Industrial flue gas multi-pollutant detection method and system

Abstract

The invention discloses an industrial flue gas multi-pollutant detection method and system, relates to the technical field of industrial flue gas detection, and aims to solve the problem that an existing industrial flue gas multi-pollutant detection system has difficult-to-predict systematic deviation on a key pollutant real-time concentration measurement result. The method comprises the following steps: acquiring a current real-time signal output by a sensor for various pollutants in flue gas, and determining a current signal feature corresponding to the current real-time signal; comparing the current signal characteristic with a normal signal reference to identify whether there is an abnormal signal pattern caused by an unknown interferent; in response to the identified abnormal signal mode, determining an interference deviation value generated by an unknown interferent to the target pollutant measurement signal according to a deviation degree of a current signal feature relative to a normal signal reference; and correcting the measurement signal of the target pollutant according to the interference deviation value so as to output the corrected concentration of the target pollutant.

Description

A method and system for detecting multiple pollutants in industrial flue gas Technical Field

[0001] This application relates to the field of industrial flue gas detection technology, and in particular to a method and system for detecting multiple pollutants in industrial flue gas. Background Technology

[0002] In industrial production processes, combustion emissions often contain a variety of pollutants, including sulfur oxides, nitrogen oxides, and particulate matter. Traditional detection methods rely on single-index analysis, making it difficult to simultaneously assess the impact of pollutant interactions on the treatment system. Existing technologies lack the ability to monitor complex gaseous pollutant components in real time, leading to delays in subsequent control strategies and potentially causing efficiency fluctuations in resource utilization. Summary of the Invention

[0003] This application provides a method and system for detecting multiple pollutants in industrial flue gas, aiming to solve the problem that existing industrial flue gas multi-pollutant detection systems cannot identify and compensate for the direct interference of these unknown components on the sensors when long-term wear of industrial combustion equipment components leads to the appearance of unknown trace interference components in the flue gas, and when the factory production load fluctuates. This results in unpredictable systematic deviations in the real-time concentration measurement results of key pollutants.

[0004] In a first aspect, to address the aforementioned technical problems, this invention provides a method for detecting multiple pollutants in industrial flue gas. The method includes: establishing a normal signal reference to characterize the signal features of multiple flue gas pollutant sensors under conditions free of unknown interference; acquiring the current real-time signals output by the multiple pollutant sensors in the flue gas and determining the current signal features corresponding to the current real-time signals; comparing the current signal features with the normal signal reference to identify whether there are abnormal signal patterns caused by unknown interference; in response to the identified abnormal signal patterns, determining the interference deviation amount caused by the unknown interference to the target pollutant measurement signal based on the degree of deviation of the current signal features from the normal signal reference; and correcting the target pollutant measurement signal based on the interference deviation amount to output the corrected target pollutant concentration.

[0005] Secondly, this application provides an industrial flue gas multi-pollutant detection system, which includes: an establishment unit for establishing a normal signal reference to characterize the signal features of multiple flue gas pollutant sensors under conditions without unknown interference; an acquisition unit for acquiring the current real-time signals output by multiple pollutant sensors in the flue gas and determining the current signal features corresponding to the current real-time signals; an identification unit for comparing the current signal features with the normal signal reference to identify whether there is an abnormal signal pattern caused by unknown interference; a determination unit for determining, in response to the identified abnormal signal pattern, the amount of interference deviation caused by unknown interference to the target pollutant measurement signal based on the degree of deviation of the current signal features from the normal signal reference; and an output unit for correcting the target pollutant measurement signal based on the amount of interference deviation to output the corrected target pollutant concentration.

[0006] This application has at least the following beneficial effects: The industrial flue gas multi-pollutant detection method disclosed in this application, by establishing a dynamically adjusted normal signal reference and comparing the current signal characteristics with this reference in real time, can effectively identify abnormal signal patterns caused by unknown interference. Based on this, the interference deviation is determined according to the degree of deviation, and the target pollutant measurement signal is corrected, thereby outputting the corrected target pollutant concentration. This method solves the problem in the prior art where, due to long-term wear of combustion furnace components leading to the appearance of unknown trace interference components in the flue gas, and under fluctuating factory production loads, existing industrial flue gas multi-pollutant detection systems cannot identify and compensate for the direct interference of these unknown components on the sensor, resulting in unpredictable systematic deviations in the real-time concentration measurement results of key pollutants. Through this technical solution, this application can significantly improve the accuracy and reliability of industrial flue gas multi-pollutant measurement, providing an accurate basis for adjusting the control strategy of subsequent flue gas treatment devices, thereby optimizing the efficiency of resource utilization, reducing the risk of excessive emissions, and having significant implications for environmental protection. Attached Figure Description

[0007] Figure 1 is a flowchart illustrating a method for detecting multiple pollutants in industrial flue gas provided in this application. Detailed Implementation

[0008] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments. The components of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0009] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0010] Traditional industrial flue gas multi-pollutant detection systems are ill-equipped to handle situations where long-term wear and tear on combustion furnace components introduces unknown trace interference components into the flue gas, or when factory production loads fluctuate. Their calibration parameter sets are unable to identify and compensate for the direct interference of these unknown components on the sensors, leading to unpredictable and systematic biases in the real-time concentration measurements of key pollutants such as sulfur oxides and nitrogen oxides. This bias lacks accurate basis for adjusting the control strategies of subsequent flue gas treatment devices, hindering efficiency optimization in resource utilization and potentially leading to excessive emissions due to misjudgments, thus posing a potential environmental impact.

[0011] In view of the above problems, this application provides a method for detecting multiple pollutants in industrial flue gas. This method can effectively identify and compensate for the influence of unknown interfering substances on the measurement signal of the target pollutant, thereby improving the accuracy and reliability of industrial flue gas monitoring. The core of this method lies in establishing and dynamically adjusting a normal signal reference. By comparing the current real-time signal characteristics with the normal signal reference, abnormal signal patterns are identified, and then the interference deviation of unknown interfering substances is quantified, and the measurement signal of the target pollutant is corrected.

[0012] The following specific embodiments will provide a detailed introduction and explanation of the precision fertilization program control method and system for intelligent agricultural equipment provided in this application.

[0013] Referring to Figure 1, an embodiment of this application provides a method for detecting multiple pollutants in industrial flue gas. The method may include the following steps: S1, establishing a normal signal reference for characterizing the sensor signal characteristics of multiple flue gas pollutants under conditions without unknown interference.

[0014] Among them, multiple flue gas pollutant sensors can refer to sensor arrays used to detect different pollutant components in flue gas, such as sulfur dioxide (SO2) sensors, nitrogen oxide (NOx) sensors, carbon monoxide (CO) sensors, and oxygen (O2) sensors. These sensors work together to provide comprehensive information on flue gas components. Normal signal reference can refer to the typical characteristics exhibited by these sensor signals under ideal or baseline operating conditions without the influence of unknown interference, including but not limited to the signal mean, fluctuation range, and correlation between different sensor signals.

[0015] This application first requires establishing a normal signal reference to characterize the sensor signal features of various flue gas pollutants under conditions free of unknown interference. This reference forms the basis for subsequent identification of abnormal signal patterns. For example, it can be constructed by collecting a large amount of historical sensor data when the industrial combustion equipment is in stable operation and there are known to be no abnormal interferences. This data can include the output values, fluctuation ranges, and interrelationships of each sensor. The reference can be a multi-dimensional statistical model, such as a Gaussian mixture model or principal component analysis model, used to describe the distribution characteristics of sensor signals under normal operating conditions. Alternatively, the reference can be a rule-based system that defines the normal threshold range for each sensor signal and the logical relationships between different sensor signals. For example, when SO2 concentration increases, O2 concentration usually decreases; this correlation can be incorporated into the normal signal reference.

[0016] S2. Obtain the current real-time signals output by multiple pollutant sensors in the flue gas, and determine the current signal characteristics corresponding to the current real-time signals.

[0017] As one possible implementation, this application requires acquiring the current real-time signals output by multiple pollutant sensors in the flue gas and determining the current signal characteristics corresponding to the current real-time signals. For example, analog or digital signals from multiple sensors such as SO2, NOx, CO, and O2 sensors can be acquired synchronously in real time via a data acquisition module. These raw signals may contain noise, therefore preprocessing is required, such as digital filtering to remove high-frequency noise and range conversion to unify the signals to comparable units. Subsequently, feature parameters are extracted from the preprocessed signals. These feature parameters can be the stable output values ​​of each sensor signal (e.g., the average value within a certain time window) and / or the rate of change (e.g., the derivative of the signal with respect to time). These feature parameters are combined to form a multi-dimensional current signal feature vector for subsequent comparison with a normal signal reference.

[0018] S3. Compare the current signal characteristics with a normal signal reference to identify whether there are abnormal signal patterns caused by unknown interference.

