.

[0072] Example 3: Based on Example 1 or Example 2, the particulate matter concentration collector includes:

[0073] A multi-wavelength laser emitting unit is used to emit lasers of at least two different wavelengths into the flue gas under test;

[0074] The multi-angle scattered light acquisition unit includes multiple photoelectric sensors arranged at different scattering angles, which are used to synchronously acquire the scattered light signals generated after the particulate matter is irradiated by lasers of different wavelengths.

[0075] The signal processing unit is used to condition and convert the scattered light signal from analog to digital before inputting it into the microprocessor, which then outputs the mass concentration of particles in different particle size ranges.

[0076] In this embodiment, the multi-wavelength laser emitting unit includes a laser array and a laser control circuit; the laser control circuit is connected to a microprocessor and is used to receive power control commands and provide feedback control over the output power of each laser in the laser array to keep it stable.

[0077] In this embodiment, the signal processing unit includes a signal conditioning circuit, an analog-to-digital converter, and a microprocessor connected in sequence;

[0078] The signal conditioning circuit includes an I / V conversion circuit, a filtering circuit, and a differential amplifier circuit, which are used to convert the current signal output by the photoelectric sensor into a stable voltage signal.

[0079] The microprocessor has a pre-stored calibration database, which contains parameters relating the intensity of scattered light to the mass concentration of particles in different particle sizes.

[0080] In this embodiment, the plurality of photoelectric sensors include at least one photoelectric sensor 1 set at the backscattering angle, one photoelectric sensor 2 set at the sidescattering angle, and one photoelectric sensor 3 for detecting ambient background light.

[0081] In this embodiment, the particulate matter concentration collector uses a microcontroller as its microprocessor. The laser array (laser one, laser two, and laser three) in the multi-wavelength laser emitting unit synchronously emits lasers of different wavelengths into the flue gas to be tested. The laser power supply circuit (laser power supply circuit one, laser power supply circuit two, and laser power supply circuit three) outputs a voltage signal that is linearly related to the power of each laser in real time. After A / D conversion, the signal is input to the microprocessor. Based on this signal, the microprocessor dynamically adjusts the parameters of the laser power supply circuit through a digital potentiometer to form a closed-loop control, thereby ensuring the independence and stability of the output power of each laser.

[0082] When multi-sized particles in flue gas are irradiated by lasers of various wavelengths, a complex scattered light field is generated. The multi-angle scattered light acquisition unit is equipped with multiple photoelectric sensors at different scattering angles (at least one photoelectric sensor for backscattering, one photoelectric sensor for side scattering, and one photoelectric sensor for detecting ambient background light) to synchronously acquire these scattered light signals. The photoelectric sensors convert the scattered light signals into weak current signals, which are then processed by the signal processing unit: first, they are converted into voltage signals by an I / V conversion circuit; then, they are filtered to suppress power frequency and other noise interference; finally, a differential amplifier circuit compares the signals containing particulate matter information obtained from photoelectric sensors one and two with the pure background signal obtained from photoelectric sensor three. Differential operation effectively eliminates the influence of ambient background light and common-mode noise, obtaining the net scattered light intensity signal. The processed stable voltage signal is converted into a digital signal by an analog-to-digital converter and input to the microprocessor. The microprocessor calls its internally stored calibration database, which stores precise mapping parameters between multi-wavelength, multi-angle scattered light intensity combinations and the mass concentration of particulate matter in different particle size ranges. By running the built-in inversion algorithm, the microprocessor comprehensively calculates the received multi-channel net scattered light intensity digital signals, and finally outputs the mass concentration of different particle size ranges such as ultrafine particles, fine particles, medium particles, and coarse particles. The data is then uploaded to the host computer via the communication module (USART), thereby realizing the accurate measurement of the particle size classification concentration of particulate matter in sintering flue gas. The circuit principle of the particulate matter concentration acquisition device can be found in [reference needed]. Figure 2 .

[0083] The beneficial effects of the above technical solution are as follows: Through the multi-wavelength laser emission unit, multi-angle scattered light acquisition unit, and signal processing unit in the particulate matter concentration collector, accurate measurement of the mass concentration of particulate matter in different particle size ranges is achieved. The multi-wavelength laser emission unit emits at least two different wavelengths of laser light, which, combined with photoelectric sensors with different scattering angles arranged in the multi-angle scattered light acquisition unit, synchronously acquires the scattered light signals, effectively capturing the scattering characteristics of particles of different sizes. The signal processing unit conditions, performs analog-to-digital conversion, and microprocessor processing on the scattered light signals, using a pre-stored calibration database to retrieve the mass concentration of different particle size ranges. This design eliminates ambient background light interference, improves measurement accuracy and stability, and is particularly suitable for particle size classification concentration detection under complex compositions of sintering flue gas, providing a reliable data foundation for subsequent process optimization and dust control.

[0084] Example 4: Based on Example 1 or Example 2, the sintering process optimization module includes:

[0085] The curve construction submodule is used to construct time curves of particulate matter concentration in different particle size segments of flue gas in the large flue based on the mass concentration of particulate matter in different particle size segments of flue gas.

