Aluminum ash disposal exhaust gas monitoring and early warning method and system of a gas sensor

Abstract

The application discloses an aluminum ash treatment waste gas monitoring and early warning method and system of a gas sensor, and belongs to the technical field of aluminum ash treatment waste gas monitoring. The method first acquires multi-dimensional targeted data of waste gas in the whole process of aluminum ash treatment, then completes feature extraction and redundancy elimination through a first analysis model to obtain a core parameter set of waste gas state, inputs the core parameter set into a second optimization model to acquire a precise quantitative parameter set of pollutants, and then generates multi-level early warning signals and process real-time adjustment instructions through a third early warning and regulation model. Finally, the signals and instructions are pushed through 5G and edge computing nodes to realize real-time monitoring, dynamic early warning and closed-loop regulation of waste gas. The application realizes precise extraction of waste gas characteristics, precise quantification of pollutant concentration, and dynamic early warning and precise regulation through three core models, builds a closed-loop system relying on 5G and edge computing, ensures reliable monitoring, timely early warning and precise regulation, improves the overall efficiency of waste gas treatment, and helps to achieve environmental protection standards and stable production.

Description

Technical Field

[0001] This invention relates to the field of aluminum ash disposal waste gas monitoring technology, specifically to a gas sensor-based method and system for monitoring and early warning of aluminum ash disposal waste gas. Background Technology

[0002] In the process of aluminum ash disposal and storage, aluminum ash easily releases harmful gases such as ammonia and fluorine. If absorption and purification are not timely and early warning is not accurate, it will not only cause air pollution, but may also cause health risks to personnel and corrosion problems to equipment. Therefore, gas treatment and safety early warning in aluminum ash hazardous waste warehouses and storage warehouses have become key industry requirements.

[0003] Chinese patent (publication number: CN220159634U) discloses an ammonia and fluorine absorption and purification device for aluminum ash hazardous waste storage. Although it achieves the purification of harmful gases through the absorption tower structure, it focuses on static gas absorption and treatment and does not combine the dynamic characteristics of the gases released by aluminum ash to design an adaptive adjustment function. When faced with changes in the amount of aluminum ash piled up and sudden increases in gas concentration caused by temperature and humidity fluctuations, the purification efficiency is easily affected, making it difficult to achieve high-efficiency purification with real-time matching.

[0004] Chinese patent (publication number: CN216053274U) discloses a safety early warning system for aluminum ash slag storage warehouse. Although it can monitor parameters such as gas concentration, temperature and humidity in the warehouse and trigger early warnings, the early warning mechanism relies on fixed thresholds, does not consider the differences in gas release patterns during the aluminum ash storage period, and does not form a linkage control with the gas purification device. It can only realize the monitoring and early warning functions, and cannot simultaneously complete the closed-loop response of early warning and purification, making it difficult to meet the dual requirements of safety and environmental protection throughout the entire aluminum ash storage cycle.

[0005] Existing technologies generally suffer from problems such as disconnect between gas purification and early warning systems, inability of purification parameters to dynamically adapt to changes in operating conditions, and lack of graded handling mechanisms for early warning signals. These technologies are insufficient to address the complexity and uncertainty of gas release in aluminum ash hazardous waste storage facilities and warehouses. Therefore, there is an urgent need for a gas sensor-based method and system solution for monitoring and early warning of aluminum ash disposal waste gas. Summary of the Invention

[0006] Based on the above-mentioned technical problems, this application discloses a method and system for monitoring and early warning of aluminum ash disposal waste gas using a gas sensor, specifically including: Acquire multi-dimensional targeted data of waste gas from the entire aluminum ash disposal process, including first core data and second core data; The first analytical model is used to extract features and remove redundancy from the first core data and the second core data to obtain the set of core parameters of the exhaust gas state. The first analytical model is constructed based on the physical and chemical characteristics of pollution and the dynamic spatiotemporal correlation mechanism of multi-source signals. The core parameter set of the exhaust gas state is input into the second optimization model to obtain the accurate quantification parameter set of pollutants. The second optimization model is constructed based on deep belief network and least squares support vector machine. The precise quantification parameter set of the pollutants is input into the third early warning and control model to generate multi-level early warning signals and real-time process adjustment instructions. The third early warning and control model is constructed based on fuzzy PID control logic and sliding window dynamic threshold algorithm. By pushing multi-level early warning signals to the monitoring terminal through 5G and edge computing nodes, and simultaneously issuing real-time process adjustment instructions to the actuators of the aluminum ash disposal device, real-time monitoring, dynamic early warning and closed-loop control of waste gas can be achieved.

[0007] Preferably, the targeted data includes first core data and second core data. The first core data consists of spectral absorption signals, electrochemical response signals, and instantaneous concentration fluctuation data of fluorides, nitrogen oxides, and particulate matter in the waste gas. The second core data consists of gradual temperature and humidity change data, stable air pressure values, airflow velocity vector data, and equipment operating load time-series data of the treatment scenario. The instantaneous concentration fluctuation data is quantified by sliding standard deviation, and the equipment operating load time-series data is the coupled data of real-time power and feeding rate of the aluminum ash treatment furnace.

