A method and system for monitoring industrial waste gas based on intelligent sensing

Abstract

This invention provides an industrial waste gas monitoring method and system based on intelligent sensing. The method includes: collecting waste gas composition data and gas flow rate data, and determining predicted waste gas composition data through a waste gas treatment efficiency prediction model; determining a waste gas treatment time window, and then determining waste gas composition treatment efficiency indicators and waste gas early warning indicators, generating waste gas treatment prompt information and determining the waste gas early warning type. According to this invention, by setting waste gas composition sensors at multiple stages and considering the impact of gas flow rate on treatment efficiency, the waste gas composition treatment efficiency indicators can be determined through a waste gas treatment efficiency prediction model. This provides a basis for judging whether there are faults in each stage, and can also calculate waste gas early warning indicators to determine whether the overall waste gas treatment status meets the standards. In the event that waste gas treatment does not meet the standards, a prompt message can be generated, indicating the faulty stage, which is beneficial for accurately troubleshooting and reducing pollutant emissions.

Description

Technical Field

[0001] This invention relates to the field of waste gas monitoring technology, and in particular to an industrial waste gas monitoring method and system based on intelligent sensing. Background Technology

[0002] In related technologies, monitoring of industrial waste gas usually only involves setting up sensors at the waste gas emission outlet for monitoring, that is, directly monitoring the industrial waste gas emitted into the atmosphere. However, the treatment of industrial waste gas may involve multiple links. Problems or malfunctions in any link may lead to the pollutant content in the waste gas emitted into the atmosphere exceeding the standard. However, it is difficult to accurately determine the cause of the excessive pollutant content by only monitoring the industrial waste gas at the outlet, so it is difficult to accurately eliminate the fault and reduce the amount of pollutant emissions. Summary of the Invention

[0003] This invention provides an industrial waste gas monitoring method and system based on intelligent sensing, which can solve the technical problem that related technologies have difficulty in accurately determining the cause of pollutant content exceeding the standard.

[0004] According to a first aspect of the present invention, an industrial waste gas monitoring method based on intelligent sensing is provided, comprising:

[0005] Waste gas composition sensors are installed at the starting point and in multiple stages of industrial waste gas treatment, and a gas flow sensor is installed at the starting point of industrial waste gas treatment.

[0006] At multiple monitoring times, acquire data on various exhaust gas components collected by different exhaust gas component sensors, as well as gas flow data collected by gas flow sensors.

[0007] Based on the gas flow rate data and exhaust gas composition data at the starting end, as well as the trained exhaust gas treatment efficiency prediction model, the predicted exhaust gas composition data at multiple stages are determined.

[0008] Based on the gas flow data, determine the waste gas treatment time window for each stage;

[0009] Based on the predicted data of exhaust gas composition, the time window for exhaust gas treatment, and the data of exhaust gas composition, the efficiency index of exhaust gas composition treatment at each stage is determined.

[0010] Based on the waste gas composition treatment efficiency index, waste gas composition data, and gas flow rate data, determine the waste gas early warning index;

[0011] Based on the waste gas composition treatment efficiency index and waste gas early warning index, determine whether to generate waste gas treatment prompt information and determine the type of waste gas early warning;

[0012] If it is determined that an exhaust gas treatment prompt message will be generated, then the exhaust gas treatment prompt message will be generated according to the exhaust gas warning type.

[0013] According to the present invention, the training steps of the waste gas treatment efficiency prediction model include:

[0014] Set multiple stages of industrial waste gas treatment to an ideal state;

[0015] According to the preset flow rate change pattern, industrial waste gas containing multiple components is input into multiple stages of industrial waste gas treatment.

[0016] At multiple training moments, the exhaust gas composition data and training gas flow data at each stage are detected by exhaust gas composition sensors.

[0017] By using a waste gas treatment efficiency prediction model, the training gas flow rate data and training waste gas composition data at the starting point of industrial waste gas treatment are processed to obtain waste gas composition output data for each stage.

[0018] Based on the training gas flow rate data, the preset flow rate variation law, the training exhaust gas composition data, and the exhaust gas composition output data, the loss function of the exhaust gas treatment efficiency prediction model is determined.

[0019] Based on the loss function of the waste gas treatment efficiency prediction model, the waste gas treatment efficiency prediction model is trained to obtain the trained waste gas treatment efficiency prediction model.

[0020] According to the present invention, a waste gas treatment efficiency prediction model is used to process training gas flow rate data and training waste gas composition data at the starting point of industrial waste gas treatment to obtain waste gas composition output data for each stage, including:

[0021] The training gas flow rate data and training exhaust gas composition data at the beginning of multiple training moments are concatenated to obtain the training exhaust gas state vector for each training moment.

[0022] The training exhaust gas state vector is input into the self-attention mechanism corresponding to the first stage in the exhaust gas treatment efficiency prediction model for processing, so as to obtain the output training exhaust gas state vector at multiple training times corresponding to the first stage.

[0023] The training exhaust gas state vector at the corresponding training time point and the output training exhaust gas state vector at the first stage are concatenated to obtain the concatenated state vector, which is then input into the self-attention mechanism at the second stage for processing to obtain the output training exhaust gas state vectors at multiple training times at the second stage.

[0024] The output training exhaust gas state vector corresponding to the (i-1)th stage and the output training exhaust gas state vector corresponding to the ith stage are concatenated to obtain the concatenated state vector, which is then input into the self-attention mechanism corresponding to the (i+1)th stage for processing to obtain the output training exhaust gas state vectors for multiple training times corresponding to the (i+1)th stage.

[0025] The output training exhaust gas state vectors corresponding to each stage are input into the corresponding multilayer perceptron layers for processing to obtain exhaust gas composition output data at multiple training times for each stage.

[0026] According to the present invention, based on training gas flow rate data, a preset flow rate variation law, training exhaust gas composition data, and exhaust gas composition output data, a loss function for an exhaust gas treatment efficiency prediction model is determined, including:

[0027] Based on the training gas flow data and the preset flow change pattern, determine the training exhaust gas composition data corresponding to the training exhaust gas composition data at the beginning of the process at multiple moments in each stage.

[0028] According to the formula

[0029]

[0030] Determine the loss function LOSS for the waste gas treatment efficiency prediction model, where, Let J be the content of the j-th exhaust gas component in the training exhaust gas component data corresponding to the starting point of the training exhaust gas component data at the t-th training time, which is the i-th stage. The content of the j-th exhaust gas component in the exhaust gas composition output data at the t-th training time of the i-th stage. Let J be the content of the j-th exhaust gas component in the training exhaust gas component data corresponding to the starting point of the training exhaust gas component data at the k-th stage and the t-th training time. Let n be the content of the j-th exhaust gas component in the exhaust gas composition output data at the t-th training time of the k-th stage, m be the number of training time points detected at the starting end, and N be the number of stages. When i=1, ... Set to 1.