[0086] The feature matrix construction submodule is used to extract the multidimensional feature values ​​corresponding to each detection cycle of the particulate matter concentration time curve of different particle size segments in the large flue, and generate the feature vector corresponding to each detection cycle of the particulate matter concentration time curve of different particle size segments. Based on the feature vector corresponding to each detection cycle of the particulate matter concentration time curve of different particle size segments, a process analysis comparison matrix of different particle size segments is constructed.

[0087] The sintering process comparison submodule is used to perform multi-feature fusion and weighted analysis on the process analysis comparison matrix of each particle size range, generate a comprehensive process feature matrix, and compare the comprehensive process feature matrix with the pre-stored benchmark process matrix in multiple dimensions. Based on the comparison results, the current sintering process status level is identified and the main sources of anomalies are located.

[0088] The process optimization and control submodule generates targeted sintering process control strategies based on the current sintering process status level and the identified main sources of anomalies through a multi-level rule base.

[0089] In this embodiment, the particle size range includes at least an ultrafine particle range (e.g., particle size of 0.1-1 μm), a fine particle range (e.g., particle size of 1-2.5 μm), a medium particle range (e.g., particle size of 2.5-10 μm), and a coarse particle range (e.g., particle size >10 μm).

[0090] In this embodiment, the horizontal axis of the particulate matter concentration-time curve for different particle size segments in the main flue is time, and the vertical axis is the mass concentration of the corresponding particle size segment.

[0091] In this embodiment, the time curves of particulate matter concentration in different particle size segments of the large flue are constructed based on continuous time series data. The horizontal axis of the curve includes several detection cycles. Assuming that the horizontal axis has a sampling time interval of 1 minute and a detection cycle of 5 minutes, then each detection cycle has 5 evenly spaced sampling points.

[0092] In this embodiment, the multidimensional feature values ​​include the mean concentration (the average concentration of the particle size segment in each detection period), the standard deviation of the concentration (the standard deviation of the concentration of the particle size segment in each detection period), the linear regression slope (the slope is taken by fitting the concentration trend of the particle size segment in each detection period using linear regression), the Fourier transform dominant frequency (the frequency with the largest amplitude is taken by performing a Fourier transform on the concentration sequence of each detection period), and the particle proportion of the corresponding particle size segment (the ratio of the average concentration of the particle size segment to the average concentration of total particulate matter in each detection period). The mean concentration, the standard deviation of the concentration, the linear regression slope, the Fourier transform dominant frequency, and the particle proportion of the corresponding particle size segment represent the concentration level dimension, the fluctuation dimension, the trend dimension, the frequency domain dimension, and the distribution feature dimension, respectively.

[0093] In this embodiment, the feature vector corresponding to each detection cycle of the particulate matter concentration time curve for different particle size ranges is obtained by sorting all feature values ​​for each detection cycle according to the fixed sorting principle of feature values. Assuming the sorting principle is the mean concentration, standard deviation of concentration, linear regression slope, Fourier transform main frequency, and the proportion of particles in the corresponding particle size range, after obtaining the sequence of mean concentration, standard deviation of concentration, linear regression slope, Fourier transform main frequency, and proportion of particles in the corresponding particle size range for each detection cycle, this feature value sequence is used as the feature vector of each detection cycle of the particulate matter concentration time curve for different particle size ranges.

[0094] In this embodiment, the process analysis comparison matrix for different particle size ranges is as follows: when the feature vector is a row vector or a column vector, the feature vectors corresponding to each detection cycle of different particle size ranges are sorted by row or column based on the time sequence, and the array obtained by sorting is used as the matrix element while retaining the relative positional relationship between each element to generate the matrix.

[0095] In this embodiment, the comprehensive process feature matrix is ​​compared with the pre-stored benchmark process matrix in multiple dimensions. The comparison dimensions include element similarity dimension, feature similarity dimension and structural similarity dimension.

[0096] In this embodiment, the multi-level rule base includes:

[0097] When the sintering process status level is "Caution":

[0098] If the main source of anomalies is ultrafine particles, the rule base generation strategy is to slightly increase the voltage of the electrostatic precipitator to enhance the capture efficiency of ultrafine particles.

[0099] If the main source of anomalies is fine particulate matter, the rule base generation strategy is: check the spray pressure of the wet scrubber and appropriately increase the spray volume;

[0100] If the main source of anomalies is the medium particle segment, then the rule base generation strategy is to adjust the airflow parameters of the sintering machine to reduce particulate matter carrying.

[0101] If the main source of anomalies is the coarse-grained segment, the rule base generation strategy is to check the sealing of the main flue and clean up the accumulated ash.

[0102] When the sintering process status level is "abnormal":

[0103] If the main source of anomalies is ultrafine particles, the rule base generation strategy is to simultaneously optimize both electrostatic precipitators and wet scrubbers, increasing the voltage of the electrostatic precipitator and the amount of chemical additives in the wet scrubber.

[0104] If the main source of anomalies is fine particulate matter, the rule base generation strategy is to adjust the operating parameters of the wet desulfurization equipment, such as increasing the desulfurizing agent flow rate, and checking the ammonia injection volume of the SCR denitrification tower.

[0105] If the main source of anomalies is the medium particle size, the rule base generation strategy is to reduce the sintering machine trolley speed and increase the moisture content of the sintering mixture.

[0106] If the main source of the anomaly is the coarse-grained segment, the rule base generation strategy is: stop the machine to check the sintering machine grate bars and clean the main flue.