[0008] Preferably, the dynamic adaptation spatiotemporal correlation mechanism of the first analytical model is implemented through a spatiotemporal weight matrix. First, the signal correlation baseline weight is determined based on the physical and chemical characteristics of pollution. Then, it is dynamically adjusted by combining the temporal correlation and spatial distribution characteristics of the data. The formula for feature extraction and redundancy removal is as follows:

[0009] in, For the first One core parameter of exhaust gas status, For the first The first core data of the class and the first Dynamically adapting weights for the second core data of the class. For the feature values ​​of the corresponding data, The covariance of this feature with other features, It is the maximum value of the covariance of all features. Number is the first core data type. It is the second core data type.

[0010] Preferably, the dynamic adaptation weight Based on the dynamic calculation of pollution physicochemical characteristics and data credibility, data credibility is quantified by signal-to-noise ratio. Pollution physicochemical characteristics include the spectral absorption cross section, electrochemical activity, solubility and reactivity of pollutants. The calculation of dynamic adaptation weights must meet the normalization constraint of all data association weights.

[0011] Preferably, in the second optimization model, the Deep Belief Network (DBN) includes multiple Restricted Boltzmann Machine (RBM) layers and an output layer. The number of neurons in each RBM layer is configured in descending order of gradient. First, the core parameter set of the exhaust gas state is pre-trained unsupervised. During the pre-training process, the contrastive divergence algorithm is used to minimize the reconstruction error. Then, high-dimensional feature dimensionality reduction and reconstruction are achieved through supervised fine-tuning.

[0012] Preferably, the second optimization model inputs the reconstructed features output by DBN into the least squares support vector machine (LSSVM), and achieves accurate fitting of pollutant concentration through radial basis kernel function mapping. The fitting formula is as follows:

[0013] in, For the first Accurately quantify pollutant concentrations at all times. For the Lagrange multipliers of LSSVM, For DBN output Features are constantly being reconstructed. For the first Reconstructed features of each training sample For kernel function parameters, For bias terms, This represents the number of training samples.

[0014] Preferably, the sliding window dynamic threshold of the third early warning and control model is determined by the statistical characteristics of recent pollutant concentrations, and the window length is adaptively set according to the aluminum ash treatment process cycle. The dynamic early warning threshold formula is:

[0015] in, For the first Real-time dynamic warning thresholds The average concentration within the sliding window. The standard deviation of concentration within the sliding window. The early warning sensitivity coefficient is adjusted in a positive correlation with the toxicity level of the pollutant.

[0016] Preferably, the fuzzy PID control logic of the third early warning and control model includes two input variables and three output variables, with the input variable being the concentration deviation. and the rate of change of deviation The output variables are PID parameter correction values: proportional coefficient correction value, integral coefficient correction value, and derivative coefficient correction value. Both input and output variables are configured with multiple fuzzy subsets. The parameter correction values ​​are generated through a preset fuzzy rule base. The fuzzy rule base is defuzzified using the centroid method to obtain the final correction value.

[0017] Preferably, the actuator of the aluminum ash disposal device includes a waste gas purification spray pump, a ventilation regulating valve, a burner power controller, and a bag filter differential pressure regulator; the generation of real-time process adjustment commands must be combined with environmental emission standards and equipment operating constraints, and the command parameters must match the rated operating range of the actuator to ensure safe equipment operation. A gas sensor-based aluminum ash disposal waste gas monitoring and early warning system includes a data acquisition module, a model calculation module, an early warning and control module, a communication transmission module, and an execution control module. The data acquisition module integrates a spectral sensor, an electrochemical sensor, a temperature and humidity sensor, a barometric pressure sensor, a wind speed sensor, and an equipment load monitoring unit. Each sensor collects data synchronously at a uniform sampling frequency to meet the monitoring accuracy requirements. The model computation module is deployed on edge computing nodes and supports online model parameter updates and offline training. The communication transmission module adopts 5G communication technology to ensure data transmission rate and timely command issuance. The execution control module has a built-in drive circuit and feedback acquisition unit, which can collect the operating status of the actuator in real time and feed it back to the model calculation module to form a closed-loop control.

[0018] Compared with the prior art, the technical solution of this application has the following technical effects: This invention collects multi-dimensional targeted data from the entire aluminum ash disposal process and constructs a first analytical model by combining the physical and chemical characteristics of pollution with a multi-source signal dynamic adaptation spatiotemporal correlation mechanism. This model enables the accurate extraction of waste gas characteristics and the efficient removal of redundant information, providing high-quality core parameters for subsequent pollutant quantitative analysis. It effectively improves the reliability and relevance of monitoring data and provides solid data support for waste gas treatment.

[0019] This invention relies on a second optimization model constructed using deep belief networks and least squares support vector machines. Through unsupervised pre-training and supervised fine-tuning, it optimizes the dimensionality reduction and reconstruction capabilities of high-dimensional features. By leveraging radial basis function kernels, it achieves accurate fitting of pollutant concentrations, significantly enhancing the accuracy of pollutant quantification. This provides precise data for subsequent early warning and control decisions, ensuring the scientific validity of monitoring results.

[0020] This invention presents a third early warning and control model that integrates fuzzy PID control logic and a sliding window dynamic threshold algorithm. This model can dynamically adjust the early warning threshold based on the statistical characteristics of pollutant concentration, intelligently generate multi-level early warning signals and process adjustment instructions, and balance the timeliness of early warning response with the accuracy of control operation. It effectively adapts to fluctuations in operating conditions and improves the dynamic control capability of waste gas treatment.