[0031] According to the present invention, the waste gas treatment time window for each stage is determined based on gas flow data, including:

[0032] Based on the gas flow data, determine the gas flow time difference between each stage and the starting time;

[0033] Based on the monitoring time and the gas flow time difference, the expected time of exhaust gas output for each stage is determined.

[0034] Based on the expected time of exhaust gas output, a time window for exhaust gas treatment corresponding to the monitoring time is set.

[0035] According to the present invention, the waste gas component treatment efficiency index for each stage is determined based on waste gas component prediction data, waste gas treatment time window, and waste gas component data, including:

[0036] Based on the exhaust gas composition data at each moment within the exhaust gas treatment time window, obtain the exhaust gas composition vector at each moment.

[0037] Based on the predicted data of exhaust gas composition, obtain the predicted vector of exhaust gas composition;

[0038] Determine the maximum value of the cosine similarity between the exhaust gas component vector and the exhaust gas component prediction vector at each time point, and determine the target exhaust gas component vector corresponding to the maximum value.

[0039] Determine the target waste gas composition categories for each stage of treatment;

[0040] According to the formula

[0041]

[0042] Determine the waste gas composition treatment efficiency index for the i-th stage. ,in, Let x be the exhaust gas component data of the x-th target exhaust gas component category in the target exhaust gas component vector corresponding to the y-th monitoring time in the i-th stage. This refers to the exhaust gas component data of the x-th target exhaust gas component category in the target exhaust gas component vector corresponding to the y-th monitoring time in the (i-1)-th stage. This refers to the predicted exhaust gas composition data for the x-th target exhaust gas component category corresponding to the y-th monitoring time at the ith stage. This refers to the predicted exhaust gas composition data for the x-th target exhaust gas component category corresponding to the y-th monitoring time at the (i-1)-th stage. M represents the number of target exhaust gas component categories processed in the i-th stage, and M represents the number of multiple monitoring times.

[0043] According to the present invention, exhaust gas early warning indicators are determined based on exhaust gas component treatment efficiency indicators, exhaust gas component data, and gas flow rate data, including:

[0044] According to the formula

[0045]

[0046] Determine the early warning index for the xth component of the exhaust gas. ,in, This refers to the exhaust gas composition data for the xth type of exhaust gas component in the final stage, where F represents the gas flow rate. Let x be the allowable emission amount of the xth type of exhaust gas component. Let be the waste gas component treatment efficiency index for the s-th stage, targeting waste gas component category x. The number of stages where the xth type of waste gas component is the target waste gas component category.

[0047] According to a second aspect of the present invention, an industrial waste gas monitoring system based on intelligent sensing is provided, comprising:

[0048] The module is used to install waste gas composition sensors at the starting point and multiple stages of industrial waste gas treatment, and to install a gas flow sensor at the starting point of industrial waste gas treatment.

[0049] The monitoring module is used to acquire multiple exhaust gas component data collected by various exhaust gas component sensors and gas flow data collected by gas flow sensors at multiple monitoring times.

[0050] The prediction module is used to determine the predicted data of the exhaust gas composition at multiple stages based on the gas flow rate data and exhaust gas composition data at the starting end, as well as the trained exhaust gas treatment efficiency prediction model.

[0051] The window module is used to determine the waste gas treatment time window for each stage based on the gas flow data.

[0052] The efficiency module is used to determine the waste gas composition treatment efficiency index for each stage based on waste gas composition prediction data, waste gas treatment time window, and waste gas composition data.

[0053] The early warning module is used to determine the early warning indicators for waste gas based on the waste gas composition treatment efficiency index, waste gas composition data, and gas flow rate data.

[0054] The judgment module is used to determine whether to generate waste gas treatment prompt information and determine the type of waste gas warning based on the waste gas composition treatment efficiency index and waste gas warning index.

[0055] The generation module is used to generate exhaust gas treatment prompt information based on the exhaust gas warning type if it is determined that exhaust gas treatment prompt information should be generated.

[0056] By adopting the above technical solution, the present invention can achieve the following technical effects:

[0057] According to the present invention, by setting exhaust gas composition sensors at multiple stages and considering the impact of gas flow rate on treatment efficiency, an exhaust gas treatment efficiency prediction model can be used to determine the exhaust gas composition treatment efficiency index, thereby providing a basis for judging whether there are faults in each stage. Furthermore, an exhaust gas early warning index can be calculated to determine whether the overall exhaust gas treatment status meets the standards. A prompt message can be generated when exhaust gas treatment fails to meet the standards, and the faulty stage can be identified, accurately pinpointing the cause of pollutant exceedances, which is beneficial for accurately eliminating faults and reducing pollutant emissions. During the processing of the exhaust gas treatment efficiency prediction model, a self-attention mechanism corresponding to multiple stages in the model, as well as vector concatenation processing, can be used to match the processing process of the exhaust gas treatment efficiency prediction model with the actual industrial exhaust gas treatment process. The information considered and applied during the processing also matches the actual processing process. That is, it can comprehensively consider the treatment efficiency of each stage, the exhaust gas state and flow rate at the starting end, and the state of the exhaust gas output from the previous stage to obtain an output training exhaust gas state vector, improving the accuracy and objectivity of the exhaust gas composition output data, as well as its matching degree with the actual exhaust gas treatment process. Furthermore, during training, the influence of the prediction accuracy of each preceding stage can be considered to determine the corrected prediction error for each stage. This ensures that the loss function setting matches the actual factors considered in the waste gas treatment process, thereby improving the accuracy and objectivity of the loss function. This, in turn, enhances the matching between the training process and the actual waste gas treatment process, ultimately improving training efficiency and the accuracy of the waste gas treatment efficiency prediction model. When determining the waste gas component treatment efficiency index, errors in the airflow process can be considered. A waste gas treatment time window can be set, and the actual corresponding moment to the monitoring moment can be found within the waste gas treatment time window. The ratio of the measured decrease to the theoretical decrease at the actual corresponding moment can then be used to determine the waste gas component treatment efficiency index for each stage, improving the accuracy of the waste gas component treatment efficiency index. It also allows for determining the degree of closeness between the actual and theoretical states of each stage, thus accurately identifying whether there are problems or malfunctions in each stage. When determining the waste gas early warning index, it can be determined from two aspects: the weight of waste gas components emitted into the atmosphere per unit time and the efficiency of each stage in the industrial waste gas treatment process. This objectively and accurately reflects the efficiency problems of the treatment stages and the problem of exceeding emission standards. Attached Figure Description

[0058] Figure 1 A schematic flowchart of an industrial waste gas monitoring method based on intelligent sensing according to an embodiment of the present invention is shown as an example.

[0059] Figure 2 A schematic diagram illustrating the determination of exhaust gas composition output data according to an embodiment of the present invention is shown.