[0107] When the sintering process condition level is "severely abnormal":

[0108] Regardless of the particle size range from which the main source of the anomaly is, the rule base generation strategy is to immediately trigger the alarm system and recommend a complete shutdown of all equipment for a comprehensive inspection (such as sintering machines, electrostatic precipitators, wet scrubbers, etc.), while simultaneously notifying maintenance personnel to intervene.

[0109] The beneficial effects of the above technical solution are as follows: Through the curve construction submodule, feature matrix construction submodule, sintering process comparison submodule, and process optimization control submodule in the sintering process optimization module, refined process state analysis and optimization based on particulate matter concentration data of different particle size ranges are achieved. The curve construction submodule constructs time curves of particulate matter concentration for different particle size ranges, providing a foundation for time-series analysis; the feature matrix construction submodule extracts multi-dimensional feature values ​​for each detection cycle and constructs a process analysis comparison matrix, capturing multi-dimensional features of the concentration data (such as concentration level, fluctuation, trend, frequency domain, and distribution); the sintering process comparison submodule generates a comprehensive process feature matrix through multi-feature fusion and weighted analysis, and compares it with the benchmark process matrix in multiple dimensions to accurately identify the current sintering process state level and locate the main sources of anomalies; the process optimization control submodule generates targeted sintering process control strategies based on the current sintering process state level and the main sources of anomalies through a multi-level rule base. These submodules work together to solve the problem that existing technologies cannot accurately locate the sources of anomalies based on particle size range data, realizing intelligent optimization and control of the sintering process from the source, improving production efficiency and environmental protection levels.

[0110] Example 5: Based on Example 4, the sintering process comparison sub-module includes:

[0111] The feature extraction and encoding unit is used to extract four types of features from the process analysis comparison matrix of each particle size segment: statistical features, structural features, dynamic features, and distribution features. The four types of features corresponding to each particle size segment are then concatenated into a feature vector corresponding to each particle size segment.

[0112] The attention weight learning unit is used to input the feature vector corresponding to each particle size segment into the trained attention network, and output the attention weight values ​​of ultrafine particle size segment, fine particle size segment, medium particle size segment and coarse particle size segment through the trained attention network.

[0113] The dynamic matrix fusion unit is used to standardize the process analysis comparison matrix of each particle size segment, and to perform weighted summation on the standardized process analysis comparison matrix of each particle size segment based on the attention weight values ​​of ultrafine particles, fine particles, medium particles, and coarse particles to generate a comprehensive process feature matrix.

[0114] The multi-granularity similarity calculation unit is used to calculate the element-level similarity, feature-level similarity, and structural-level similarity between the integrated process feature matrix and the pre-stored benchmark process matrix, and to calculate the overall similarity based on the element-level similarity, feature-level similarity, and structural-level similarity.

[0115] The process status identification unit is used to determine the current sintering process status level based on the overall similarity. The sintering process status level includes five levels: excellent, good, attention, abnormal, and serious abnormal. If the current sintering process status level is attention, abnormal, or serious abnormal, the particle size segment with the largest attention weight value in each particle size segment is located as the main source of abnormality.

[0116] In this embodiment, the statistical features include: calculating the mean (representing the overall level), standard deviation (representing the magnitude of fluctuation), skewness (representing the asymmetry of data distribution), and kurtosis (representing the steepness of data distribution) of all elements in the process analysis comparison matrix for each particle size range.

[0117] Structural features include: the rank (representing the effective information dimension), condition number (representing the stability of the matrix), and eigenvalue distribution entropy (representing the degree of disorder of eigenvalues) of the process analysis comparison matrix for each particle size range.

[0118] Dynamic features include: calculating the autocorrelation coefficient (representing continuity over time) of the row vectors and the coefficient of variation (representing relative fluctuations between different features) of the process analysis comparison matrix for each particle size range.

[0119] Distribution characteristics include: histogram distribution of matrix elements of the process analysis comparison matrix for each particle size range (indicating which interval the values ​​are concentrated in) and quantile characteristics (such as median and quartiles).

[0120] The above four types of features are concatenated into a one-dimensional array in a fixed order of [statistical features, structural features, dynamic features, distribution features], which is the feature vector corresponding to each particle size segment.

[0121] In this embodiment, the trained attention network is a deep learning-based feature importance evaluation model. Its core function is to adaptively learn the relative importance weights of particles of different sizes in process status evaluation. The attention network takes the feature vectors corresponding to each particle size segment output by the front-end feature extraction and encoding unit as input, and outputs the attention weight values ​​corresponding to the four particle size segments through complex internal nonlinear transformations.

[0122] The attention network is trained in a supervised manner using a large amount of historical normal sintering process data. The specific training mechanism is as follows:

[0123] Training data construction:

[0124] Training samples: Feature vectors corresponding to each particle size segment are extracted from historical normal process data. Each training sample contains feature vectors corresponding to the four particle size segments: ultrafine, fine, medium, and coarse.