[0019] Among them, unknown interference refers to substances that are not identified or considered in routine monitoring but can affect the sensor measurement signals, such as incomplete combustion products and secondary pollutants.

[0020] As one possible implementation, this application compares the current signal characteristics with a normal signal reference to identify whether there are abnormal signal patterns caused by unknown interference. For example, the current signal feature vector can be input into a pre-established statistical model to calculate the statistical distance between the current signal characteristics and the multidimensional signal distribution center defined by the normal signal reference, such as Mahalanobis distance. Mahalanobis distance can take into account the correlation between different sensor signals, thus more accurately measuring the degree of deviation. As another implementation, a rule-based comparison method can also be used. For example, if the stable output value of any sensor signal exceeds its normal range, or the correlation between different sensor signals changes significantly (e.g., SO2 increases but O2 does not decrease), then an abnormal signal pattern is determined to exist.

[0021] S4. In response to the identified abnormal signal pattern, determine the amount of interference deviation of the unknown interfering substance on the target pollutant measurement signal based on the degree of deviation of the current signal characteristics from the normal signal reference.

[0022] In some embodiments, in response to an identified abnormal signal pattern, this application determines the amount of interference deviation of the unknown interfering substance on the target pollutant measurement signal based on the degree of deviation of the current signal characteristics relative to a normal signal reference. For example, when an abnormal signal pattern is identified, a first interference intensity value can be determined based on the statistical distance between the current signal characteristics and the normal signal reference. The larger the statistical distance, the greater the deviation and the higher the interference intensity. Simultaneously, the abnormal signal pattern can be categorized into a preset interference category based on its specific manifestation (e.g., whether the SO2 signal is abnormally high or the NOx signal is abnormally low). For different interference categories and target pollutants, corresponding compensation coefficients can be queried from a preset interference deviation mapping relationship. Finally, multiplying the first interference intensity value by the compensation coefficient yields the amount of interference deviation of the unknown interfering substance on the target pollutant measurement signal.

[0023] S5. Based on the interference deviation, correct the measurement signal of the target pollutant to output the corrected target pollutant concentration.

[0024] Among them, the target pollutant refers to the main pollutant whose concentration needs to be accurately measured, such as SO2 or NOx.

[0025] In some embodiments, this application corrects the measurement signal of the target pollutant based on the interference deviation to output a corrected target pollutant concentration. For example, if the target pollutant is SO2, its original measurement signal is S_SO2, and the calculated interference deviation is D_SO2, then the corrected SO2 concentration can be expressed as C_SO2 = f(S_SO2 - D_SO2), where f is a function that converts the sensor signal into concentration. In this way, even in the presence of unknown interfering substances, a more accurate measurement result of the target pollutant concentration can be obtained.

[0026] The proposed method for detecting multiple pollutants in industrial flue gas effectively identifies abnormal signal patterns caused by unknown interfering substances by establishing and dynamically adjusting a normal signal reference. Compared to existing technologies, the core innovation of this method lies in its ability to proactively identify and quantify the impact of unknown interfering substances on the measurement signal of the target pollutant and then correct it. Traditional methods often rely on preset calibration parameter sets, which fail when faced with unknown or unexpected interfering substances, leading to systematic deviations in the measurement results. This method, by comparing the current signal characteristics with a dynamically adjusted normal signal reference in real time, can promptly detect such deviations and calculate the interference deviation based on the degree of deviation, thereby correcting the measurement signal of the target pollutant. For example, when wear of combustion furnace components leads to the presence of trace amounts of secondary pollutants in the flue gas, these secondary pollutants can cause abnormal patterns in the sensor signal. This method can identify such anomalies and calculate the interference deviation on the SO2 or NOx measurement signal based on the degree of deviation, thus correcting the measurement results. This enables the acquisition of more accurate pollutant concentration data even under complex industrial conditions, providing a reliable basis for the optimized control of subsequent flue gas treatment devices and effectively avoiding the risk of reduced resource utilization efficiency and excessive emissions due to measurement errors.

[0027] In some embodiments described above, a normal signal reference is proposed to characterize the sensor signal features of various flue gas pollutants under conditions free of unknown interference. However, in the actual operation of industrial combustion equipment, the condition of equipment components deteriorates over time, and production loads and operating conditions may change frequently. If the normal signal reference is static or cannot adapt to these dynamic changes, it may lead to a decrease in the accuracy of identifying unknown interference, such as misjudging normal operating condition fluctuations as abnormalities, or failing to identify them in a timely manner when real interference occurs.

[0028] In response, this application further proposes the following steps for establishing a normal signal reference to characterize the signal characteristics of flue gas pollutant sensors under conditions without unknown interference: when the industrial combustion equipment is in a preset healthy state and the production load is stable, first historical signal data output by the various pollutant sensors is collected; based on the first historical signal data, an initial multi-dimensional signal reference range is constructed to characterize the mean, fluctuation range, and correlation of each sensor signal under conditions without unknown interference; the initial multi-dimensional signal reference range includes the normal value range of each sensor signal; during the subsequent operation of the industrial combustion equipment, first auxiliary data characterizing the state of the combustion equipment components and second auxiliary data characterizing the operating conditions of the production process are acquired in real time; based on the first auxiliary data and the second auxiliary data, the statistical characteristics of the initial multi-dimensional signal reference range are dynamically adjusted to generate the normal signal reference adapted to the current operating conditions.

[0029] Specifically, when the industrial combustion equipment is in a preset healthy state and the production load is stable, the first historical signal data output by the various pollutant sensors is collected. Here, "preset healthy state" means that the industrial combustion equipment has undergone comprehensive maintenance and debugging, and all its performance indicators meet design requirements, with no obvious signs of faults or abnormal wear. "Stable production load" means that the combustion equipment operates continuously under a fixed or narrow range of load to avoid sensor signal fluctuations caused by drastic load changes. The first historical signal data is baseline data collected under these ideal conditions, and its purpose is to provide a clean and reliable baseline for subsequently constructing the initial multi-dimensional signal reference range.

[0030] Furthermore, based on the first historical signal data, an initial multi-dimensional signal reference range is constructed to characterize the mean, fluctuation range, and correlation of each sensor signal under conditions free from unknown interference. This initial multi-dimensional signal reference range can be understood as a baseline model built upon historical data, which describes the central tendency (mean), dispersion (fluctuation range, such as standard deviation or variance), and interrelationships (correlation, such as covariance or correlation coefficient) of each sensor signal under interference-free conditions. The initial multi-dimensional signal reference range includes the normal numerical range of each sensor signal, i.e., the expected value range of each sensor signal under interference-free conditions.

[0031] During the subsequent operation of industrial combustion equipment, first auxiliary data characterizing the status of combustion equipment components and second auxiliary data characterizing the operating conditions of the production process are acquired in real time. The first auxiliary data may include, but is not limited to, equipment operating time, wear degree of key components, maintenance records, vibration data, bearing temperature, etc., which reflect the long-term changing trend of the physical state of the combustion equipment. The second auxiliary data may include, but is not limited to, fuel type, feed rate, air volume, furnace temperature, pressure, flue gas velocity, etc., which reflect the real-time operating conditions of the production process. The acquisition of this auxiliary data aims to provide comprehensive input information for dynamically adjusting normal signal references.

[0032] Therefore, based on the first auxiliary data and the second auxiliary data, the statistical characteristics of the initial multi-dimensional signal reference range are dynamically adjusted to generate a normal signal reference adapted to the current operating conditions. This dynamic adjustment refers to correcting the statistical parameters (such as mean, variance, covariance, etc.) of the initial multi-dimensional signal reference range based on real-time acquired auxiliary data, enabling it to match the "normal" signal performance under the current equipment status and operating conditions in real time. The purpose is to ensure that the normal signal reference accurately reflects the true distribution of sensor signals under the current operating conditions without unknown interference, thereby improving the accuracy of abnormal signal pattern recognition.

[0033] This application's solution effectively addresses the limitations of traditional static references under complex and variable operating conditions in industrial combustion equipment by introducing a dynamic adjustment mechanism for the normal signal reference. Specifically, firstly, when the equipment is in a preset healthy state and the production load is stable, first historical signal data is collected. This lays the foundation for constructing a reliable initial multi-dimensional signal reference range, ensuring the accuracy of the baseline reference. Subsequently, during the subsequent operation of the equipment, by acquiring first auxiliary data characterizing the state of combustion equipment components and second auxiliary data characterizing the operating conditions of the production process in real time, internal and external factors affecting the flue gas pollutant sensor signals can be comprehensively captured. Based on these auxiliary data, the statistical characteristics of the initial multi-dimensional signal reference range are dynamically adjusted, enabling the normal signal reference to adapt in real time to operating condition drifts such as equipment aging and load changes. This ensures that at any given moment, the normal signal reference accurately reflects the true distribution of sensor signals under the current operating conditions without unknown interference. Therefore, when the actual signal deviates from this dynamically adjusted reference, abnormal signal patterns caused by unknown interference can be more accurately identified, avoiding false alarms or missed alarms due to changes in operating conditions.