[0060] Figure 3A block diagram of an industrial exhaust gas monitoring system based on intelligent sensing according to an embodiment of the present invention is shown as an example. Detailed Implementation

[0061] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0062] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0063] Figure 1 An exemplary flowchart of an industrial waste gas monitoring method based on intelligent sensing according to an embodiment of the present invention is shown, the method comprising:

[0064] Step S1: Install waste gas composition sensors at the starting point and multiple stages of industrial waste gas treatment, and install a gas flow sensor at the starting point of industrial waste gas treatment.

[0065] Step S2: At multiple monitoring times, acquire multiple exhaust gas component data collected by each exhaust gas component sensor and gas flow data collected by the gas flow sensor.

[0066] Step S3: Based on the gas flow rate data and exhaust gas composition data at the starting end, as well as the trained exhaust gas treatment efficiency prediction model, determine the predicted exhaust gas composition data at multiple stages.

[0067] Step S4: Determine the waste gas treatment time window for each stage based on the gas flow data.

[0068] Step S5: Determine the waste gas composition treatment efficiency index for each stage based on the waste gas composition prediction data, waste gas treatment time window, and waste gas composition data.

[0069] Step S6: Determine the waste gas early warning index based on the waste gas component treatment efficiency index, waste gas component data, and gas flow rate data;

[0070] Step S7: Based on the waste gas component treatment efficiency index and waste gas early warning index, determine whether to generate waste gas treatment prompt information and determine the waste gas early warning type;

[0071] Step S8: If it is determined that an exhaust gas treatment prompt message will be generated, then the exhaust gas treatment prompt message will be generated according to the exhaust gas warning type.

[0072] The industrial waste gas monitoring method based on intelligent sensing according to an embodiment of the present invention can determine the waste gas component treatment efficiency index by setting waste gas component sensors at multiple stages and considering the influence of gas flow on treatment efficiency through a waste gas treatment efficiency prediction model. This provides a basis for judging whether there is a fault in each stage. Furthermore, it can also calculate waste gas early warning index to determine whether the overall waste gas treatment status meets the standards. It can generate a prompt message when the waste gas treatment does not meet the standards and can indicate the stage with the fault, accurately find the cause of pollutant exceeding the standard, and help to accurately eliminate the fault and reduce pollutant emissions.

[0073] According to an embodiment of the present invention, in step S1, after the industrial waste gas is generated and collected, before entering the treatment stage, it can serve as the starting point. The multiple stages may include an electrostatic dust removal stage, a combustion stage, an adsorption stage (e.g., using activated carbon adsorption), etc. A gas flow sensor can be set at the starting point, and a waste gas composition sensor can be set at the starting point and at the outlet after each stage to detect waste gas composition data, that is, the content of multiple pollutant components in the waste gas.

[0074] According to one embodiment of the present invention, in step S2, each monitoring moment can be spaced 5 minutes, 10 minutes, etc., and the present invention does not limit the time interval between adjacent monitoring moments. At multiple monitoring moments, data collected by each sensor can be acquired, namely, data on various exhaust gas components collected by the exhaust gas composition sensor, and gas flow data collected by the gas flow sensor.

[0075] According to one embodiment of the present invention, in step S3, the waste gas composition data is related to the type of process. In an example, the purpose of setting up a certain process is to remove a certain waste gas component. After treatment by this process, the composition of that component in the industrial waste gas may decrease. The electrostatic precipitator has a removal effect on solid particulate matter; that is, the electrostatic precipitator takes solid particulate matter as the target waste gas component category. Therefore, after the industrial waste gas passes through the electrostatic precipitator, the content of solid particulate matter in the industrial waste gas will decrease. However, during the treatment process, the treatment effect is related to whether there are problems or malfunctions in the treatment process. For example, if the adsorption capacity of the electrostatic precipitator has reached its limit, this problem will cause a decrease in the treatment efficiency of the electrostatic precipitator for solid particulate matter. Another example is that in the desulfurization and denitrification process, lime slurry can be used to absorb sulfur dioxide in the industrial waste gas. However, if the effective components in the lime slurry in this process have been consumed, the treatment efficiency for sulfur dioxide will decrease. On the other hand, the treatment effect is also related to the gas flow rate. For example, when the gas flow rate is large, the industrial waste gas velocity is also fast, and the treatment time for industrial waste gas in each stage is short. For instance, when the gas velocity is fast, the electrostatic precipitator is less effective at capturing fast-moving solid particles, while when the gas velocity is slow, the electrostatic precipitator is more effective at capturing slow-moving solid particles. For another example, in the desulfurization and denitrification process, if the airflow velocity is slow, the reaction time between sulfur dioxide and lime slurry is longer, the reaction is more complete, and the removal effect of sulfur dioxide is better. Conversely, if the airflow velocity is fast, the reaction time between sulfur dioxide and lime slurry is shorter, the reaction is less complete, and the removal effect of sulfur dioxide is poor.

[0076] According to an embodiment of the present invention, based on the above analysis, the waste gas composition data detected at each stage are related to the stage type, fault condition, and gas flow rate data. Therefore, a waste gas treatment efficiency prediction model can be used to predict the waste gas composition prediction data at each subsequent stage based on the gas flow rate data and the waste gas composition data at the starting end. In this way, by combining the rules of the removal efficiency of each stage under different waste gas compositions and different gas flow rates, the waste gas composition after the subsequent stages are completed can be predicted.

[0077] According to an embodiment of the present invention, the training steps of the waste gas treatment efficiency prediction model include: setting multiple stages of industrial waste gas treatment to an ideal state; inputting industrial waste gas containing multiple components into the multiple stages of industrial waste gas treatment according to a preset flow rate change law; detecting the training waste gas component data and the training gas flow rate data of each stage at multiple training moments using a waste gas component sensor; processing the training gas flow rate data and training waste gas component data at the starting end of industrial waste gas treatment using the waste gas treatment efficiency prediction model to obtain the waste gas component output data of each stage; determining the loss function of the waste gas treatment efficiency prediction model based on the training gas flow rate data, the preset flow rate change law, the training waste gas component data, and the waste gas component output data; and training the waste gas treatment efficiency prediction model based on the loss function to obtain the trained waste gas treatment efficiency prediction model.

[0078] According to one embodiment of the present invention, when training the waste gas treatment efficiency prediction model, training can be carried out under the condition that each link reaches an ideal state (i.e., no problems and no faults), so as to determine the treatment situation when the waste gas flow rate is of various types under the condition that there are no faults and no problems in each link, that is, the predicted data of waste gas composition after treatment in each link.

[0079] According to one embodiment of the present invention, since the flow of gas in each stage requires a certain amount of time, industrial waste gas can be input into multiple stages of industrial waste gas treatment according to a preset flow rate change pattern. For example, the flow rate changes according to a sine curve or a specific function pattern, so that in subsequent stages, the industrial waste gas can be detected based on the flow rate change pattern, and the composition of the industrial waste gas can be determined, thereby determining the treatment status of each stage.