[0021] This invention utilizes 5G communication technology and edge computing nodes to construct a closed-loop control system, enabling rapid push of early warning signals and efficient issuance of process adjustment instructions. Simultaneously, through real-time feedback of actuator status and a secondary adjustment mechanism, it ensures safe equipment operation, reduces the risk of failure, and improves the overall efficiency of monitoring, early warning, and treatment of aluminum ash disposal waste gas, thereby contributing to environmental compliance and stable production.

[0022] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the preferred embodiments of this application are described in detail below with reference to the accompanying drawings.

[0023] The above and other objects, advantages and features of this application will become more apparent to those skilled in the art from the following detailed description of specific embodiments in conjunction with the accompanying drawings. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In all drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.

[0025] Based on the description of the figures and their corresponding technical content in the document, the titles of the figures are as follows: Figure 1 A schematic diagram of the overall process of a gas sensor-based method for monitoring and early warning of aluminum ash disposal waste gas; Figure 2 First analytical model architecture diagram; Figure 3 Second optimization model architecture diagram; Figure 4 : Architecture diagram of an aluminum ash disposal waste gas monitoring and early warning system using a gas sensor; Figure 5 Fitting curves of the quantified concentrations and measured concentrations of the three pollutants; Figure 6 : Under slightly excessive operating conditions Concentration changes versus dynamic threshold curves; Figure 7 Concentration recovery curve and actuator control curve under severely excessive operating conditions; Figure 8 : Pre-training iteration curve of the DBN model; Figure 9 : Kernel function parameter optimization curve of LSSVM model; Figure 10 Parameter correction curves for fuzzy PID control. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. In the following description, specific details such as specific configurations and components are provided merely to help fully understand the embodiments of this application. Therefore, those skilled in the art should understand that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this application. In addition, for clarity and brevity, descriptions of known functions and structures are omitted in the embodiments.

[0027] It should be understood that the phrase "an embodiment" or "this embodiment" throughout the specification means that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of this application. Therefore, "an embodiment" or "this embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments.

[0028] Furthermore, reference numerals and / or letters may be repeated in different examples within this application. Such repetition is for the purpose of simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or settings discussed.

[0029] In this article, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can mean: A exists alone, B exists alone, and A and B exist simultaneously. The term " / and" in this article describes another type of relationship between related objects, indicating that two relationships can exist. For example, A / and B can mean: A exists alone, and A and B exist alone. In addition, the character " / " in this article generally indicates that the related objects before and after it are in an "or" relationship.

[0030] In this article, the term "at least one" is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, "at least one of A and B" can mean: A exists alone, A and B exist simultaneously, or B exists alone.

[0031] It should also be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion.

[0032] Example 1 This embodiment mainly describes a method for monitoring and early warning of waste gas from aluminum ash disposal using a gas sensor, such as... Figure 1 As shown, it specifically includes: Acquire multi-dimensional targeted data of waste gas from the entire aluminum ash disposal process, including first core data and second core data; The first analytical model is used to extract features and remove redundancy from the first core data and the second core data to obtain the set of core parameters of the exhaust gas state. The first analytical model is constructed based on the physical and chemical characteristics of pollution and the dynamic spatiotemporal correlation mechanism of multi-source signals. The core parameter set of the exhaust gas state is input into the second optimization model to obtain the accurate quantification parameter set of pollutants. The second optimization model is constructed based on deep belief network and least squares support vector machine. The precise quantification parameter set of the pollutants is input into the third early warning and control model to generate multi-level early warning signals and real-time process adjustment instructions. The third early warning and control model is constructed based on fuzzy PID control logic and sliding window dynamic threshold algorithm. By pushing multi-level early warning signals to the monitoring terminal through 5G and edge computing nodes, and simultaneously issuing real-time process adjustment instructions to the actuators of the aluminum ash disposal device, real-time monitoring, dynamic early warning and closed-loop control of waste gas can be achieved.

[0033] Furthermore, multi-dimensional targeted data from the entire aluminum ash disposal process is collected through an integrated sensor array, specifically including: The first core data acquisition involves using a near-infrared spectral sensor to collect the spectral absorption signals of fluorides, nitrogen oxides, and particulate matter (wavelengths covering the characteristic absorption bands of the target pollutants). Electrochemical sensors are used to collect the electrochemical response signals of the corresponding pollutants. Based on a uniform sampling frequency across the sensors, instantaneous fluctuations in pollutant concentrations are recorded synchronously. These instantaneous concentration fluctuations are quantified using a moving standard deviation, with the following formula: In the formula, for Instantaneous fluctuations in pollutant concentration at any given time. For the first Raw data of pollutant concentrations at each sampling point This represents the average concentration within the sliding window. The length of the sliding window is calculated for fluctuations (adaptively set according to the stability of the aluminum ash treatment process). The second core data acquisition involves: collecting temperature and humidity gradient data of the treatment scenario through temperature and humidity sensors, collecting real-time stable air pressure values ​​through air pressure sensors, collecting airflow velocity vector data (including wind speed and direction information) through wind speed sensors, and collecting real-time power and feeding rate of the aluminum ash treatment furnace through the equipment controller. Finally, it generates equipment operating load time-series data through a coupling formula. In the formula, for Constant equipment operating load, For real-time power, For feed rate, , This is a weighting coefficient (calibrated based on the equipment's rated parameters and process requirements).

