A tobacco processing environment-friendly monitoring and prediction method based on multi-source data fusion

Abstract

This invention discloses a method for environmental monitoring and prediction in tobacco processing based on multi-source data fusion. The method includes: collecting multi-source monitoring data during the tobacco processing process and classifying and encrypting it for storage in a unified data storage platform; inputting the multi-source monitoring data into a target prediction model based on a multi-task learning architecture; and performing joint calculations using a shared monitoring data encoding layer, so that the three prediction tasks of outlet chemical oxygen demand water quality, single-box pollutant emissions, and exhaust gas nitrogen oxide emissions can share process characteristic information and form mutual constraints; by comparing the three types of prediction results with corresponding dynamic emission thresholds, the risk of exceeding various emission standards is determined; when an exceedance risk is determined, pollution control process control instructions matching the exceedance type are generated and issued to reduce the probability of exceeding emission standards, thereby realizing intelligent and refined environmental monitoring and prediction in tobacco processing and improving the environmental management level and emission reduction and consumption reduction capabilities of tobacco processing.

Description

Technical Field

[0001] The present invention relates to the technical fields of cigarette production and waste gas and waste treatment, and particularly to an environmental protection monitoring and prediction method for tobacco processing based on multi-source data fusion. Background Technique

[0002] With the intelligent and green development of the tobacco processing industry, the environmental protection monitoring system has gradually realized the collection, aggregation and visualization display of environmental protection-related data, and the data collection, transmission and basic monitoring system have become increasingly perfect. In order to further improve the refined level of pollution control and meet the increasingly strict environmental protection control requirements, it is urgent in the industry to rely on multi-source data fusion and intelligent modeling analysis to build a more forward-looking and collaborative environmental protection monitoring and prediction mechanism, and promote the coordinated optimization of production operation and environmental protection governance.

[0003] Currently, independent environmental protection monitoring systems are mostly used for environmental protection control in tobacco processing, mainly to realize the collection and intuitive display of environmental protection data. Some systems have basic energy consumption prediction functions, and at the same time rely on external emission concentration indicators to carry out emission supervision. The overall control mode is mainly post-event monitoring and passive management.

[0004] However, the existing environmental protection control methods can only achieve simple collection and display of scattered data, and the integrity and reliability of monitoring data are insufficient; mostly use a single model for independent prediction, unable to explore the internal relationship between processes and emissions, and the accuracy of risk prediction is low; mostly use fixed thresholds for emission determination, the threshold adaptability is poor, and it is difficult to accurately identify various emission over-standard risks; usually only have monitoring and alarm functions, and the environmental protection control accuracy and collaborative regulation ability are insufficient. Summary of the Invention

[0005] The present invention provides an environmental protection monitoring and prediction method for tobacco processing based on multi-source data fusion to solve the problems of insufficient integrity and reliability of monitoring data, low accuracy of risk prediction, and insufficient environmental protection control accuracy and collaborative regulation ability.

[0006] According to one aspect of the embodiments of the present invention, an environmental protection monitoring and prediction method for tobacco processing based on multi-source data fusion is provided, including: Collect multi-source monitoring data during the tobacco processing process, and classify and encrypt it for storage in a unified data storage platform; Input the multi-source monitoring data in the unified data storage platform into a target prediction model constructed based on a multi-task learning architecture, and perform joint calculations through a shared monitoring data encoding layer to obtain the predicted results of the outlet chemical oxygen demand water quality, the predicted results of the pollutant emissions per box, and the predicted results of the waste gas nitrogen oxide emissions; The predicted results of chemical oxygen demand (COD) water quality at the outlet, the predicted results of pollutant emissions per container, and the predicted results of nitrogen oxide emissions in exhaust gas are compared with the corresponding dynamic emission thresholds to determine whether there is a risk of exceeding emission standards. When it is determined that at least one of the following is at risk of exceeding the standard: chemical oxygen demand (COD) water quality at the outlet, pollutant emissions per tank, and nitrogen oxide emissions in exhaust gas, a pollution control process control instruction matching the type of exceedance will be generated and issued.

[0007] According to another aspect of the present invention, an environmental monitoring and prediction device for tobacco processing based on multi-source data fusion is provided, comprising: The data acquisition and storage module is used to collect multi-source monitoring data during the tobacco processing process and categorize and encrypt the data for storage to a unified data storage platform. The model prediction module is used to input multi-source monitoring data from the unified data storage platform into the target prediction model built on a multi-task learning architecture. Through joint calculation via a shared monitoring data encoding layer, the predicted results of outlet chemical oxygen demand water quality, single-box pollutant emissions, and exhaust gas nitrogen oxide emissions are obtained. The threshold determination module is used to compare the predicted results of chemical oxygen demand water quality at the outlet, the predicted results of pollutant emissions per tank, and the predicted results of nitrogen oxide emissions in exhaust gas with the corresponding dynamic emission thresholds to determine whether there is a risk of exceeding emission standards. The process control module is used to generate and issue pollution control process control instructions that match the type of exceedance when it is determined that at least one of the following is at risk of exceeding the standard: chemical oxygen demand water quality at the outlet, pollutant discharge in a single tank, and nitrogen oxide emissions in exhaust gas.

[0008] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising: At least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores a computer program executable by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to execute the tobacco processing environmental monitoring and prediction method based on multi-source data fusion as described in any embodiment of the present invention.

[0009] According to another aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing computer instructions, the computer instructions being configured to cause a processor to execute and implement the tobacco processing environmental monitoring and prediction method based on multi-source data fusion as described in any embodiment of the present invention.

[0010] According to another aspect of the present invention, a computer program product is also provided, including a computer program that, when executed by a processor, implements the steps of the method as described in any embodiment of the present invention.

[0011] The technical solution of this invention, by collecting multi-source monitoring data during tobacco processing and classifying and encrypting it for storage in a unified data storage platform, enables centralized management and secure storage of multi-type and multi-dimensional environmentally related data. This avoids management chaos and leakage risks caused by data dispersion, providing complete and reliable data support for subsequent predictive analysis. The multi-source monitoring data is input into a target prediction model based on a multi-task learning architecture, and joint calculations are performed using a shared monitoring data encoding layer. This allows the three prediction tasks—outlet chemical oxygen demand (COD) water quality, single-container pollutant emissions, and nitrogen oxide emissions—to share process characteristic information and form mutual constraints. Compared to a single prediction model, this not only improves the accuracy and consistency of the prediction results but also fully explores the relationship between tobacco processing technology and environmental performance. The inherent correlation between pollutant emissions solves the prediction bias problem caused by neglecting process characteristics in traditional prediction methods. By comparing the three types of prediction results with the corresponding dynamic emission thresholds, the risk of exceeding emission standards for various types can be accurately determined, avoiding the shortcomings of insufficient adaptability of fixed thresholds. When the risk of exceeding standards is determined, pollution control process control instructions matching the type of exceeding standard are generated and issued, enabling precise control and timely coordinated regulation of exceeding standards risks. Compared with the traditional method of only monitoring and alarming without targeted regulation, it can effectively reduce the probability of exceeding emission standards, while reducing the situation of over-treatment or under-treatment. It realizes intelligent and refined environmental monitoring and prediction in tobacco processing, taking into account both environmental compliance and production efficiency, and further improving the environmental management level and emission reduction and consumption reduction capabilities of tobacco processing.

[0012] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0014] Figure 1 This is a flowchart of a method for environmental monitoring and prediction of tobacco processing based on multi-source data fusion, provided in Embodiment 1 of the present invention. Figure 2This is a flowchart of another method for environmental monitoring and prediction of tobacco processing based on multi-source data fusion, provided in Embodiment 2 of the present invention. Figure 3 This is a flowchart of another method for environmental monitoring and prediction of tobacco processing based on multi-source data fusion, provided in Embodiment 3 of the present invention; Figure 4 This is a schematic diagram of an environmental protection management business architecture applicable to an embodiment of the present invention; Figure 5 This is a schematic diagram of an environmental protection management application architecture applicable to an embodiment of the present invention; Figure 6 This is a schematic diagram of an overall application architecture for a cigarette factory applicable to an embodiment of the present invention; Figure 7 This is a schematic diagram of an application architecture for environmental protection management in a cigarette factory, applicable to an embodiment of the present invention. Figure 8 This is a schematic diagram of a data architecture applicable to an embodiment of the present invention; Figure 9 This is a schematic diagram of a technical architecture applicable to an embodiment of the present invention; Figure 10 This is a schematic diagram of a deployment architecture applicable to an embodiment of the present invention; Figure 11 This is a schematic diagram of the structure of a tobacco processing environmental monitoring and prediction device based on multi-source data fusion, according to Embodiment 4 of the present invention. Figure 12 This is a schematic diagram of the structure of an electronic device that implements the tobacco processing environmental monitoring and prediction method based on multi-source data fusion according to embodiments of the present invention. Detailed Implementation

[0015] To enable those skilled in the art to better understand the present invention, 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. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0016] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0017] Example 1 Figure 1 This is a flowchart of a method for environmental monitoring and prediction of tobacco processing based on multi-source data fusion, provided in Embodiment 1 of the present invention. This embodiment is applicable to monitoring the environmental status of tobacco processing and predicting environmental risks. The method can be executed by a tobacco processing environmental monitoring and prediction device based on multi-source data fusion. This device can be implemented in hardware and / or software and is generally configured in electronic devices. Figure 1 As shown, the method includes: S110. Collect multi-source monitoring data during the tobacco processing process, and classify and encrypt the data for storage in a unified data storage platform.

[0018] In this embodiment of the invention, multi-source monitoring data can be specifically understood as: online and manual monitoring data on production energy consumption, pollution control conditions, and pollutant emissions in tobacco processing processes such as packaging, tobacco processing, power generation, pollution control, and pollution monitoring.

[0019] A unified data storage platform can be specifically understood as a basic data platform used for data collection, storage, processing and analysis, realizing centralized aggregation and management of multi-source environmental protection and production data.

[0020] Specifically, it involves collecting multi-dimensional and multi-type monitoring data generated throughout the entire tobacco processing process, such as online and manual monitoring data on production energy consumption, pollution control operating conditions, and pollutant emissions generated in the cigarette production, tobacco processing, power production, pollution treatment, and pollution monitoring business processes.

[0021] Production energy consumption data can include real-time and historical summary data on water, electricity, gas, and steam consumption from departments such as the power plant, yarn processing plant, and coil packaging plant. This includes real-time values, baseline values, predicted values, and deviation values. It can also include statistical indicators such as overall energy consumption and unit energy consumption. Pollution control operating data can include wastewater treatment monitoring data, such as real-time operating indicators like influent pollutant concentration, aeration time, dissolved oxygen concentration, sludge concentration, sludge return ratio, hydraulic retention time, sedimentation time, and decanting time.

[0022] Pollutant emission data can include basic information such as monitoring points, pollutants, monitoring equipment, and operation and maintenance information, as well as indicators such as real-time values, average values, maximum values, minimum values, and emission volumes of pollutants. For example, wastewater emission data can include factors such as COD (Chemical Oxygen Demand, an indicator of the degree of organic pollution in water), ammonia nitrogen, total phosphorus, total nitrogen, and pH (potential of hydrogen, a hydrogen ion concentration index), while exhaust gas emission data can include factors such as sulfur dioxide, nitrogen oxides, and particulate matter.