[0034] Through the above technical solution, this application can significantly improve the accuracy and robustness of the multi-pollutant detection method for industrial flue gas under complex and variable operating conditions. By dynamically adjusting the normal signal reference, the method can effectively distinguish between signal drift caused by changes in the equipment's own state or fluctuations in production operating conditions and real anomalies caused by unknown interference, thereby greatly reducing the false alarm rate. Furthermore, this adaptive reference mechanism enables the system to operate stably for a long period without frequent manual calibration or model retraining, reducing maintenance costs and ensuring that the corrected target pollutant concentration output is more accurate and reliable, providing a more solid data foundation for the optimized control of industrial production processes and environmental emission monitoring.

[0035] In some preferred embodiments, a specific example is given below. Assume this method is applied to an industrial combustion system in a coal-fired power plant. First, when the power plant unit has completed a major overhaul and is operating at a stable load (e.g., 500MW), first historical signal data from sulfur dioxide, nitrogen oxides, carbon monoxide, and oxygen content sensors are collected under conditions free of known interference. Based on this data, an initial multi-dimensional signal reference range is constructed, defining the normal mean, fluctuation range, and interrelationships of the sensor signals under ideal health conditions and stable load.

[0036] Subsequently, during the daily operation of the power plant, primary auxiliary data is acquired in real time, such as the wear degree of boiler heating surfaces, fan vibration data, and burner efficiency decline trends. These data characterize the physical wear trends of key components of the combustion equipment. Simultaneously, secondary auxiliary data is acquired in real time, such as the current unit's power generation load, coal composition (e.g., sulfur content, ash content), and the ratio of primary to secondary air volume. This data characterizes the operating conditions of the production process.

[0037] Using a pre-defined nonlinear mapping model (such as a neural network model trained on historical operating data), the real-time acquired first and second auxiliary data are used as inputs to calculate the adjustment amounts for the central mean vector and covariance matrix of the initial multidimensional signal reference range. For example, when the boiler load increases from 500MW to 600MW, the model predicts that the normal baseline mean values ​​of nitrogen oxides and sulfur dioxide should increase slightly, and their fluctuation range may also expand slightly. When the sulfur content of the coal changes, the normal baseline mean value of sulfur dioxide will be adjusted accordingly. Applying these adjustments to the initial multidimensional signal reference range generates a normal signal reference adapted to the current power generation load, coal type, and equipment aging status. When subsequent real-time sensor signals are compared with this dynamically adjusted normal signal reference, abnormal signal patterns caused by unknown interferences such as desulfurizer leakage or the introduction of new fuel additives can be identified more accurately, rather than misjudging normal operating condition changes as interference.

[0038] In some embodiments described above, this application proposes dynamically adjusting the statistical characteristics of the initial multi-dimensional signal reference range based on first and second auxiliary data to generate a normal signal reference adapted to the current operating conditions. However, in actual industrial operation, the component states and production operating conditions of the combustion equipment may affect the sensor's signal characteristics in a complex and non-linear manner. Simple statistical adjustments alone may not accurately capture these complex dynamic changes, resulting in insufficient adaptability of the normal signal reference and affecting the accuracy of abnormal signal pattern recognition. Therefore, this application further proposes a specific method for dynamically adjusting the statistical characteristics of the aforementioned initial multi-dimensional signal reference range to more accurately adapt to the operating states of industrial combustion equipment under different conditions.

[0039] In response, this application further proposes the above-mentioned dynamic adjustment of the statistical characteristics of the initial multidimensional signal reference range based on the first auxiliary data and the second auxiliary data, including: extracting component state feature vectors from the first auxiliary data to characterize the physical wear trend of key components of the combustion equipment; extracting operation mode feature vectors from the second auxiliary data to characterize production load and combustion control parameters; calculating the adjustment amount of the central mean vector and covariance matrix of the initial multidimensional signal reference range based on the component state feature vectors and operation mode feature vectors through a preset nonlinear mapping model; and applying the adjustment amount to the initial multidimensional signal reference range to complete the dynamic adjustment of the statistical characteristics of the initial multidimensional signal reference range.

[0040] Specifically, the first auxiliary data can be understood as a dataset reflecting the internal physical condition and health status of industrial combustion equipment. The component state feature vector extracted from the first auxiliary data refers to a set of quantitative indicators extracted from the original auxiliary data using data analysis methods (e.g., principal component analysis, feature engineering), which effectively characterize the physical loss trends such as wear, corrosion, and fatigue of key components of the combustion equipment (e.g., boiler heating surfaces, fans, pumps, valves). For example, this vector may include parameters such as vibration frequency changes, abnormal temperature trends, material thickness attenuation rates, and cumulative operating time of key components. Its purpose is to transform complex equipment physical state information into structured data that can be used as model input.

[0041] The second auxiliary data can be understood as a dataset describing the external operating conditions and control strategies of industrial combustion equipment. The operating mode feature vector extracted from the second auxiliary data refers to a set of quantitative indicators extracted from the original auxiliary data using similar data analysis methods. These indicators effectively characterize the current production load (such as power generation, steam output, product output, etc.) and combustion control parameters (such as fuel type and flow rate, primary air volume, secondary air volume, oxygen content setpoint, furnace pressure, etc.). For example, this vector may include parameters such as equipment load rate, fuel calorific value, air-fuel ratio, and burner tilt angle. Its purpose is to transform variable production operating condition information into structured data that can be used as model input.

[0042] In practical applications, the pre-defined nonlinear mapping model can be a model built based on machine learning or deep learning algorithms, such as a Multilayer Perceptron (MLP), Recurrent Neural Network (RNN), Long Short-Term Memory (LSTM), Support Vector Regression (SVR), or Gaussian Process Regression (GPR) model. This model is trained to learn the complex nonlinear relationship between the component state feature vector and operating mode feature vector and the adjustment amounts of the central mean vector and covariance matrix of the initial multidimensional signal reference range. The adjustment amount of the central mean vector is used to correct the expected average value of each sensor signal to reflect baseline drift caused by equipment aging or changes in operating conditions; the adjustment amount of the covariance matrix is ​​used to correct the correlation between each sensor signal and their respective fluctuation range to reflect changes in signal noise and mutual influence under different operating conditions. The aim is to achieve refined and adaptive adjustment of the normal signal reference by capturing these nonlinear relationships.

[0043] Therefore, the calculated adjustment is applied to the initial multidimensional signal reference range. Specifically, the adjustment of the central mean vector is superimposed on the central mean of the initial multidimensional signal reference range, and the adjustment of the covariance matrix is ​​applied to the covariance matrix of the initial multidimensional signal reference range. In this way, the statistical characteristics of the initial multidimensional signal reference range (including mean, variance, covariance, etc.) are dynamically updated, thereby generating a normal signal reference that accurately reflects the current physical state and operating mode of the equipment.

[0044] This application's solution quantifies and structures the internal physical losses and external operating conditions affecting flue gas pollutant sensor signals by introducing component state feature vectors and operating mode feature vectors. It is precisely because these feature vectors can comprehensively and concisely characterize the actual operating state of the equipment that subsequent dynamic adjustments have a solid data foundation. Based on this, a pre-defined nonlinear mapping model effectively captures the complex, nonlinear intrinsic relationship between these feature vectors and the statistical characteristics of the sensor signals. Traditional linear models may struggle to accurately describe the subtle influences of factors such as equipment aging and load fluctuations on sensor responses. However, through nonlinear models, it is possible to learn and predict how precisely the mean and fluctuation range of each pollutant sensor signal should deviate from their initial healthy reference values ​​under specific component wear levels and operating modes. Therefore, by calculating and applying the adjustment amounts of the central mean vector and covariance matrix, the statistical characteristics of the initial multi-dimensional signal reference range can be dynamically adjusted in a refined manner, ensuring that the generated normal signal reference is highly adaptable to the current operating conditions and accurately reflects the true signal distribution when there are no unknown interferences.

[0045] Through the above technical solution, this application overcomes the problem of insufficient adaptability of normal signal references in complex industrial environments using traditional methods. Specifically, by extracting features from the state and operating mode of equipment components and dynamically adjusting them using a nonlinear mapping model, the generated normal signal reference can more accurately reflect the actual operating state of industrial combustion equipment under different aging levels and variable operating conditions. This significantly improves the accuracy and robustness of the normal signal reference, enabling the subsequent abnormal signal pattern recognition process to more effectively identify anomalies caused by unknown interferences, reducing false alarms and missed alarms caused by changes in operating conditions or equipment aging, and thus improving the overall reliability and accuracy of multi-pollutant detection in industrial flue gas.

