Online monitoring method and system for multiple pollutants in flue gas of small coke oven

Abstract

The invention relates to the technical field of small coke oven pollution monitoring, in particular to an online monitoring method and system for multiple pollutants in flue gas of a small coke oven, and the method comprises the steps: obtaining real-time working condition data and target pollutant monitoring concentration of the small coke oven according to a preset frequency; inputting the working condition data into a pre-trained pollutant emission concentration prediction model to obtain a pollutant prediction concentration, and matching each pollutant emission safety concentration threshold from a preset risk feature library in combination with the working condition data; then calculating the deviation degree of each pollutant monitoring and prediction concentration relative threshold value, taking the maximum value as the instantaneous risk deviation degree, and synthesizing the deviation degrees of all pollutants to generate a flue gas emission comprehensive risk entropy; and finally, carrying out time sequence change trend analysis on the flue gas emission comprehensive risk entropies corresponding to a plurality of continuous and adjacent moments to obtain small coke oven pollution standard exceeding risk early warning information, and prompting the small coke oven pollution standard exceeding risk early warning information. According to the invention, accurate and timely early warning of the multi-pollutant emission risk of the small coke oven flue gas is realized, and ineffective alarm interference is avoided.

Description

Technical Field

[0001] This invention relates to the technical field of pollution monitoring in small coke ovens, and in particular to an online monitoring method and system for multiple pollutants in flue gas from small coke ovens. Background Technology

[0002] As an important piece of equipment in the coking production of the chemical industry, the small coke oven contains a variety of pollutants such as sulfur dioxide and dust in its flue gas. The production process of the small coke oven is characterized by periodicity and intermittency. Its flue gas emission concentration is strongly correlated with the operating conditions such as coal charging, coke pushing, and oven shut-down, and exhibits rapid and large fluctuations.

[0003] Existing online monitoring methods for multiple pollutants in small coke oven flue gas generally rely on emission standard thresholds based on fixed or simple operating conditions for determining exceedances. These methods monitor pollutant concentrations in real time and directly compare them to preset limits; if the limit is exceeded, an alarm is triggered. However, these methods are highly susceptible to misinterpreting instantaneous concentration peaks caused by normal process switching as abnormal emissions, leading to frequent false alarms. This not only disrupts the continuity of production operations but also means that genuine emission risks may be buried under a large number of invalid alarm messages, reducing the accuracy of early warnings and making it difficult to identify true pollution risks in a timely and precise manner. Summary of the Invention

[0004] This invention provides a method and system for online monitoring of multiple pollutants in flue gas from small coke ovens, which can effectively solve the problems in the background art.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A method for online monitoring of multiple pollutants in flue gas from a small coke oven includes: Based on the preset monitoring frequency, real-time operating condition data of small coke ovens and real-time monitoring concentration values ​​of each target pollutant are obtained. Real-time operating condition data is input into a pre-trained pollutant emission concentration prediction model to obtain real-time predicted concentration values ​​for each target pollutant. Based on real-time operating data, a matching search is performed in a pre-set pollutant risk feature database to determine the emission safety concentration threshold corresponding to each target pollutant at the current moment. For each target pollutant, the deviation between the real-time predicted concentration value and the real-time monitored concentration value and their corresponding emission safety concentration threshold is calculated, and the maximum deviation value is taken as the instantaneous risk deviation of the pollutant. A comprehensive assessment of the instantaneous risk deviation of each target pollutant is conducted to generate the comprehensive risk entropy of flue gas emissions from the small coke oven at the current moment. A time-series variation trend analysis of the comprehensive risk entropy of flue gas emissions at multiple consecutive and adjacent time points was conducted to obtain and provide early warning information on the risk of pollution exceeding standards in small coke ovens.

[0006] Furthermore, real-time operating condition data includes raw material characteristic parameters, equipment service status parameters, production load parameters, and flue gas operating condition parameters collected in real time. The target pollutants include sulfur dioxide, nitrogen oxides, and dust.

[0007] Furthermore, the raw material characteristic parameters include the volatile matter content, ash content, and moisture content of the blended coal; Equipment service status parameters include sampling probe differential pressure, filter membrane dust accumulation saturation, and detector response time; Production load parameters include coal charging rate, coke pushing cycle, and oven shut-in time; flue gas operating parameters include flue gas temperature, pressure, humidity, and flow rate.

[0008] Furthermore, the training methods for the pollutant emission concentration prediction model include: Collect historical operating condition data of small coke ovens and corresponding historical emission concentration data of each target pollutant to construct a sample dataset of operating condition concentrations. Preprocessing of the working condition concentration sample dataset includes outlier removal, data standardization, and feature dimension reduction; Long Short-Term Memory Neural Network was selected as the basic model framework. Preprocessed historical operating condition data was used as input features, and corresponding historical emission concentration data was used as output labels to establish a mapping relationship between operating conditions and pollutant concentrations. The Adam optimizer was used to train the model. By adjusting the number of hidden layer nodes, learning rate and number of iterations, the model's prediction error was made to meet the preset accuracy requirements. The real-time collected operating data, predicted concentration values, and actual monitored concentration values ​​are combined to form a new sample, and the pollutant emission concentration prediction model is updated periodically.

[0009] Furthermore, the real-time operating condition data is preprocessed in the same way as the preprocessing process for the operating condition concentration sample dataset.

[0010] Furthermore, the methods for constructing the pollutant risk characteristic database include: Typical operating condition combinations for small coke ovens are defined by the range of raw material characteristic parameters, the range of equipment service status parameters, the range of production load parameters, and the range of flue gas operating condition parameters. For each typical operating condition combination, and in conjunction with the set pollutant emission standards, the emission risk levels of each target pollutant are classified, including safe level, warning level, and exceedance level. Based on the emission risk level, a corresponding emission safety concentration threshold is set, a mapping relationship between typical operating condition combinations and emission safety concentration thresholds is established, and the relationship is stored in the pollutant risk feature database.

[0011] Furthermore, based on real-time operating condition data, a matching search is performed in a pre-defined pollutant risk characteristic database, including: Extract raw material characteristic parameters, equipment service status parameters, production load parameters, and flue gas condition parameters from real-time operating condition data, determine the parameter range to which each type of parameter belongs, and determine the combination of four types of parameter ranges corresponding to the real-time operating condition. By comparing the four types of parameter range combinations with typical operating condition combinations in the pollutant risk characteristic database, a completely consistent typical operating condition combination was obtained. If the real-time operating parameters are within the boundary of a certain parameter range, the nearest matching principle is used to match to the closest parameter range; if the real-time operating parameters exceed all preset parameter ranges, it is determined to be an abnormal operating condition, and the closest typical operating condition combination is matched and marked as an abnormal condition.

[0012] Furthermore, a time-series trend analysis was conducted on the comprehensive risk entropy of flue gas emissions at multiple consecutive and adjacent time points, including: Select a sliding time window of a preset duration to obtain the time series data of comprehensive risk entropy of flue gas emissions within the time window; Calculate the slope and variance of time series data. The slope represents the trend of risk entropy change, and the variance represents the stability of risk entropy. If the slope of change is greater than the preset slope threshold and the variance of fluctuation is greater than the preset variance threshold, it is determined to be an upward trend in the risk of pollution exceeding the standard. If the slope of change is less than zero and the absolute value is greater than the preset slope threshold, it is determined to be a downward trend in the risk of pollution exceeding the standard.