[0034] Furthermore, such as Figure 2 As shown, the first analytical model, based on the physical and chemical characteristics of pollution and the dynamic spatiotemporal correlation mechanism of multi-source signals, achieves core parameter extraction, as detailed below: Dynamic adaptation weight calculation: Dynamic adaptation weight The calculation needs to integrate the physicochemical characteristics of pollution with data reliability. The formula is: In the formula, For the first The first core data of the class and the first The physical correlation coefficients of the second core data (correlation coefficients of spectral absorption cross sections corresponding to fluorides, correlation coefficients of electrochemical activity corresponding to nitrogen oxides, etc.). This is the original data combination type sequence number. For the first The reliability of the data source corresponding to each parameter is calculated using the signal-to-noise ratio (SNR). , The signal-to-noise ratio of the current original data combination. (where is the maximum signal-to-noise ratio threshold of the sensor), this formula naturally satisfies the normalization constraint of all data association weights; Feature extraction and redundancy removal: Based on dynamic adaptation weights and covariance analysis, core parameter extraction is achieved through the following formula: In the formula, For the first One core parameter of exhaust gas status, For the first The first core data of the class and the first Dynamically adapting weights for the second core data of the class. For the feature values ​​of the corresponding data, The covariance of this feature with other features, It is the maximum value of the covariance of all features. This is the first core data type (corresponding to spectral absorption signals, electrochemical response signals, and instantaneous concentration fluctuation data). The second core data type (corresponding to temperature and humidity gradient data, air pressure stability value, airflow velocity vector data, and equipment operating load time series data) is used to filter out core parameters from the original multi-dimensional data, forming a set of core parameters for exhaust gas status, and eliminating redundant information and noise interference.

[0035] Furthermore, such as Figure 3 As shown, the second optimization model is based on a coupled architecture of Deep Belief Network (DBN) and Least Squares Support Vector Machine (LSSVM) to achieve dimensionality reduction of high-dimensional parameters and accurate fitting of pollutant concentrations, as detailed below: DBN Model Training and Feature Reconstruction: The DBN consists of multiple Restricted Boltzmann Machines (RBMs) and one output layer. The number of neurons in each RBM layer is configured in descending order of the gradient. First, unsupervised pre-training is performed on the core parameter set of the exhaust gas state. The contrastive divergence algorithm (CD-k algorithm) is used to minimize the reconstruction error. The pre-training objective function is: In the formula, This represents the total number of core parameters for the exhaust gas state. For the first Original values ​​of the core parameters The parameters are reconstructed from the RBM. After pre-training, the entire network parameters are optimized through supervised fine-tuning (using measured pollutant concentrations as labels), and the reconstructed features are finally output. The mathematical expression for the reconstruction process is:

[0036] In the formula, , , These are the weight matrices for each layer of the RBM. , , These are the bias terms for each layer. The Sigmoid activation function ( ), for The set of core parameters for the exhaust gas state at any given time; LSSVM concentration fitting: reconstructing features from DBN output Input the LSSVM model and use the radial basis function kernel to achieve an accurate fit of pollutant concentrations. The fitting formula is as follows: The output pollutant concentration is accurately quantified. The relative error between the measured concentration and the actual concentration is controlled within a reasonable range, meeting the monitoring accuracy requirements.

[0037] Furthermore, the third early warning and control model, based on fuzzy PID control logic and a sliding window dynamic threshold algorithm, generates multi-level early warning signals and real-time process adjustment instructions, as specifically implemented as follows: Sliding window dynamic threshold calculation: dynamic early warning threshold The calculation needs to take into account the recent concentration statistics: ,in, , In the formula, This represents the average pollutant concentration within the sliding window. The length of the sliding window. The standard deviation of pollutant concentration within the moving window. The early warning sensitivity coefficient (calibrated according to the toxicity level of pollutants; the higher the toxicity, the lower the sensitivity). The higher the value, the more sensitive the early warning of highly toxic pollutants; based on dynamic thresholds. A three-tiered early warning mechanism has been established: when 0.8 At that time, there was no warning; When 0.8 1.2 At that time, a Level 1 warning was issued; when 1.2 At that time, a level-two warning was issued; Fuzzy PID control logic implementation: Fuzzy PID control is based on concentration deviation. and the rate of change of deviation ( The input is the sampling interval, and the PID parameter correction is the proportional coefficient correction. Integral coefficient correction amount Differential coefficient correction amount For the output, the specifics are as follows: Input and output fuzzification: and The fuzzy subsets are all set as {Negative Large (NB), Negative Medium (NM), Negative Small (NS), Zero (ZO), Positive Small (PS), Positive Medium (PM), Positive Large (PB)}, and the quantization universe is [-6, 6]. , , The fuzzy subset and quantization domain are the same as above; fuzzy rule base construction: based on industrial control experience, 49 fuzzy rules are constructed (if... For PB and If it is PB, then For PB, For NB, For PS); Defuzzing: The centroid method is used to defuzzify the fuzzy output to obtain the precise values ​​of the PID parameter corrections. The final PID parameters are: In the formula, , , These are the proportional coefficient, integral coefficient, and differential coefficient, respectively. , , These are the initial parameters for the PID controller (calibrated based on the initial operating conditions of the process); based on the PID control output, real-time process adjustment commands are generated, including the flow rate adjustment value of the exhaust gas purification spray pump, the opening degree of the ventilation regulating valve, the burner power correction amount, and the differential pressure setting value of the bag filter. All command parameters are matched with the rated operating range of the actuator.