[0125] Output labels: Based on process expert knowledge or automatic labeling mechanisms, an ideal importance weight distribution is assigned to each training sample to ensure that each particle size segment receives reasonable attention under excellent process conditions;

[0126] A well-trained attention network includes:

[0127] Feature splicing layer: used to sequentially splice the feature vectors corresponding to the four particle size segments—ultrafine, fine, medium, and coarse—into a single comprehensive feature vector;

[0128] Feature encoding layer: It consists of at least two cascaded fully connected layers, with ReLU activation function used for non-linear transformation between layers, which is used to perform high-order feature interaction and encoding on the comprehensive feature vector;

[0129] Attention Weight Output Layer: Composed of a fully connected layer and a Softmax activation function, it maps the output of the feature encoding layer into a four-dimensional weight vector. The four elements in the four-dimensional weight vector correspond to the ultrafine-grained attention weight value, fine-grained attention weight value, medium-grained attention weight value, and coarse-grained attention weight value, respectively, and the sum of the four weight values ​​is 1.

[0130] In this embodiment, the attention weight values ​​for ultrafine particles, fine particles, medium particles, and coarse particles reflect the relative importance of the corresponding particle size segments in the process status assessment. The higher the attention weight value, the greater the influence of the corresponding particle size segment on the process status.

[0131] In this embodiment, the process analysis comparison matrix for each particle size range is standardized, specifically as follows:

[0132] Process analysis comparison matrix for any particle size range Standardization process, after standardization for ,in, ;in, for The mean of all elements in the set. for The standard deviations of all elements in the ultrafine, fine, medium, and coarse particle sizes. and The values ​​of i in the equation are 1, 2, 3, and 4.

[0133] In this embodiment, the process analysis comparison matrix for each particle size segment after standardization is weighted and summed based on the attention weight values ​​for ultrafine particles, fine particles, medium particles, and coarse particles to generate a comprehensive process feature matrix. :

[0134] Among them, the ultrafine particle segment, fine particle segment, medium particle segment, and coarse particle segment correspond to The values ​​of i are 1, 2, 3, and 4. This represents the attention weight value corresponding to the i-th particle size segment, where , , , These correspond to attention weight values ​​for ultra-fine granular segments, fine granular segments, medium granular segments, and coarse granular segments, respectively.

[0135] In this embodiment, the pre-stored baseline process matrix is ​​a matrix constructed using the same construction method as the comprehensive process feature matrix when the sintering process is in an "excellent" state.

[0136] In this embodiment, element-level similarity represents the degree of similarity between two matrices in corresponding element values, i.e., point-to-point numerical proximity; feature-level similarity represents the similarity between two matrices in the main feature directions, capturing the similarity of global feature patterns; and structure-level similarity represents the similarity of the graph structures represented by the two matrices.

[0137] In this embodiment, the sintering process status is divided into five levels—Excellent, Good, Note, Abnormal, and Severe Abnormal—based on the numerical range of the overall similarity.

[0138] If the overall similarity If the current sintering process status is excellent, it means that the sintering process is operating in an ideal state and the particle concentration characteristics of all particle size ranges are highly consistent with the baseline state.

[0139] if If the current sintering process status is good, it means that the sintering process is operating normally, with slight fluctuations but still within a controllable range.

[0140] if If the current sintering process status level is "Caution," it indicates that the sintering process has deviated to a certain extent and needs to be monitored and adjusted in a timely manner.

[0141] if If the current sintering process status is abnormal, it indicates that the sintering process is obviously abnormal and immediate measures need to be taken to adjust it.

[0142] if If the current sintering process status is severely abnormal, it indicates that the sintering process has deviated significantly from the normal state, and there may be equipment failure or major process problems.

[0143] The beneficial effects of the above technical solution are as follows: by using the feature extraction and encoding unit, attention weight learning unit, dynamic matrix fusion unit, multi-granularity similarity calculation unit and process status identification unit in the sintering process comparison submodule, the process status evaluation process is further refined. The feature extraction and encoding unit extracts four types of features—statistical features, structural features, dynamic features, and distribution features—from the process analysis comparison matrix for each particle size segment and encodes them into feature vectors, ensuring the comprehensiveness of feature information. The attention weight learning unit adaptively learns the attention weight values ​​for each particle size segment through a trained attention network, reflecting the relative importance of different particle size segments in process status assessment and enhancing the model's interpretability and adaptability. The dynamic matrix fusion unit performs a weighted summation of the standardized process analysis comparison matrices for each particle size segment based on the attention weight values, generating a comprehensive process feature matrix and achieving effective fusion of multi-source data. The multi-granularity similarity calculation unit calculates element-level similarity, feature-level similarity, and structural-level similarity, capturing the similarity between matrices from different dimensions to ensure the comprehensiveness and accuracy of the comparison. The process status identification unit determines the current sintering process status level and locates the main sources of anomalies based on the overall similarity, providing a clear operational status assessment. These units work together to make process status assessment more scientific, reliable, and refined, effectively supporting real-time optimization and fault diagnosis of the sintering process.