[0013] Furthermore, a comprehensive assessment is conducted on the instantaneous risk deviation of each target pollutant, including: Based on the pollution control requirements for small coke oven flue gas, the environmental hazard level of each target pollutant and historical emission data, the risk weight of each target pollutant is determined. The weighted instantaneous risk deviation of each target pollutant is obtained by multiplying the instantaneous risk deviation of each pollutant by its corresponding risk weight. The weighted instantaneous risk deviations of all target pollutants are superimposed to obtain the comprehensive risk entropy of flue gas emissions from the small coke oven at the current moment.

[0014] On the other hand, the present invention also provides an online monitoring system for multiple pollutants in flue gas from small coke ovens, comprising: The data acquisition module is used to collect real-time operating condition data of the small coke oven and real-time monitoring concentration values ​​of each target pollutant based on a preset monitoring frequency. The concentration prediction module is used to input real-time operating condition data into a pre-trained pollutant emission concentration prediction model to obtain the real-time predicted concentration values ​​of each target pollutant. The threshold matching module is used to perform matching and retrieval in a preset pollutant risk feature database based on real-time operating condition data to determine the emission safety concentration threshold corresponding to each target pollutant at the current moment. The deviation calculation module is used to calculate the deviation between the real-time predicted concentration value and the real-time monitored concentration value and their corresponding emission safety concentration threshold for each target pollutant, and take the maximum deviation as the instantaneous risk deviation of the pollutant. The risk entropy generation module is used to comprehensively assess the instantaneous risk deviation of each target pollutant and generate the comprehensive risk entropy of flue gas emissions from the small coke oven at the current moment. The risk warning module is used to analyze the time-series change trend of the comprehensive risk entropy of flue gas emissions at multiple consecutive and adjacent time points, obtain and issue warning information on the risk of pollution exceeding standards in small coke ovens.

[0015] The technical solution of this invention can achieve the following technical effects: By using operating data as a guide to link the concentration prediction of the forecasting model, the matching of dynamic thresholds, and the production process, the standard for judging pollution concentration is transformed from a static fixed value into a dynamic reference system that fits the actual process, thereby eliminating the basis for misjudgment caused by instantaneous peaks during process switching. By calculating the deviation between real-time predicted concentration and real-time monitored concentration, the model's predicted value compensates for the accidental fluctuation deviation of a single monitoring value, while the actual monitoring value reflects the true emission status. The two are cross-verified, and extreme values ​​are taken to quantify instantaneous risk, improving the objectivity and accuracy of judging abnormal deviations in pollutant concentration. By quantifying and integrating the instantaneous risk at a single moment into a comprehensive risk entropy, a normalized assessment of the risks of multiple pollutants can be achieved, and the overall pollution risk can be coordinated. At the same time, by conducting time-series trend analysis on the comprehensive risk entropy at continuous moments, risk judgment is upgraded from single-point-of-moment judgment to time-series trend analysis, accurately identifying potential exceedance risks with slowly rising concentrations, false risks with short-term fluctuations, and real exceedance risks with instantaneous changes. It can identify immediate exceedances, predict potential risks, and effectively filter out random fluctuations without trends, realizing proactive early warning, trend analysis, and comprehensive assessment of multiple pollutants.

[0016] 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 to 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 following are specific embodiments of this application. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention 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 only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a schematic flowchart of an online monitoring method for multiple pollutants in flue gas from a small coke oven according to the present invention. Figure 2 This is a schematic diagram of the structure of an online monitoring system for multiple pollutants in flue gas from a small coke oven, according to the present invention. Detailed Implementation

[0019] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0020] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0021] like Figure 1 As shown, the present invention provides an online monitoring method for multiple pollutants in flue gas from a small coke oven, which specifically includes the following steps: Step S1: Based on the preset monitoring frequency, obtain the real-time operating condition data of the small coke oven and the real-time monitoring concentration values ​​of each target pollutant; Step S2: Input the real-time operating condition data into the pre-trained pollutant emission concentration prediction model to obtain the real-time predicted concentration values ​​of each target pollutant. Step S3: Based on real-time operating data, perform matching and retrieval in the preset pollutant risk feature database to determine the emission safety concentration threshold corresponding to each target pollutant at the current moment. Step S4: For each target pollutant, calculate the deviation between the real-time predicted concentration value and the real-time monitored concentration value and their corresponding emission safety concentration threshold, and take the maximum deviation as the instantaneous risk deviation of the pollutant. Step S5: Conduct a comprehensive assessment of the instantaneous risk deviation of each target pollutant to generate the comprehensive risk entropy of flue gas emissions from the small coke oven at the current moment; Step S6: Perform time-series change trend analysis on the comprehensive risk entropy of flue gas emissions corresponding to multiple consecutive and adjacent time points to obtain and issue early warning information on the risk of pollution exceeding standards in small coke ovens.

[0022] In this embodiment, by combining a prediction model driven by operating condition data with a dynamic safety threshold that matches the operating condition, the judgment of pollution exceeding the standard is adapted to the periodic and intermittent production process of small coke ovens, reducing invalid alarms and highlighting the real pollution risks. At the same time, by combining the calculation of dual concentration deviation and the time-series trend analysis of comprehensive risk entropy, the limitations of single concentration comparison and single-moment judgment are overcome, realizing multi-dimensional and full-time-series accurate identification of pollution risks, and simultaneously improving the accuracy and timeliness of early warning.