[0038] Furthermore, the adjustment instructions for the exhaust gas purification spray pump must specify the spray liquid flow rate adjustment parameters. The flow rate adjustment should be based on the rated flow rate of the spray system, taking into account the gas-liquid contact efficiency between the absorbent liquid and the exhaust gas. This avoids insufficient pollutant absorption due to excessively low flow rate, or excessively high flow rate exceeding the pump's rated head and causing pipeline pressure overload. The adjustment instructions for the ventilation regulating valve focus on valve opening control. The opening range must match the design ventilation volume of the exhaust gas treatment system. This involves both increasing the exhaust gas flow rate by increasing the opening to reduce local pollutant accumulation and preventing sudden changes in opening that could lead to abnormal negative pressure in the system and affect the combustion stability of the aluminum ash treatment furnace. Qualitatively, the adjustment commands of the burner power controller need to accurately correct the combustion power. The power correction needs to be combined with the thermal conditions required for aluminum ash disposal to avoid excessive power causing fuel waste and secondary pollutant generation, or excessive power causing insufficient disposal temperature and reduced exhaust gas purification efficiency. The adjustment commands of the bag filter differential pressure regulator take the pressure difference between the inlet and outlet of the dust collector as the control target. The set differential pressure threshold needs to be adapted to the filtration accuracy and dust holding capacity of the filter bags. By adjusting the cleaning frequency or cleaning pressure, it is necessary to avoid excessive pressure difference leading to increased airflow resistance and increased system energy consumption, or excessive pressure difference leading to excessive cleaning of the filter bags and shortened service life. All parameters for real-time process adjustment commands must undergo boundary verification to ensure they do not exceed safety thresholds such as rated voltage, rated power, and mechanical stroke of the actuators. Simultaneously, the actual operating status of each actuator (such as actual flow rate of the spray pump, actual valve opening, actual burner power, and actual differential pressure of the dust collector) is monitored in real time through the feedback acquisition unit. If any parameter deviates from the command value or approaches the safety threshold, a secondary adjustment is triggered immediately, forming a closed-loop control of command issuance, status feedback, and parameter calibration. This ensures that exhaust emissions meet standards while avoiding the risk of equipment failure due to overload or malfunction.

[0039] Example 2 This embodiment mainly describes a gas sensor-based aluminum ash disposal waste gas monitoring and early warning system, such as... Figure 4 As shown, it specifically includes: Data acquisition module: integrates spectral sensor, electrochemical sensor, temperature and humidity sensor, air pressure sensor, wind speed sensor and equipment load monitoring unit. Among them, the wavelength resolution of spectral sensor, the concentration resolution of electrochemical sensor and the measurement error of environmental sensor all meet the monitoring accuracy requirements. All sensors collect data synchronously at a unified sampling frequency. After the raw data is pre-processed by the data preprocessing unit (including filtering and normalization modules), it is transmitted to the model calculation module. Model computation module: Deployed on edge computing nodes (using industrial-grade edge servers, supporting GPU-accelerated computation), integrating the algorithm programs of the first analytical model, the second optimization model, and the third early warning and control model. This module supports online updating of model parameters (iterative optimization based on sample data under new working conditions) and offline training (completing initial model training and parameter calibration in a laboratory environment), and the computation latency meets real-time requirements. The early warning and control module includes an early warning signal generation unit and an instruction encoding unit. The early warning signal generation unit generates visualized early warning information (including pollutant type, concentration value, and early warning level) based on the output of the third early warning and control model and pushes it to the monitoring terminal (such as the display screen of the industrial control center or the mobile terminal of the management personnel). The instruction encoding unit converts the real-time process adjustment instructions into control signals that can be recognized by the actuators. Communication transmission module: Adopting 5G communication technology, a dedicated communication link is established to ensure data transmission rate and command issuance timeliness, while supporting encrypted data transmission to ensure industrial control security; The execution control module has a built-in drive circuit and feedback acquisition unit. The drive circuit is directly connected to the actuators of the aluminum ash disposal device (exhaust gas purification spray pump, ventilation regulating valve, burner power controller, bag filter differential pressure regulator) to realize the execution of control signals. The feedback acquisition unit collects the real-time operating status of the actuators (such as flow rate, opening degree, power, etc.) through sensors and transmits the feedback data to the model calculation module to form a closed-loop control.

[0040] Based on Examples 1 and 2, this example verifies the practicality, accuracy, and stability of the gas sensor-based aluminum ash disposal waste gas monitoring and early warning method and system proposed in this application. A 50,000-ton-per-year aluminum ash disposal production line of an aluminum company was selected as the verification scenario. An experimental platform including a spectral sensor, an electrochemical sensor, an environmental sensor, and a 5G edge computing node was built and operated continuously for 30 days (720 hours), covering the entire process stage of aluminum ash disposal (feeding, combustion, purification, and emission). Real-time monitoring data, early warning response data, and equipment control data were collected simultaneously. The system was compared and verified with a traditional monitoring system (based on a single electrochemical sensor and fixed threshold early warning). The verification indicators included the accuracy of pollutant concentration quantification, early warning response timeliness, control closed-loop effect, and equipment operation stability.