[0080] According to one embodiment of the present invention, training gas flow data at each stage can be detected at multiple training moments, and the flow change pattern can be used to match the training gas flow data detected at multiple moments at each stage with the training gas flow data at the starting end, thereby determining the moment when the same part of industrial waste gas is detected at different stages, and thus determining the effect of the same part of industrial waste gas being treated by each stage.

[0081] According to one embodiment of the present invention, a waste gas treatment efficiency prediction model can be used to process the training gas flow data and training waste gas composition data at the starting end of industrial waste gas treatment to determine the waste gas composition output data of each stage. The waste gas composition output data is the composition prediction data of industrial waste gas belonging to the same part as the industrial waste gas at the starting end. For example, if the industrial waste gas of the same part enters the starting end at the first time, is processed by the first stage at the tenth time, and is processed by the second stage at the twentieth time, then the waste gas composition output data is the predicted value of the waste gas composition at the tenth time at the first stage and the predicted value at the twentieth time at the second stage.

[0082] Figure 2 A schematic diagram illustrating the determination of exhaust gas composition output data according to an embodiment of the present invention is shown.

[0083] According to one embodiment of the present invention, a waste gas treatment efficiency prediction model is used to process training gas flow data and training waste gas composition data at the starting point of industrial waste gas treatment to obtain waste gas composition output data for each stage. This includes: concatenating training gas flow data and training waste gas composition data at the starting point of multiple training times to obtain a training waste gas state vector for each training time; inputting the training waste gas state vector into the self-attention mechanism corresponding to the first stage in the waste gas treatment efficiency prediction model for processing to obtain output training waste gas state vectors for multiple training times corresponding to the first stage; and concatenating the training waste gas state vectors corresponding to the training times and the output training waste gas state vectors corresponding to the first stage. The concatenated state vector is obtained and processed by the self-attention mechanism corresponding to the second stage to obtain the output training exhaust gas state vectors for multiple training times corresponding to the second stage. The output training exhaust gas state vectors corresponding to the (i-1)th stage and the output training exhaust gas state vectors corresponding to the ith stage are concatenated to obtain the concatenated state vector, which is then processed by the self-attention mechanism corresponding to the (i+1)th stage to obtain the output training exhaust gas state vectors for multiple training times corresponding to the (i+1)th stage. The output training exhaust gas state vectors corresponding to each stage are then processed by the multilayer perceptron layer corresponding to each stage to obtain the exhaust gas component output data for multiple training times of each stage.

[0084] According to one embodiment of the present invention, since the influence of gas flow rate is not instantaneous—for example, the gas flow rate at a previous moment may affect the processing efficiency at a next moment. For instance, if the gas flow rate is high and the velocity is fast at a previous moment, consuming more effective components of the lime slurry, even if the gas flow rate is low at a next moment, the processing efficiency may still be low. Therefore, comprehensive training can be performed using training gas flow rate data and training exhaust gas composition data from the beginning of multiple training moments. During training, not only the influence of the current gas flow rate on the processing efficiency can be considered, but the influence of gas flow rates at other moments can also be comprehensively considered, thereby improving the prediction accuracy of exhaust gas composition at each stage.

[0085] According to one embodiment of the present invention, the training gas flow rate data and training exhaust gas composition data at the initial stage can be concatenated to obtain the training exhaust gas state vector at each training moment. This vector can be used to describe the flow rate and composition of the industrial exhaust gas at each training moment, representing the overall condition of the industrial exhaust gas. A multilayer sensing network can be used to perform dimensionality-upgrading processing on the training exhaust gas state vector to more comprehensively describe the characteristics of the industrial exhaust gas from multiple aspects, obtaining the processed training exhaust gas state vector.

[0086] According to one embodiment of the present invention, the self-attention mechanism corresponding to the first step can be used to process the training exhaust state vectors at multiple training times. For example, the training exhaust state vectors at multiple training times are multiplied by the key-value matrix of the self-attention mechanism to obtain the key-value vector corresponding to each training exhaust state vector. The training exhaust state vectors at multiple training times are multiplied by the query matrix of the self-attention mechanism to obtain the query vector corresponding to each training exhaust state vector. The training exhaust state vectors at multiple training times are multiplied by the weight matrix of the self-attention mechanism to obtain the weight vector corresponding to each training exhaust state vector. The query vector corresponding to a training exhaust gas state vector is multiplied by all key-value vectors, and the result is divided by the square root of the dimension of the key-value vectors. A softmax activation function is then applied to obtain the association weights between the training exhaust gas state vector and each training exhaust gas state vector. Using these association weights, the weight vectors corresponding to each training exhaust gas state vector are weighted and summed to obtain the output training exhaust gas state vector corresponding to the training exhaust gas state vector. Similarly, the output training exhaust gas state vectors corresponding to each training exhaust gas state vector can be obtained, which can be used to represent the industrial waste output of the first stage at each training time. The state of the gas is defined, and the training time corresponds to the time of the training exhaust gas state vector. For example, if the training exhaust gas state vector is the state vector of the 1st to 5th training time, then the number of output training exhaust gas state vectors is also 5, corresponding to the state vectors of the 1st to 5th training time respectively. In the example, if the first stage takes 10 time to complete, that is, the exhaust gas at the beginning of the first stage is the same exhaust gas as the exhaust gas at the outlet of the first stage from the 11th to 15th time, then the output training exhaust gas state vectors of the multiple training time corresponding to the first stage are the output training exhaust gas state vectors of the 11th to 15th time.

[0087] According to one embodiment of the present invention, the training exhaust gas state vector at the corresponding training moment and the output training exhaust gas state vector corresponding to the first stage can be concatenated to obtain a concatenated state vector. This concatenated state vector can include the exhaust gas state at the beginning and the exhaust gas state after treatment by the first stage. The exhaust gas treated by the second stage is the exhaust gas after treatment by the first stage, and the treatment efficiency of the second stage is also affected by the treatment efficiency of the first stage. For example, if the second stage is a desulfurization and denitrification stage and the first stage is an electrostatic dust removal stage, if the treatment efficiency of the first stage is low, the exhaust gas output by the first stage will still contain a large number of solid particles. These solid particles may be retained in the lime slurry of the second stage, affecting the treatment of the second stage. Therefore, the vectors before and after treatment by the first stage (i.e., the training exhaust gas state vector and the output training exhaust gas state vector corresponding to the first stage) can be concatenated to include the state of the exhaust gas output by the first stage and the state of the exhaust gas before treatment by the first stage, thereby including the characteristics of the treatment efficiency of the first stage. Furthermore, the self-attention mechanism corresponding to the second stage can be used to process the spliced ​​state vector to obtain the output training exhaust gas state vectors at multiple training times corresponding to the second stage. The times of these output training exhaust gas state vectors still correspond to the times of the training exhaust gas state vectors.