[0023] In a specific implementation, as one example, given the periodic and intermittent production characteristics of small coke ovens, the concentration of flue gas pollutants fluctuates rapidly and significantly with changes in operating conditions. Therefore, it is necessary to simultaneously collect multi-dimensional operating condition data and pollutant concentration data, and to match the concentration fluctuation cycle to set the monitoring frequency to ensure data validity. This example achieves synchronous collection of operating condition and concentration data through unified time-series control, selects targeted operating condition parameters, and controls the monitoring frequency based on concentration fluctuation characteristics, ensuring that the data remains consistent and completely correlated over time. Specifically, as follows: Step S11: Determine the monitoring frequency based on the instantaneous switching characteristics of small coke oven coal charging and coke pushing operations. The monitoring frequency should be able to capture the complete fluctuation curve of pollutant concentration during a single operation switching process, avoiding the omission of key data due to too low a frequency, or data redundancy and increased processing load due to too high a frequency. The selection principle is to match the pollutant concentration fluctuation cycle. For example, a monitoring frequency of 50Hz can cover the concentration surge process within 1-2 seconds during the coal charging stage, while controlling the data volume within the equipment's processing capacity. A unified time sequence control mechanism can be adopted, that is, to control all acquisition devices to start and stop synchronously according to the preset frequency, realize monitoring frequency coordination, and ensure that the operation data and pollutant concentration data correspond one-to-one in the time dimension, eliminating the asynchronous data deviation caused by inconsistent equipment acquisition rhythms. Step S12: The target pollutants are identified as sulfur dioxide, nitrogen oxides, and particulate matter. Sulfur dioxide originates from the combustion and conversion of sulfides in the coal. Nitrogen oxides are generated by the thermal and fuel reactions during combustion. Particulate matter mainly consists of coal dust, coke dust, and ash particles. All three are key pollutants controlled in the coking industry, and their concentration changes are strongly correlated with the operating conditions of small coke ovens, reflecting the degree of flue gas pollution. Real-time concentration values ​​of each target pollutant are collected using corresponding online detection equipment. Sulfur dioxide and nitrogen oxides are detected using an ultraviolet differential absorption spectrometer. This instrument is installed at the sampling point of the main flue gas pipeline of the small coke oven. The concentration is calculated by the characteristic absorption intensity of ultraviolet light of a specific wavelength by the pollutant. This method can resist interference from complex components in the flue gas and avoid distortion of the concentration detection value due to non-specific absorption or scattering. Particulate matter is detected using a laser scattering method instrument. This instrument emits a laser to irradiate dust particles in the flue gas and uses the laser signal scattered by the particles to convert it into dust concentration. It is installed at the same cross section as the previous two instruments to avoid deviations caused by differences in sampling locations. Step S13: Determine the real-time operating condition data as raw material characteristic parameters, equipment service status parameters, production load parameters, and flue gas operating condition parameters, which are respectively associated with pollutant generation mechanism, monitoring equipment reliability, production load intensity, and flue gas environmental characteristics. All of these directly affect the generation and monitoring accuracy of pollutant concentration and reflect the real-time operating status of the small coke oven. The raw material characteristic parameters include the volatile matter content, ash content, and moisture content of the blended coal, as the components of the blended coal directly determine the amount of pollutants generated during combustion: volatile matter content affects the intensity of combustion; higher volatile matter content makes it easier to generate localized high temperatures, promoting increased nitrogen oxide generation. Incomplete combustion of volatile matter also leads to increased dust emissions. Ash content is one of the main sources of dust, and the ash percentage is directly related to the baseline dust emission. Moisture content affects combustion temperature and flue gas humidity, thus indirectly altering the generation efficiency of sulfur dioxide and nitrogen oxides, as well as the adsorption characteristics of dust. Volatile matter content is collected using a near-infrared online detector installed at the end of the blended coal conveyor belt. This detector scans the coal material on the belt surface in real time, and the data is converted using the correlation between near-infrared spectral characteristics and volatile matter content. Ash and moisture content are simultaneously collected using a microwave moisture and ash analyzer installed downstream of the near-infrared detector. After the microwave signal penetrates the coal material, the ash and moisture percentages are calculated based on the signal attenuation. The equipment's service status parameters include sampling probe differential pressure, filter membrane dust accumulation saturation, and detector response time. The purpose is to eliminate the influence of the monitoring equipment's own condition on the accuracy of concentration data: Sampling probe differential pressure reflects the degree of probe blockage; blockage leads to insufficient flue gas sampling, resulting in lower concentration monitoring values. Filter membrane dust accumulation saturation affects dust filtration efficiency; excessively high saturation allows dust to penetrate the filter membrane, causing distortion in dust monitoring values. Detector response time reflects equipment sensitivity; abnormal response leads to delayed or inaccurate concentration data. Sampling probe differential pressure is acquired through a differential pressure sensor, with the sensor connected to the probe's inlet and outlet, directly outputting a differential pressure signal. When the differential pressure exceeds a preset range, the corresponding concentration data is marked as suspicious. Filter membrane dust accumulation saturation is derived from the change in differential pressure before and after the filter membrane, calculated by combining dust monitoring concentration and sampling time, and compared with the filter membrane's rated capacity. Detector response time is obtained through periodic standard gas injection calibration. After each operating condition switch, the corresponding pollutant standard gas is injected, and the duration of the detector outputting a stable concentration value is recorded as the response time parameter for that period. Production load parameters, including coal charging rate, coke pushing cycle, and oven shut-in time, are adapted to the intermittent production characteristics of small coke ovens and directly determine the reaction conditions inside the furnace: The coal charging rate affects the bulk density of coal and the combustion space inside the furnace; too fast a rate can easily cause insufficient local oxygen supply, leading to increased sulfur dioxide emissions, while too slow a rate prolongs the coal charging time, causing periodic fluctuations in concentration; the coke pushing cycle reflects the coking time of coal inside the furnace, and periodic changes will alter the completeness of coal combustion, thus affecting the amount of pollutants generated; the oven shut-in time determines the degree of combustion of residual combustible components inside the furnace, and insufficient time will... This results in unburned components being discharged with the flue gas, accompanied by an increase in dust and nitrogen oxide concentrations; the coal charging rate is collected in real time by a belt scale, which is linked to the coal conveying belt to convert the weight of coal conveyed per unit time into the coal charging rate; the coke pushing cycle is recorded by a coke pushing car stroke sensor, marking the start time of two coke pushing operations and calculating the time interval as the coke pushing cycle; the oven shut-in time is determined by an oven temperature sensor. When coke pushing is completed, the oven temperature stabilizes within a preset range and there is no coal charging operation, the timer starts and stops when the next coal charging operation begins; Flue gas operating parameters include flue gas temperature, pressure, humidity, and flow velocity. The physical state of the flue gas itself affects the accuracy of pollutant concentration monitoring and its formation characteristics: temperature affects pollutant activity; nitrogen oxide formation reactions are more intense at high temperatures, which also alters the detection accuracy of the instrument; pressure affects the flue gas sampling volume, and pressure fluctuations can lead to deviations in the calculated pollutant concentration per unit volume; humidity causes dust particles to absorb moisture, altering the dust particle size distribution and affecting dust concentration monitoring results; humidity changes also interfere with the detection signals of sulfur dioxide and nitrogen oxides; and flow velocity determines the residence time of flue gas in the pipeline, affecting the pollutant emission rate, and flow velocity instability... This will inevitably lead to uneven concentration distribution at sampling points; therefore, a combination of multi-point acquisition and averaging is used to obtain data. Multiple sets of sensors are arranged at the same cross-section of the sampling point in the main flue gas pipeline to collect temperature, pressure, humidity, and flow velocity data respectively. Temperature is collected by thermocouple sensors, which are suitable for high-temperature flue gas environments; pressure is collected by pressure transmitters, which can withstand corrosive components in flue gas; humidity is collected by resistive-capacitive humidity sensors, which have the ability to resist dust interference; and flow velocity is collected by Pitot hydrostatic tube flow meters, which are suitable for flue gas velocity fluctuations in the pipeline. The spatial distribution differences are eliminated by arithmetic mean calculation of the acquired sensor data to obtain representative values ​​of the operating parameters at that cross-section.

[0024] In this embodiment, raw material characteristic parameters are used to provide a component basis for predicting pollutant generation, equipment service status parameters are used to provide a basis for concentration data calibration, production load parameters are used to provide support for operating condition identification and risk threshold matching, and flue gas operating condition parameters are used to provide conditions for concentration monitoring accuracy correction and generation characteristic analysis. The interference of time deviation on model training can be eliminated through complete and synchronous data acquisition.