[0023] Various types of data can be aggregated into a unified data storage platform through methods such as direct device transmission, service bus access, application programming interface (API) integration, and manual data entry. Data collection channels can also be expanded by using edge gateways and IoT terminal reporting, depending on the on-site equipment conditions.

[0024] After data is imported into the platform, it is categorized and organized according to data type. At the same time, encryption and security control measures are used to store it on a unified data storage platform. For example, production energy consumption data such as steam consumption and electricity load in the power workshop, pollution control operation data such as COD concentration and aeration time in the wastewater treatment process, and pollution monitoring data such as nitrogen oxide concentration and particulate matter content in flue gas emissions can be classified and stored according to the business type of production energy consumption, treatment operation, emission monitoring, and data dimensions such as real-time collected data, historical statistical data, and analysis and prediction data. At the same time, encryption algorithms are used to protect the confidentiality of data involving production monitoring data and core pollution indicators. Verification mechanisms ensure the integrity of data transmission and storage. Full or incremental data backups are performed daily to achieve rapid recovery after failure. Different data access, modification, and export permissions are set according to the job roles of operation and maintenance personnel, management personnel, and supervisory personnel to achieve hierarchical permission control, thereby achieving stable collection, reliable storage, and efficient management of environmental protection and production-related data throughout the tobacco processing process.

[0025] S120. Input the multi-source monitoring data in the unified data storage platform into the target prediction model built on the multi-task learning architecture, and perform joint calculations through the shared monitoring data encoding layer to obtain the predicted results of outlet chemical oxygen demand water quality, single-box pollutant emission, and exhaust gas nitrogen oxide emission.

[0026] In this embodiment of the invention, the multi-task learning architecture can be specifically understood as: simultaneously supporting wastewater COD, single-tank sewage discharge, and exhaust gas nitrogen oxides (NOx).x This model integrates three prediction tasks: wastewater COD prediction (primarily originating from boiler combustion, and is one of the air pollutants), and waste gas NO2 prediction. The shared monitoring data encoding layer can be understood as a unified extraction and encoding of common monitoring features related to production, energy consumption, and pollution control, serving as a shared feature processing layer for all three prediction tasks. The target prediction model built on a multi-task learning architecture can be implemented using a hard parameter sharing approach (i.e., all three prediction tasks share the same shared monitoring data encoding layer). Using tobacco processing monitoring data as a common input, the shared monitoring data encoding layer performs unified feature extraction and encoding on common process data such as drying temperature, moisture regain, production line speed, and production grade, generating a globally shared process feature vector that is simultaneously distributed to wastewater COD prediction, single-box pollutant emission prediction, and waste gas NO2 prediction. x The emission prediction system consists of three task branches, each of which combines its own specific monitoring data characteristics for fusion and inference calculations, achieving simultaneous and accurate prediction of multiple tasks while ensuring consistency of process characteristics.

[0027] The predicted chemical oxygen demand (COD) water quality at the outlet can be specifically understood as: the predicted COD concentration at the wastewater discharge point. The predicted pollutant emissions per unit container can be specifically understood as: the predicted total pollutant emissions per unit of cigarette production. The predicted nitrogen oxide emissions from exhaust gas can be specifically understood as: the predicted total nitrogen oxide emissions from boiler and process exhaust gas.

[0028] Specifically, real-time collected production operation data, energy consumption data, pollution control operating condition data, and other multi-source monitoring data gathered in the unified data storage platform are input into the target prediction model built using a multi-task learning architecture. The shared monitoring data encoding layer within the model performs unified feature extraction and joint calculation on the data. For example, after extracting unified process features such as drying temperature, production line speed, and production shifts through the shared monitoring data encoding layer, the model will perform differentiated special inference calculations for different prediction objects. For example, for wastewater discharge branches, the model will combine influent COD, pH value, aeration time, and other treatment operating condition features to infer the trend of effluent water quality changes; for single-box sewage discharge branches, the model will combine current output, unit load, production grade, and other data to calculate the pollutant emission level per unit output; and for exhaust gas discharge branches, the model will combine natural gas consumption, boiler load, flue gas treatment equipment operating status, and other parameters to estimate nitrogen oxide emission concentration and total amount. Thus, while sharing process features, the model can achieve accurate predictions for different scenarios and indicators.

[0029] The final output includes three predicted results: chemical oxygen demand (COD) water quality at the outlet, pollutant emissions per tank, and nitrogen oxide emissions in exhaust gas. These results will provide a basis for subsequent determination and control of exceedances.

[0030] S130. The predicted results of chemical oxygen demand (COD) water quality at the outlet, the predicted results of pollutant emissions per container, and the predicted results of nitrogen oxide emissions in exhaust gas are compared with the corresponding dynamic emission thresholds to determine whether there is a risk of exceeding emission standards.

[0031] Specifically, dynamic emission thresholds for corresponding indicators are adaptively calculated and generated based on current operating conditions such as production grade, process path, and raw material origin. In addition to adjustments based on the above conditions, the threshold values ​​can be further optimized by combining production load, seasonal environment, and equipment operating years. Then, the predicted results of chemical oxygen demand water quality at the outlet, the predicted results of pollutant emissions per tank, and the predicted results of nitrogen oxide emissions in exhaust gas are compared with their respective dynamic emission thresholds one by one. If a certain predicted value is higher than the corresponding dynamic threshold, it is determined that there is a risk of exceeding the emission standard for that indicator, thereby achieving accurate risk identification that fits the actual production status.

[0032] S140. When it is determined that at least one of the following is at risk of exceeding the standard: chemical oxygen demand water quality at the outlet, pollutant discharge per tank, and nitrogen oxide discharge in exhaust gas, a pollution control process control instruction matching the type of exceedance shall be generated and issued.

[0033] In this embodiment of the invention, the pollution control process control command can be specifically understood as: a control command to adjust the operating parameters of production and pollution control equipment.

[0034] Specifically, when it is determined that any of the following is at risk of exceeding the standard: chemical oxygen demand (COD) of the outlet water quality, pollutant discharge from a single tank, or nitrogen oxide emissions from exhaust gas, the system generates and issues corresponding pollution control process control instructions based on the type of exceedance indicator. This proactive control is implemented in advance. For example, for wastewater exceeding the standard, parameters such as aeration time, dissolved oxygen concentration, and sludge return ratio can be adjusted; for exhaust gas exceeding the standard, parameters such as boiler combustion air volume, natural gas supply, and flue gas circulation volume can be adjusted; and for single tank discharge exceeding the standard, parameters such as production load, material input, and output per unit time can be adjusted. In addition, the system can also adaptively add control parameters such as reagent dosage, sewage pump operating frequency, and desulfurization and denitrification equipment operating levels based on equipment operating status. At the same time, the prediction results, risk assessment results, control instructions, and execution status are simultaneously pushed to the pollution discharge monitoring visualization analysis system for process flow, prediction comparison, and control record display. Corresponding exceedance warning information is generated in the early warning center, thereby fully realizing the closed-loop operation of real-time environmental monitoring, prediction and early warning, and intelligent control.

[0035] Optionally, based on the above embodiments, generating and issuing pollution control process control instructions that match the type of exceedance may include: Based on the predicted results of chemical oxygen demand (COD) water quality at the outlet, the predicted results of pollutant emissions per tank, and the predicted results of nitrogen oxide emissions in exhaust gas, an emission anomaly propagation model is constructed. When any predicted result is determined to have a risk of exceeding the standard, the emission anomaly propagation model is used to trace back to the corresponding specific process unit and equipment operating status to locate the root cause of the emission anomaly. Based on the located root cause, a comprehensive control scheme is generated that coordinates front-end production process optimization instructions and end-of-pipe pollution control equipment control instructions, and a pre-evaluation of the emission change trend after the implementation of the comprehensive control scheme is conducted. Based on the pre-evaluation results, pollution control process control instructions matching the type of exceedance are determined and issued.

[0036] In this embodiment of the invention, the emission anomaly propagation model can be specifically understood as an analytical model used to characterize the causal transmission relationship between production process units, equipment operating parameters, and pollutant emissions. The front-end production process optimization command can be specifically understood as adjusting parameters in production stages such as yarn making, winding and packaging, and power. The end-of-pipe pollution control equipment control command can be specifically understood as adjusting the operating parameters of wastewater treatment and flue gas treatment facilities.

[0037] Specifically, when generating control instructions for pollution control processes that match the types of exceedances, an emission anomaly propagation graph model is constructed based on the predicted results of outlet chemical oxygen demand (COD) water quality, single-tank pollutant emissions, and exhaust nitrogen oxide emissions. This model reflects the transmission relationship of monitoring data. In addition to tracing back based on the prediction results, it can also be combined with a historical anomaly case database to assist in locating the root cause. For example, when generating control instructions for pollution control processes that match the types of exceedances, a Bayesian network or causal inference graph is used to construct an emission anomaly propagation graph model that reflects the transmission relationship between production monitoring data, equipment operating status, and pollutant emissions. The predicted results of outlet COD water quality, single-tank pollutant emissions, and exhaust nitrogen oxide emissions are used as input nodes of the model. Process units such as the silk-making process, power boiler, and wastewater treatment system are used as intermediate transmission nodes, and various operating parameters are used as associated edges to construct a complete causal link. When locating the root cause of the anomaly, in addition to tracing back along the causal link based on the predicted exceedance results, historical data of similar operating conditions and similar exceedance phenomena in the historical anomaly case database are also retrieved for matching and comparison to help identify and accurately locate the core process unit and key parameters of the emission anomaly.

[0038] When any indicator is determined to have a risk of exceeding the standard, this model traces back from the predicted exceedance result to the operating status of specific process units and corresponding equipment, such as silk making, power, and wastewater treatment, to determine the root cause of the emission anomaly. For example, when any indicator such as wastewater COD, single-box discharge, or exhaust gas nitrogen oxides is determined to have a risk of exceeding the standard, a reverse path search and causal reasoning algorithm is used. Starting from the exceedance result node in the emission anomaly propagation graph model, the model traces back along the monitoring data transmission link, sequentially checking the material moisture content and drying temperature of the silk making production line, the combustion air volume and gas supply of the boiler in the power workshop, and the aeration time and sludge return ratio of the wastewater treatment system, as well as the operating status parameters of specific process units and equipment. Combined with parameter deviation thresholds and correlation influence weights, the core process links and key equipment anomalies causing the emission exceedance are finally located.