[0088] According to one embodiment of the present invention, generally, the output training exhaust gas state vector corresponding to the (i-1)th stage at the corresponding training time and the output training exhaust gas state vector corresponding to the ith stage are concatenated, so that the concatenated state vector includes the state of the exhaust gas output by the ith stage and the state of the exhaust gas before processing by the ith stage, thereby including the characteristics of the processing efficiency of the ith stage. Since the output training exhaust gas state vector corresponding to the (i-1)th stage takes into account the output training exhaust gas state vectors of the (i-2)th and (i-3)th stages during computation, and the output training exhaust gas state vector corresponding to the (i-2)th stage takes into account the output training exhaust gas state vectors of the (i-3)th and (i-4)th stages during computation, and so on, when the self-attention mechanism corresponding to the (i+1)th stage processes the data, its processing object is the concatenated state vector of the output training exhaust gas state vector corresponding to the ith stage and the output training exhaust gas state vector corresponding to the (i-1)th stage. This allows its computation to take into account the processing efficiency of each previous stage and the exhaust gas state output by the ith stage, thereby obtaining an output training exhaust gas state vector that integrates the processing efficiency of each stage, the exhaust gas state at the starting end, and the flow rate.

[0089] According to one embodiment of the present invention, the output training exhaust gas state vector corresponding to each stage can be input into the multilayer sensing network layer corresponding to each stage for processing to obtain exhaust gas component output data at multiple training times of each stage. The exhaust gas component output data can describe the content data of various exhaust gas components in the form of vectors.

[0090] In this way, the self-attention mechanism corresponding to multiple stages in the waste gas treatment efficiency prediction model, as well as vector concatenation processing, can make the processing process of the waste gas treatment efficiency prediction model match the actual treatment process of industrial waste gas. The information considered and applied in the processing process also matches the actual treatment process. That is, it can comprehensively consider the treatment efficiency of each stage, the waste gas state and flow rate at the starting end, and the state of the waste gas output from the previous stage to obtain the output training waste gas state vector, thereby improving the accuracy and objectivity of the waste gas component output data and the degree of matching with the actual waste gas treatment process.

[0091] According to one embodiment of the present invention, during the training process, the loss function of the waste gas treatment efficiency prediction model can be determined based on the waste gas composition output data and measured data at each stage.

[0092] According to one embodiment of the present invention, the loss function of the waste gas treatment efficiency prediction model is determined based on training gas flow data, a preset flow change law, training waste gas composition data, and waste gas composition output data. This includes: determining, based on the training gas flow data and the preset flow change law, the training waste gas composition data corresponding to the initial training waste gas composition data in the training waste gas composition data at multiple moments in each stage; and determining the loss function LOSS of the waste gas treatment efficiency prediction model according to formula (1).

[0093] (1)

[0094] in, Let J be the content of the j-th exhaust gas component in the training exhaust gas component data corresponding to the starting point of the training exhaust gas component data at the t-th training time, which is the i-th stage. The content of the j-th exhaust gas component in the exhaust gas composition output data at the t-th training time of the i-th stage. Let J be the content of the j-th exhaust gas component in the training exhaust gas component data corresponding to the starting point of the training exhaust gas component data at the k-th stage and the t-th training time. Let n be the content of the j-th exhaust gas component in the exhaust gas composition output data at the t-th training time of the k-th stage, m be the number of training time points detected at the starting end, and N be the number of stages. When i=1, ... Set to 1.

[0095] According to an embodiment of the present invention, in formula (1), Let be the difference between the content of the j-th exhaust gas component in the training exhaust gas component data corresponding to the starting point of the training exhaust gas component data at the i-th stage and the measured content of the j-th exhaust gas component. The total difference between the content of each type of exhaust gas component at each training time of the i-th stage and the measured content can be used to represent the prediction error of the i-th stage. However, as mentioned above, the exhaust gas treatment in the i-th stage is affected by the exhaust gas treatment efficiency of the previous stages, and the prediction process of the exhaust gas component output data in the i-th stage is also affected by the prediction data of the previous stages. Therefore, when determining the prediction error of the i-th stage, the prediction accuracy of the previous stages can also be considered to improve the objectivity of the prediction error of the i-th stage. Let be the relative error between the training exhaust gas composition data and the measured data for the k-th stage preceding the i-th stage. Let be the average relative error of various exhaust gas components at multiple training moments in the k-th stage. Therefore, Let be the average prediction accuracy of various exhaust gas components at multiple training moments in the k-th stage. The product of the accuracy rates of the stages preceding the i-th stage can be used to represent the impact of these stages on the prediction accuracy. Therefore, This can be expressed as the corrected prediction error of the i-th stage, considering the influence of the prediction accuracy of previous stages. Alternatively, it can represent the prediction error of the i-th stage when only the prediction accuracy of previous stages is considered (e.g., the prediction result is a random value). In other words, it represents the prediction error of the i-th stage after excluding the influence of the accuracy of previous stages. Training with this corrected prediction error can not only reduce the prediction error but also... and The gap narrows, and can also make It is closer to 1, that is, it not only improves the accuracy of the prediction result of the i-th stage, but also further improves the accuracy of the prediction results of each previous stage. This represents the sum of prediction errors after corrections from multiple stages, and can be used as a loss function.

[0096] According to one embodiment of the present invention, during the training process, the loss function can be backpropagated using the gradient descent method to adjust the parameters of the exhaust gas treatment efficiency prediction model, thereby reducing the loss function and training the exhaust gas treatment efficiency prediction model. After multiple training sessions, the trained exhaust gas treatment efficiency prediction model can be obtained.

[0097] In this way, the influence of the prediction accuracy of each preceding step can be considered during the training process to determine the corrected prediction error for each step. This allows the loss function to be set in accordance with the actual factors considered in the waste gas treatment process, thereby improving the accuracy and objectivity of the loss function. Consequently, the matching between the training process and the actual waste gas treatment process is improved, thus enhancing training efficiency and the accuracy of the waste gas treatment efficiency prediction model.

[0098] According to one embodiment of the present invention, after training is completed, the predicted exhaust gas composition data at multiple stages can be predicted based on the gas flow rate data and exhaust gas composition data at multiple monitoring times at the starting end. The times corresponding to the predicted exhaust gas composition data at the multiple stages correspond to the multiple monitoring times. For example, if the first stage requires 10 times to complete, that is, the first to fifth monitoring times at the starting end correspond to the eleventh to fifteenth monitoring times at the outlet of the first stage.

[0099] According to an embodiment of the present invention, in step S4, the corresponding times of the above monitoring times at each stage can be determined based on gas flow data. Determining the exhaust gas treatment time window for each stage based on the gas flow data includes: determining the gas flow time difference between each stage and the starting time based on the gas flow data; determining the expected exhaust gas output time for each stage based on the monitoring time and the gas flow time difference; and setting the exhaust gas treatment time window corresponding to the monitoring time based on the expected exhaust gas output time.