[0144] Example 6: Based on Example 5, the multi-granularity similarity calculation unit includes:

[0145] The element-level similarity calculation subunit calculates the element-level similarity between the integrated process feature matrix and the pre-stored benchmark process matrix based on the element values ​​of the integrated process feature matrix and the pre-stored benchmark process matrix. ;

[0146] The feature-level similarity calculation subunit performs dimensionality reduction on the integrated process feature matrix and the pre-stored benchmark process matrix based on principal component analysis (PCA). This generates the integrated process principal component eigenvectors and the benchmark process principal component eigenvectors corresponding to the integrated process feature matrix and the pre-stored benchmark process matrix, respectively. The cosine similarity between the integrated process principal component eigenvectors and the benchmark process principal component eigenvectors is then calculated and used as the feature-level similarity between the integrated process feature matrix and the pre-stored benchmark process matrix. ;

[0147] The structural similarity calculation subunit treats both the integrated process feature matrix and the pre-stored benchmark process matrix as weighted adjacency matrices, constructs weighted graphs corresponding to the integrated process feature matrix and the pre-stored benchmark process matrix respectively, and obtains the integrated process feature weighted graph (e.g., Figure 3 (as shown) and baseline process weighted chart (as shown) Figure 4 (As shown), calculate the graph edit distance D between the integrated process feature weighted graph and the baseline process weighted graph, and calculate the structural similarity between the integrated process feature matrix and the pre-stored baseline process matrix based on the graph edit distance D. ;

[0148] The overall similarity calculation subunit calculates the overall similarity based on element-level similarity, feature-level similarity, and structural-level similarity and their corresponding weight coefficients. .

[0149] In this embodiment, "dimensionality reduction of the integrated process feature matrix and the pre-stored benchmark process matrix based on principal component analysis technology, and generation of integrated process principal component eigenvectors and benchmark process principal component eigenvectors corresponding to the integrated process feature matrix and the pre-stored benchmark process matrix respectively" is a mature existing technology. Specifically, it includes: centering the integrated process feature matrix and the pre-stored benchmark process matrix, then calculating the covariance matrix of the centered integrated process feature matrix and the pre-stored benchmark process matrix, then performing eigenvalue decomposition on the covariance matrix of the centered integrated process feature matrix and the pre-stored benchmark process matrix, solving for eigenvalues ​​and corresponding eigenvectors, and taking the eigenvector corresponding to the largest eigenvalue as the integrated process principal component eigenvector and the benchmark process principal component eigenvector.

[0150] In this embodiment, the formula for calculating the feature-level similarity between the integrated process feature matrix and the pre-stored baseline process matrix is ​​as follows:

[0151] ;in, and These are the principal component eigenvectors of the integrated process and the principal component eigenvectors of the baseline process, respectively. The dot product of the principal component eigenvectors of the integrated process and the principal component eigenvectors of the baseline process is used. and These represent the modulus of the principal component eigenvectors of the integrated process and the modulus of the principal component eigenvectors of the baseline process, respectively.

[0152] In this embodiment, the graph edit distance D between the integrated process feature weighted graph and the baseline process weighted graph is the minimum editing operation cost to convert the integrated process feature weighted graph into the baseline process weighted graph. The calculation of the graph edit distance is a mature existing technology.

[0153] The beneficial effects of the above technical solution are as follows: Through the element-level similarity calculation subunit, feature-level similarity calculation subunit, structure-level similarity calculation subunit, and overall similarity calculation subunit within the multi-granularity similarity calculation unit, multi-granularity similarity calculation between the comprehensive process feature matrix and the benchmark process matrix is ​​achieved. The element-level similarity calculation subunit calculates point-to-point similarity based on matrix element values, directly reflecting numerical closeness and ensuring consistent evaluation at the detail level. The feature-level similarity calculation subunit uses principal component analysis (PCA) to reduce dimensionality and calculate cosine similarity, capturing the similarity of global feature patterns and avoiding noise interference. The structure-level similarity calculation subunit treats the matrix as a weighted adjacency matrix, calculates graph edit distance, and derives structural similarity, reflecting the consistency of the data structure represented by the matrix and enhancing the depth of comparison. The overall similarity calculation subunit integrates similarities at each level based on weight coefficients to generate an overall similarity, providing a comprehensive evaluation basis. The necessity of these sub-units lies in their comprehensive assessment of similarity from different dimensions (elements, features, structure), overcoming the limitations of single-dimensional comparison, making process status identification more accurate and robust, and providing solid technical support for the stable operation and anomaly early warning of the sintering process.

[0154] Example 7: Based on Example 6, the element-level similarity calculation subunit calculates the element-level similarity between the integrated process feature matrix and the pre-stored benchmark process matrix based on the element values ​​of the integrated process feature matrix and the pre-stored benchmark process matrix.

[0155] ;in, The element-level similarity between the comprehensive process feature matrix and the pre-stored benchmark process matrix is ​​given by m and n, respectively, where the comprehensive process feature matrix and the pre-stored benchmark process matrix have the same number of rows and columns. This represents the value of the element in the j-th row and k-th column of the current integrated process feature matrix. The value of the element in the j-th row and k-th column of the pre-stored baseline process matrix.

[0156] The beneficial effects of the above technical solution are as follows: By providing a specific formula for calculating element-level similarity, this formula calculates the element-level similarity between the comprehensive process feature matrix and the pre-stored benchmark process matrix based on the relative differences and sums of matrix element values. This not only considers the closeness of element values ​​but also eliminates the influence of dimensions through normalization, ensuring the fairness and accuracy of the similarity calculation. This calculation method is simple and easy to implement, suitable for real-time processing, and provides a reliable element-level foundation for multi-granularity similarity calculation, further enhancing the accuracy and practicality of overall process status assessment.