[0039] Based on the root cause, a comprehensive control plan is formulated that balances production-side parameter optimization with pollution control equipment adjustment. A preliminary assessment is conducted by simulating the pollutant emission trends after the plan's implementation. If the assessment results show that exceeding standards can be effectively avoided, the plan is confirmed and the final control command is issued. For example, based on the identified root cause of emission anomalies, a comprehensive control plan is formulated using a multi-objective optimization algorithm. At the production end, parameters such as the silk-making feeding rate, production line speed, or production load are adjusted accordingly. At the pollution control end, equipment parameters such as aeration time, chemical dosage, boiler oxygen content, or flue gas recirculation are simultaneously adjusted. Then, a time-series prediction model or numerical simulation model is used to simulate the changes in outlet COD, single-box emissions, and nitrogen oxide emissions after the plan's execution. The predicted values ​​are compared with the dynamic emission thresholds. If the simulation results show that pollutant emissions can fall back to the compliant range, the plan is confirmed as effective, and the final control command is issued to the field control system for execution.

[0040] By constructing an emission anomaly propagation model and combining it with prediction results for reverse tracing, we can accurately locate the process units and equipment root causes corresponding to emission anomalies, avoid blind regulation, and generate a comprehensive regulation plan that coordinates production and pollution control. Furthermore, by conducting a pre-implementation assessment of emission trends after the plan is implemented, we can verify the regulation effect in advance and select the optimal strategy. This not only enables proactive and precise prevention and control, reducing the risk of exceeding standards, but also reduces production fluctuations and governance costs caused by unreasonable regulation, and improves the level of intelligence and closed-loop disposal capabilities of environmental protection management.

[0041] The technical solution of this invention, by collecting multi-source monitoring data during tobacco processing and classifying and encrypting it for storage in a unified data storage platform, enables centralized management and secure storage of multi-type and multi-dimensional environmentally related data. This avoids management chaos and leakage risks caused by data dispersion, providing complete and reliable data support for subsequent predictive analysis. The multi-source monitoring data is input into a target prediction model based on a multi-task learning architecture, and joint calculations are performed using a shared monitoring data encoding layer. This allows the three prediction tasks—outlet chemical oxygen demand (COD) water quality, single-container pollutant emissions, and nitrogen oxide emissions—to share process characteristic information and form mutual constraints. Compared to a single prediction model, this not only improves the accuracy and consistency of the prediction results but also fully explores the relationship between tobacco processing technology and environmental performance. The inherent correlation between pollutant emissions solves the prediction bias problem caused by neglecting process characteristics in traditional prediction methods. By comparing the three types of prediction results with the corresponding dynamic emission thresholds, the risk of exceeding emission standards for various types can be accurately determined, avoiding the shortcomings of insufficient adaptability of fixed thresholds. When the risk of exceeding standards is determined, pollution control process control instructions matching the type of exceeding standard are generated and issued, enabling precise control and timely coordinated regulation of exceeding standards risks. Compared with the traditional method of only monitoring and alarming without targeted regulation, it can effectively reduce the probability of exceeding emission standards, while reducing the situation of over-treatment or under-treatment. It realizes intelligent and refined environmental monitoring and prediction in tobacco processing, taking into account both environmental compliance and production efficiency, and further improving the environmental management level and emission reduction and consumption reduction capabilities of tobacco processing.

[0042] Example 2 Figure 2 This is a flowchart of another method for environmental monitoring and prediction of tobacco processing based on multi-source data fusion, provided in Embodiment 2 of the present invention. This embodiment is a refinement of the environmental monitoring and prediction method for tobacco processing based on multi-source data fusion described in the above embodiments. Figure 2 As shown, the method includes: S210. Collect multi-source monitoring data during the tobacco processing process, and categorize and encrypt the data for storage in a unified data storage platform.

[0043] S220. Based on historical multi-source monitoring data and tobacco processing monitoring data in a unified data storage platform, a training dataset is constructed.

[0044] The training dataset includes production conditions, treatment parameters, emission data, and process indicators.

[0045] S230. Based on the training dataset, the prediction model using a multi-task learning architecture is iteratively trained to obtain the target prediction model.

[0046] The target prediction model includes a shared monitoring data encoding layer, a first dedicated feature extraction layer, a second dedicated feature extraction layer, a third dedicated feature extraction layer, a first independent output layer, a second independent output layer, and a third independent output layer. The shared monitoring data encoding layer performs unified feature extraction and vector encoding on the input historical multi-source monitoring data and tobacco processing monitoring data, generating feature vectors for wastewater treatment operating conditions monitoring data, production and wastewater discharge statistics monitoring data, production energy consumption and equipment operation monitoring data, and a shared process feature vector. These feature vectors are then synchronously input into the matching first, second, and third dedicated feature extraction layers. The first dedicated feature extraction layer fuses the wastewater treatment operating conditions monitoring data feature vectors with the shared process feature vectors, outputting a value for export chemical demand. The system comprises three layers: a first dedicated feature layer for predicting oxygen demand and water quality; a second dedicated feature extraction layer that fuses the feature vectors of production and wastewater discharge statistical monitoring data with the shared process feature vectors to output dedicated feature features for predicting pollutant emissions from a single tank; a third dedicated feature extraction layer that fuses the feature vectors of production energy consumption and equipment operation monitoring data with the shared process feature vectors to output dedicated feature features for predicting nitrogen oxide emissions from exhaust gas; a first independent output layer that calculates based on the dedicated feature features output by the first dedicated feature extraction layer to output the outlet chemical oxygen demand (COD) water quality prediction result; a second independent output layer that calculates based on the dedicated feature features output by the second dedicated feature extraction layer to output the single tank pollutant emission prediction result; and a third independent output layer that calculates based on the dedicated feature features output by the third dedicated feature extraction layer to output the exhaust nitrogen oxide emission prediction result.

[0047] In this embodiment of the invention, historical multi-source monitoring data can be specifically understood as: historical data such as production conditions, treatment parameters, emission data, and process indicators accumulated in a unified data storage platform. Tobacco processing monitoring data can be specifically understood as: core monitoring data of process stages such as drying temperature, moisture regain, feeding accuracy, cutting width, and leaf filling value.

[0048] The first dedicated feature extraction layer and the first independent output layer, the second dedicated feature extraction layer and the second independent output layer, and the third dedicated feature extraction layer and the third independent output layer can be specifically understood as: corresponding to the prediction of effluent COD water quality, the prediction of pollutant emissions per tank, and the prediction of NO in exhaust gas, respectively. x The three tasks of emission prediction are network structures used to complete the specific feature fusion and prediction result calculation for each prediction task. Here, the first, second, and third are only used to distinguish the network structures corresponding to different tasks and do not represent the execution order or priority.

[0049] Specifically, before inputting multi-source monitoring data into the target prediction model, a training dataset is constructed using historical multi-source monitoring data and tobacco processing monitoring data from a unified data storage platform. This dataset includes production conditions (which may include output, energy consumption, and equipment operation data), governance parameters, emission data, and process indicators.

[0050] The tobacco processing monitoring data may include: drying temperature, moisture content, feeding accuracy, cutting width, and leaf filling value.

[0051] Production conditions, tobacco processing monitoring data, and process indicators are used together for model training to share process characteristics. Treatment parameters are attributed to wastewater treatment condition monitoring data and used for COD water quality prediction training. Production and emission data are attributed to production and sewage discharge statistics monitoring data and used for single-container pollutant emission prediction training. Production energy consumption and equipment operation data are attributed to production energy consumption and equipment operation monitoring data and used for waste gas nitrogen oxide emission prediction training.

[0052] This dataset was used to iteratively train a multi-task learning model with a shared monitoring data encoding layer, three sets of dedicated feature extraction layers, and independent output layers. The shared monitoring data encoding layer uniformly encodes production conditions, treatment parameters, emission data, and process indicators to form a shared process feature vector. This vector is then synchronously distributed to the first task branch corresponding to the prediction of effluent COD water quality, the second task branch corresponding to the prediction of single-tank pollutant emissions, and the third task branch corresponding to the prediction of nitrogen oxide emissions in exhaust gas. This allows the three prediction tasks to share the same process features and achieve mutual constraints and joint optimization. The first, second, and third dedicated feature extraction layers correspond to the fusion of wastewater treatment condition monitoring features, production and discharge statistics monitoring features, production energy consumption and equipment operation monitoring features, and shared process features, respectively. The first, second, and third independent output layers perform specific prediction calculations for effluent COD water quality, single-tank pollutant emissions, and nitrogen oxide emissions in exhaust gas based on their respective fused features. The trained target prediction model is stored in the model management system for easy online real-time retrieval.

[0053] In the iterative training process of the multi-task learning model, the parameter reuse mechanism of the shared monitoring data encoding layer and the collaborative constraints of the multi-task joint loss function achieve mutual constraints and joint optimization of the three prediction tasks: The shared monitoring data encoding layer uniformly encodes production conditions, treatment parameters, emission data, and process indicators to form a shared process feature vector. This vector is synchronously distributed to the three task branches: outlet COD water quality prediction, single-tank pollutant emission prediction, and exhaust gas nitrogen oxide emission prediction. This allows the three tasks to share the same underlying process features, realizing the correlation and constraint at the feature level. At the same time, a weighted joint loss function is constructed, which includes the outlet COD water quality prediction loss, the single-tank pollutant emission prediction loss, and the exhaust gas nitrogen oxide emission prediction loss. During the model backpropagation iteration, the joint loss gradient will be applied synchronously to the shared encoding layer and the task-specific layers. If a certain If the prediction deviation of one task is too large, its loss gradient will adjust the parameters of the shared layer in reverse, thereby constraining the feature extraction and prediction logic of the other two tasks and achieving mutual checks and balances among the three tasks. In addition, the first, second and third dedicated feature extraction layers respectively integrate wastewater treatment operation monitoring features, production and pollution discharge statistics monitoring features, production energy consumption and equipment operation monitoring features and shared process features. While retaining the unique business rules of each task, knowledge transfer and deviation correction between tasks are achieved by relying on the linkage of shared features. The first, second and third independent output layers complete the special prediction calculation based on the fused features. The target prediction model after training is stored in the model management system for easy online real-time access. Thus, while ensuring the accuracy of each special prediction, the three environmental protection prediction tasks are coordinated and mutually constrained, improving the overall generalization ability and prediction reliability of the model.

[0054] In a specific example, a dataset containing historical production conditions, treatment parameters, emission data, and monitoring data is used to iteratively train a neural network model built on a hard parameter-shared multi-task learning architecture. The model sets a shared monitoring data encoding layer as a common feature extraction module, and separate layers corresponding to COD, single-container emissions, and NOx. xThe prediction system comprises three dedicated feature extraction layers and three independent output layers. During training, the mean squared error can be used as the loss function for each task, and an adaptive moment estimation optimizer can be used for joint optimization. The shared monitoring data encoding layer uniformly encodes data such as drying temperature, moisture regain, aeration time, and energy consumption to form a shared process feature vector, which is then simultaneously distributed to the three task branches. Each dedicated feature extraction layer concatenates or weights the corresponding monitoring data features with the shared process features. Each independent output layer can use a fully connected layer combined with linear or nonlinear activation functions to complete the specific prediction calculation. In addition, an early stopping strategy can be introduced to terminate training based on the validation set loss to prevent overfitting, and regularization can be used to constrain the model weights to improve generalization ability. After training, the target prediction model that reaches the preset accuracy is stored in the model management system, supporting subsequent real-time data access and online inference.