[0046] In some preferred embodiments, a specific example is given below. Suppose that in a multi-pollutant detection scenario of industrial flue gas from a coal-fired power plant, it is necessary to monitor the signals output by a sulfur dioxide sensor, a nitrogen oxide sensor, a carbon monoxide sensor, and an oxygen content sensor.

[0047] First, component state feature vectors are extracted from the primary auxiliary data, such as boiler heating surface tube wall thickness data, induced draft fan bearing vibration data, and desulfurization tower slurry pH sensor wear data. These vectors may include tube wall thinning rate, root mean square value of vibration acceleration, and pH sensor response time drift.

[0048] Secondly, an operating mode feature vector is extracted from the second auxiliary data, such as boiler load rate, pulverized coal feed rate, primary air volume, secondary air volume, furnace temperature, and flue gas velocity. This vector may include the current load percentage, air-fuel ratio, burner tilt angle setpoint, etc.

[0049] Next, a pre-trained deep neural network is used as the nonlinear mapping model. The input layer of this neural network receives the aforementioned component state feature vectors and operating mode feature vectors, while the output layer outputs the adjustment amount of the center mean vector and covariance matrix of the initial multi-dimensional signal reference range. For example, when the boiler load increases from 50% to 80%, the model predicts that the mean values ​​of the sulfur dioxide and nitrogen oxide sensor signals will increase slightly, and their fluctuation range may expand slightly, and outputs the corresponding adjustment amount.

[0050] Finally, these adjustments are applied to the initial multidimensional signal reference range. For example, if the initial normal mean of the sulfur dioxide sensor signal is 500 ppm, and the model predicts an adjustment of +20 ppm, the corrected normal mean becomes 520 ppm. Similarly, the covariance matrix is ​​updated based on the model output to reflect a more accurate statistical correlation between different sensor signals under the current operating conditions. In this way, the normal signal reference is dynamically adjusted to accurately adapt to the current operating load and equipment health of the power plant, thus providing a highly accurate benchmark for subsequent identification of unknown interference.

[0051] Specifically, the steps described above for acquiring the current real-time signals output by multiple pollutant sensors in flue gas and determining the current signal characteristics corresponding to the current real-time signals can be further refined into the following operations: acquiring the current real-time signals output by multiple pollutant sensors in flue gas and determining the current signal characteristics corresponding to the current real-time signals includes: simultaneously acquiring first real-time measurement signals output by sulfur dioxide sensors, nitrogen oxide sensors, carbon monoxide sensors, and oxygen content sensors in flue gas; performing digital filtering and range conversion on the first real-time measurement signals to obtain a preprocessed second real-time measurement signal; extracting feature parameters from the second real-time measurement signal for pattern comparison with the normal signal reference, the feature parameters including the stable output value and / or rate of change of each sensor signal; and combining the feature parameters to obtain the current signal characteristics.

[0052] The simultaneous acquisition of the first real-time measurement signals output by sulfur dioxide, nitrogen oxides, carbon monoxide, and oxygen content sensors in flue gas refers to obtaining raw, unprocessed measurement data from multiple different flue gas pollutant sensors at the same time point or within a very short time window. Specifically, sulfur dioxide, nitrogen oxides, carbon monoxide, and oxygen content sensors are commonly used in industrial flue gas monitoring, measuring the concentrations of sulfur dioxide, nitrogen oxides, carbon monoxide, and oxygen content in flue gas, respectively. The purpose of simultaneous acquisition is to ensure temporal consistency of all sensor data, thereby accurately reflecting the overall state of the flue gas at a given moment and providing a reliable time reference for subsequent signal processing and pattern comparison.

[0053] Furthermore, the first real-time measurement signal is digitally filtered and range-converted to obtain a preprocessed second real-time measurement signal. Digital filtering aims to remove random noise, high-frequency interference, or transient spikes present in the first real-time measurement signal, thereby improving signal smoothness and accuracy. For example, techniques such as moving average filtering, Kalman filtering, low-pass filtering, or median filtering can be used. Range conversion transforms the filtered signal from the sensor's original output form (e.g., voltage, current, or digital count) into units with actual physical meaning (e.g., ppm, mg / Nm³, or percentage), unifying the dimensions of different sensor signals for easier subsequent comparison and analysis. Thus, the preprocessed second real-time measurement signal has a higher signal-to-noise ratio and unified physical dimensions, making it more suitable for feature extraction.

[0054] Based on this, feature parameters are extracted from the second real-time measurement signal for pattern comparison with the normal signal reference. These feature parameters include the stable output value and / or rate of change of each sensor signal. The extraction of feature parameters aims to extract key information that effectively characterizes the flue gas state from the complex raw signal. The stable output value can be understood as the average, median, or steady-state value of the sensor signal within a certain time window, reflecting the basic level of pollutant concentration. The rate of change can be understood as the first derivative or slope of the sensor signal within a specific time period, reflecting the dynamic trend of pollutant concentration changes. In practical applications, depending on the specific application scenario and interference characteristics, one can choose to extract the stable output value alone, extract the rate of change alone, or combine both, to more comprehensively capture the characteristics of the flue gas signal.

[0055] Finally, the feature parameters are combined to obtain the current signal feature. This combination process integrates multiple extracted feature parameters into a multi-dimensional vector or matrix, forming a comprehensive data representation used to fully characterize the real-time state of various pollutants in the flue gas. For example, the stable output values ​​and change rates of each sensor can be concatenated into a feature vector. This current signal feature will serve as input for subsequent comparison with a normal signal reference to identify whether there are abnormal signal patterns caused by unknown interference.

[0056] The proposed solution ensures that the acquired signal characteristics accurately and comprehensively reflect the real-time status of various pollutants in flue gas by synchronously acquiring, preprocessing, and extracting and combining feature parameters from the raw sensor signals. Synchronous acquisition guarantees the time alignment of data from different sensors, providing a reliable basis for subsequent pattern comparisons. Digital filtering and range conversion effectively remove signal noise, unify the data format, and improve signal usability and accuracy. By extracting feature parameters such as stable output values ​​and / or rates of change, key information can be extracted from the complex raw signal. This information is crucial for characterizing the static concentration levels and dynamic trends of pollutants in flue gas, thus providing accurate data input for subsequent identification of abnormal signal patterns caused by unknown interfering substances.

[0057] The above technical solution ensures the quality and accuracy of the acquired real-time signal and allows for the extraction of representative feature parameters. This not only improves the sensitivity and reliability of subsequent abnormal signal pattern recognition but also effectively reduces the dimensionality of the original data and simplifies the complexity of pattern comparison through signal preprocessing and feature extraction, thereby enhancing the real-time performance and computational efficiency of the entire detection method. Furthermore, by comprehensively considering stable output values ​​and rates of change, the characterization of the flue gas state becomes more comprehensive, enabling better capture of subtle or dynamic abnormal changes caused by unknown interfering substances, thus laying a solid foundation for accurately determining the amount of interference deviation.

[0058] Specifically, the step of comparing the current signal characteristics with a normal signal reference to identify whether an abnormal signal pattern caused by an unknown interference exists can be implemented as follows: The step of comparing the current signal characteristics with the normal signal reference to identify whether an abnormal signal pattern caused by an unknown interference exists includes: calculating the statistical distance between the current signal characteristics and the multidimensional signal distribution center defined by the normal signal reference; comparing the statistical distance with a preset dynamic distance threshold; if the statistical distance is greater than the preset dynamic distance threshold, then it is determined that the abnormal signal pattern exists.

[0059] Calculating the statistical distance between the current signal characteristics and the multidimensional signal distribution center defined by the normal signal reference refers to quantifying the difference between the characteristics of the current real-time signal and the expected behavior represented by the normal signal reference using mathematical methods. Specifically, the statistical distance can be calculated using Mahalanobis distance, Euclidean distance, or other suitable statistical metrics. The aim is to comprehensively consider the overall distribution and interrelationships of multidimensional sensor signals, rather than focusing solely on the deviation of a single signal, thereby more accurately reflecting the degree of anomaly in the signal pattern under current operating conditions. The multidimensional signal distribution center defined by the normal signal reference typically refers to the mean vector or central tendency of each sensor signal characteristic under conditions without unknown interference, and may include its covariance matrix to comprehensively describe the statistical characteristics of the normal signal.

[0060] Furthermore, the statistical distance is compared with a preset dynamic distance threshold to determine whether the current signal deviation exceeds the normal fluctuation range. The preset dynamic distance threshold is not fixed but can be adjusted in real time based on factors such as the operating conditions of the industrial combustion equipment, production load, and equipment aging. For example, the threshold can be appropriately relaxed during equipment startup or drastic load changes; while during stable operation, the threshold can be tightened to improve detection sensitivity. This avoids false alarms caused by fluctuations in normal operating conditions and improves the accuracy of anomaly identification. If the statistical distance is greater than the preset dynamic distance threshold, an abnormal signal pattern is determined to exist, meaning that the current sensor signal combination pattern differs significantly from the normal signal reference and is highly likely to be affected by unknown interference.