[0157] Example 8: Based on Example 6 or Example 7, the structural similarity calculation subunit calculates the structural similarity between the integrated process feature matrix and the pre-stored baseline process matrix based on the graph edit distance D. The formula is:

[0158] ;in, To edit the distance in the image, This represents the theoretical upper limit of the graph editing distance.

[0159] The overall similarity calculation subunit calculates the overall similarity based on element-level similarity, feature-level similarity, and structural-level similarity, along with their corresponding weight coefficients. The formula is:

[0160] ;in, , and These are the weight coefficients corresponding to element-level similarity, feature-level similarity, and structural-level similarity, respectively.

[0161] The beneficial effects of the above technical solution are as follows: By providing specific calculation formulas for structural similarity and overall similarity, the multi-granularity similarity calculation system is improved. The structural similarity calculation formula, based on graph edit distance and its theoretical upper limit, transforms structural differences into similarity values, intuitively reflecting the degree of similarity of matrix structures. The overall similarity calculation formula integrates element-level similarity, feature-level similarity, and structural similarity through weight coefficients, achieving a weighted fusion of multi-dimensional similarities. This ensures the comprehensiveness and balance of the overall evaluation, not only improving the scientificity and operability of similarity calculation but also making the process state level classification more reasonable and reliable, providing key algorithmic support for the intelligent monitoring and optimization of sintering processes.

[0162] Example 9: Based on Example 1, the dust removal optimization module includes:

[0163] The dust removal efficiency calculation submodule is used to calculate the graded dust removal efficiency of each particle size segment under the corresponding dust collector based on the mass concentration of particulate matter in different particle size segments collected at the inlet and outlet of the electrostatic precipitator and the inlet and outlet of the wet dust collector.

[0164] The dynamic optimization submodule is used to generate adjustment strategies for electrostatic precipitators and wet dust collectors based on the graded dust removal efficiency of particles of different sizes under the corresponding dust collectors.

[0165] In this embodiment, the formula for calculating the staged dust removal efficiency of each particle size range under the electrostatic precipitator is as follows:

[0166] ;in, Let be the classification and dust removal efficiency of the i-th particle size segment in the electrostatic precipitator. and These are the mass concentrations of particulate matter in the i-th particle size segment of the flue gas at the inlet and outlet of the electrostatic precipitator, respectively.

[0167] The formula for calculating the staged dust removal efficiency of each particle size range in a wet scrubber is as follows:

[0168] ;in, Let be the staged dust removal efficiency of the i-th particle size segment in a wet scrubber. and These are the mass concentrations of particulate matter in the i-th particle size segment of the flue gas at the inlet and outlet of the wet scrubber, respectively.

[0169] In this embodiment, based on the graded dust removal efficiency of each particle size range under the corresponding dust collector, adjustment strategies for electrostatic precipitators and wet scrubbers are generated, which may specifically include:

[0170] When the grading and dust removal efficiency of ultrafine particles (i=1) or fine particles (i=2) in the electrostatic precipitator is lower than its corresponding threshold, a strategy to increase the operating voltage of the electrostatic precipitator is generated.

[0171] When the efficiency of the particle segment (i=3) in the electrostatic precipitator is lower than its corresponding threshold and its inlet concentration exceeds the preset attention concentration value, a strategy is generated to enhance the rapping intensity of the electrostatic precipitator or adjust the rapping cycle.

[0172] When the efficiency of the coarse particle segment (i=4) in the electrostatic precipitator is lower than its corresponding threshold, a strategy is generated to check whether the airflow distribution plate at the inlet of the electrostatic precipitator is blocked or damaged, and to suggest cleaning the ash accumulation in the main flue.

[0173] When the efficiency of the ultrafine particle segment (i=1) or fine particle segment (i=2) in the wet scrubber is lower than its corresponding threshold, a strategy is generated to increase the flow rate of the wet scrubber spray liquid or add a chemical agglomerator.

[0174] When the grading dust removal efficiency of the medium particle segment (i=3) in the wet scrubber is abnormally high and the inlet concentration of the medium particle segment exceeds the standard, a strategy is generated to check whether the internal components of the wet scrubber are blocked or scaled.

[0175] When the concentration of coarse particles (i=4) at the inlet of the wet scrubber exceeds a preset threshold, a coordinated control strategy is generated to reduce the inlet velocity of the wet scrubber or activate the pre-settling facility.

[0176] The beneficial effects of the above technical solution are as follows: Through the dust removal efficiency calculation submodule and the dynamic optimization submodule within the dust removal optimization module, precise optimization of electrostatic precipitators and wet scrubbers is achieved. The dust removal efficiency calculation submodule calculates the graded dust removal efficiency for each particle size range based on the inlet and outlet particulate matter mass concentration, accurately evaluating the dust collector's removal effect on particles of different sizes. The dynamic optimization submodule generates adjustment strategies based on the graded dust removal efficiency, such as increasing voltage and adjusting spray volume, to specifically optimize the dust collector's operating parameters. This design solves the problem of the disconnect between concentration data and application in existing technologies, achieving precise and coordinated control of the dust removal system, improving overall dust removal efficiency and energy efficiency ratio, while extending equipment lifespan.