[0100] For example, based on gas flow data and the cross-sectional area of ​​the pipes in the processing stage, the gas flow rate can be determined, thereby determining the time required for the gas to flow to each stage and pass through each stage of processing, thus determining the corresponding time (i.e., the expected time of exhaust gas output). Considering the errors in the gas flow process (e.g., the error may be caused by changes in the pipe cross-sectional area, the resistance to gas flow during the processing process, and thus the delay), the corresponding time can be used as the center to set the exhaust gas treatment time window. For example, if the duration of the exhaust gas treatment time window is the duration of 2 time points, and the corresponding time of the first monitoring time at the outlet of the first stage is the 11th time point, then the exhaust gas treatment time window at the outlet of the first stage with the first monitoring time point is the 10th to 12th time points.

[0101] According to an embodiment of the present invention, in step S5, the waste gas component treatment efficiency index of each stage can be determined based on the measured waste gas component data within the waste gas treatment time window and the waste gas component prediction data predicted by the waste gas treatment efficiency prediction model.

[0102] According to an embodiment of the present invention, the waste gas component treatment efficiency index of each stage is determined based on waste gas component prediction data, waste gas treatment time window, and waste gas component data, including: obtaining waste gas component vectors at each time point based on waste gas component data at each time point within the waste gas treatment time window; obtaining waste gas component prediction vectors based on waste gas component prediction data; determining the maximum value among the cosine similarities between the waste gas component vectors at each time point and the waste gas component prediction vectors, and determining the target waste gas component vector corresponding to the maximum value; determining the target waste gas component category treated in each stage; and determining the waste gas component treatment efficiency index of the i-th stage according to formula (2). ,

[0103] (2)

[0104] in, Let x be the exhaust gas component data of the x-th target exhaust gas component category in the target exhaust gas component vector corresponding to the y-th monitoring time in the i-th stage. This refers to the exhaust gas component data of the x-th target exhaust gas component category in the target exhaust gas component vector corresponding to the y-th monitoring time in the (i-1)-th stage. This refers to the predicted exhaust gas composition data for the x-th target exhaust gas component category corresponding to the y-th monitoring time at the ith stage. This refers to the predicted exhaust gas composition data for the x-th target exhaust gas component category corresponding to the y-th monitoring time at the (i-1)-th stage. M represents the number of target exhaust gas component categories processed in the i-th stage, and M represents the number of multiple monitoring times.

[0105] According to one embodiment of the present invention, as described above, the actual flow rate of the gas may differ from the theoretical flow rate, resulting in an error between the calculated expected time of exhaust gas output and the actual corresponding time of the monitoring time. Therefore, the actual corresponding time of the monitoring time can be found within the exhaust gas treatment time window corresponding to the detection time. The time corresponding to the maximum value of the cosine similarity between the exhaust gas component vector and the exhaust gas component prediction vector at each time within the exhaust gas treatment time window can be taken as the actual corresponding time of the monitoring time, and the exhaust gas component vector corresponding to this time is the target exhaust gas component vector.

[0106] According to one embodiment of the present invention, each stage has a corresponding target exhaust gas component category. For example, the electrostatic precipitator stage is mainly used to adsorb solid particulate matter, therefore, its target exhaust gas component category is a variety of solid particulate matter.

[0107] According to an embodiment of the present invention, in formula (2), This represents the measured decrease in the data of the x-th target waste gas component category in the waste gas at the y-th monitoring time after processing in the ith stage. This represents the predicted decrease in the exhaust gas component data of the x-th target exhaust gas component category at the y-th monitoring time. Since the exhaust gas treatment efficiency prediction model is trained under ideal conditions at each stage, this predicted decrease is the theoretical decrease under ideal conditions at each stage. Therefore, It is the ratio of the measured decrease to the theoretical decrease, representing the treatment efficiency of a single waste gas component at a single moment. This represents the average treatment efficiency of the i-th treatment stage for various target waste gas components in the industrial waste gas entering the waste gas treatment stage at the M-th monitoring time and multiple times prior to it (times 1 to M). This efficiency can be used as an indicator of the waste gas component treatment efficiency of the i-th stage to represent the health status of the i-th treatment stage. The closer the waste gas component treatment efficiency index is to 1, the closer the i-th stage is to an ideal state, and the lower the probability of malfunctions or problems.

[0108] In this way, errors in the airflow process can be taken into account, a waste gas treatment time window can be set, and the actual corresponding time to the monitoring time can be found within the waste gas treatment time window. Then, the waste gas component treatment efficiency index of each link can be determined by the ratio of the measured decrease to the theoretical decrease at the actual corresponding time, thereby improving the accuracy of the waste gas component treatment efficiency index. It can also determine the degree of closeness between the actual state and the theoretical state of each link, thereby accurately determining whether there are problems or faults in each link.

[0109] According to an embodiment of the present invention, in step S6, the exhaust gas early warning index is determined based on the exhaust gas component treatment efficiency index, exhaust gas component data, and gas flow rate data, including: determining the exhaust gas early warning index of the xth exhaust gas component according to formula (3). ,

[0110] (3)

[0111] in, This refers to the exhaust gas composition data for the xth type of exhaust gas component in the final stage, where F represents the gas flow rate. Let x be the allowable emission amount of the xth type of exhaust gas component. Let be the waste gas component treatment efficiency index for the s-th stage, targeting waste gas component category x. The number of stages where the xth type of waste gas component is the target waste gas component category.

[0112] According to one embodiment of the present invention, in formula (3), It can represent the weight of the x-th component of industrial waste gas per unit volume. It can represent the actual weight of the xth type of waste gas component emitted into the atmosphere per unit time. This represents the ratio of the actual weight of the xth type of exhaust gas component emitted into the atmosphere per unit time to the allowable weight. If this ratio is high, it indicates that the xth type of exhaust gas component is close to or has already exceeded the standard. This represents the overall efficiency of each stage targeting the xth type of waste gas component. This can represent the efficiency index of the entire waste gas treatment process for treating the xth waste gas component. The higher the index, the lower the treatment efficiency for the xth waste gas component. This is the early warning indicator for the xth type of waste gas component. If this indicator is high, it indicates that there is a problem with the treatment of the xth type of waste gas component, possibly because there is too much of the xth type of waste gas component, exceeding the treatment capacity of each stage, even if the treatment efficiency of each stage is high. If emissions are close to 1 and still exceed the standard, it may be due to problems or malfunctions in the process of treating the xth type of exhaust gas component, resulting in low treatment efficiency, or both.

[0113] In this way, the early warning indicators for waste gas can be determined from two aspects: the weight of the waste gas components emitted into the atmosphere per unit time and the efficiency of each stage of industrial waste gas treatment. This can objectively and accurately reflect the efficiency of the treatment process and the problem of exceeding emission standards.