[0055] S240. Input the multi-source monitoring data in the unified data storage platform into the target prediction model built on the multi-task learning architecture, and perform joint calculations through the shared monitoring data encoding layer to obtain the predicted results of outlet chemical oxygen demand water quality, single-box pollutant emission, and exhaust gas nitrogen oxide emission.

[0056] Optionally, based on the above embodiments, inputting multi-source monitoring data from the unified data storage platform into the target prediction model constructed based on a multi-task learning architecture, and performing joint calculations through a shared monitoring data encoding layer to obtain the outlet chemical oxygen demand (COD) water quality prediction result, may include: Data on influent pollutant concentration, aeration time, dissolved oxygen concentration, sludge concentration, sludge return ratio, hydraulic retention time, sedimentation time, and decanting time from the unified data storage platform during the wastewater treatment process are input into a target prediction model constructed based on a multi-task learning architecture. The input data is feature-encoded through the shared monitoring data encoding layer of the target prediction model to obtain corresponding wastewater treatment operating condition monitoring data feature vectors and shared process feature vectors. These feature vectors are then input into the first dedicated feature extraction layer of the target prediction model for feature fusion. The fused features are then input into the first independent output layer of the target prediction model for calculation, yielding the predicted chemical oxygen demand (COD) water quality at the wastewater discharge outlet. The first dedicated feature extraction layer and the first independent output layer are neural network prediction models constructed based on the circulating activated sludge process.

[0057] In this embodiment of the invention, the circulating activated sludge process can be specifically understood as: a wastewater treatment process that achieves wastewater purification through circulating aeration, sedimentation, decanting and other processes, and is the process basis for constructing the first exclusive feature extraction layer and the first independent output layer.

[0058] Specifically, multi-source monitoring data from the unified data storage platform, including influent pollutant concentration, aeration time, dissolved oxygen concentration, sludge concentration, sludge return ratio, hydraulic retention time, sedimentation time, and decanting time during the wastewater treatment process, are input into the target prediction model (i.e., the multi-task learning model) built on a multi-task learning architecture. The wastewater treatment process adopts the Cyclic Activated Sludge System (CASS).

[0059] The input data is feature-encoded by the shared monitoring data encoding layer of the target prediction model to obtain the corresponding wastewater treatment operating condition monitoring data feature vector and shared process feature vector. The encoding process can use an attention mechanism to strengthen the feature weights of key parameters (such as influent pollutant concentration and dissolved oxygen concentration). The two types of feature vectors are input to the first dedicated feature extraction layer of the target prediction model for feature fusion. The fusion method can be weighted summation or feature concatenation. The fused features are then input to the first independent output layer for calculation to obtain the predicted chemical oxygen demand (COD) water quality at the wastewater discharge outlet. The first dedicated feature extraction layer and the first independent output layer together constitute a neural network prediction model based on the CASS process. This model takes key wastewater treatment parameters such as influent concentration and aeration time as input and uses a neural network algorithm (such as backpropagation neural network or long short-term memory neural network, etc., without specific restrictions) to achieve accurate prediction of effluent COD concentration. At the same time, data normalization processing can be introduced to improve the model calculation accuracy and ensure that the prediction results are consistent with the actual wastewater treatment operating conditions.

[0060] By inputting multi-dimensional wastewater treatment operation data under the CASS process into a multi-task learning model, and using the shared monitoring data encoding layer to uniformly extract common process features, and then combining the first dedicated feature extraction layer to deeply integrate operating condition features and shared features, the intrinsic correlation between wastewater treatment parameters and effluent COD can be fully explored, improving the accuracy and stability of effluent chemical oxygen demand (COD) water quality prediction. At the same time, a dedicated neural network prediction branch is constructed based on the actual wastewater treatment process, which makes the prediction results more consistent with the on-site operation patterns, enabling accurate prediction of effluent water quality, providing a reliable basis for pre-emptive control and prevention of exceedance risks, and improving the intelligent management and control level of the wastewater treatment process.

[0061] Optionally, based on the above embodiments, inputting multi-source monitoring data from the unified data storage platform into the target prediction model constructed based on a multi-task learning architecture, and performing joint calculations through a shared monitoring data encoding layer to obtain the predicted pollutant emissions for a single container, may include: Cigarette production data, historical chemical oxygen demand (COD) emission data, production shift and seasonal information from a unified data storage platform are input into a target prediction model built on a multi-task learning architecture. The input data is feature-encoded through the shared monitoring data encoding layer of the target prediction model to obtain corresponding production and pollution discharge statistical monitoring data feature vectors and shared process feature vectors. These feature vectors are then input into the second dedicated feature extraction layer of the target prediction model for feature fusion. The fused features are then input into the second independent output layer of the target prediction model for calculation, yielding a single-box pollutant emission prediction result. The second dedicated feature extraction layer incorporates regression analysis into the second independent output layer to establish a mapping relationship between production and emissions.

[0062] In this embodiment of the invention, the production and pollution discharge statistical monitoring data can be specifically understood as: statistical data related to cigarette production and pollution discharge in a unified data storage platform, which may include cigarette production data, historical data of chemical oxygen demand emissions, production shifts and seasonal information, etc.

[0063] Specifically, cigarette production data, historical chemical oxygen demand (COD) emission data, production shift and seasonal information from the unified data storage platform are input into the target prediction model built on a multi-task learning architecture. The input data is feature-encoded by the shared monitoring data encoding layer of the target prediction model (the encoding process can be combined with feature normalization to improve the accuracy of feature extraction), resulting in the corresponding production and pollution discharge statistical monitoring data feature vectors and shared process feature vectors. The two types of feature vectors are then input into the second dedicated feature extraction layer of the target prediction model for feature fusion. The fusion method can be feature concatenation or weighted fusion. After fusion, the fused features are input into the second independent output layer for calculation to obtain the single-box pollutant emission prediction result.

[0064] This result can accurately output the predicted value of single-box chemical oxygen demand (COD) emissions based on the actual production data of the current period. At the same time, it can also adaptively determine the matching dynamic single-box pollutant emission threshold by combining production brand, process route, and raw material origin information. Among them, the second dedicated feature extraction layer and the second independent output layer establish a mapping relationship between production and emissions based on regression analysis, forming a single-box pollutant emission prediction model. This model establishes a regression model between COD emissions and cigarette production (multiple regression algorithms such as linear regression, ridge regression, and random forest regression can be selected, not limited to a single regression method), which supports the prediction of corresponding pollutant emissions based on cigarette production data. At the same time, historical abnormal emission data can be introduced to optimize model parameters, further improving the reliability and fit of the prediction results, and providing accurate data support for the assessment of single-box emission exceedance risk.

[0065] By inputting historical data on cigarette production, COD emissions, production shifts, and seasonal information into a multi-task learning model, and using a shared monitoring data encoding layer to uniformly encode features and generate common process features, a second dedicated feature extraction layer achieves deep integration of pollution discharge statistics features and shared process features. This fully explores the intrinsic correlation between production output, production conditions, and pollutant emissions. Combined with regression analysis, a precise mapping relationship is constructed, thereby improving the accuracy and reliability of single-box pollutant emission prediction, enabling quantitative prediction of pollution intensity, and providing data support for production scheduling and pollution control. At the same time, relying on the multi-task architecture, feature reuse and joint optimization are achieved, improving the overall generalization ability and prediction efficiency of the model.

[0066] Optionally, based on the above embodiments, multi-source monitoring data from a unified data storage platform can be input into a target prediction model constructed based on a multi-task learning architecture. Joint calculations can then be performed through a shared monitoring data encoding layer to obtain the predicted nitrogen oxide emissions. This can include: The natural gas consumption, steam consumption, water and electricity consumption, and production output data of the power workshop, silk-making workshop, and rolling and packaging workshop in the unified data storage platform are input into the target prediction model constructed based on a multi-task learning architecture. The input data is feature-encoded through the shared monitoring data encoding layer of the target prediction model to obtain the corresponding production energy consumption and equipment operation monitoring data feature vectors and shared process feature vectors. These feature vectors are then input into the third dedicated feature extraction layer of the target prediction model for feature fusion. The fused features are then input into the third independent output layer of the target prediction model for calculation to obtain the predicted nitrogen oxide emissions. The third dedicated feature extraction layer and the third independent output layer establish the correlation between production energy consumption and nitrogen oxide emissions based on multiple regression analysis.

[0067] In this embodiment of the invention, the emission of nitrogen oxides in exhaust gas can be specifically understood as: nitrogen oxides (NOx) generated during the combustion and energy consumption in the tobacco production process (focusing on the power workshop, tobacco processing workshop, and cigarette packaging workshop). x Total emissions are a core indicator for air pollution control. Production energy consumption and equipment operation monitoring data can be specifically understood as data related to production energy consumption and equipment operation collected in a unified data storage platform, specifically including natural gas consumption, steam consumption, water and electricity consumption, and production data for the power workshop, silk-making workshop, and packaging workshop.

[0068] Multiple regression analysis can be specifically understood as a data analysis method used to establish the relationship between multiple independent variables (here, production energy consumption data and output data) and a dependent variable (here, nitrogen oxide emissions), which can achieve prediction under the synergistic influence of multiple factors.

[0069] Specifically, the natural gas consumption, steam consumption, water and electricity consumption, and production data of the power workshop, silk-making workshop, and rolling and packaging workshop in the unified data storage platform are input into the target prediction model built on a multi-task learning architecture. The input data is feature-encoded through the shared monitoring data encoding layer of the target prediction model (the encoding process can introduce a feature screening mechanism to strengthen the feature weights of key energy parameters such as natural gas consumption and improve the effectiveness of encoding). The corresponding production energy consumption and equipment operation monitoring data feature vectors and shared process feature vectors are obtained. The two types of feature vectors are then input into the third dedicated feature extraction layer of the target prediction model for feature fusion. The fusion method can be weighted fusion or feature splicing. After the fusion is completed, the fused features are input into the third independent output layer for calculation to obtain the predicted result of nitrogen oxide emissions in waste gas.

[0070] Among them, the third dedicated feature extraction layer and the third independent output layer establish the correlation between production energy consumption and nitrogen oxide emissions based on multiple regression analysis, forming a nitrogen oxide emission and production energy consumption analysis model. The core of this model is to establish a relationship model between nitrogen oxide emissions and parameters such as natural gas consumption, steam consumption, and output through multiple regression analysis (multiple regression analysis can use multiple linear regression, stepwise regression, robust regression, etc., and is not limited to a single regression algorithm). Additional production monitoring data (such as wire drying temperature and production load) can also be introduced to assist in optimizing the correlation model and further improve the prediction accuracy.

[0071] After predicting nitrogen oxide emissions, multi-objective decision analysis and linear programming methods can be used to conduct a collaborative optimization assessment of energy consumption and pollution discharge. Using production energy consumption data such as natural gas consumption, steam consumption, and electricity load as optimization variables, nitrogen oxide emission compliance as a constraint, and minimum overall energy consumption as the optimization objective, a collaborative optimization model is established. By comparing predicted pollution discharge values ​​with emission limits and real-time energy consumption data with benchmark energy consumption indicators, the matching degree between pollution discharge and energy consumption under the current production conditions is determined. If the predicted nitrogen oxide emissions exceed the standard, the model automatically outputs an energy adjustment plan, such as reducing the natural gas supply to the power workshop and optimizing the steam load in the silk-making workshop. This ensures that pollution discharge meets standards while minimizing production energy consumption, thus completing the integrated collaborative optimization assessment of energy consumption and pollution discharge.