[0025] In some embodiments of the present invention, the reason why existing methods cannot distinguish between instantaneous peak concentrations caused by normal process switching and actual excessive emissions, and which judge exceedances by comparing real-time concentrations with fixed thresholds, is that they have not established a mapping relationship between operating conditions and pollutant concentrations, and therefore cannot predict the concentration fluctuation patterns corresponding to changes in operating conditions. Based on this, this embodiment constructs a pollutant emission concentration prediction model, trains the model with historical operating conditions and concentration data, and realizes real-time prediction of pollutant concentrations under changing operating conditions. At the same time, a periodic iteration mechanism is designed to ensure that the model adapts to the periodic and intermittent changes in the operating conditions of small coke ovens. Specifically, the following operations are performed: Step S21: Invoke the pre-trained pollutant emission concentration prediction model; the training method of this model includes: a. Collect historical operating condition data of small coke ovens and corresponding historical emission concentration data of each target pollutant to construct a sample dataset of operating condition concentrations. The collected data covers the entire production cycle of small coke ovens, including all typical operating conditions such as coal charging, coke pushing, and oven shut-down, ensuring that the sample dataset can reflect the pollutant concentration variation patterns under different operating condition combinations. The historical operating condition data is completely consistent with the data type of the real-time collected operating condition data, including raw material characteristic parameters, equipment service status parameters, production load parameters, and flue gas operating condition parameters. The corresponding data sources are the same as those for real-time acquisition, all taken from the historical storage data of the small coke oven production control system and online monitoring equipment, ensuring data consistency and correlation. The historical emission concentration data of each target pollutant corresponds to sulfur dioxide, nitrogen oxides, and dust, and is taken from the historical records of the same online monitoring equipment as the real-time monitoring. Each set of historical operating condition data corresponds to the pollutant concentration data at the same moment, avoiding sample failure due to data misalignment. The operating condition type at the time of data collection is recorded synchronously during the collection process. b. Preprocessing of the working condition concentration sample dataset includes outlier removal, data standardization, and feature dimension reduction to ensure the accuracy and standardization of the sample data input to the model and avoid invalid data interfering with model training. Outlier removal targets abnormal data caused by equipment failure, sampling deviation, or data transmission errors during the collection process. The judgment criteria combine the small coke oven operating logic and data rationality. For example, differential pressure data collected when the sampling probe fails and concentration data recorded during the calibration of the detection equipment are all identified as outliers and removed. Data in the concentration data that exceeds the reasonable fluctuation range under normal operating conditions and has no corresponding operating condition changes to support it are also removed as outliers. After removal, missing data is supplemented by interpolation of data from adjacent time points to ensure the integrity of the sample dataset. Data standardization... Standardization is used to eliminate dimensional differences between different types of parameters. For example, the moisture content in raw material characteristic parameters and the flow rate in flue gas operating parameters have large dimensional differences, which can lead to an imbalance in feature weights. The standardization process uses a uniform scaling method to adjust all sample data to the same numerical range, ensuring that the influence of each feature parameter on model training is balanced. Feature dimension reduction is used to address redundant features in operating parameters. Some operating parameters are correlated, such as the positive correlation between coal charging rate and production load. Redundant features will increase the computational load of model training and may also cause overfitting. The dimension reduction process analyzes the correlation between each operating parameter and pollutant concentration, retains feature parameters with strong correlation that can directly affect concentration changes, eliminates redundant features, simplifies the model structure, and improves model training efficiency. c. A Long Short-Term Memory (LSTM) neural network is selected as the basic model framework. Preprocessed historical operating condition data is used as input features, and corresponding historical emission concentration data is used as output labels to establish a mapping relationship between operating conditions and pollutant concentrations. The small coke oven production process is periodic and intermittent, and there is a clear temporal dependency between changes in operating conditions and changes in pollutant concentrations. For example, the cumulative change in the oven simmering time will gradually affect subsequent concentration fluctuations. Traditional neural network models cannot capture this long-term temporal correlation, which can easily lead to prediction bias. The LTM neural network has a gating mechanism, which can effectively capture the long-term dependency in the time series data, adapt to the temporal characteristics of small coke oven operating conditions and concentration changes, and can accurately learn the corresponding rules between different combinations of operating conditions and pollutant concentrations. In the process of establishing the mapping relationship, the preprocessed historical operating condition data is used as the input layer, and a hidden layer is set to extract the correlation features between operating condition features and concentration changes. The output layer corresponds to the concentration prediction values ​​of three target pollutants. Independent mapping relationships are established between raw material characteristics, equipment status, production load, flue gas conditions and sulfur dioxide, nitrogen oxides, and dust concentrations, respectively, to ensure that the prediction of each target pollutant can be specifically adapted to its concentration change pattern. d. The Adam optimizer is used to train the model. By adjusting the number of hidden layer nodes, learning rate, and number of iterations, the model's prediction error is made to meet the preset accuracy requirements. The Adam optimizer combines the advantages of momentum gradient descent and adaptive learning rate adjustment, which can alleviate the gradient vanishing or gradient exploding problems during model training, adapt to the training requirements of long short-term memory neural networks, and has a stable convergence speed, which can reduce parameter oscillations during training and ensure the stability of model training. During model training, the preprocessed working condition concentration sample dataset is divided into a training set and a validation set. The training set is used for model parameter learning, and the validation set is used to verify the model's prediction accuracy in real time. After each iteration, the prediction error on the validation set is calculated, and the model parameters are adjusted based on the error results. The number of hidden layer nodes is adjusted to adapt to the feature extraction requirements. Too few nodes will lead to insufficient feature extraction and insufficient prediction accuracy, while too many nodes will increase model complexity and cause overfitting. The learning rate is adjusted to control the parameter update speed. Too high a learning rate will cause the model to converge and oscillate, failing to reach the preset accuracy, while too low a learning rate will prolong the training cycle and reduce training efficiency. The number of iterations is adjusted until the prediction error on the validation set stabilizes within the preset range, ensuring that the model can accurately learn the mapping relationship between working conditions and concentration, and that the prediction results can reflect the true concentration change pattern. e. Combine the real-time collected operating data with the predicted concentration values ​​and the actual monitored concentration values ​​to form a new sample, and periodically iterate and update the pollutant emission concentration prediction model. During the long-term operation of small coke ovens, equipment aging, changes in raw material composition, and adjustments to production processes may occur, causing changes in the mapping relationship between operating conditions and concentration. If the pre-trained model is not updated for a long time, the prediction accuracy will decrease and it will be unable to adapt to new changes in operating conditions. The periodic iteration update cycle is set in conjunction with the small coke oven production cycle. For example, an iteration is performed after each complete production cycle. During the update process, new samples are added to the original sample dataset, and the model is re-preprocessed and retrained. The model parameters are adjusted so that the model can adapt to the concentration change patterns caused by changes in operating conditions, continuously maintain the preset prediction accuracy, and ensure the accuracy of real-time predicted concentration values. Step S22: Preprocess the real-time operating condition data. This preprocessing process is the same as the preprocessing process of the operating condition concentration sample dataset during model training. This ensures that the data format and numerical range are consistent with the model training samples, and avoids prediction deviations caused by non-standard data. Step S23: Input the preprocessed real-time operating condition data into the pre-trained pollutant emission concentration prediction model. The model outputs the real-time predicted concentration values ​​of each target pollutant at the corresponding time by using the learned mapping relationship between operating conditions and concentration. The prediction process is carried out simultaneously with real-time monitoring to ensure the timeliness of the predicted concentration.

[0026] In this embodiment, the real-time predicted concentration value obtained by the model, in conjunction with the real-time monitored concentration value, can be used to calculate the deviation and distinguish between the concentration peak caused by the switching of normal operating conditions and the actual excessive emissions. Through the application of long short-term memory neural networks, it can be ensured that the prediction model is adapted to the time-series operating characteristics of small coke ovens and the prediction accuracy meets the requirements. Through a regular iterative update mechanism, the model can adapt to long-term changes in operating conditions, continuously output accurate prediction results, and ensure the accuracy of risk warning.