[0177] Example 10: This invention provides a method for detecting particulate matter concentration in sintering flue gas from a steel plant, which is performed using any one of the particulate matter concentration detection devices for sintering flue gas from steel plants in Examples 1-9, including:

[0178] S1. By installing several particulate matter concentration collectors at the main flue of the sintering flue gas treatment system, the inlet and outlet of the electrostatic precipitator, and the inlet and outlet of the wet scrubber, the mass concentration of particulate matter of different particle sizes in the flue gas in the main flue, the inlet and outlet of the electrostatic precipitator, and the inlet and outlet of the wet scrubber is collected.

[0179] S2. Perform process status analysis on the particulate matter mass concentration data of different particle size ranges collected at the main flue, identify the current sintering process status level and locate the main sources of anomalies, and generate sintering process control strategies.

[0180] S3. Based on the mass concentration of particulate matter of different particle size ranges collected at the inlet and outlet of the electrostatic precipitator and the inlet and outlet of the wet dust collector, calculate the graded dust removal efficiency of particulate matter of each particle size range under the corresponding dust collector, and generate the corresponding dust collector adjustment strategy.

[0181] S4. Integrate and render the real-time data collected in step S1, the current sintering process status level, main sources of abnormalities and sintering process control strategies analyzed in step S2, and the graded dust removal efficiency and generated dust collector adjustment strategies calculated in step S3, and display them in the form of a graphical interface.

[0182] The beneficial effects of the above technical solution are as follows: by providing a method for detecting particulate matter concentration in sintering flue gas from steel plants, all functions of the device are realized accordingly. The method has clear steps, is highly operable, and fully covers the entire process from data acquisition, analysis, optimization to display, ensuring the systematic nature and effectiveness of particulate matter concentration detection in sintering flue gas.

[0183] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A device for detecting concentration of particulate matters in sintering flue gas of a steel mill, characterized in that, The sintering process optimization module includes: A curve construction submodule is configured to construct a concentration-time curve of the particulate matter of different particle sizes in the large flue based on the mass concentration of the particulate matter of different particle sizes in the flue gas in the large flue. A feature matrix construction submodule is configured to extract multi-dimensional feature values corresponding to each detection period of the concentration-time curve of the particulate matter of different particle sizes in the large flue, generate a feature vector corresponding to each detection period of the concentration-time curve of the particulate matter of different particle sizes, and construct a process analysis comparison matrix of different particle sizes based on the feature vector corresponding to each detection period of the concentration-time curve of the particulate matter of different particle sizes. A sintering process comparison submodule is configured to perform multi-feature fusion and weighted analysis on the process analysis comparison matrix of each particle size, generate a comprehensive process feature matrix, compare the comprehensive process feature matrix with a pre-stored benchmark process matrix in multiple dimensions, identify the current sintering process state level and locate the main abnormal source according to the comparison result, and generate a targeted sintering process control strategy based on the current sintering process state level and the located main abnormal source. A process optimization control submodule is configured to generate a targeted sintering process control strategy based on the current sintering process state level and the located main abnormal source through a multi-level rule base. The sintering flue gas treatment system includes a sintering machine, a large flue, an electrostatic precipitator, a wet desulfurization device, a wet precipitator, an SCR denitration tower, and a chimney connected in sequence. The particulate matter concentration collector includes: A multi-wavelength laser emission unit is configured to emit at least two different wavelengths of laser light to the flue gas to be measured. A multi-angle scattered light collection unit includes a plurality of photoelectric sensors arranged at different scattering angles and is configured to synchronously collect scattered light signals generated after the particulate matter is irradiated by the different wavelengths of laser light. A signal processing unit is configured to input the scattered light signals into a microprocessor after conditioning and analog-to-digital conversion, and output the mass concentration of the particulate matter of different particle sizes through the microprocessor.

2. The steel mill sintering flue gas particulate matter concentration detection device according to claim 1, characterized by, The sintering process comparison submodule includes:

3. The steel plant sintering off-gas particulate matter concentration detecting device according to claim 1 or 2, characterized by, ​ ​ ​ ​ 4. The steel mill sintering flue gas particulate matter concentration detection device according to claim 1, characterized by, ​ The feature extraction and coding unit is configured to extract four types of features, i.e., statistical features, structural features, dynamic features and distribution features, of a process analysis comparison matrix of each particle size segment, and splice the four types of features corresponding to each particle size segment into a feature vector corresponding to each particle size segment; The attention weight learning unit is configured to input the feature vector corresponding to each particle size segment into the trained attention network, and output an ultra-fine particle segment attention weight value, a fine particle segment attention weight value, a medium particle segment attention weight value and a coarse particle segment attention weight value through the trained attention network; The dynamic matrix fusion unit is configured to perform standardization processing on the process analysis comparison matrices of the particle size segments, and perform weighted summation on the standardized process analysis comparison matrices of the particle size segments based on the ultra-fine particle segment attention weight value, the fine particle segment attention weight value, the medium particle segment attention weight value and the coarse particle segment attention weight value, to generate a comprehensive process feature matrix; The multi-granularity similarity calculation unit is configured to calculate element-level similarity, feature-level similarity and structure-level similarity between the comprehensive process feature matrix and a pre-stored benchmark process matrix, and calculate an overall similarity based on the element-level similarity, the feature-level similarity and the structure-level similarity; The process state recognition unit is configured to determine a current sintering process state level based on the overall similarity, and the sintering process state level includes five levels, i.e., excellent, good, attention, abnormal and serious abnormal. If the current sintering process state level is attention, abnormal or serious abnormal, the particle size segment with the maximum attention weight value of the particle size segments is positioned as a main abnormal source.