[0072] By inputting energy consumption and output data from multiple workshops into a multi-task learning model, and using a shared monitoring data encoding layer to uniformly complete feature encoding and form a shared process feature vector, and then using a third dedicated feature extraction layer to deeply integrate production energy consumption features and shared process features, the inherent correlation between multi-dimensional energy consumption parameters, output, and nitrogen oxide emissions can be fully explored. By combining multiple regression analysis to construct a correlation model, the accuracy and stability of nitrogen oxide emission prediction can be improved, providing a reliable basis for early prediction, risk warning, and control decisions of nitrogen oxide emissions in the production process. At the same time, relying on the multi-task architecture to achieve feature reuse and joint optimization, the model's generalization ability and prediction efficiency can be improved, providing technical support for the coordinated management of energy consumption and pollution discharge.

[0073] S250. The predicted results of chemical oxygen demand (COD) water quality at the outlet, the predicted results of pollutant emissions per container, and the predicted results of nitrogen oxide emissions in exhaust gas are compared with the corresponding dynamic emission thresholds to determine whether there is a risk of exceeding emission standards.

[0074] S260. When it is determined that at least one of the following is at risk of exceeding the standard: chemical oxygen demand water quality at the outlet, pollutant discharge per tank, and nitrogen oxide discharge in exhaust gas, a pollution control process control instruction matching the type of exceedance is generated and issued.

[0075] The technical solution of this invention collects multi-source monitoring data during tobacco processing and stores it in a unified data storage platform with encryption and classification. By pre-integrating historical multi-source monitoring data with tobacco processing monitoring data to construct a training dataset, it can provide comprehensive samples covering production conditions, treatment parameters, emission data, and process indicators for model training, ensuring that the model learns real and complete emission patterns. By adopting a multi-task learning architecture and setting a shared monitoring data encoding layer, it can uniformly encode multi-dimensional data and form a shared process feature vector, enabling the three prediction tasks of wastewater, single-box emissions, and nitrogen oxides to share process features. This reduces redundant feature extraction, lowers model complexity, and achieves mutual constraints and information complementarity between tasks. Furthermore, by fusing corresponding monitoring data features and shared process features through each dedicated feature extraction layer, and combining them with an independent output layer for calculation, it can improve the accuracy and consistency of the prediction of outlet chemical oxygen demand water quality, single-box pollutant emissions, and exhaust gas nitrogen oxide emissions. This solves the prediction bias problem caused by traditional models ignoring process correlations, laying a reliable foundation for subsequent accurate risk assessment and intelligent control. By inputting multi-source monitoring data into a target prediction model based on a multi-task learning architecture, and using a shared monitoring data encoding layer for joint calculation, the prediction tasks of outlet chemical oxygen demand (COD) water quality, single-container pollutant emissions, and exhaust nitrogen oxide emissions can share process characteristic information and form mutual constraints. By comparing the three types of prediction results with the corresponding dynamic emission thresholds, when an exceedance risk is determined, pollution control process control instructions matching the exceedance type are generated and issued. This enables precise control and timely coordinated regulation of exceedance risks, while reducing over-treatment or under-treatment. It achieves intelligent and refined environmental monitoring and prediction in tobacco processing, balancing environmental compliance and production efficiency, and further improves the environmental management level and emission reduction and consumption reduction capabilities of tobacco processing.

[0076] Example 3 Figure 3 This is a flowchart of another method for environmental monitoring and prediction of tobacco processing based on multi-source data fusion provided in Embodiment 3 of the present invention. This embodiment is a refinement of the above embodiment's step of "collecting multi-source monitoring data during the tobacco processing process and classifying and encrypting it for storage in a unified data storage platform." Figure 3 As shown, the method includes: S310. Collect online monitoring data, shift-level laboratory data, and batch-level production execution data during tobacco processing as multi-source monitoring data.

[0077] In this embodiment of the invention, online monitoring data can be specifically understood as: continuous operating data collected in real time by production equipment, such as temperature, flow rate, concentration, energy consumption, etc. Shift-level laboratory data can be specifically understood as: test results of water quality, emissions, material indicators, etc., tested according to the production shift cycle. Batch-level production execution data can be specifically understood as: monitoring data, output, grade, raw materials, etc., recorded on a batch-by-batch basis.

[0078] S320. The multi-source monitoring data is adaptively time-aligned and feature-weighted through an attention mechanism, and then classified, encrypted, and stored in a unified data storage platform.

[0079] In this embodiment of the invention, the attention mechanism can be specifically understood as: an adaptive weighting algorithm used to highlight important data, suppress redundant information, and achieve temporal alignment and feature enhancement. Adaptive time alignment can be specifically understood as: automatically matching and aligning data with different acquisition frequencies and timestamps on the timeline.

[0080] Specifically, during tobacco processing, real-time online monitoring data, laboratory data generated according to shift cycles, and production execution data by production batch are collected to form multi-dimensional, multi-frequency, and multi-source monitoring data. An attention mechanism is used to adaptively align data with different sampling frequencies and timestamps. Simultaneously, feature weighting is performed based on the importance of the data to pollution discharge prediction and process control, thereby strengthening key features and weakening noise interference. For example, a method combining time-series attention mechanism and dynamic time warping is used to normalize and adaptively align timestamps of high-frequency online monitoring data, mid-frequency shift laboratory data, and low-frequency batch production data. Then, scaling dot product attention is used to calculate the contribution weight of each feature to COD and nitrogen oxide pollution discharge prediction and process control. Higher weights are assigned to high-contribution features such as influent pollutant concentration, natural gas consumption, and cigarette production, while lower weights are assigned to fluctuating noise such as instantaneous current and ambient temperature and humidity. This achieves the strengthening of key features and suppression of redundant noise, providing high-quality time-series data for subsequent unified storage and model prediction.

[0081] After preprocessing, the data is classified into online monitoring, laboratory testing, and production execution categories, and encrypted using either symmetric or hash encryption methods, and finally securely stored on a unified data storage platform.

[0082] S330. Input the multi-source monitoring data in the unified data storage platform into the target prediction model built on the multi-task learning architecture, and perform joint calculations through the shared monitoring data encoding layer to obtain the predicted results of outlet chemical oxygen demand water quality, single-box pollutant emission, and exhaust gas nitrogen oxide emission.

[0083] S340. The predicted results of chemical oxygen demand (COD) water quality at the outlet, the predicted results of pollutant emissions per container, and the predicted results of nitrogen oxide emissions in exhaust gas are compared with the corresponding dynamic emission thresholds to determine whether there is a risk of exceeding emission standards.

[0084] S350: When it is determined that at least one of the following is at risk of exceeding the standard: chemical oxygen demand water quality at the outlet, pollutant discharge per tank, and nitrogen oxide discharge in exhaust gas, a pollution control process control instruction matching the type of exceedance is generated and issued.

[0085] In a specific example, a tobacco processing environmental monitoring and prediction system based on multi-source data fusion includes a data foundation module, an algorithm model module, and a business system module. The data foundation module integrates multi-source monitoring data, including tobacco processing production line business data, product output data, energy consumption monitoring data, pollution control operating condition data, online pollutant monitoring data, and manual monitoring data. The algorithm model module includes an outlet COD water quality prediction model based on CASS technology and neural networks, a single-box pollutant emission prediction model based on the correlation between output and discharge, and a nitrogen oxide emission and production energy consumption analysis model based on multiple regression analysis. The business system module includes a data analysis model management unit with model training, parameter configuration, and early warning strategy management functions, as well as a discharge monitoring visualization analysis unit that provides process flow diagram display, prediction result comparison analysis, and an early warning center. Through multi-source data fusion and intelligent algorithm model collaboration, this system achieves real-time prediction and early warning of wastewater and exhaust gas emissions, effectively improving the intelligent level of environmental management in tobacco processing, reducing pollutant emission risks, and providing reliable support for energy conservation and emission reduction decisions.

[0086] In a specific example Figure 4 This is a schematic diagram of an environmental protection management business architecture applicable to an embodiment of the present invention, such as... Figure 4As shown, the environmental protection management business architecture for cigarette factories applicable to this embodiment of the invention takes establishing a management system that meets environmental protection requirements and reducing negative impacts on the surrounding environment as its top-level goal. It constructs a complete system from four dimensions: the goal layer, the business layer, the support layer, and the perception layer. The goal layer includes goal management, which includes system management and work arrangements (covering planning management, fund management, personnel management, and performance management) and emission standard management (covering wastewater discharge standards and exhaust gas emission standards), providing top-level development guidance for environmental management. The business layer, as the core execution link, is divided into five business modules: environmental compliance management (including environmental management system construction, environmental compliance inspection, and employee training), environmental risk assessment (including environmental risk identification, environmental risk analysis, and environmental risk evaluation), waste management (including wastewater collection and treatment, exhaust gas collection and treatment, and hazardous waste collection and disposal), environmental monitoring and improvement (including wastewater monitoring and improvement, exhaust gas monitoring and improvement, and fugitive emissions monitoring and improvement), and environmental emergency response (including emergency resource management, emergency plan management, and emergency drills), comprehensively covering... The system encompasses the entire environmental management process; the support layer provides support for system operation, including data integration management (covering data collection and aggregation, equipment connection and control), equipment operation and maintenance management (covering internal and third-party operation and maintenance management), data model management (covering emission reduction and energy consumption analysis models, and effluent water quality prediction models), and network security management (covering infrastructure and security software), providing data, equipment, models, and security assurance for business operations; the perception layer, as the data source, is divided into production energy consumption data collection (covering water consumption monitoring, electricity consumption monitoring, gas consumption monitoring, and steam monitoring), treatment operation data collection (covering wastewater treatment operation and exhaust gas treatment operation), and online discharge data collection (covering wastewater online monitoring, exhaust gas online monitoring, fugitive online monitoring, and manual monitoring), realizing real-time perception and collection of data across all dimensions of production energy consumption, pollution treatment, and discharge. Through the synergistic linkage of the four-layer architecture, a comprehensive environmental protection management business system for cigarette factories is formed, covering the entire chain of objectives, business, assurance, and perception, supporting the standardized and intelligent operation of environmental management.