[0027] In a specific implementation, as one example, existing methods use a fixed or simply adjusted uniform emission standard threshold when determining the emission safety concentration threshold for each target pollutant at the current moment. This fails to consider the real-time operating conditions of the small coke oven at the current moment, nor does it distinguish the concentration variation patterns of each target pollutant. Consequently, it cannot match the reasonable concentration fluctuations of each target pollutant under the current operating conditions, resulting in a lack of accuracy and specificity in determining the emission safety concentration threshold for each target pollutant at the current moment. This embodiment constructs a pollutant risk feature database, establishes a mapping between operating condition combinations and thresholds by dividing typical operating condition combinations and setting graded safety concentration thresholds. Then, through matching and searching the feature database with real-time operating conditions, it determines the dynamic threshold corresponding to the current operating condition, achieving operating condition adaptability for exceeding the standard judgment and avoiding false alarms. The specific implementation steps are as follows: Step S31: Call the preset pollutant risk feature library; the method for constructing this feature library includes: a. A typical operating condition combination for small coke ovens is defined, which is jointly calibrated by raw material characteristic parameter ranges, equipment service status parameter ranges, production load parameter ranges, and flue gas condition parameter ranges. These four types of parameters jointly determine the real-time operating status of the small coke oven and directly affect the pollutant generation and monitoring accuracy. A single parameter or a partial parameter range cannot fully characterize the operating condition characteristics; only through the combination of these four parameter ranges can different operating scenarios be accurately defined. The division of each parameter range is based on the historical operating data of the small coke oven and production process requirements, combined with the range of various parameters collected in real time. Continuous intervals are divided according to the parameter variation patterns and the degree of influence on pollutant concentrations: Raw material characteristic parameter ranges are for the volatile matter content, ash content, and moisture content of the blended coal, divided according to the normal fluctuation range of coal components and the differences in the influence of different components on pollutant generation. For example, volatile matter content is divided into low, medium, and high ranges, corresponding to different potentials for nitrogen oxide and dust generation; Equipment service status... The parameter ranges are divided according to the normal operating thresholds and performance degradation patterns of the equipment, targeting the differential pressure of the sampling probe, the cumulative dust saturation of the filter membrane, and the detector response time. For example, the differential pressure of the sampling probe is divided into normal, slightly blocked, and severely blocked ranges, corresponding to different sampling accuracies. The production load parameter ranges are divided according to the coal charging rate, coke pushing cycle, and oven shut-in time, based on the conventional operating conditions of small coke ovens during periodic production. For example, the coke pushing cycle is divided into short, medium, and long ranges, corresponding to different coal coking saturation. The flue gas operating condition parameter ranges are divided according to the flue gas temperature, pressure, humidity, and flow rate, based on the conventional operating range within the flue gas pipeline and the differences in their impact on pollutant concentration monitoring. For example, the flue gas temperature is divided into low temperature, normal temperature, and high temperature ranges, corresponding to different pollutant activity and detection accuracy. Each typical operating condition combination is a unique combination of the four parameter ranges, covering all typical operating conditions of small coke ovens, such as coal charging, coke pushing, and oven shut-in, ensuring that any real-time operating condition can find a matching typical combination in the feature library. b. For each typical operating condition combination, and in conjunction with the established pollutant emission standards, classify the emission risk levels of each target pollutant. These risk levels include safe, warning, and exceedance levels. The established pollutant emission standards adopt the current national or industry emission standards for small coke oven flue gas pollutants as the benchmark for risk level classification, ensuring that the classification complies with environmental control requirements. The safe level corresponds to scenarios where pollutant generation is low, concentration is stable, and far below emission standards under the operating condition combination, requiring no warning to be triggered. The safe level concentration range is the interval where historical concentration distribution is concentrated and below emission standards by a certain range, ensuring that pollutant concentrations remain stable at safe levels under this operating condition combination. The warning level corresponds to scenarios where pollutant concentrations are close to emission standards under the operating condition combination, posing a potential exceedance risk but not yet meeting the exceedance requirements. Staff should be alerted to these situations; the warning level concentration range is between the upper limit of the safety level and the emission standard. Within this range, pollutant concentrations are close to exceeding the standard, and monitoring changes in operating conditions is necessary to avoid exceeding the standard; the exceeding level corresponds to a scenario where pollutant concentrations exceed the emission standard under the corresponding operating condition combination, posing a real risk of pollution emissions, requiring immediate triggering of the exceeding level warning; the exceeding level concentration range is a range higher than the emission standard, where pollutant concentrations have reached the exceeding requirements, requiring immediate control measures; the risk level classification for each target pollutant corresponding to each typical operating condition combination is conducted independently. For example, the typical combination corresponding to the coal charging condition has a different sulfur dioxide warning level concentration range than the typical combination corresponding to the furnace shut-in condition, adapting to the pollutant generation patterns under different operating conditions to ensure the targeted nature of the risk level classification; c. Based on the risk levels, set emission safety concentration thresholds for each target pollutant under each typical operating condition combination. The emission safety concentration threshold for the safe level is set as the upper limit of the safe level concentration range under that operating condition combination, ensuring that pollutant emissions are in a safe state when the concentration is below this threshold. The emission safety concentration threshold for the warning level is set as the upper limit of the warning level concentration range under that operating condition combination, i.e., setting the concentration value corresponding to the pollutant emission standard, ensuring that a warning is triggered when the concentration reaches this threshold. The emission safety concentration threshold for the exceeding level is set as the lower limit of the exceeding level concentration range under that operating condition combination, i.e., setting the concentration value corresponding to the pollutant emission standard, consistent with the upper limit of the warning level, ensuring that an exceeding warning is triggered when the concentration exceeds this threshold. d. Establish a mapping relationship between typical operating condition combinations and emission safety concentration thresholds, and store it in the pollutant risk feature database. The mapping relationship is established in the form of key-value pairs, with typical operating condition combinations as keys and emission safety concentration thresholds corresponding to different risk levels of each target pollutant under the combination as values, to ensure that the corresponding thresholds can be quickly retrieved through typical combinations. Step S32: Based on real-time operating condition data, perform matching and retrieval in the preset pollutant risk feature database. The specific retrieval method is as follows: extract four types of parameters from the real-time operating condition data: raw material characteristics, equipment service status, production load, and flue gas conditions. Determine the parameter range to which each type of parameter belongs, determine the combination of the four parameter ranges corresponding to the real-time operating condition, and then compare this combination with the typical operating condition combinations stored in the pollutant risk feature database to find a completely consistent typical combination. If the real-time operating condition parameter is at the boundary of a certain parameter range, the nearest matching principle is adopted to match to the closest parameter range, ensuring that any real-time operating condition can complete the matching retrieval and avoiding matching failure due to parameter boundary issues. If the real-time operating condition parameter exceeds all preset parameter ranges, it is determined to be an abnormal operating condition, and the closest typical operating condition combination is matched. At the same time, the operating condition is marked as abnormal, prompting staff to investigate the cause of the abnormal operating condition. Step S33: Using the matched typical operating condition combination, retrieve the emission safety concentration thresholds for different risk levels of each target pollutant corresponding to the combination from the pollutant risk feature database. Combine this with the preliminary judgment of the real-time monitoring concentration at the current moment to determine the safety concentration threshold for the corresponding risk level. If the real-time monitoring concentration is lower than the safety level threshold, retrieve the safety level emission safety concentration threshold as the current threshold. If the real-time monitoring concentration is between the safety level and the warning level threshold, retrieve the warning level emission safety concentration threshold as the current threshold. If the real-time monitoring concentration is higher than the warning level threshold, retrieve the exceeding level emission safety concentration threshold as the current threshold.

[0028] In this embodiment, by dividing typical operating condition combinations, it is ensured that matching thresholds can be found for all real-time operating conditions, adapting to the fluctuating characteristics of small coke oven operating conditions; by setting three risk levels and corresponding thresholds, combined with historical data and emission standards, it is ensured that the threshold settings are accurate and comply with environmental control requirements; by constructing and matching the pollutant risk feature database, the thresholds can be quickly retrieved, meeting the timeliness requirements of real-time monitoring; by determining dynamic thresholds, it is avoided that instantaneous concentration peaks caused by normal process switching are misjudged as exceeding standards, while ensuring that the real risk of exceeding emission standards can be accurately identified.