[0114] According to one embodiment of the present invention, in step S7, if the exhaust gas warning indicator exceeds a threshold (e.g., 1.5), or if there is a stage where the exhaust gas component treatment efficiency indicator is lower than an efficiency threshold (e.g., 0.6), then a corresponding type of exhaust gas treatment prompt information is generated. The exhaust gas warning type includes emission problems and problems or malfunctions in the treatment stage. In step S8, if the exhaust gas warning indicator exceeds the threshold, then an exhaust gas treatment prompt information indicating emission problems is generated; if there is a stage where the exhaust gas component treatment efficiency indicator is lower than an efficiency threshold, then an exhaust gas treatment prompt information indicating problems or malfunctions in the treatment stage is generated. This helps staff find problems or malfunctions and facilitates troubleshooting. In the example, if only an exhaust gas treatment prompt information indicating emission problems is generated, and no exhaust gas treatment prompt information indicating problems or malfunctions in the treatment stage is generated, it indicates that there is no obvious problem in the treatment stage, but the exhaust gas flow may be too high, exceeding the treatment capacity of each stage, and the exhaust gas flow needs to be reduced. If only a problem or malfunction is generated in the waste gas treatment process, and no problem is generated for emissions, it indicates a problem exists in a certain treatment stage, but the emissions are not currently exceeding the standard. Each stage can be investigated based on its waste gas component treatment efficiency indicators. If both types of waste gas treatment alerts are generated, waste gas emissions can be reduced, and each stage can be investigated based on its waste gas component treatment efficiency indicators.

[0115] The industrial waste gas monitoring method based on intelligent sensing according to embodiments of the present invention can determine the waste gas component treatment efficiency index by setting waste gas component sensors at multiple stages and considering the impact of gas flow rate on treatment efficiency through a waste gas treatment efficiency prediction model. This provides a basis for judging whether there are faults in each stage. Furthermore, it can also calculate waste gas early warning indicators to determine whether the overall waste gas treatment status meets the standards. In the event of substandard waste gas treatment, a prompt message can be generated, indicating the faulty stage and accurately identifying the cause of pollutant exceedance, which is beneficial for accurately eliminating faults and reducing pollutant emissions. During the processing of the waste gas treatment efficiency prediction model, the self-attention mechanism corresponding to multiple stages in the model and vector concatenation processing can be used to match the processing process of the waste gas treatment efficiency prediction model with the actual industrial waste gas treatment process. The information considered and applied during the processing also matches the actual processing process. That is, it can comprehensively consider the treatment efficiency of each stage, the waste gas state and flow rate at the starting end, and the state of the waste gas output from the previous stage to obtain an output training waste gas state vector, improving the accuracy and objectivity of the waste gas component output data and its matching degree with the actual waste gas treatment process. Furthermore, during training, the influence of the prediction accuracy of each preceding stage can be considered to determine the corrected prediction error for each stage. This ensures that the loss function setting matches the actual factors considered in the waste gas treatment process, thereby improving the accuracy and objectivity of the loss function. This, in turn, enhances the matching between the training process and the actual waste gas treatment process, ultimately improving training efficiency and the accuracy of the waste gas treatment efficiency prediction model. When determining the waste gas component treatment efficiency index, errors in the airflow process can be considered. A waste gas treatment time window can be set, and the actual corresponding moment to the monitoring moment can be found within the waste gas treatment time window. The ratio of the measured decrease to the theoretical decrease at the actual corresponding moment can then be used to determine the waste gas component treatment efficiency index for each stage, improving the accuracy of the waste gas component treatment efficiency index. It also allows for determining the degree of closeness between the actual and theoretical states of each stage, thus accurately identifying whether there are problems or malfunctions in each stage. When determining the waste gas early warning index, it can be determined from two aspects: the weight of waste gas components emitted into the atmosphere per unit time and the efficiency of each stage in the industrial waste gas treatment process. This objectively and accurately reflects the efficiency problems of the treatment stages and the problem of exceeding emission standards.

[0116] Figure 3 An exemplary block diagram of an industrial waste gas monitoring system based on intelligent sensing according to an embodiment of the present invention is shown, the system comprising:

[0117] The module is used to install waste gas composition sensors at the starting point and multiple stages of industrial waste gas treatment, and to install a gas flow sensor at the starting point of industrial waste gas treatment.

[0118] The monitoring module is used to acquire multiple exhaust gas component data collected by various exhaust gas component sensors and gas flow data collected by gas flow sensors at multiple monitoring times.

[0119] The prediction module is used to determine the predicted data of the exhaust gas composition at multiple stages based on the gas flow rate data and exhaust gas composition data at the starting end, as well as the trained exhaust gas treatment efficiency prediction model.

[0120] The window module is used to determine the waste gas treatment time window for each stage based on the gas flow data.

[0121] The efficiency module is used to determine the waste gas composition treatment efficiency index for each stage based on waste gas composition prediction data, waste gas treatment time window, and waste gas composition data.

[0122] The early warning module is used to determine the early warning indicators for waste gas based on the waste gas composition treatment efficiency index, waste gas composition data, and gas flow rate data.

[0123] The judgment module is used to determine whether to generate waste gas treatment prompt information and determine the type of waste gas warning based on the waste gas composition treatment efficiency index and waste gas warning index.

[0124] The generation module is used to generate exhaust gas treatment prompt information based on the exhaust gas warning type if it is determined that exhaust gas treatment prompt information should be generated.

[0125] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0126] Those skilled in the art should understand that the embodiments of the present invention described above and shown in the accompanying drawings are merely examples and do not limit the present invention. The objectives of the present invention have been fully and effectively achieved. The functions and structural principles of the present invention have been demonstrated and explained in the embodiments, and any variations or modifications may be made to the implementation of the present invention without departing from the stated principles.