[0087] In a specific example Figure 5 This is a schematic diagram of an environmental protection management application architecture applicable to an embodiment of the present invention, such as... Figure 5As shown, the environmental protection management application architecture applicable to this embodiment of the invention revolves around the business processes of roll packaging production, yarn production, power production, sewage treatment, and sewage monitoring. It is constructed and planned from four levels: intelligent devices, edge applications, edge management, and cloud management. The bottom layer is the intelligent device layer, encompassing energy consumption collection devices, sewage treatment operation condition monitoring data collection and control equipment, waste gas treatment operation condition monitoring data collection and control equipment, environmental data acquisition instruments, COD (chemical oxygen demand) analyzers, total nitrogen analyzers, etc., providing hardware support for the entire process of environmental data collection. Above this is the edge application layer, including systems such as production energy consumption monitoring systems, sewage treatment operation condition monitoring systems, waste gas treatment operation condition monitoring systems, wastewater online monitoring data collection systems, and flue gas continuous online monitoring systems, realizing edge data collection and monitoring of environmental data for each business process. The next layer is the edge management layer, divided into two modules: business applications and data models. The modules include environmental compliance management (covering compliance inspections and employee training), environmental risk assessment (covering risk identification and risk evaluation), "three wastes" management (covering online monitoring and data analysis of wastewater and waste gas in this project, as well as planned monitoring and data analysis of fugitive emissions and a hazardous waste lifecycle management system), environmental emergency response (covering emergency information management and emergency drill management), and visualization analysis (covering the wastewater monitoring center in this project and planned environmental risk management). The data model module includes data analysis models, online monitoring data anomaly analysis models, wastewater discharge outlet effluent quality prediction models, and emission reduction and energy consumption reduction analysis models constructed in this invention embodiment, which together constitute a target prediction model based on a multi-task learning architecture to achieve edge-side business control and intelligent analysis. The top layer is the cloud-side management layer, which is divided into AI (Artificial Intelligence Management System). The system comprises three modules: AI (Artificial Intelligence) modeling and analysis, cloud MES (Manufacturing Execution System), and applications. The AI ​​modeling and analysis module includes cloud computing, industry-wide models, and a knowledge base. The cloud MES module includes production management, process management, quality management, equipment management, energy management, and workshop management. The application module includes smart environmental protection and smart industrial parks. This system enables cloud-based AI modeling and analysis, production business collaboration, and regulatory integration. The architecture uses diagrams to distinguish between existing systems, systems currently under construction, and planned systems, covering the entire chain of environmental protection management applications from underlying hardware data collection to cloud-based intelligent analysis, supporting the intelligent upgrade of environmental management in cigarette factories.

[0088] In a specific example Figure 6 This is a schematic diagram of an overall application architecture for a cigarette factory applicable to an embodiment of the present invention, as shown below. Figure 6As shown, the overall application architecture of the cigarette factory applicable to this embodiment of the invention aims to meet the needs of environmental protection management. The architecture is planned from four levels: cloud-side management, edge-side management, terminal applications, and intelligent equipment, comprehensively considering technical feasibility, deployment strategies, and operation and maintenance models. Among them, cloud-side management is the top-level architecture, divided into business management modules such as operation and decision-making, supply chain collaboration, and R&D-production-sales collaboration, as well as a cloud MES system (covering functional modules such as production management, process management, quality management, equipment management, energy management, workshop management, logistics management, and procurement management). It also includes extended applications such as intelligent park, comprehensive management, and smart security, providing cloud support for business decision-making and production collaboration. Edge-side management is the core execution layer, divided into the smallest unit of the MES factory (covering on-site management, collaborative control, quality traceability, new tobacco products, and SPC (Statistical Process Control)). The system encompasses various functionalities including statistical process control (SPC), auxiliary material barcode scanning, work order management, and an integrated industry platform (covering production and operation data management system, cigarette coding subsystem, specialty auxiliary material coding subsystem, and box condition association management subsystem). It also includes business segments such as production planning and scheduling, production execution control, process quality control, intelligent technology service platform, data integration platform, equipment operation and maintenance management, production warehousing and logistics, energy supply and management, and safety and environmental management. The safety and environmental management segment includes real-time safety risk monitoring and linkage, power workshop safety management system, intelligent safety management analysis, and carbon asset and waste management, fully covering all environmental management-related business functions. Each business segment is further subdivided into corresponding systems; for example, production planning and scheduling includes intelligent production scheduling, lean production control optimization, and digital twins for tobacco processing lines 4 and 5. Production execution control... The architecture includes a workshop intelligent inspection and analysis system, and tobacco processing process control. Process quality control encompasses intelligent management of quality inspection throughout the entire process. The edge application layer includes equipment connection and control, edge computing, data acquisition, real-time status monitoring, and existing systems such as centralized tobacco processing control, discrete logistics, cigarette pack data acquisition, power management, logistics overhead systems, carton and carton association, conditional association, cigarette coding, and exclusive auxiliary material coding, providing support for underlying data acquisition and edge computing. The bottom layer is the intelligent equipment layer, encompassing intelligent tobacco machines, intelligent production lines, intelligent testing equipment, warehousing and logistics equipment, and auxiliary equipment and facilities, providing the hardware foundation for full-process production and environmental management. The architecture uses diagrams to distinguish between industry applications, business management, intelligent production, and existing systems, covering the entire chain of cigarette factory application architecture from underlying hardware to cloud-based decision-making, effectively supplementing and supporting environmental protection management.

[0089] In a specific example Figure 7 This is a schematic diagram of an application architecture for environmental protection management in a cigarette factory, applicable to an embodiment of the present invention. Figure 7As shown, the environmental protection management application architecture for cigarette factories applicable to this embodiment of the invention revolves around the business links of cigarette packaging production, tobacco processing, power production, pollution treatment, and pollution monitoring. It is planned and constructed from four levels: intelligent equipment, edge applications, edge management, and cloud management. The architecture distinguishes between existing systems, systems under construction, and planned systems through diagrams, covering the entire chain of environmental management application construction. The lowest level is the intelligent equipment layer, which includes energy consumption data acquisition devices, wastewater treatment operation condition monitoring data acquisition and control equipment, exhaust gas treatment operation condition monitoring data acquisition and control equipment, environmental data acquisition instruments, COD (chemical oxygen demand) analyzers, etc. Total nitrogen analyzers and other equipment provide hardware support for the entire process of environmental data collection. Above this is the end-side application layer, which includes existing systems such as the production energy consumption monitoring system, wastewater treatment operation monitoring system, exhaust gas treatment operation monitoring system, wastewater online monitoring data acquisition system, and flue gas continuous online monitoring system, enabling end-side collection and monitoring of environmental data in various business processes. Above that is the edge management layer, which is divided into two modules: business applications and data models. The business application module includes environmental compliance management (covering compliance inspections and employee training), environmental risk assessment (covering risk identification and risk evaluation), and "three wastes" management (covering the wastewater online monitoring system in this project). Together with data analysis, online monitoring and data analysis of exhaust gas, as well as the planned and constructed monitoring and data analysis of fugitive emissions, hazardous waste life cycle management system, environmental emergency response (covering emergency information management and emergency drill management), and visualization analysis (covering the sewage discharge monitoring center under construction and the planned environmental risk management system), the data model module includes data analysis and judgment models, online monitoring data anomaly analysis and judgment models, sewage discharge outlet effluent quality prediction models and emission reduction and energy consumption reduction analysis and judgment models, etc., to form a target prediction model based on a multi-task learning architecture, realizing edge-side business control and intelligence. It can perform analysis; the top layer is the cloud-side management layer, which is divided into three modules: AI modeling and analysis, cloud MES, and regulatory department applications. The AI ​​modeling and analysis module includes cloud computing, industry big data models, and knowledge bases. The cloud MES module includes production management, process management, quality management, equipment management, energy management, and workshop management. The regulatory department application module includes smart environmental protection and smart parks. It realizes cloud-based AI modeling and analysis, production business collaboration, and connection with regulatory departments. Through the collaborative linkage of the four-layer architecture, it covers the entire chain of environmental protection management application construction from the underlying hardware data collection to cloud-based intelligent analysis, supporting the intelligent upgrade of the cigarette factory's environmental management.

[0090] In a specific example Figure 8 This is a schematic diagram of a data architecture applicable to an embodiment of the present invention, such as... Figure 8As shown, the data architecture for environmental protection management in cigarette factories applicable to this embodiment of the invention revolves around five business types: cigarette packaging production, tobacco processing production, power production, pollution control, and pollution monitoring. It is divided into four levels: business type, data type, data details, and data application. The business type layer covers cigarette packaging production, tobacco processing production, power production, pollution control, and pollution monitoring, providing the business sources for the data architecture. The data type layer inherits the business types and is divided into three categories: production energy consumption data, treatment operation data, and online emission data. Production energy consumption data includes water consumption monitoring, electricity consumption monitoring, gas consumption monitoring, and steam monitoring; treatment operation data includes wastewater treatment operation data and waste gas treatment operation data; and online emission data includes wastewater online monitoring, waste gas online monitoring, fugitive emissions online monitoring, and manual monitoring, achieving data classification and organization. The data details layer further breaks down each type of data. Production energy consumption data details include real-time data... According to statistical indicators such as data collection, historical data, measured values, benchmark values, predicted values, deviation values, comprehensive energy consumption, and unit energy consumption, the detailed data items for treatment operation include indicators such as influent pollutant concentration, aeration time, dissolved oxygen concentration, sludge concentration, sludge return ratio, hydraulic retention time, sedimentation time, and decanting time. The detailed data items for online emissions include indicators such as monitoring points, pollutants, monitoring equipment, operation and maintenance information, real-time values, average values, maximum values, minimum values, and emission volumes of pollutants, providing detailed support for data collection and analysis. The bottom layer is the data application layer, which includes a data analysis model management system and a sewage discharge monitoring visualization analysis system, realizing the modeling analysis and visualization management of all environmental protection data. Through the hierarchical progression and data flow of the four-layer architecture, a data architecture for the environmental protection management of cigarette factories covering the entire chain of business, type, details, and applications is formed, providing data support for intelligent manufacturing and intelligent environmental management.

[0091] In a specific example Figure 9 This is a schematic diagram of a technical architecture applicable to an embodiment of the present invention, such as... Figure 9As shown, the environmental protection management technology architecture for cigarette factories applicable to this embodiment of the invention is based on the industrial internet system and the cybersecurity system as a whole. It is constructed from three levels: intelligent equipment and facilities, digital infrastructure, and environmental protection management. The bottom layer is the intelligent equipment and facilities layer, which covers intelligent production lines (for monitoring energy consumption in tobacco processing, cigarette packaging, and power production lines), intelligent treatment (for monitoring the operating conditions of wastewater and exhaust gas treatment facilities), and intelligent sewage discharge (for online monitoring of wastewater, exhaust gas, and fugitive emissions), providing hardware support for the entire process of environmental protection management. Above this is the digital infrastructure layer, which is divided into infrastructure, network, and technology platform modules. The infrastructure module includes the relational database MySQL (My Structured Query Language), the key-value database Redis (Remote Dictionary Server), the operating system OpenEuler (open source Euler operating system), the container Tomcat (Tomcat server), the message queue Kafka (Kafka message queue), and the underlying facilities HCS (cloud stack), edge hyper-convergence, and end-sensing devices. The network module includes connection devices such as routers, switches, firewalls, and data acquisition gateways, as well as the network protocol HTTP (HyperText Transfer Protocol). Protocols (Hypertext Transfer Protocol), TCP / IP (Transmission Control Protocol / Internet Protocol), MQTT (Message Queuing Telemetry Transport), and HJ212 (HJ212, data transmission standard for online pollutant monitoring systems) provide underlying hardware, software, and network support for system operation. The technology platform section is divided into development languages ​​(Java, development frameworks SpringMVC (Spring Model-View-Controller), MyBatis (persistence layer framework), MiniUI (Mini User Interface framework)), IT (Information Technology) data integration (API (Application Programming Interface) gateway / ESB (Enterprise Service Bus) and data transmission channels), and intelligent technology services (AI training, AI execution, AI optimization, AI reporting), providing technical support for system development, data integration, and intelligent analysis.The top layer is the environmental protection management layer, which includes a data analysis model management platform and a pollution discharge monitoring visualization analysis platform. The data analysis model management platform covers effluent water quality prediction models, water quality early warnings, emission reduction and energy conservation analysis models, and over-emission early warnings. The pollution discharge monitoring visualization analysis platform covers monitoring data viewing, prediction result querying, key indicator analysis, and early warning center alerts, enabling intelligent analysis and visualized management of environmental protection operations. The architecture is guided by SOA (Service-Oriented Architecture) design principles, adhering to high cohesion and low coupling principles. It adopts a multi-layered B / S (Browser / Server) architecture and component development technology, combined with microservice architecture, Redis caching, and other technologies to ensure system compatibility with the factory's existing hardware and software environment and the overall technical specifications of China Tobacco, supporting the intelligent operation of the enterprise's environmental management.