[0029] In a specific implementation, as one example, existing methods calculate deviations by simply comparing real-time monitored concentration values ​​with fixed thresholds during the determination of instantaneous risk deviation. This fails to meet the dual requirements of operational condition prediction and actual monitoring, and is prone to misjudgment of risk due to instantaneous fluctuations in monitoring data or prediction deviations. Therefore, this embodiment calculates the deviations between the real-time predicted concentration value, the real-time monitored concentration value, and the corresponding emission safety concentration threshold for each target pollutant, selecting the maximum value of the two as the instantaneous risk deviation for that pollutant. This achieves comprehensive and accurate capture of instantaneous risk, and is implemented through the following steps: Step S41: For each target pollutant, extract the real-time predicted concentration value, the real-time monitored concentration value, and the emission safety concentration threshold corresponding to the pollutant at the current moment, ensuring the temporal consistency and correspondence of the three sets of data. The real-time predicted concentration value reflects the predicted level of pollutant concentration under the current operating conditions. Based on the model output trained and iteratively updated using real-time operating data of the small coke oven, it can reflect the trend of the impact of operating condition changes on pollutant concentration. The real-time monitored concentration value reflects the actual emission concentration of the pollutant at the current moment. It comes directly from the real-time acquisition of online detection equipment and can reflect the true emission status of the pollutant. The emission safety concentration threshold is a dynamic threshold adapted to the pollutant under the current operating conditions. It is obtained based on typical operating condition combinations and can reflect the reasonable concentration control boundary of the pollutant under the current operating conditions. All three sets of data correspond to the same target pollutant and the same moment. The temporal consistency can be ensured through a unified temporal control mechanism to avoid calculation deviations caused by data misalignment. Step S42: For each target pollutant, calculate the first deviation between the real-time predicted concentration value and the corresponding safe emission concentration threshold, and the second deviation between the real-time monitored concentration value and the corresponding safe emission concentration threshold. The deviation is used to quantify the degree of deviation between the pollutant concentration and the current operating condition adaptation threshold. The degree of deviation directly reflects the instantaneous risk level of the pollutant; the larger the deviation, the further the pollutant concentration deviates from the reasonable control boundary, and the higher the instantaneous risk. The first deviation captures the predicted deviation of the pollutant concentration under the changing trend of the operating condition, and can predict the possible deviation risk of the pollutant concentration in advance, compensating for the real-time... The monitoring data may contain lags or instantaneous errors. For example, when a small coke oven switches from a shut-in operation to a coal charging operation, the real-time monitoring concentration value may not respond promptly to a sudden increase in concentration, while the real-time predicted concentration value has already made a prediction based on the change in operating conditions. The first deviation can detect such risks in advance. The second deviation captures the deviation between the actual emission concentration of pollutants and the current threshold, which can reflect the true emission risk of pollutants and make up for the model bias that may exist in the real-time predicted concentration value. For example, when the online detection equipment is operating normally and the monitoring data is accurate, the second deviation can accurately reflect whether the actual emission of pollutants is close to or exceeds the control boundary. Step S43: For each target pollutant, compare the values ​​of the first deviation and the second deviation, and select the maximum value of the two as the instantaneous risk deviation of the pollutant. The real-time predicted concentration value and the real-time monitored concentration value reflect the deviation between the pollutant concentration and the threshold from the two dimensions of predicted trend and actual state, respectively. The maximum value of the two can capture the highest instantaneous risk of the pollutant at the current moment, avoiding risk omission caused by single-dimensional calculation.

[0030] In this embodiment, by processing target pollutants independently, the deviation calculation of each pollutant is ensured to be consistent with its own characteristics, avoiding cross-interference; by calculating the two deviations separately, both the predicted trend and the actual emission status are taken into account, which can make up for the limitations of single data and avoid risk omission; by selecting the maximum value, the highest instantaneous risk of each pollutant can be captured, ensuring the comprehensiveness of risk assessment; by extracting data with consistent time sequence, calculation deviations caused by data misalignment are avoided, ensuring accurate results.

[0031] In a specific implementation, as one example, to take into account the risk differences and synergistic effects of various pollutants, it is necessary to comprehensively assess the instantaneous risk deviation of each target pollutant and introduce reasonable quantitative indicators to accurately characterize the risk of synergistic emissions of multiple pollutants from the small coke oven at the current moment. This example comprehensively assesses the instantaneous risk deviation of each target pollutant and introduces the comprehensive risk entropy of flue gas emissions as a quantitative indicator of comprehensive risk to quantify the risk of synergistic emissions of multiple pollutants from the small coke oven at the current moment. This approach takes into account the independent risk characteristics of each pollutant while also intuitively reflecting the overall risk level. The specific implementation steps are as follows: Step S51: Determine the risk weight of each target pollutant to distinguish the degree of influence of different pollutants in the overall risk. The weight value is determined based on the pollution control requirements of small coke oven flue gas, the environmental hazard degree of each pollutant, and historical emission data. The weight value can be dynamically adjusted according to the adjustment of small coke oven production process, the update of emission standards, and changes in surrounding environmental control requirements. Step S52: Multiply the instantaneous risk deviation of each target pollutant by its corresponding risk weight to obtain the weighted instantaneous risk deviation of each pollutant; this value can reflect the contribution of a single pollutant to the current comprehensive risk. The larger the weighted value, the greater the impact of the pollutant on the current comprehensive risk. Step S53: Superimpose the three sets of weighted instantaneous risk deviations to obtain the comprehensive risk entropy value; the comprehensive risk entropy can transform multiple independent single pollutant risk indicators into an intuitive and quantifiable comprehensive indicator, which can reflect both the independent risk level of each pollutant and the synergistic effect of the risks of each pollutant, and is suitable for the risk characteristics of multi-pollutant synergistic emissions from small coke ovens; at the same time, the quantitative characteristics of the comprehensive risk entropy can accurately distinguish the differences in comprehensive risk at different times.

[0032] In this embodiment, the determination of risk weights is aligned with actual control requirements and pollutant characteristics, distinguishing the degree of impact of different pollutants to ensure the pertinence of the comprehensive assessment, and dynamically adjusting the characteristics to adapt to changes in operating conditions and control requirements; the weighted instantaneous risk deviation is superimposed to generate a comprehensive risk entropy, achieving accurate quantification of the comprehensive risk of multiple pollutants, taking into account both single pollutant risk and synergistic risk, and avoiding the defects of simple superposition.