[0041] Validation of the accuracy of pollutant concentration quantification: Concentration quantification accuracy is a core performance indicator of a monitoring system. This verification compares the pollutant concentration accuracy output by the second optimized model of this invention with that of the pollutant quantification model. The relative error (RE), mean relative error (MRE), and coefficient of determination (R²) were calculated by comparing the measured concentrations of the sample with those of an online gas chromatograph (detection accuracy ±0.05 ppm, used as a standard reference). 2 The results were compared with those of traditional systems. The experiment selected fluorides (HF) and nitrogen oxides (NH4O2). ), particulate matter ( Three target pollutants were sampled, and one set of valid data was recorded every hour, for a total of 720 sets of samples. The statistical results are shown in Table 1.

[0042] Table 1 Comparison of Pollutant Concentration Quantification Accuracy

[0043] As shown in Table 1, the average relative error of the system in this application for the three pollutants is controlled within 3.5%, the coefficient of determination is greater than 0.987, and the maximum relative error does not exceed 5.82%. In contrast, the average relative error of traditional systems is generally above 8%, and the maximum relative error exceeds 15%. This indicates that the coupled model of DBN and LSSVM in this application effectively achieves dimensionality reduction and accurate fitting of high-dimensional features, significantly improving the concentration quantification accuracy and providing reliable data support for subsequent early warning and control. To intuitively demonstrate the quantification effect, fitting curves of the quantified concentration and measured concentration of the three pollutants are plotted, as shown in Table 1. Figure 5 As shown, the figure contains 6 subgraphs, namely HF, , The three fitted curves and their corresponding confidence intervals (95%) are presented in the form of a scatter plot combined with the fitted line. All three curves are close to the ideal fitted line. Furthermore, the scatter points around the fitted line are evenly distributed with no significant deviation. It can be seen that the quantification results of this application almost coincide with the measured values, with a narrow confidence interval, while the fitted curves of traditional systems deviate significantly and have a wide confidence interval, further verifying the superiority of the concentration quantification of this application.

[0044] Verification of the timeliness and accuracy of early warning response: The timeliness of the early warning response directly affects the timeliness of handling after the exhaust gas exceeds the standard, while the accuracy avoids false and missed warnings. This verification statistically analyzed the early warning response time (the time from the concentration exceeding the standard to the early warning signal being pushed to the monitoring terminal) under different pollution conditions, and calculated the early warning accuracy rate (number of correct warnings / total number of warnings × 100%), false warning rate (number of false warnings / total number of warnings × 100%), and missed warning rate (number of missed warnings / total number of exceedances × 100%). The verification data are shown in Table 2. In the experiment, by adjusting the aluminum ash feeding rate and combustion power, three typical operating conditions were simulated: normal operating condition (concentration below 80% of the dynamic threshold), slightly exceeding the standard operating condition (concentration between 80% and 120% of the dynamic threshold), and severely exceeding the standard operating condition (concentration above 120% of the dynamic threshold). Each operating condition was run continuously for 10 hours, and the relevant early warning data were recorded.

[0045] Table 2 Comparison of Early Warning Performance

[0046] As shown in Table 2, the average warning response time of the system in this application is only 40.6ms, far lower than the 202.5ms of the traditional system. This is attributed to the combination of 5G communication technology and edge computing, enabling local data processing and high-speed transmission. Regarding warning accuracy, the average warning accuracy rate of this application is as high as 97.89%, with a false warning rate of only 2.16% and a missed warning rate as low as 1.56%. In contrast, the warning accuracy rate of the traditional system is less than 85%, and both the false warning rate and the missed warning rate exceed 14%. This is because this application uses a sliding window dynamic threshold and fuzzy PID control logic, which can dynamically adjust the warning threshold according to the statistical characteristics and toxicity level of pollutant concentration, effectively adapting to fluctuations in operating conditions and reducing misjudgments and missed judgments. To demonstrate the adaptability of the dynamic threshold, a graph showing the concentration change and dynamic threshold under slightly exceeding operating conditions is plotted. Figure 6 As shown, with For example, the diagram contains Real-time concentration curve, dynamic threshold curve and the three curves of the traditional fixed threshold line, The real-time concentration change curve exhibits a periodic fluctuation trend, ranging from 1.5 to 13.2 ppm; the dynamic threshold curve... The value is dynamically adjusted based on the real-time concentration mean and standard deviation, with an adjustment range between 2.8 and 12.5 ppm. The real-time concentration curve maintains synchronous fluctuations; the traditional fixed threshold is set at 8.5 ppm. As can be observed from the figure, when the real-time concentration briefly exceeds the fixed threshold but does not exceed the dynamic threshold, the traditional system will generate false warnings (the three false warning points marked in the figure), while the system of this application does not generate false warnings; when the real-time concentration continuously exceeds the dynamic threshold, the system of this application responds quickly to the warning, and the response time is within 45ms. It can be seen that the dynamic threshold can adaptively adjust with concentration fluctuations, while the fixed threshold is prone to generating false warnings or missed warnings when the concentration fluctuations are large.