[0092] In a specific example Figure 10 This is a schematic diagram of a deployment architecture applicable to an embodiment of the present invention, such as... Figure 10As shown, the environmental protection management deployment architecture for cigarette factories applicable to this embodiment of the invention is deployed in four layers: intelligent equipment and facilities, IoT (Internet of Things), data integration platform, and environmental protection management platform. The bottom layer is the intelligent equipment and facilities layer, encompassing intelligent production lines (for monitoring energy consumption in tobacco processing, packaging, and power production lines), intelligent treatment (for monitoring the operating conditions of wastewater and exhaust gas treatment facilities), and intelligent sewage discharge (for online monitoring of wastewater, exhaust gas, and fugitive emissions), providing underlying data acquisition hardware support for the system. The intelligent equipment and facilities achieve data interaction through the IoT layer, which includes equipment connection control, data acquisition, and data on production energy consumption, treatment conditions, and online emissions, completing the collection, aggregation, and transmission of data from the underlying equipment. The data integration platform receives data from the IoT layer, including data resource management and operation, data resource sharing services, data resource synchronization management, data storage, data aggregation, and data standards, achieving unified management, allocation, storage, and sharing of all environmental protection data, providing data support for upper-layer applications. The top layer is... The environmental protection management platform layer is divided into three modules: a data analysis model management platform, a pollution discharge monitoring visualization analysis platform, and an intelligent technology service platform. The data analysis model management platform covers effluent water quality prediction models, emission reduction and energy conservation analysis models, water quality early warnings, and over-discharge early warnings. The pollution discharge monitoring visualization analysis platform covers monitoring data viewing, prediction result querying, key indicator analysis, and early warning center alerts. The intelligent technology service platform covers AI training, AI execution, AI optimization, and AI reporting, enabling intelligent analysis, visualized monitoring, and AI-driven full-process management of environmental protection operations. Simultaneously, the architecture integrates with the enterprise's existing cloud-network-edge-device hyperconverged system. Application servers and database servers are configured according to application scale, data volume, and computing power, meeting the requirements of Level 3 Information Security Protection. Through multi-dimensional security strategies encompassing equipment security, control security, network security, application security, and data security, the system ensures safe and stable operation throughout the entire process, supporting the intelligent upgrade of the cigarette factory's environmental management.

[0093] The technical solution of this invention, by collecting online monitoring data, shift-level test data, and batch-level production execution data respectively, can comprehensively cover environmental protection and production-related information throughout the entire tobacco processing process, avoiding analytical bias caused by single data sources. Utilizing an attention mechanism to adaptively align and weight features of multi-source monitoring data across multiple time scales solves the problems of inconsistent time sequences and significant differences in importance among different types of data, making data characteristics more closely aligned with actual emission patterns. Furthermore, the processed data is categorized, encrypted, and stored on a unified data storage platform, enabling centralized and standardized management of multi-source data, ensuring data security, and providing high-quality, highly available data input for subsequent prediction models, further improving the accuracy and stability of overall prediction and control. By inputting multi-source monitoring data into a target prediction model based on a multi-task learning architecture and jointly calculating it using a shared monitoring data encoding layer, the prediction tasks for effluent chemical oxygen demand (COD) water quality, single-container pollutant emissions, and nitrogen oxide emissions can share process characteristic information and form mutual constraints. Compared with a single prediction model, this not only improves the accuracy and consistency of the prediction results but also fully explores the intrinsic relationship between tobacco processing technology and pollutant emissions, solving the prediction bias problem caused by neglecting process characteristics in traditional prediction methods. By comparing the three types of prediction results with corresponding dynamic emission thresholds, the risk of exceeding emission standards can be accurately determined, avoiding the shortcomings of insufficient adaptability of fixed thresholds. When an exceedance risk is determined, pollution control process control instructions matching the exceedance type are generated and issued, enabling precise control and timely coordinated regulation of exceedance risks. Compared with the traditional method of only monitoring and alarming without targeted regulation, this can effectively reduce the probability of exceeding emission standards and reduce over-treatment or under-treatment. This achieves intelligent and refined environmental monitoring and prediction for tobacco processing, balancing environmental compliance and production efficiency, and further improving the environmental management level and emission reduction and consumption reduction capabilities of tobacco processing.

[0094] Example 4 Figure 11 This is a schematic diagram of the structure of a tobacco processing environmental monitoring and prediction device based on multi-source data fusion, provided in Embodiment 4 of the present invention. Figure 11 As shown, the device includes: a data acquisition and storage module 1110, a model prediction module 1120, a threshold determination module 1130, and a process control module 1140, wherein: The data acquisition and storage module 1110 is used to collect multi-source monitoring data during the tobacco processing process and classify and encrypt the data for storage to a unified data storage platform. The model prediction module 1120 is used to input multi-source monitoring data from the unified data storage platform into the target prediction model built on a multi-task learning architecture, and perform joint calculations through a shared monitoring data encoding layer to obtain the predicted results of outlet chemical oxygen demand water quality, single-box pollutant emission, and exhaust gas nitrogen oxide emission. The threshold determination module 1130 is used to compare the predicted results of chemical oxygen demand water quality at the outlet, the predicted results of pollutant emissions per tank, and the predicted results of nitrogen oxide emissions in exhaust gas with the corresponding dynamic emission thresholds to determine whether there is a risk of exceeding emission standards. The process control module 1140 is used to generate and issue pollution control process control instructions that match the type of exceedance when it is determined that at least one of the following is at risk of exceeding the standard: the outlet chemical oxygen demand water quality, the single tank pollutant emission, and the exhaust gas nitrogen oxide emission.

[0095] The technical solution of this invention, by collecting multi-source monitoring data during tobacco processing and classifying and encrypting it for storage in a unified data storage platform, enables centralized management and secure storage of multi-type and multi-dimensional environmentally related data. This avoids management chaos and leakage risks caused by data dispersion, providing complete and reliable data support for subsequent predictive analysis. The multi-source monitoring data is input into a target prediction model based on a multi-task learning architecture, and joint calculations are performed using a shared monitoring data encoding layer. This allows the three prediction tasks—outlet chemical oxygen demand (COD) water quality, single-container pollutant emissions, and nitrogen oxide emissions—to share process characteristic information and form mutual constraints. Compared to a single prediction model, this not only improves the accuracy and consistency of the prediction results but also fully explores the relationship between tobacco processing technology and environmental performance. The inherent correlation between pollutant emissions solves the prediction bias problem caused by neglecting process characteristics in traditional prediction methods. By comparing the three types of prediction results with the corresponding dynamic emission thresholds, the risk of exceeding emission standards for various types can be accurately determined, avoiding the shortcomings of insufficient adaptability of fixed thresholds. When the risk of exceeding standards is determined, pollution control process control instructions matching the type of exceeding standard are generated and issued, enabling precise control and timely coordinated regulation of exceeding standards risks. Compared with the traditional method of only monitoring and alarming without targeted regulation, it can effectively reduce the probability of exceeding emission standards, while reducing the situation of over-treatment or under-treatment. It realizes intelligent and refined environmental monitoring and prediction in tobacco processing, taking into account both environmental compliance and production efficiency, and further improving the environmental management level and emission reduction and consumption reduction capabilities of tobacco processing.

[0096] Furthermore, based on the above embodiments, the tobacco processing environmental monitoring and prediction device based on multi-source data fusion may further include: a training construction module and an iterative training module, wherein; The training module is used to construct a training dataset based on historical multi-source monitoring data and tobacco processing monitoring data in the unified data storage platform before inputting the multi-source monitoring data in the unified data storage platform into the target prediction model built based on the multi-task learning architecture. The training dataset includes production conditions, governance parameters, emission data, and process indicators. The iterative training module is used to iteratively train the prediction model using a multi-task learning architecture based on the training dataset to obtain the target prediction model. The target prediction model includes a shared monitoring data encoding layer, a first dedicated feature extraction layer, a second dedicated feature extraction layer, a third dedicated feature extraction layer, a first independent output layer, a second independent output layer, and a third independent output layer. The shared monitoring data encoding layer performs unified feature extraction and vector encoding on the input historical multi-source monitoring data and tobacco processing monitoring data, generating feature vectors for wastewater treatment operating conditions monitoring data, production and wastewater discharge statistics monitoring data, production energy consumption and equipment operation monitoring data, and a shared process feature vector. These feature vectors are then synchronously input into the matching first, second, and third dedicated feature extraction layers. The first dedicated feature extraction layer fuses the wastewater treatment operating conditions monitoring data feature vectors with the shared process feature vectors, outputting a value for export chemical demand. The system comprises three layers: a first dedicated feature layer for predicting oxygen demand and water quality; a second dedicated feature extraction layer that fuses the feature vectors of production and wastewater discharge statistical monitoring data with the shared process feature vectors to output dedicated feature features for predicting pollutant emissions from a single tank; a third dedicated feature extraction layer that fuses the feature vectors of production energy consumption and equipment operation monitoring data with the shared process feature vectors to output dedicated feature features for predicting nitrogen oxide emissions from exhaust gas; a first independent output layer that calculates based on the dedicated feature features output by the first dedicated feature extraction layer to output the outlet chemical oxygen demand (COD) water quality prediction result; a second independent output layer that calculates based on the dedicated feature features output by the second dedicated feature extraction layer to output the single tank pollutant emission prediction result; and a third independent output layer that calculates based on the dedicated feature features output by the third dedicated feature extraction layer to output the exhaust nitrogen oxide emission prediction result.