[0061] The proposed solution quantifies the deviation between the current flue gas pollutant sensor signal and the normal signal under conditions free of unknown interference by calculating the statistical distance between the current signal characteristics and the multidimensional signal distribution center defined by a normal signal reference. The introduction of statistical distance enables a comprehensive assessment of the overall deviation of the multidimensional signal characteristics, rather than relying solely on anomalies in a single sensor. Furthermore, comparing this statistical distance with a preset dynamic distance threshold effectively determines whether the deviation of the current signal has reached a level sufficient to be considered abnormal. The setting of the dynamic distance threshold allows the system to adapt to normal fluctuations in industrial combustion equipment under different operating conditions, avoiding false alarms and ensuring the accuracy and robustness of abnormal signal pattern recognition.

[0062] The above technical solution enables accurate identification of abnormal signal patterns caused by unknown interfering substances in industrial flue gas. Using statistical distance comparison allows for comprehensive consideration of the correlation between signals from multiple pollutant sensors, improving the sensitivity of anomaly detection. Simultaneously, the introduction of a dynamic distance threshold allows the system to adaptively adjust detection standards according to changes in actual operating conditions, effectively reducing false alarm and false negative rates. This enhances the reliability and accuracy of the entire multi-pollutant detection method, providing a solid foundation for subsequent determination of interference deviation and measurement signal correction.

[0063] In some embodiments described above, the presence of abnormal signal patterns caused by unknown interference is identified by comparing the current signal characteristics with a normal signal reference. However, in some cases, unknown interference may cause small or gradual changes in the signal, which may be misjudged as normal operating condition fluctuations, or the abnormal patterns caused by them may not be significant enough to be effectively identified, thus affecting the sensitivity and accuracy of interference identification.

[0064] In response, this application further proposes a method that includes the following steps before comparing the current signal characteristics with the normal signal reference: periodically applying controlled micro-operational perturbations to the combustion process during the operation of the industrial combustion equipment; acquiring perturbation response signals output by the multiple pollutant sensors during the application of the micro-operational perturbations; and generating actual perturbation response characteristics based on the perturbation response signals for comparison with a preset normal perturbation response range.

[0065] Specifically, the "periodic application of controlled minor operational disturbances to the combustion process" refers to making small, brief, and reversible adjustments to key operational parameters of the combustion process (such as fuel supply, primary air volume, secondary air volume, furnace pressure, or burner tilt angle) at predetermined time intervals or under specific operating conditions during the normal operation of industrial combustion equipment. These disturbances are designed not to significantly affect the normal production operation of the equipment, but are sufficient to induce observable and predictable responses in the flue gas pollutant sensor signals. The purpose is to actively detect the system's response characteristics to known inputs, so that its response characteristics will deviate in the presence of unknown interference.

[0066] The phrase "collecting the disturbance response signals output by the various pollutant sensors" refers to simultaneously recording the trajectory or pattern of the output signals of various pollutant sensors, such as the sulfur dioxide sensor, nitrogen oxide sensor, carbon monoxide sensor, and oxygen content sensor, changing over time under the influence of the disturbance, either simultaneously or immediately after applying the aforementioned minor operational disturbance. These signal changes reflect the dynamic response of the combustion system to the disturbance.

[0067] In practical applications, the phrase "generating actual disturbance response features based on the disturbance response signal for comparison with a preset normal disturbance response range" can be understood as performing data processing and feature extraction on the acquired disturbance response signal to quantify its key characteristics. These actual disturbance response features may include, but are not limited to, the amplitude of the response, the delay time of the response, the decay rate of the response, the frequency components of the response, or the shape parameters of the response. These feature parameters are combined to form a feature vector that represents the current system's response to disturbances. The "preset normal disturbance response range" is established under conditions of no unknown interference, through historical data or model prediction, and is used to characterize the statistical range or pattern of the normal response features that the system should have under the same disturbance.

[0068] This application's solution effectively addresses the limitations of traditional methods in identifying weak or similar-to-normal-condition unknown disturbances by actively applying controlled, minute operational disturbances during the operation of industrial combustion equipment and analyzing the system's response to these disturbances. Because unknown disturbances typically alter the physicochemical properties of the combustion process or the sensor's own response mechanism, the dynamic response pattern of the system under external disturbances differs significantly from the normal response pattern under undisturbed conditions. By acquiring disturbance response signals and generating actual disturbance response characteristics, these dynamic behavioral changes caused by unknown disturbances, which may be inconspicuous in the steady-state signal, can be captured. This active detection mechanism allows for more sensitive identification of the presence of unknown disturbances by analyzing the system's "abnormal" response to disturbances, even when the steady-state signal deviation is small, thereby improving the accuracy and reliability of abnormal signal pattern recognition.

[0069] Through the above technical solution, this application can significantly improve the ability of industrial flue gas multi-pollutant detection methods to identify unknown interfering substances. By actively applying controlled disturbances and analyzing their responses, the dynamic characteristic changes of the system caused by unknown interfering substances can be captured more sensitively, even if these disturbances are not obvious in the steady-state signal. This makes the identification of abnormal signal patterns more accurate and timely, thereby avoiding misjudging weak or gradual disturbances as normal operating condition fluctuations, and effectively improving the robustness and reliability of interference identification. In addition, by analyzing the disturbance response characteristics, richer and more discriminative information can be provided for the subsequent determination of interference deviation, further optimizing the correction accuracy of the target pollutant measurement signal.

[0070] In some preferred embodiments, a specific example is given below. Suppose that during the operation of a boiler in a coal-fired power plant, to detect the presence of unknown interfering substances, the system periodically (e.g., every hour) applies a small, five-minute step disturbance to the secondary air volume. During this period, sulfur dioxide, nitrogen oxide, carbon monoxide, and oxygen sensors simultaneously acquire their output signals. Under normal circumstances, when the secondary air volume increases, the oxygen content initially rises and then stabilizes, while the carbon monoxide initially decreases and then stabilizes. The concentration changes of sulfur dioxide and nitrogen oxides may be smaller or exhibit specific trends. The system extracts the peak value of the response, the time to reach the peak value, and the time for the signal to decay to a stable state as actual disturbance response characteristics based on the changes in these sensor signals. For example, if, after the disturbance is applied, the oxygen content sensor's response speed is significantly slower than normal, or the carbon monoxide sensor's decrease is much smaller than expected, this may indicate the presence of some unknown interfering substance (e.g., sensor drift, flue blockage, or the introduction of new chemical reactants) that alters the normal response of the combustion process to changes in secondary air volume. At this point, the actual disturbance response characteristics of the anomaly will be used for subsequent anomaly signal pattern recognition, thereby discovering potential interference problems earlier and more accurately.

[0071] In some embodiments described above, controlled minor operational disturbances are applied to the combustion process, and actual disturbance response characteristics are generated based on the disturbance response signal. These characteristics are then compared with a preset normal disturbance response range to identify abnormal signal patterns. However, in actual implementation, the operating state and production load of industrial combustion equipment are dynamically changing. The preset fixed normal disturbance response range may not adequately adapt to these changes, potentially leading to false alarms or missed alarms, thus affecting the accuracy of abnormal signal pattern identification. To address this, this application further proposes an optimization scheme: dynamically predicting the expected response trajectory of sensor signals and determining a normal disturbance response range suitable for the current operating conditions, and compensating for the normal signal reference based on this, to more accurately identify abnormal disturbance response patterns caused by unknown interference.

[0072] In some embodiments, the method further includes: predicting the expected response trajectory of the signals from the multiple pollutant sensors after applying the minor operational disturbance based on the first auxiliary data and the second auxiliary data; determining the preset normal disturbance response range based on the expected response trajectory; and comparing the current signal features with the normal signal reference, including: using the actual disturbance response features as the current signal features, compensating the normal signal reference based on the preset normal disturbance response range, and comparing the actual disturbance response features with the compensated normal signal reference to identify whether there is an abnormal disturbance response pattern caused by an unknown interfering substance.

[0073] Specifically, the first and second auxiliary data can be understood as key parameters characterizing the operating status of industrial combustion equipment and the operating conditions of the production process. The first auxiliary data may include physical wear trends, operating time, and maintenance records of key components of the combustion equipment, while the second auxiliary data may include control parameters such as production load, fuel type, combustion air volume, and flue gas recirculation volume. This data is used to build or drive a predictive model that can simulate or predict how various pollutant sensor signals will respond under current operating conditions when minor operational disturbances are applied.

[0074] Predicting the expected response trajectory of various pollutant sensor signals after a minor operational disturbance involves using historical operating data and / or physical models, combined with current first and second auxiliary data, to calculate the dynamic change path of the sensor signals after the disturbance occurs, assuming no unknown interference. This prediction process can employ machine learning models, such as recurrent neural networks (RNNs), long short-term memory networks (LSTMs), or models based on Gaussian process regression. These models can learn and capture the dynamic response characteristics of sensor signals under different operating conditions.