[0047] Verification of the closed-loop control effect and equipment operational stability: The closed-loop control effect was measured by the concentration recovery time (time from exceeding the standard to recovering to below the dynamic threshold) and the concentration fluctuation amplitude during the recovery process under the excessive operating conditions. The equipment operation stability was evaluated by the fluctuation range of the actuator's operating parameters, energy consumption, and number of failures. The verification data are shown in Table 3. The experiment selected a severe exceeding condition (NO). x The concentration reached 18.5 ppm. The system of this application and the traditional system were used for regulation, and the concentration recovery process and equipment operation data were recorded. As shown in Table 3, the concentration recovery time of the system of this application is only 48.6 s, which is 61.2% shorter than that of the traditional system. The concentration fluctuation range during the recovery process is 1.85 ppm, which is only 42.8% of that of the traditional system. This shows that the fuzzy PID control logic of this application can dynamically adjust the PID parameters to achieve precise regulation of the actuator and quickly stabilize the pollutant concentration. In terms of equipment operation stability, the fluctuation range of the actuator operating parameters of the system of this application is significantly smaller than that of the traditional system. The average daily energy consumption is reduced by 24.2%, and there are no equipment failures within 30 days. In contrast, the traditional system has 5 failures (2 spray pump overloads and 3 ventilation valve jams). This shows that the process adjustment command of this application strictly matches the rated working range of the actuator and forms a closed-loop control through the feedback acquisition unit to avoid equipment overload and malfunction.

[0048] Table 3 Comparison of Control Effect and Equipment Operation Stability

[0049] To demonstrate the dynamic characteristics of the control process, concentration recovery curves and actuator control curves under severely excessive operating conditions are plotted, such as... Figure 7 As shown, the figure contains Concentration recovery curve, spray pump flow control curve, and ventilation valve opening control curve are shown in the figure. The concentration recovery curve decreased rapidly from an initial 18.5 ppm, dropping to 10.2 ppm (below the dynamic threshold) at 48.6 s, and then stabilized between 8.5 and 10.0 ppm. The spray pump flow rate control curve gradually increased to 78 L / min, with a smooth adjustment process and no abrupt changes, and a fluctuation range of ±5.2 L / min. The ventilation valve opening control curve gradually increased to 68%, also with a smooth adjustment and a fluctuation range of ±3.1%. The spray pump flow rate control curve, ventilation valve opening control curve, and concentration recovery curve showed good synergy, demonstrating the effectiveness of closed-loop control.

[0050] To further verify the training effect of the model, a pre-training iteration curve of the DBN model was plotted, as shown below. Figure 8 As shown in the figure, the reconstruction error curve and the training time curve are included. The reconstruction error curve shows a rapid decrease in error during the first 100 iterations, from an initial 0.156 to 0.032. The error decreases slowly from 100 to 500 iterations, and stabilizes at around 0.021 after 500 iterations without significant fluctuations. The training time curve shows a linear increase with the number of iterations. The training time is 118.5s after 500 iterations and 235.2s after 1000 iterations, indicating that the model has high training efficiency and good convergence.

[0051] Plot the kernel function parameter optimization curves of the LSSVM model, such as... Figure 9 As shown in the figure, different kernel function parameters are included. The prediction error curve corresponding to the value shows a trend of first decreasing and then increasing. When the value is 0.8, the prediction error reaches its minimum of 0.018%. At 0.8, the prediction error increases with... The increase followed by a rapid decrease indicates that the kernel function's mapping capability is gradually enhanced; At 0.8, the prediction error increases with... The increase is gradual and indicates overfitting; therefore, cross-validation was used to select the appropriate method. The rationale for 0.8; Plot the parameter correction curves for fuzzy PID control, such as... Figure 10 As shown, the figure contains , , Three correction curves The correction curve rises as the concentration deviation increases, with a maximum correction of 1.82, and gradually decreases as the deviation decreases. The trend of the correction curve and Consistent, but with smaller adjustments, the maximum correction being 0.45; The correction curve rises as the deviation change rate increases, with a maximum correction of 0.98. All three curves show no abnormal fluctuations and are smoothly adjusted. This indicates that the correction amount is dynamically adjusted with the concentration deviation and the deviation change rate. The adjustment range is reasonable and there are no abnormal fluctuations, ensuring the stability and accuracy of the control.

[0052] In summary, through 30 days of continuous operation and verification, the system of this application is significantly superior to the traditional system in terms of pollutant concentration quantification accuracy, early warning response timeliness and accuracy, control closed-loop effect and equipment operation stability. All performance indicators meet the actual needs of aluminum ash disposal waste gas monitoring and early warning, and have good engineering application value.

[0053] The above are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. For those skilled in the art, the present invention can have various modifications and variations. Any changes, modifications, substitutions, integrations, and parameter changes made to these embodiments within the spirit and principles of the present invention, without departing from the principles and spirit of the present invention, through conventional substitutions or to achieve the same function, fall within the scope of protection of the present invention.