5. The steel mill sintering flue gas particulate matter concentration detection device according to claim 4, characterized by, The multi-granularity similarity calculation unit includes: The element-level similarity calculation subunit calculates the element-level similarity between the comprehensive process feature matrix and the pre-stored benchmark process matrix based on element values of the comprehensive process feature matrix and the pre-stored benchmark process matrix ; The feature level similarity calculation subunit reduces dimensions of the comprehensive process feature matrix and the pre-stored benchmark process matrix based on a principal component analysis technique, respectively generates comprehensive process principal component feature vectors and benchmark process principal component feature vectors corresponding to the comprehensive process feature matrix and the pre-stored benchmark process matrix, calculates the cosine similarity between the comprehensive process principal component feature vectors and the benchmark process principal component feature vectors, and takes the cosine similarity as the feature level similarity between the comprehensive process feature matrix and the pre-stored benchmark process matrix ; The structure level similarity calculation subunit takes the comprehensive process feature matrix and the pre-stored benchmark process matrix as weighted adjacency matrices, respectively constructs weighted graphs corresponding to the comprehensive process feature matrix and the pre-stored benchmark process matrix, obtains a comprehensive process feature weighted graph and a benchmark process weighted graph, calculates a graph edit distance D between the comprehensive process feature weighted graph and the benchmark process weighted graph, and calculates the structure level similarity between the comprehensive process feature matrix and the pre-stored benchmark process matrix based on the graph edit distance D ; The whole similarity calculation subunit calculates the whole similarity based on the element-level similarity, the feature-level similarity and the structure-level similarity and the corresponding weight coefficients .

6. The steel mill sintering flue gas particulate matter concentration detection device according to claim 5, characterized by, The element-level similarity calculation sub-unit calculates the element-level similarity between the comprehensive process feature matrix and the pre-stored benchmark process matrix based on element values of the comprehensive process feature matrix and the pre-stored benchmark process matrix, and the formula is: ; wherein, is an element-level similarity between the comprehensive process feature matrix and the pre-stored reference process matrix, m and n are respectively the number of rows and the number of columns of the comprehensive process feature matrix, wherein the number of rows and the number of columns of the comprehensive process feature matrix and the pre-stored reference process matrix are the same, is an element value of the jth row and the kth column of the comprehensive process feature matrix, is an element value of the jth row and the kth column of the pre-stored reference process matrix.

7. The steel plant sintering off-gas particulate matter concentration detecting device according to claim 5 or 6, characterized by, The structure-level similarity calculation subunit calculates the structure-level similarity between the comprehensive process feature matrix and the pre-stored reference process matrix based on a graph edit distance D The formula is: ; wherein, is the graph edit distance, is the theoretical upper bound value of the graph edit distance; The overall similarity calculation subunit calculates the overall similarity based on the element-level similarity, the feature-level similarity, and the structure-level similarity and corresponding weight coefficients The formula is: ; wherein, , and are weight coefficients corresponding to the element-level similarity, the feature-level similarity, and the structure-level similarity, respectively.

8. The steel mill sintering flue gas particulate matter concentration detection device according to claim 1, characterized by, The dust removal optimization module includes: The dust removal efficiency calculation sub-module is configured to calculate the classification dust removal efficiency of each particle size segment under the corresponding dust remover based on the mass concentration of the particles of different particle size segments collected at the inlet and outlet of the electrostatic dust remover and the inlet and outlet of the wet dust remover. The dynamic optimization sub-module is configured to generate an adjustment strategy of the electrostatic dust remover and the wet dust remover based on the classification dust removal efficiency of each particle size segment under the corresponding dust remover.

9. A method for detecting the concentration of particulate matters in sintering flue gas of a steel mill, characterized by, The steel plant sintering flue gas particle concentration detection device of any one of claims 1-8 is executed, including: S1, collecting the mass concentration of particles of different particle size segments in the flue gas at the large flue, the inlet and outlet of the electrostatic dust remover and the inlet and outlet of the wet dust remover through a plurality of particle concentration collectors installed at the large flue, the inlet and outlet of the electrostatic dust remover and the inlet and outlet of the wet dust remover of the sintering flue gas treatment system; S2, performing process state analysis on the mass concentration data of particles of different particle size segments collected at the large flue, identifying the current sintering process state level and positioning the main abnormal source, and generating a sintering process control strategy; S3, calculating the classification dust removal efficiency of each particle size segment under the corresponding dust remover based on the mass concentration of the particles of different particle size segments collected at the inlet and outlet of the electrostatic dust remover and the inlet and outlet of the wet dust remover, and generating an adjustment strategy of the corresponding dust remover; S4, integrating, rendering and displaying the real-time data collected in step S1, the current sintering process state level, the main abnormal source and the sintering process control strategy analyzed in step S2, the classified dust removal efficiency calculated in step S3 and the dust remover adjustment strategy generated in step S3 in the form of a graphical interface.

Images