[0033] In practical implementation, as one example, existing methods only make static judgments on risks at a single moment or judge changes by simply comparing the comprehensive risk entropy of two adjacent moments. This fails to capture the dynamic development trend of risks, resulting in early warnings lagging behind actual risk changes, or misjudging normal fluctuations as risk trends, making it difficult to achieve timely and accurate control of pollution exceeding standards risks in small coke ovens. Based on this, this embodiment selects a sliding time window of a preset duration to obtain time-series data of comprehensive risk entropy of flue gas emissions at consecutive adjacent moments, calculates the slope and variance of the time-series data, and combines it with a preset threshold to judge the time-series change trend of pollution exceeding standards risks, generating accurate early warning information and providing prompts. This adapts to the periodic and intermittent characteristics of small coke oven operating conditions and can achieve early capture of risk trends. The specific implementation steps are as follows: Step S61: Select a sliding time window of a preset duration and acquire time-series data of the comprehensive risk entropy of flue gas emissions within the time window. The sliding time window can continuously and in real-time capture time-series data of adjacent moments, track the continuous changes in the comprehensive risk entropy, adapt to the periodic fluctuations of small coke oven operating conditions, and avoid the lag in trend analysis caused by the inability of a fixed window to update data in real time. The preset duration of the sliding time window is determined based on the production cycle of the small coke oven, the characteristics of pollutant concentration fluctuations, and historical records of exceeding standards. For example, a duration equivalent to one shutdown cycle of the small coke oven can be selected as the basic duration of the window to ensure that the window can contain a sufficient number of data. Continuous monitoring captures the comprehensive risk entropy change pattern corresponding to a single operating condition fluctuation, while avoiding excessively long window durations that lead to lag in trend analysis or excessively short window durations that result in insufficient data and biased trend judgments. The sliding step size of the sliding time window is consistent with the preset monitoring frequency. That is, after each real-time monitoring is completed and a new comprehensive risk entropy data is generated, the sliding time window slides forward by one monitoring moment, discarding the earliest comprehensive risk entropy data in the window and incorporating the latest generated comprehensive risk entropy data. This ensures that the window always maintains continuous adjacent time-series data of the preset duration, achieving real-time trend analysis. Step S62: Calculate the slope and variance of the time-series data. The slope represents the trend of risk entropy change, i.e., whether the overall risk entropy is rising, falling, or stable. The slope is based on the time-to-value correspondence of the overall risk entropy time-series data within the window. By quantifying the magnitude of change in the overall risk entropy at each adjacent time point, the overall slope is fitted. When the slope is positive, it indicates that the overall overall risk entropy within the window is rising; the larger the slope value, the faster the rate of increase. When the slope is negative, it indicates that the overall overall risk entropy within the window is falling; the slope is absolute. The larger the value, the faster the rate of decline; when the slope is close to zero, it indicates that the overall risk entropy within the window is generally stable, without significant increase or decrease; the volatility variance characterizes the stability of the risk entropy, that is, the degree of fluctuation of the overall risk entropy under the changing trend; the volatility variance is calculated based on the dispersion of the time series data of the overall risk entropy within the window, and is obtained by quantifying the deviation of the overall risk entropy at each time point from the mean within the window. The smaller the variance value, the smaller the fluctuation of the overall risk entropy within the window and the more stable it is overall; the larger the variance value, the greater the fluctuation of the overall risk entropy within the window and the less stable it is overall. Step S63: Set slope threshold and variance threshold as the basis for judging the trend of pollution exceedance risk. Both thresholds are determined based on the time series data of historical comprehensive risk entropy of small coke ovens, historical pollution exceedance risk records, periodic characteristics of small coke oven operating conditions, and the set pollutant emission standards to ensure the scientific and targeted nature of the threshold setting. The preset slope threshold is divided into positive threshold and negative threshold, with the same absolute value, and is used to judge whether the upward or downward trend of comprehensive risk entropy reaches the level that requires early warning. The preset variance threshold is a single positive threshold and is used to judge whether the fluctuation of comprehensive risk entropy exceeds the normal operating condition range. Step S64: Compare the calculated slope and variance with preset slope thresholds and preset variance thresholds, respectively, and determine the temporal trend of pollution exceedance risk in small coke ovens based on the comparison results; the determination logic strictly follows preset conditions: If the slope of change is greater than the preset positive slope threshold and the variance of fluctuation is greater than the preset variance threshold, it indicates that the comprehensive risk entropy is not only showing a rapid upward trend, but also fluctuating violently, exceeding the normal operating condition fluctuation range. It is judged as an upward trend of pollution exceeding the standard. At this time, there is a risk that the pollutant emission concentration will continue to rise, will soon exceed the standard, or has already exceeded the standard. It is necessary to trigger the upward trend warning in time. If the slope of change is less than the preset negative slope threshold, that is, the absolute value of the slope is greater than the preset slope threshold, regardless of whether the variance of fluctuation is greater than the preset threshold, it indicates that the comprehensive risk entropy is showing a rapid downward trend and the risk of pollution exceeding the standard is continuously decreasing. It is judged as a downward trend of pollution exceeding the standard. At this time, a downward trend prompt needs to be triggered, and the risk is being alleviated. If the slope of change is within the preset slope threshold range, or the variance of fluctuation is less than or equal to the preset variance threshold, it indicates that the comprehensive risk entropy does not have a significant rapid upward or downward trend, or the fluctuation is within the normal operating range, and it is judged as a stable trend of pollution exceeding the standard risk, without the need to trigger an early warning. Step S65: Based on different trend judgment results, generate corresponding early warning information for pollution exceeding standards in small coke ovens. The content of the early warning information should be tailored to the needs of industrial production control and management, and may include the trend type, the mean value of the comprehensive risk entropy within the current sliding time window, the slope value of change, and the variance value of fluctuation. At the same time, the corresponding monitoring time and the real-time operating condition type of the small coke oven should be marked, so that staff can quickly grasp the risk dynamics and the current operating condition background and accurately determine whether control measures need to be taken. The early warning information can be prompted by pop-up prompts on industrial control terminals, on-site audible and visual alarms, and SMS notifications to staff mobile phones to ensure that staff can receive the early warning information in a timely manner and avoid control delays caused by untimely prompts.

[0034] In this embodiment, the selection of the sliding time window and the sliding method are adapted to the periodicity and intermittency of the small coke oven operating conditions, ensuring the real-time and comprehensiveness of trend analysis and avoiding trend analysis lag; the calculation of the slope of change can accurately quantify the trend and rate of change of the comprehensive risk entropy, avoiding misjudgment caused by data fluctuations at a single moment; the calculation of the fluctuation variance can distinguish between normal operating condition fluctuations and abnormal risk fluctuations, improving the accuracy of trend judgment; the setting of the preset threshold is based on historical data and operating condition characteristics, which can ensure the relevance and scientific nature of the threshold; and the multiple warning prompts can ensure the timely transmission of warning information.

[0035] Based on the same inventive concept as the online monitoring method for multiple pollutants in small coke oven flue gas described in the foregoing embodiments, this invention also provides an online monitoring system for multiple pollutants in small coke oven flue gas, such as... Figure 2 As shown, the system includes: The data acquisition module is used to collect real-time operating condition data of the small coke oven and real-time monitoring concentration values ​​of each target pollutant based on a preset monitoring frequency. The concentration prediction module is used to input real-time operating condition data into a pre-trained pollutant emission concentration prediction model to obtain the real-time predicted concentration values ​​of each target pollutant. The threshold matching module is used to perform matching and retrieval in a preset pollutant risk feature database based on real-time operating condition data to determine the emission safety concentration threshold corresponding to each target pollutant at the current moment. The deviation calculation module is used to calculate the deviation between the real-time predicted concentration value and the real-time monitored concentration value and their corresponding emission safety concentration threshold for each target pollutant, and take the maximum deviation as the instantaneous risk deviation of the pollutant. The risk entropy generation module is used to comprehensively assess the instantaneous risk deviation of each target pollutant and generate the comprehensive risk entropy of flue gas emissions from the small coke oven at the current moment. The risk warning module is used to analyze the time-series change trend of the comprehensive risk entropy of flue gas emissions at multiple consecutive and adjacent time points, obtain and issue warning information on the risk of pollution exceeding standards in small coke ovens.

[0036] The system described above in this invention can effectively realize an online monitoring method for multiple pollutants in flue gas from small coke ovens, and the technical effects it can achieve are as described in the above embodiments, and will not be repeated here.