Claims

1. A method for monitoring and warning of aluminum dross disposal exhaust gas of a gas sensor, characterized by, include: Acquire multi-dimensional targeted data of waste gas from the entire aluminum ash disposal process, including first core data and second core data; The first analytical model is used to extract features and remove redundancy from the first core data and the second core data to obtain the set of core parameters of the exhaust gas state. The first analytical model is constructed based on the physical and chemical characteristics of pollution and the dynamic spatiotemporal correlation mechanism of multi-source signals. The dynamic adaptation spatiotemporal correlation mechanism of the first analytical model is realized through a spatiotemporal weight matrix. First, the signal correlation benchmark weight is determined based on the physical and chemical characteristics of pollution, and then dynamically adjusted in combination with the temporal correlation and spatial distribution characteristics of the data. The core parameter set of the exhaust gas state is input into the second optimization model to obtain the accurate quantification parameter set of pollutants. The second optimization model is constructed based on deep belief network and least squares support vector machine. In the second optimization model, the Deep Belief Network (DBN) contains multiple Restricted Boltzmann Machine (RBM) layers and an output layer. The number of neurons in each RBM layer is configured in descending order of gradient. First, the core parameter set of the exhaust gas state is pre-trained in an unsupervised manner. During the pre-training process, the contrastive divergence algorithm is used to minimize the reconstruction error. Then, high-dimensional feature dimensionality reduction and reconstruction are achieved through supervised fine-tuning. The second optimization model inputs the reconstructed features output by DBN into the least squares support vector machine (LSSVM), and achieves accurate fitting of pollutant concentrations through radial basis function mapping. The fitting formula is as follows: ； in, For the first Accurately quantify pollutant concentrations at all times. For the Lagrange multipliers of LSSVM, For DBN output Features are constantly being reconstructed. For the first Reconstructed features of each training sample For kernel function parameters, For bias terms, This represents the number of training samples; The precise quantification parameter set of the pollutants is input into the third early warning and control model to generate multi-level early warning signals and real-time process adjustment instructions. The third early warning and control model is constructed based on fuzzy PID control logic and sliding window dynamic threshold algorithm. By pushing multi-level early warning signals to the monitoring terminal through 5G and edge computing nodes, and simultaneously issuing real-time process adjustment instructions to the actuators of the aluminum ash disposal device, real-time monitoring, dynamic early warning and closed-loop control of waste gas can be achieved.

2. The method for monitoring and early warning of aluminum ash disposal waste gas using a gas sensor according to claim 1, characterized in that, The targeted data includes a first core data and a second core data. The first core data consists of spectral absorption signals, electrochemical response signals, and instantaneous concentration fluctuation data of fluorides, nitrogen oxides, and particulate matter in the waste gas. The second core data consists of gradual temperature and humidity change data, stable air pressure values, airflow velocity vector data, and equipment operating load time series data in the treatment scenario. The instantaneous concentration fluctuation data is quantified by the sliding standard deviation, and the equipment operating load time series data is the coupled data of the real-time power and feeding rate of the aluminum ash treatment furnace.

3. The method for monitoring and early warning of aluminum ash disposal waste gas using a gas sensor according to claim 1, characterized in that, The formula for feature extraction and redundancy removal is as follows: ； in, For the first One core parameter of exhaust gas status, For the first The first core data of the class and the first Dynamically adapting weights for the second core data of the class. For the feature values ​​of the corresponding data, The covariance of this feature with other features, It is the maximum value of the covariance of all features. Number is the first core data type. It is the second core data type.

4. The method for monitoring and early warning of aluminum ash disposal waste gas using a gas sensor according to claim 3, characterized in that, The dynamic adaptation weight Based on the dynamic calculation of pollution physicochemical characteristics and data credibility, data credibility is quantified by signal-to-noise ratio. Pollution physicochemical characteristics include the spectral absorption cross section, electrochemical activity, solubility and reactivity of pollutants. The calculation of dynamic adaptation weights must meet the normalization constraint of all data association weights.

5. The method for monitoring and early warning of aluminum ash disposal waste gas using a gas sensor according to claim 1, characterized in that, The sliding window dynamic threshold of the third early warning and control model is determined by the statistical characteristics of recent pollutant concentrations, and the window length is adaptively set according to the aluminum ash treatment process cycle. The dynamic early warning threshold formula is: ； in, For the first Real-time dynamic warning thresholds The average concentration within the sliding window. The standard deviation of concentration within the sliding window. The early warning sensitivity coefficient is adjusted in a positive correlation with the toxicity level of the pollutant.

6. The method for monitoring and early warning of aluminum ash disposal waste gas using a gas sensor according to claim 1, characterized in that, The fuzzy PID control logic of the third early warning and control model includes two input variables and three output variables. The input variable is the concentration deviation. and the rate of change of deviation The output variables are PID parameter correction values: proportional coefficient correction value, integral coefficient correction value, and derivative coefficient correction value. Both input and output variables are configured with multiple fuzzy subsets. The parameter correction values ​​are generated through a preset fuzzy rule base. The fuzzy rule base is defuzzified using the centroid method to obtain the final correction value.

7. The method for monitoring and early warning of aluminum ash disposal waste gas using a gas sensor according to claim 1, characterized in that, The actuators of the aluminum ash disposal device include a waste gas purification spray pump, a ventilation regulating valve, a burner power controller, and a bag filter differential pressure regulator. The generation of real-time process adjustment commands must be combined with environmental emission standards and equipment operation constraints, and the command parameters must match the rated working range of the actuators to ensure safe equipment operation.

8. A gas sensor-based aluminum ash disposal waste gas monitoring and early warning system, used to implement the method according to any one of claims 1-7, characterized in that, It includes a data acquisition module, a model calculation module, an early warning and control module, a communication transmission module, and an execution control module; The data acquisition module integrates a spectral sensor, an electrochemical sensor, a temperature and humidity sensor, a barometric pressure sensor, a wind speed sensor, and an equipment load monitoring unit. Each sensor collects data synchronously at a uniform sampling frequency to meet the monitoring accuracy requirements. The model computation module is deployed on edge computing nodes and supports online model parameter updates and offline training. The communication transmission module adopts 5G communication technology to ensure data transmission rate and timely command issuance. The execution control module has a built-in drive circuit and feedback acquisition unit, which can collect the operating status of the actuator in real time and feed it back to the model calculation module to form a closed-loop control.

Images