[0075] Furthermore, determining the preset normal disturbance response range based on the expected response trajectory means defining a statistical range around the predicted expected response trajectory. This range represents the acceptable response interval of the sensor signal to minor disturbances under normal operating conditions. This range can be composed of the mean of the expected response trajectory and its confidence interval at different time points. For example, it can be set to a range of one or two standard deviations above or below the expected trajectory.

[0076] In practical applications, using the actual disturbance response characteristics as the current signal characteristics means that when performing abnormal pattern recognition, the focus is no longer solely on static real-time signal characteristics, but rather on the system's dynamic response to controlled disturbances as the core judgment criterion. This approach can better capture the impact of unknown interference on the system's dynamic behavior.

[0077] Furthermore, compensating the normal signal reference based on the preset normal disturbance response range means adjusting the original, possibly relatively static, normal signal reference according to the predicted dynamic response characteristics under the current operating conditions. For example, if the expected response trajectory shows a normal instantaneous increase in the sensor signal at a certain point in time, the normal signal reference will also be adjusted upwards accordingly at that point in time, making the comparison benchmark more closely reflect the actual dynamic operating conditions. This compensation mechanism ensures the fairness and accuracy of the comparison and avoids misjudgments caused by changes in normal operating conditions.

[0078] Therefore, comparing the actual disturbance response characteristics with the compensated normal signal reference to identify whether there is an abnormal disturbance response pattern caused by unknown interference is achieved by comparing the actual observed system dynamic response with the normal dynamic response benchmark after condition adaptive adjustment, thereby more accurately detecting abnormal behaviors caused by unknown interference that cannot be detected by conventional static monitoring.

[0079] This application's solution effectively addresses the limitation of the aforementioned basic solution, where the preset normal disturbance response range may not adapt to dynamic operating conditions, by introducing a prediction mechanism for the expected response trajectories of various pollutant sensor signals. Specifically, by utilizing first and second auxiliary data, the system can acquire the component status and production operating conditions of industrial combustion equipment in real time. Based on this information, it predicts the dynamic response of sensor signals after applying minor operational disturbances under current operating conditions using a preset model. Therefore, the determined normal disturbance response range is no longer statically fixed but dynamically reflects normal fluctuations under current operating conditions, providing a more accurate and adaptable benchmark for anomaly pattern recognition. Furthermore, by comparing the actual disturbance response characteristics with the normal signal reference after dynamic range compensation, the system can distinguish between signal fluctuations caused by changes in normal operating conditions and genuine anomalies caused by unknown interference, significantly improving the accuracy and robustness of anomaly signal pattern recognition.

[0080] Through the above technical solution, this application can achieve more accurate identification of abnormal signal patterns caused by unknown interfering substances in industrial flue gas. Since the normal disturbance response range is dynamically predicted and determined based on the current operating conditions and used to compensate for the normal signal reference, this method can effectively avoid misjudgments caused by changes in the operating status or production load of industrial combustion equipment, significantly reducing the false positive rate of abnormal signal pattern identification. Furthermore, by comparing the actual disturbance response characteristics with the compensated normal signal reference, this application can more sensitively capture the subtle influence of unknown interfering substances on the dynamic behavior of the system, thereby improving the detection capability of real anomalies and enhancing the robustness and reliability of the entire multi-pollutant detection system.

[0081] In some preferred embodiments, a specific example is given below. Suppose that in a coal-fired power plant, it is necessary to monitor the sulfur dioxide (SO2), nitrogen oxides (NOx), and oxygen content (O2) in the flue gas. During the operation of the industrial combustion equipment, for example when the power plant load increases from 50% to 75%, the system periodically applies a controlled, minor operational disturbance to the combustion process, such as briefly increasing or decreasing the primary air volume.

[0082] At this point, the system acquires first auxiliary data (e.g., boiler wear level, burner operating time) and second auxiliary data (e.g., current boiler load, primary air volume, secondary air volume, fuel type) in real time. Based on this auxiliary data, a pre-trained dynamic prediction model (e.g., an LSTM-based neural network model) is used to predict the expected response trajectories of SO2, NOx, and O2 sensor signals under the current 75% load condition when a small disturbance occurs in the primary air volume. By learning the disturbance response patterns under different load and operating conditions in historical data, this model can output a smooth, time-dependent expected signal change curve.

[0083] Based on this predicted response trajectory, the system dynamically determines a normal disturbance response range around that trajectory. For example, it can be set to a range of 1.5 standard deviations above and below the predicted trajectory, which represents the normal fluctuation range of SO2, NOx, and O2 sensor signals to this small airflow disturbance under the current 75% load.

[0084] Subsequently, during the actual application of a small airflow disturbance, the system collects the actual disturbance response signals output by the SO2, NOx, and O2 sensors and extracts the actual disturbance response characteristics (such as response peak value, response time, decay rate, etc.). These actual disturbance response characteristics are then used as the current signal characteristics during anomaly pattern recognition.

[0085] The key is that the system compensates for the original normal signal reference based on the dynamically determined normal disturbance response range. For example, if the prediction model shows a normal instantaneous rise in the SO2 signal 10 seconds after the disturbance, the normal signal reference will be adjusted upwards accordingly at that time. Finally, the actual disturbance response characteristics are compared with this dynamically compensated normal signal reference. If the actual disturbance response characteristics deviate significantly from the range defined by the compensated normal signal reference (e.g., the SO2 signal response peak is much higher than expected, or the response time is abnormally prolonged), an abnormal disturbance response pattern caused by an unknown interfering substance is identified. For example, if a sulfide that was not considered by the model suddenly appears in the flue gas, it may cause the SO2 sensor to respond abnormally to airflow disturbances, thus being accurately identified by this solution.

[0086] Traditional methods for detecting multiple pollutants in industrial flue gas, after identifying abnormal signal patterns caused by unknown interfering agents, can determine the interference deviation based on the degree of deviation of the current signal characteristics relative to a normal signal reference. However, this quantification of deviation can be rather general. Specifically, different unknown interfering agents may have varying degrees and properties of influence on the measurement signals of different target pollutants. Relying solely on a generalized deviation is insufficient to accurately provide a customized interference deviation for each target pollutant, potentially leading to errors in the corrected pollutant concentration.

[0087] In response to this, this application further proposes a method for determining the amount of interference deviation of an unknown interfering substance on a target pollutant measurement signal based on the degree of deviation of the current signal characteristics from a normal signal reference, in response to the aforementioned identified abnormal signal pattern. Specifically, this method includes: determining a first interference intensity value based on the magnitude of the statistical distance; querying a compensation coefficient corresponding to the target pollutant from a preset interference deviation mapping relationship based on the preset interference category to which the abnormal signal pattern belongs; and calculating the amount of interference deviation based on the first interference intensity value and the compensation coefficient.

[0088] Specifically, after identifying the abnormal signal pattern, the first step is to determine the first interference intensity value based on the statistical distance. The statistical distance is a quantitative indicator that measures the deviation between the current signal characteristics and the multidimensional signal distribution center defined by a normal signal reference; its value directly reflects the significance or intensity of the abnormal signal pattern. For example, a larger statistical distance usually indicates a higher interference intensity. The first interference intensity value can be a dimensionless numerical value or a normalized quantitative indicator, used to uniformly measure the interference level under different abnormal patterns.

[0089] Furthermore, to more accurately quantify the impact of interference on a specific target pollutant, it is necessary to query the compensation coefficient corresponding to the target pollutant from a preset interference deviation mapping relationship based on the preset interference category to which the abnormal signal pattern belongs. The preset interference category refers to different types of unknown interference or interference scenarios predefined through analysis and classification of historical abnormal signal patterns. For example, based on characteristics such as the shape, duration, and sensor combinations involved, the abnormal signal pattern can be classified as "high humidity interference," "specific chemical substance interference," or "sensor drift interference." The preset interference deviation mapping relationship can be a multidimensional mapping model stored in a database or lookup table, recording the compensation coefficients required for different target pollutants under different interference categories and intensities. The compensation coefficient is a proportional factor or function used to convert a first interference intensity value into an interference deviation amount for a specific target pollutant; its value can be calibrated based on experimental data, historical operating experience, or expert knowledge.

[0090] Finally, the interference deviation can be obtained by calculating the first interference intensity value with the retrieved compensation coefficient. For example, the calculation can be a simple product of the two, or a more complex nonlinear function relationship, depending on the definition of the preset interference deviation mapping relationship.