[0127] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for monitoring industrial waste gas based on intelligent sensing, characterized in that, include: Waste gas composition sensors are installed at the starting point and in multiple stages of industrial waste gas treatment, and a gas flow sensor is installed at the starting point of industrial waste gas treatment. At multiple monitoring times, acquire data on various exhaust gas components collected by different exhaust gas component sensors, as well as gas flow data collected by gas flow sensors. Based on the gas flow rate data and exhaust gas composition data at the starting end, as well as the trained exhaust gas treatment efficiency prediction model, the predicted exhaust gas composition data at multiple stages are determined. Based on the gas flow data, determine the waste gas treatment time window for each stage; Based on the predicted data of exhaust gas composition, the time window for exhaust gas treatment, and the data of exhaust gas composition, the efficiency index of exhaust gas composition treatment at each stage is determined. Based on the waste gas composition treatment efficiency index, waste gas composition data, and gas flow rate data, determine the waste gas early warning index; Based on the waste gas composition treatment efficiency index and waste gas early warning index, determine whether to generate waste gas treatment prompt information and determine the type of waste gas early warning; If it is determined that an exhaust gas treatment prompt message will be generated, then the exhaust gas treatment prompt message will be generated according to the exhaust gas warning type; The training steps for the waste gas treatment efficiency prediction model include: Set multiple stages of industrial waste gas treatment to an ideal state; According to the preset flow rate change pattern, industrial waste gas containing multiple components is input into multiple stages of industrial waste gas treatment. At multiple training moments, the exhaust gas composition data and training gas flow data at each stage are detected by exhaust gas composition sensors. By using a waste gas treatment efficiency prediction model, the training gas flow rate data and training waste gas composition data at the starting point of industrial waste gas treatment are processed to obtain waste gas composition output data for each stage. Based on the training gas flow rate data, the preset flow rate variation law, the training exhaust gas composition data, and the exhaust gas composition output data, the loss function of the exhaust gas treatment efficiency prediction model is determined. Based on the loss function of the waste gas treatment efficiency prediction model, the waste gas treatment efficiency prediction model is trained to obtain the trained waste gas treatment efficiency prediction model. Based on gas flow data, determine the waste gas treatment time window for each stage, including: Based on the gas flow data, determine the gas flow time difference between each stage and the starting time; Based on the monitoring time and the gas flow time difference, the expected time of exhaust gas output for each stage is determined. Based on the expected time of exhaust gas output, set an exhaust gas treatment time window corresponding to the monitoring time; Based on the predicted waste gas composition, the waste gas treatment time window, and the waste gas composition data, the waste gas composition treatment efficiency indicators for each stage are determined, including: Based on the exhaust gas composition data at each moment within the exhaust gas treatment time window, obtain the exhaust gas composition vector at each moment. Based on the predicted data of exhaust gas composition, obtain the predicted vector of exhaust gas composition; Determine the maximum value of the cosine similarity between the exhaust gas component vector and the exhaust gas component prediction vector at each time point, and determine the target exhaust gas component vector corresponding to the maximum value. Determine the target waste gas composition categories for each stage of treatment; According to the formula Determine the waste gas composition treatment efficiency index for the i-th stage. ,in, Let x be the exhaust gas component data of the x-th target exhaust gas component category in the target exhaust gas component vector corresponding to the y-th monitoring time in the i-th stage. This refers to the exhaust gas component data of the x-th target exhaust gas component category in the target exhaust gas component vector corresponding to the y-th monitoring time in the (i-1)-th stage. This refers to the predicted exhaust gas composition data for the x-th target exhaust gas component category corresponding to the y-th monitoring time at the ith stage. This refers to the predicted exhaust gas composition data for the x-th target exhaust gas component category corresponding to the y-th monitoring time at the (i-1)-th stage. M represents the number of target waste gas component categories processed in the i-th stage, and M represents the number of multiple monitoring times. Based on the waste gas composition treatment efficiency index, waste gas composition data, and gas flow rate data, waste gas early warning indicators are determined, including: According to the formula Determine the early warning index for the xth component of the exhaust gas. ,in, This refers to the exhaust gas composition data for the xth type of exhaust gas component in the final stage, where F represents the gas flow rate. Let x be the allowable emission amount of the xth type of exhaust gas component. Let be the waste gas component treatment efficiency index for the s-th stage, targeting waste gas component category x. The number of stages where the xth type of waste gas component is the target waste gas component category.

2. The industrial waste gas monitoring method based on intelligent sensing according to claim 1, characterized in that, By using a waste gas treatment efficiency prediction model, the training gas flow rate data and training waste gas composition data at the starting point of industrial waste gas treatment are processed to obtain waste gas composition output data for each stage, including: The training gas flow rate data and training exhaust gas composition data at the beginning of multiple training moments are concatenated to obtain the training exhaust gas state vector for each training moment. The training exhaust gas state vector is input into the self-attention mechanism corresponding to the first stage in the exhaust gas treatment efficiency prediction model for processing, so as to obtain the output training exhaust gas state vector at multiple training times corresponding to the first stage. The training exhaust gas state vector at the corresponding training time point and the output training exhaust gas state vector at the first stage are concatenated to obtain the concatenated state vector, which is then input into the self-attention mechanism at the second stage for processing to obtain the output training exhaust gas state vectors at multiple training times at the second stage. The output training exhaust gas state vector corresponding to the (i-1)th stage and the output training exhaust gas state vector corresponding to the ith stage are concatenated to obtain the concatenated state vector, which is then input into the self-attention mechanism corresponding to the (i+1)th stage for processing to obtain the output training exhaust gas state vectors for multiple training times corresponding to the (i+1)th stage. The output training exhaust gas state vectors corresponding to each stage are input into the corresponding multilayer perceptron layers for processing to obtain exhaust gas composition output data at multiple training times for each stage.

3. The industrial waste gas monitoring method based on intelligent sensing according to claim 2, characterized in that, Based on the training gas flow rate data, the preset flow rate variation pattern, the training exhaust gas composition data, and the exhaust gas composition output data, the loss function of the exhaust gas treatment efficiency prediction model is determined, including: Based on the training gas flow data and the preset flow change pattern, determine the training exhaust gas composition data corresponding to the training exhaust gas composition data at the beginning of the process at multiple moments in each stage. According to the formula Determine the loss function LOSS for the waste gas treatment efficiency prediction model, where, Let J be the content of the j-th exhaust gas component in the training exhaust gas component data corresponding to the starting point of the training exhaust gas component data at the t-th training time, which is the i-th stage. The content of the j-th exhaust gas component in the exhaust gas composition output data at the t-th training time of the i-th stage. Let J be the content of the j-th exhaust gas component in the training exhaust gas component data corresponding to the starting point of the training exhaust gas component data at the k-th stage and the t-th training time. Let n be the content of the j-th exhaust gas component in the exhaust gas composition output data at the t-th training time of the k-th stage, m be the number of training time points detected at the starting end, and N be the number of stages. When i=1, ... Set to 1.

4. An industrial waste gas monitoring system based on intelligent sensing, characterized in that, The method for monitoring industrial waste gas based on intelligent sensing as described in any one of claims 1-3 includes: The module is used to install waste gas composition sensors at the starting point and multiple stages of industrial waste gas treatment, and to install a gas flow sensor at the starting point of industrial waste gas treatment. The monitoring module is used to acquire multiple exhaust gas component data collected by various exhaust gas component sensors and gas flow data collected by gas flow sensors at multiple monitoring times. The prediction module is used to determine the predicted data of the exhaust gas composition at multiple stages based on the gas flow rate data and exhaust gas composition data at the starting end, as well as the trained exhaust gas treatment efficiency prediction model. The window module is used to determine the waste gas treatment time window for each stage based on the gas flow data. The efficiency module is used to determine the waste gas composition treatment efficiency index for each stage based on waste gas composition prediction data, waste gas treatment time window, and waste gas composition data. The early warning module is used to determine the early warning indicators for waste gas based on the waste gas composition treatment efficiency index, waste gas composition data, and gas flow rate data. The judgment module is used to determine whether to generate waste gas treatment prompt information and determine the type of waste gas warning based on the waste gas composition treatment efficiency index and waste gas warning index. The generation module is used to generate exhaust gas treatment prompt information based on the exhaust gas warning type if it is determined that exhaust gas treatment prompt information should be generated.

Images