[0037] Although this application has been described in conjunction with specific features and embodiments, it is obvious that various modifications and combinations can be made thereto without departing from the spirit and scope of this application. Accordingly, this specification and drawings are merely exemplary illustrations of the application as defined herein, and are to be considered as covering any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Thus, if such modifications and modifications fall within the scope of this application and its equivalents, this application intends to include such modifications and modifications.

Claims

1. A method for online monitoring of multiple pollutants in flue gas from a small coke oven, characterized in that, include: Based on the preset monitoring frequency, real-time operating condition data of small coke ovens and real-time monitoring concentration values ​​of each target pollutant are obtained. The real-time operating condition data is input into the pre-trained pollutant emission concentration prediction model to obtain the real-time predicted concentration values ​​of each target pollutant. Based on the real-time operating condition data, a matching search is performed in the preset pollutant risk feature database to determine the emission safety concentration threshold corresponding to each target pollutant at the current moment. For each target pollutant, the deviation between the real-time predicted concentration value and the real-time monitored concentration value and their corresponding emission safety concentration threshold is calculated, and the maximum deviation is taken as the instantaneous risk deviation of the pollutant. A comprehensive assessment of the instantaneous risk deviation of each target pollutant is conducted to generate the comprehensive risk entropy of flue gas emissions from the small coke oven at the current moment. A time-series change trend analysis of the comprehensive risk entropy of flue gas emissions corresponding to multiple consecutive and adjacent time points is performed to obtain and issue early warning information on the risk of pollution exceeding standards in small coke ovens.

2. The online monitoring method for multiple pollutants in flue gas from a small coke oven according to claim 1, characterized in that, The real-time operating condition data includes raw material characteristic parameters, equipment service status parameters, production load parameters, and flue gas operating condition parameters collected in real time. The target pollutants include sulfur dioxide, nitrogen oxides, and dust.

3. The online monitoring method for multiple pollutants in flue gas from a small coke oven according to claim 2, characterized in that, The raw material characteristic parameters include the volatile matter content, ash content, and moisture content of the blended coal; The equipment service status parameters include sampling probe differential pressure, filter membrane dust accumulation saturation, and detector response time. The production load parameters include coal charging rate, coke pushing cycle, and oven shut-in time; the flue gas operating parameters include flue gas temperature, pressure, humidity, and flow rate.

4. The online monitoring method for multiple pollutants in flue gas from a small coke oven according to claim 3, characterized in that, The training method for the pollutant emission concentration prediction model includes: Collect historical operating condition data of small coke ovens and corresponding historical emission concentration data of each target pollutant to construct a sample dataset of operating condition concentrations. The aforementioned working condition concentration sample dataset is preprocessed, including outlier removal, data standardization, and feature dimension reduction. Long Short-Term Memory Neural Network was selected as the basic model framework. Preprocessed historical operating condition data was used as input features, and corresponding historical emission concentration data was used as output labels to establish a mapping relationship between operating conditions and pollutant concentrations. The Adam optimizer was used to train the model. By adjusting the number of hidden layer nodes, learning rate and number of iterations, the model's prediction error was made to meet the preset accuracy requirements. The real-time collected operating data, predicted concentration values, and actual monitored concentration values ​​are combined to form a new sample, and the pollutant emission concentration prediction model is updated periodically.

5. The online monitoring method for multiple pollutants in flue gas from a small coke oven according to claim 4, characterized in that, The preprocessing of real-time operating condition data is consistent with the preprocessing process of the operating condition concentration sample dataset.

6. The online monitoring method for multiple pollutants in flue gas from a small coke oven according to claim 2, characterized in that, The method for constructing the pollutant risk characteristic database includes: Typical operating condition combinations for small coke ovens are defined by the range of raw material characteristic parameters, the range of equipment service status parameters, the range of production load parameters, and the range of flue gas operating condition parameters. For each typical operating condition combination, and in conjunction with the set pollutant emission standards, the emission risk levels of each target pollutant are classified, including safe level, warning level, and exceedance level; Based on the emission risk level, a corresponding emission safety concentration threshold is set, a mapping relationship between typical operating condition combinations and emission safety concentration thresholds is established, and the relationship is stored in the pollutant risk feature database.

7. The online monitoring method for multiple pollutants in flue gas from a small coke oven according to claim 6, characterized in that, The step of matching and searching in a preset pollutant risk feature database based on real-time operating condition data includes: Extract raw material characteristic parameters, equipment service status parameters, production load parameters, and flue gas condition parameters from real-time operating condition data, determine the parameter range to which each type of parameter belongs, and determine the combination of four types of parameter ranges corresponding to the real-time operating condition. The four types of parameter range combinations are compared with typical operating condition combinations in the pollutant risk characteristic database to retrieve completely consistent typical operating condition combinations. If the real-time operating parameters are at the boundary of a certain parameter range, the nearest matching principle is used to match to the closest parameter range; if the real-time operating parameters exceed all preset parameter ranges, it is determined to be an abnormal operating condition, and the closest typical operating condition combination is matched and marked as an abnormal operating condition.

8. The online monitoring method for multiple pollutants in flue gas from a small coke oven according to claim 1, characterized in that, The analysis of the time-series variation trend of the comprehensive risk entropy of flue gas emissions corresponding to multiple consecutive and adjacent time points includes: Select a sliding time window of a preset duration to obtain the time series data of comprehensive risk entropy of flue gas emissions within the time window; Calculate the slope of change and the variance of volatility of the time series data. The slope of change represents the trend of risk entropy change, and the variance of volatility represents the stability of risk entropy. If the slope of change is greater than a preset slope threshold and the variance of fluctuation is greater than a preset variance threshold, it is determined to be an upward trend in the risk of pollution exceeding the standard. If the slope of the change is less than zero and the absolute value is greater than the preset slope threshold, it is determined to be a downward trend in the risk of pollution exceeding the standard.

9. The online monitoring method for multiple pollutants in flue gas from a small coke oven according to claim 1, characterized in that, The comprehensive assessment of the instantaneous risk deviation of each target pollutant includes: Based on the pollution control requirements for small coke oven flue gas, the environmental hazard level of each target pollutant and historical emission data, the risk weight of each target pollutant is determined. The weighted instantaneous risk deviation of each target pollutant is obtained by multiplying the instantaneous risk deviation of each pollutant by its corresponding risk weight. The weighted instantaneous risk deviations of all target pollutants are superimposed to obtain the comprehensive risk entropy of flue gas emissions from the small coke oven at the current moment.

10. An online monitoring system for multiple pollutants in flue gas from a small coke oven, characterized in that, include: The data acquisition module is used to collect real-time operating condition data of the small coke oven and real-time monitoring concentration values ​​of each target pollutant based on a preset monitoring frequency. The concentration prediction module is used to input the real-time operating condition data into the pre-trained pollutant emission concentration prediction model to obtain the real-time predicted concentration value of each target pollutant. The threshold matching module is used to perform matching and retrieval in a preset pollutant risk feature database based on the real-time operating condition data to determine the emission safety concentration threshold corresponding to each target pollutant at the current moment. The deviation calculation module is used to calculate the deviation between the real-time predicted concentration value and the real-time monitored concentration value and the corresponding emission safety concentration threshold for each target pollutant, and take the maximum deviation as the instantaneous risk deviation of the pollutant. The risk entropy generation module is used to comprehensively assess the instantaneous risk deviation of each target pollutant and generate the comprehensive risk entropy of flue gas emissions from the small coke oven at the current moment. The risk warning module is used to perform time-series change trend analysis on the comprehensive risk entropy of flue gas emissions corresponding to multiple consecutive and adjacent time points, obtain early warning information on pollution exceeding standards in small coke ovens, and provide alerts.

Images