[0091] This application addresses the potential accuracy limitations of traditional methods in quantifying the impact of unknown interfering substances on target pollutant measurement signals by introducing a first interference intensity value, a preset interference category, and an interference deviation mapping relationship. Specifically, firstly, by converting the statistical distance into a first interference intensity value, the severity of abnormal signal patterns is directly quantified. Secondly, by classifying abnormal signal patterns and combining them with a preset interference deviation mapping relationship, the system can retrieve more accurate compensation coefficients based on the specific type of interference and the target pollutant. Because different types of unknown interfering substances have different impact mechanisms and degrees on different target pollutants, this compensation coefficient based on interference category and target pollutant characteristics can more accurately reflect the actual interference effect. Therefore, this application's solution can transform a general abnormal signal deviation into a physically meaningful interference deviation specific to a particular target pollutant, thus providing a more reliable basis for subsequent measurement signal correction.

[0092] Through the above technical solution, this application can significantly improve the measurement accuracy and robustness of multi-pollutant detection methods for industrial flue gas in the presence of unknown interfering substances. Specifically, by finely quantifying the intensity of anomalous signal patterns and combining the identification of interference categories with compensation coefficients for specific target pollutants, the system can more accurately assess the specific impact of unknown interfering substances on the measurement signal of each target pollutant. This avoids the problems of over-correction or under-correction that may occur in traditional methods, thereby ensuring that the corrected target pollutant concentration data is closer to the true value, providing more reliable data support for the operation optimization of industrial combustion equipment and environmental emission control.

[0093] As a specific implementation, suppose that during the operation of an industrial combustion device, the real-time signals output by the sulfur dioxide sensor, nitrogen oxide sensor, carbon monoxide sensor, and oxygen content sensor exhibit an abnormal pattern. Through calculation, the statistical distance between the current signal characteristics and the normal signal reference is determined to be 5.2. This value of 5.2 is then used to determine the first interference intensity value. For example, it can be directly set to 5.2, or mapped to a normalized intensity value of 0.75 using a function. Further, the system analyzes the abnormal signal pattern (e.g., through a pattern recognition algorithm) and identifies it as belonging to the "high humidity interference" category. Subsequently, the system queries a preset interference deviation mapping relationship. This mapping relationship may be a table recording the compensation coefficients for sulfur dioxide and nitrogen oxides corresponding to different intensity values ​​under the "high humidity interference" category. For example, for the "high humidity interference" with an intensity value of 0.75, the compensation coefficient for sulfur dioxide is found to be 0.15 ppm / intensity unit, and the compensation coefficient for nitrogen oxides is 0.08 ppm / intensity unit. Therefore, the interference bias for sulfur dioxide was calculated to be 0.75 * 0.15 = 0.1125 ppm, and the interference bias for nitrogen oxides was calculated to be 0.75 * 0.08 = 0.06 ppm. These calculated interference biases will then be used to correct the measurement signals for their respective target pollutants, thereby outputting more accurate sulfur dioxide and nitrogen oxide concentrations.

[0094] In some embodiments described above, this application proposes identifying anomalous disturbance response patterns by applying controlled, minute operational perturbations and analyzing the sensor response. However, after identifying the anomalous disturbance response pattern, accurately determining the amount of interference deviation caused by the unknown interfering substance on the target pollutant measurement signal based on this dynamic, time-evolving pattern remains a problem that requires further refinement. Relying solely on the static degree of deviation may not adequately capture the complexity and time-varying nature of dynamic interference, thus affecting the accuracy of determining the amount of interference deviation.

[0095] In response, this application further proposes a specific method for determining the amount of interference deviation of unknown interfering object on the target pollutant measurement signal when an abnormal disturbance response pattern is identified. This method calculates the amount of interference deviation by obtaining the second interference intensity value of the actual disturbance response characteristics relative to the compensated normal signal reference and combining it with the time evolution characteristics of the compensated normal signal reference.

[0096] The above-mentioned response to the identified abnormal signal pattern, determining the amount of interference deviation of the unknown interfering object on the target pollutant measurement signal based on the degree of deviation of the current signal characteristics relative to the normal signal reference, includes: taking the abnormal disturbance response pattern as the abnormal signal pattern; in response to identifying the abnormal disturbance response pattern, obtaining a second interference intensity value of the actual disturbance response characteristics relative to the compensated normal signal reference; and calculating the amount of interference deviation based on the second interference intensity value and the time evolution characteristics of the compensated normal signal reference.

[0097] Specifically, treating an abnormal disturbance response pattern as an anomalous signal pattern means that when an abnormal disturbance response pattern is identified by comparing the actual disturbance response characteristics with a compensated normal signal reference, this pattern is considered a special anomalous signal pattern, requiring specialized methods to calculate the interference deviation. In response to identifying the abnormal disturbance response pattern, a second interference intensity value is obtained relative to the compensated normal signal reference for the actual disturbance response characteristics. This second interference intensity value can be understood as the degree of difference or deviation between the actual disturbance response characteristics and the compensated normal signal reference in the time dimension. For example, this second interference intensity value can be obtained by calculating the integral difference between the actual disturbance response characteristics and the compensated normal signal reference within a specific time window, the maximum instantaneous deviation, or the distance obtained based on the Dynamic Time Warping (DTW) algorithm. The purpose is to quantify the influence of unknown interference on the dynamic response trajectory of the sensor.

[0098] The time evolution characteristics of the compensated normal signal reference can be understood as the expected dynamic change of the sensor signal after a minor operational disturbance, under conditions of no unknown interference. These characteristics can include time-domain or frequency-domain features such as rise time, fall time, peak value, oscillation frequency, and attenuation rate. In practical applications, these time evolution characteristics can be obtained through historical data analysis, physical model simulation, or machine learning model prediction. The purpose is to provide a dynamic benchmark to more accurately assess the degree of deviation from the actual disturbance response characteristics.

[0099] This application's solution treats abnormal disturbance response patterns as a special type of abnormal signal pattern and introduces the time evolution characteristics of a second interference intensity value and a compensated normal signal reference. This allows for a more refined capture of the impact of unknown interfering substances under dynamic disturbance conditions. Because the actual disturbance response characteristics contain information about the system's dynamic response to the disturbance, and the time evolution characteristics of the compensated normal signal reference provide a normal benchmark for this dynamic response, comparing the two and quantifying their difference (i.e., the second interference intensity value) allows for a more accurate reflection of the dynamic interference generated by unknown interfering substances on the target pollutant measurement signal. Furthermore, combining calculations with time evolution characteristics avoids potential misjudgments based solely on instantaneous deviations or static thresholds, thereby improving the robustness and accuracy of determining the interference deviation.

[0100] Through the above technical solution, this application provides a more accurate and dynamic method for determining interference deviation in response patterns caused by minor operational disturbances. Compared to relying solely on the deviation degree of static signal characteristics, this solution, by fully utilizing the dynamic information of actual disturbance response characteristics and the time evolution characteristics of normal signal references, can more effectively identify and quantify the dynamic interference impact of unknown interfering substances on the target pollutant measurement signal. Therefore, it can significantly improve the accuracy and reliability of target pollutant concentration measurement under complex industrial conditions, especially in the presence of dynamic interference sources, providing strong support for the precise monitoring and control of industrial flue gas emissions.

[0101] In some preferred embodiments, a specific example is given below. Suppose that during the operation of an industrial combustion equipment, a small, step-like operational disturbance, such as a momentary, slight increase in fuel flow, is periodically applied to the combustion process. At this time, various pollutant sensors (such as sulfur dioxide sensors, nitrogen oxide sensors, etc.) will output a dynamic disturbance response signal. First, based on first and second auxiliary data, the expected response trajectory of each sensor signal after the application of this small operational disturbance is predicted, and a preset normal disturbance response range is determined accordingly. Then, the normal signal reference is compensated.

[0102] When the actual disturbance response characteristics formed by the collected disturbance response signal are compared with the compensated normal signal reference, if an abnormal disturbance response pattern is identified, such as a significantly higher-than-expected peak value and an abnormally slow decay rate in the sulfur dioxide sensor signal, this indicates the possible presence of an unknown interfering substance. In this case, the abnormal disturbance response pattern is considered an abnormal signal pattern. Next, the second interference intensity value is obtained by calculating the difference between the actual disturbance response characteristics (e.g., the integral value of the sulfur dioxide signal within 30 seconds after the disturbance is applied) and the compensated normal signal reference (the integral value within the corresponding time window). Simultaneously, combining the time evolution characteristics of the compensated normal signal reference (e.g., the half-life of the sulfur dioxide signal is normally 10 seconds), a preset nonlinear model or lookup table is used to combine the second interference intensity value with the time evolution characteristics to calculate the interference deviation of the unknown interfering substance on the target pollutant (e.g., sulfur dioxide) measurement signal. For example, if the second interference intensity value is large and the time evolution characteristics show a long abnormal duration, the calculated interference deviation will increase accordingly, resulting in a more significant correction to the sulfur dioxide measurement signal.

[0103] Traditional industrial flue gas multi-pollutant detection systems are ill-equipped to handle situations where long-term wear and tear on combustion furnace components introduces unknown trace interference components into the flue gas, or when factory production loads fluctuate. Their internal calibration parameter sets are unable to identify and compensate for the direct interference of these unknown components on the sensors, leading to unpredictable and systematic biases in the real-time concentration measurements of key pollutants such as sulfur oxides and nitrogen oxides. This bias lacks accurate basis for adjusting the control strategies of subsequent flue gas treatment devices, hindering efficiency optimization in resource utilization and potentially leading to excessive emissions due to misjudgments, thus posing a potential environmental impact.

[0104] In some embodiments, this application proposes an industrial flue gas multi-pollutant detection system, comprising: an establishment unit for establishing a normal signal reference for characterizing the signal features of multiple flue gas pollutant sensors under conditions free of unknown interference; an acquisition unit for acquiring current real-time signals output by multiple pollutant sensors in the flue gas and determining the current signal features corresponding to the current real-time signals; an identification unit for comparing the current signal features with the normal signal reference to identify whether there is an abnormal signal pattern caused by an unknown interference; a determination unit for determining, in response to the identified abnormal signal pattern, the interference deviation amount caused by the unknown interference to the target pollutant measurement signal based on the degree of deviation of the current signal features relative to the normal signal reference; and an output unit for correcting the target pollutant measurement signal based on the interference deviation amount to output the corrected target pollutant concentration.

[0105] The industrial flue gas multi-pollutant detection system proposed in this application, through its modular establishment unit, acquisition unit, identification unit, determination unit, and output unit, can effectively identify and compensate for the influence of unknown interfering substances on the measurement signal of target pollutants. Compared with traditional systems, this system no longer relies solely on a preset, static set of calibration parameters, but can actively establish and dynamically adjust a normal signal reference, comparing the current signal characteristics with this reference in real time, thereby identifying and quantifying abnormal signal patterns caused by unknown interfering substances. For example, when long-term wear of combustion furnace components leads to the appearance of trace secondary pollutants in the flue gas, traditional systems, due to their calibration parameter sets, cannot identify these new interfering substances, resulting in systematic deviations in measurement results. This system, however, can detect such anomalies through the identification unit, calculate the interference deviation of the target pollutant measurement signal through the determination unit, and finally correct it through the output unit. Therefore, this system can provide more accurate and reliable pollutant concentration data, providing a solid basis for the optimized control of industrial flue gas treatment processes, effectively avoiding reduced resource utilization efficiency and potential risks of exceeding emission standards due to measurement errors, and significantly improving the intelligence level of environmental monitoring and its ability to cope with complex operating conditions.

[0106] The above are merely embodiments of this application and are not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for detecting multiple pollutants in industrial flue gas, characterized in that, include: A normal signal reference is established to characterize the signal features of multiple flue gas pollutant sensors under conditions without unknown interference; the current real-time signals output by multiple pollutant sensors in flue gas are acquired, and the current signal features corresponding to the current real-time signals are determined; the current signal features are compared with the normal signal reference to identify whether there are abnormal signal patterns caused by unknown interference. In response to the identified abnormal signal pattern, the interference deviation of the unknown interfering substance on the target pollutant measurement signal is determined based on the degree of deviation of the current signal characteristics relative to the normal signal reference; the target pollutant measurement signal is corrected based on the interference deviation to output the corrected target pollutant concentration.

2. The method according to claim 1, characterized in that, The establishment of a normal signal reference for characterizing the signal characteristics of multiple flue gas pollutant sensors under conditions without unknown interference includes: collecting first historical signal data output by the multiple pollutant sensors when the industrial combustion equipment is in a preset healthy state and the production load is stable; constructing an initial multi-dimensional signal reference range based on the first historical signal data to characterize the mean, fluctuation range, and correlation of each sensor signal under conditions without unknown interference; the initial multi-dimensional signal reference range includes the normal value range of each sensor signal; during the subsequent operation of the industrial combustion equipment, acquiring first auxiliary data characterizing the state of the combustion equipment components and second auxiliary data characterizing the operating conditions of the production process in real time; and dynamically adjusting the statistical characteristics of the initial multi-dimensional signal reference range based on the first auxiliary data and the second auxiliary data to generate the normal signal reference adapted to the current operating conditions.

3. The method according to claim 2, characterized in that, The step of dynamically adjusting the statistical characteristics of the initial multidimensional signal reference range based on the first auxiliary data and the second auxiliary data includes: extracting component state feature vectors from the first auxiliary data to characterize the physical wear trend of key components of the combustion equipment; extracting operation mode feature vectors from the second auxiliary data to characterize production load and combustion control parameters; calculating the adjustment amount of the central mean vector and covariance matrix of the initial multidimensional signal reference range based on the component state feature vectors and the operation mode feature vectors through a preset nonlinear mapping model; and applying the adjustment amount to the initial multidimensional signal reference range to complete the dynamic adjustment of the statistical characteristics of the initial multidimensional signal reference range.

4. The method according to claim 1, characterized in that, The step of acquiring the current real-time signals output by multiple pollutant sensors in flue gas and determining the current signal characteristics corresponding to the current real-time signals includes: simultaneously acquiring first real-time measurement signals output by sulfur dioxide sensors, nitrogen oxide sensors, carbon monoxide sensors, and oxygen content sensors in flue gas; performing digital filtering and range conversion on the first real-time measurement signals to obtain a preprocessed second real-time measurement signal; extracting feature parameters from the second real-time measurement signals for pattern comparison with the normal signal reference, the feature parameters including the stable output value and / or rate of change of each sensor signal; and combining the feature parameters to obtain the current signal characteristics.

5. The method according to claim 1, characterized in that, The step of comparing the current signal feature with the normal signal reference to identify whether there is an abnormal signal pattern caused by unknown interference includes: calculating the statistical distance between the current signal feature and the multidimensional signal distribution center defined by the normal signal reference; comparing the statistical distance with a preset dynamic distance threshold; and determining that the abnormal signal pattern exists if the statistical distance is greater than the preset dynamic distance threshold.

6. The method according to claim 2, characterized in that, Before comparing the current signal characteristics with the normal signal reference, the method further includes: periodically applying controlled minor operational disturbances to the combustion process during operation of the industrial combustion equipment; acquiring disturbance response signals output by the multiple pollutant sensors during the application of the minor operational disturbances; and generating actual disturbance response characteristics based on the disturbance response signals for comparison with a preset normal disturbance response range.

7. The method according to claim 6, characterized in that, The method further includes: predicting the expected response trajectory of multiple pollutant sensor signals after applying the minor operational disturbance based on the first auxiliary data and the second auxiliary data; determining the preset normal disturbance response range based on the expected response trajectory; and comparing the current signal feature with the normal signal reference, including: using the actual disturbance response feature as the current signal feature, compensating the normal signal reference based on the preset normal disturbance response range, and comparing the actual disturbance response feature with the compensated normal signal reference to identify whether there is an abnormal disturbance response pattern caused by an unknown interfering object.

8. The method according to claim 5, characterized in that, In response to the identified abnormal signal pattern, the method determines the interference deviation of the unknown interfering object on the target pollutant measurement signal based on the degree of deviation of the current signal characteristics relative to the normal signal reference. This includes: determining a first interference intensity value based on the magnitude of the statistical distance; querying a compensation coefficient corresponding to the target pollutant from a preset interference deviation mapping relationship based on the preset interference category to which the abnormal signal pattern belongs; and calculating the interference deviation based on the first interference intensity value and the compensation coefficient.

9. The method according to claim 7, characterized in that, The step of determining the interference deviation of the unknown interfering substance on the target pollutant measurement signal in response to the identified abnormal signal pattern, based on the degree of deviation of the current signal feature relative to the normal signal reference, includes: taking the abnormal disturbance response pattern as the abnormal signal pattern; in response to identifying the abnormal disturbance response pattern, obtaining a second interference intensity value of the actual disturbance response feature relative to the compensated normal signal reference; and calculating the interference deviation based on the second interference intensity value and the time evolution characteristics of the compensated normal signal reference.

10. A multi-pollutant detection system for industrial flue gas, characterized in that, include: Establishment unit, used to establish a normal signal reference for characterizing the sensor signal characteristics of various flue gas pollutants under conditions without unknown interference; The acquisition unit is used to acquire the current real-time signal output by multiple pollutant sensors in flue gas and determine the current signal characteristics corresponding to the current real-time signal; the identification unit is used to compare the current signal characteristics with the normal signal reference to identify whether there is an abnormal signal pattern caused by an unknown interfering substance; the determination unit is used to determine the amount of interference deviation of the unknown interfering substance on the target pollutant measurement signal according to the degree of deviation of the current signal characteristics from the normal signal reference in response to the identified abnormal signal pattern. The output unit is used to correct the measurement signal of the target pollutant based on the interference deviation, so as to output the corrected target pollutant concentration.

Images