[0097] Based on the above embodiments, the model prediction module 1120 is specifically used for: The data on influent pollutant concentration, aeration time, dissolved oxygen concentration, sludge concentration, sludge return ratio, hydraulic retention time, sedimentation time, and decanting time in the wastewater treatment process from the unified data storage platform are input into the target prediction model built on a multi-task learning architecture. Through the shared monitoring data encoding layer of the target prediction model, the input data is feature-encoded to obtain the corresponding wastewater treatment operating condition monitoring data feature vector and shared process feature vector. The feature vector of the wastewater treatment operating condition monitoring data and the feature vector of the shared process are input into the first dedicated feature extraction layer of the target prediction model for feature fusion, and the fused features are input into the first independent output layer of the target prediction model for calculation to obtain the predicted result of chemical oxygen demand water quality at the wastewater discharge outlet. The first dedicated feature extraction layer and the first independent output layer are neural network prediction models constructed based on the cyclic activated sludge process.

[0098] Based on the above embodiments, the model prediction module 1120 is further configured to: The cigarette production data, historical chemical oxygen demand emission data, production shift and seasonal information in the unified data storage platform are input into the target prediction model built on a multi-task learning architecture. Through the shared monitoring data encoding layer of the target prediction model, the input data is feature-encoded to obtain the corresponding production and pollution discharge statistical monitoring data feature vectors and shared process feature vectors. The feature vectors of the production and pollution discharge statistical monitoring data and the shared process feature vector are input into the second dedicated feature extraction layer of the target prediction model for feature fusion, and the fused features are input into the second independent output layer of the target prediction model for calculation to obtain the single-box pollutant emission prediction result. In this process, the second dedicated feature extraction layer is mixed into the second independent output layer to establish a mapping relationship between production and emissions based on regression analysis.

[0099] Based on the above embodiments, the model prediction module 1120 is further configured to: The natural gas consumption, steam consumption, water and electricity consumption, and production data of the power workshop, silk-making workshop, and rolling and packaging workshop in the unified data storage platform are input into the target prediction model built based on a multi-task learning architecture. Through the shared monitoring data encoding layer of the target prediction model, the input data is feature-encoded to obtain the corresponding production energy consumption and equipment operation monitoring data feature vector and shared process feature vector. The feature vectors of the production energy consumption and equipment operation monitoring data and the shared process feature vectors are input into the third dedicated feature extraction layer of the target prediction model for feature fusion, and the fused features are input into the third independent output layer of the target prediction model for calculation to obtain the predicted result of nitrogen oxide emissions in waste gas. The third dedicated feature extraction layer and the third independent output layer establish the correlation between production energy consumption and nitrogen oxide emissions based on multiple regression analysis.

[0100] Based on the above embodiments, the data acquisition and storage module 1110 is specifically used for: Online monitoring data, shift-level laboratory data, and batch-level production execution data were collected during the tobacco processing process to serve as multi-source monitoring data. The multi-source monitoring data is adaptively time-aligned and feature-weighted using an attention mechanism, and then classified, encrypted, and stored in a unified data storage platform.

[0101] Based on the above embodiments, the process control module 1140 is specifically used for: Based on the predicted results of chemical oxygen demand (COD) water quality at the outlet, the predicted results of pollutant emissions per tank, and the predicted results of nitrogen oxide emissions in exhaust gas, an emission anomaly propagation diagram model is constructed. When any prediction result is determined to have a risk of exceeding the standard, the emission anomaly propagation model is used to trace back to the corresponding specific process unit and equipment operating status to locate the root cause of the emission anomaly. Based on the identified root causes, a comprehensive control scheme is generated that coordinates front-end production process optimization instructions with end-of-pipe pollution control equipment control instructions, and the emission change trend after the implementation of the comprehensive control scheme is pre-evaluated. Based on the results of the preliminary assessment, control instructions for pollution control processes that match the types of pollution exceeding the standards are determined and issued.

[0102] The tobacco processing environmental monitoring and prediction device based on multi-source data fusion provided in the embodiments of the present invention can execute the tobacco processing environmental monitoring and prediction method based on multi-source data fusion provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0103] The collection, storage, use, processing, transmission, provision, and disclosure of user personal information involved in the technical solution disclosed herein comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0104] Example 5 Figure 12 A schematic diagram of an electronic device 10, which can be used to implement embodiments of the present invention, is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (e.g., helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0105] like Figure 12As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded from storage unit 18 into the RAM 13. The RAM 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0106] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0107] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, central processing unit (CPU), graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, digital signal processors (DSPs), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as the tobacco processing environmental monitoring and prediction method based on multi-source data fusion, i.e.: Collect multi-source monitoring data during the tobacco processing process, and categorize and encrypt the data before storing it in a unified data storage platform; Multi-source monitoring data from a unified data storage platform is input into a target prediction model built on a multi-task learning architecture. Joint calculations are performed through a shared monitoring data encoding layer to obtain prediction results for outlet chemical oxygen demand water quality, single-box pollutant emissions, and exhaust gas nitrogen oxide emissions. The predicted results of chemical oxygen demand (COD) water quality at the outlet, the predicted results of pollutant emissions per container, and the predicted results of nitrogen oxide emissions in exhaust gas are compared with the corresponding dynamic emission thresholds to determine whether there is a risk of exceeding emission standards. When it is determined that at least one of the following is at risk of exceeding the standard: chemical oxygen demand (COD) water quality at the outlet, pollutant emissions per tank, and nitrogen oxide emissions in exhaust gas, a pollution control process control instruction matching the type of exceedance will be generated and issued.

[0108] In some embodiments, the tobacco processing environmental monitoring and prediction method based on multi-source data fusion can be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program can be loaded and / or installed on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the tobacco processing environmental monitoring and prediction method based on multi-source data fusion described above can be performed. Alternatively, in other embodiments, processor 11 can be configured to perform the tobacco processing environmental monitoring and prediction method based on multi-source data fusion by any other suitable means (e.g., by means of firmware).

[0109] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0110] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0111] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0112] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0113] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or computing systems that include middleware components (e.g., application servers), or computing systems that include frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0114] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0115] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0116] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for environmental monitoring and prediction of tobacco processing based on multi-source data fusion, characterized in that, include: Collect multi-source monitoring data during the tobacco processing process, and categorize and encrypt the data before storing it in a unified data storage platform; Multi-source monitoring data from a unified data storage platform is input into a target prediction model built on a multi-task learning architecture. Joint calculations are performed through a shared monitoring data encoding layer to obtain prediction results for outlet chemical oxygen demand water quality, single-box pollutant emissions, and exhaust gas nitrogen oxide emissions. The predicted results of chemical oxygen demand (COD) water quality at the outlet, the predicted results of pollutant emissions per container, and the predicted results of nitrogen oxide emissions in exhaust gas are compared with the corresponding dynamic emission thresholds to determine whether there is a risk of exceeding emission standards. When it is determined that at least one of the following is at risk of exceeding the standard: chemical oxygen demand (COD) water quality at the outlet, pollutant emissions per tank, and nitrogen oxide emissions in exhaust gas, a pollution control process control instruction matching the type of exceedance will be generated and issued.

2. The method according to claim 1, characterized in that, Before inputting multi-source monitoring data from the unified data storage platform into the target prediction model built on a multi-task learning architecture, the following steps are also included: A training dataset is constructed based on historical multi-source monitoring data and tobacco processing monitoring data in a unified data storage platform. The training dataset includes production conditions, treatment parameters, emission data, and process indicators. The prediction model using a multi-task learning architecture is iteratively trained based on the training dataset to obtain the target prediction model. The target prediction model includes a shared monitoring data encoding layer, a first dedicated feature extraction layer, a second dedicated feature extraction layer, a third dedicated feature extraction layer, a first independent output layer, a second independent output layer, and a third independent output layer. The shared monitoring data encoding layer is used to perform unified feature extraction and vector encoding on the input historical multi-source monitoring data and tobacco processing monitoring data, respectively generating feature vectors for wastewater treatment operating condition monitoring data, production and sewage discharge statistics monitoring data, production energy consumption and equipment operation monitoring data, and shared process feature vectors. The above feature vectors are then synchronously input into the matching first dedicated feature extraction layer, second dedicated feature extraction layer, and third dedicated feature extraction layer. The first dedicated feature extraction layer fuses the feature vector of the wastewater treatment operation monitoring data with the shared process feature vector to output dedicated task features for predicting the chemical oxygen demand (COD) water quality at the outlet. The second dedicated feature extraction layer fuses the feature vectors of production and pollution discharge statistical monitoring data with the shared process feature vectors to output dedicated task features for predicting pollutant emissions from a single container. The third dedicated feature extraction layer fuses the feature vectors of production energy consumption and equipment operation monitoring data with the shared process feature vectors to output dedicated task features for predicting nitrogen oxide emissions in waste gas. The first independent output layer calculates based on the exclusive task features output by the first exclusive feature extraction layer and outputs the predicted chemical oxygen demand (COD) water quality at the outlet; the second independent output layer calculates based on the exclusive task features output by the second exclusive feature extraction layer and outputs the predicted pollutant emissions per tank; the third independent output layer calculates based on the exclusive task features output by the third exclusive feature extraction layer and outputs the predicted nitrogen oxide emissions in exhaust gas.

3. The method according to claim 2, characterized in that, Multi-source monitoring data from a unified data storage platform is input into a target prediction model built on a multi-task learning architecture. Joint calculations are performed through a shared monitoring data encoding layer to obtain the predicted chemical oxygen demand (COD) water quality at the outlet, including: The data on influent pollutant concentration, aeration time, dissolved oxygen concentration, sludge concentration, sludge return ratio, hydraulic retention time, sedimentation time, and decanting time in the wastewater treatment process from the unified data storage platform are input into the target prediction model built on a multi-task learning architecture. Through the shared monitoring data encoding layer of the target prediction model, the input data is feature-encoded to obtain the corresponding wastewater treatment operating condition monitoring data feature vector and shared process feature vector. The feature vector of the wastewater treatment operating condition monitoring data and the feature vector of the shared process are input into the first dedicated feature extraction layer of the target prediction model for feature fusion, and the fused features are input into the first independent output layer of the target prediction model for calculation to obtain the predicted result of chemical oxygen demand water quality at the wastewater discharge outlet. The first dedicated feature extraction layer and the first independent output layer are neural network prediction models constructed based on the cyclic activated sludge process.

4. The method according to claim 2, characterized in that, Multi-source monitoring data from a unified data storage platform is input into a target prediction model built on a multi-task learning architecture. Joint calculations are performed through a shared monitoring data encoding layer to obtain the predicted pollutant emissions for a single container, including: The cigarette production data, historical chemical oxygen demand emission data, production shift and seasonal information in the unified data storage platform are input into the target prediction model built on a multi-task learning architecture. Through the shared monitoring data encoding layer of the target prediction model, the input data is feature-encoded to obtain the corresponding production and pollution discharge statistical monitoring data feature vectors and shared process feature vectors. The feature vectors of the production and pollution discharge statistical monitoring data and the shared process feature vector are input into the second dedicated feature extraction layer of the target prediction model for feature fusion, and the fused features are input into the second independent output layer of the target prediction model for calculation to obtain the single-box pollutant emission prediction result. In this process, the second dedicated feature extraction layer is mixed into the second independent output layer to establish a mapping relationship between production and emissions based on regression analysis.

5. The method according to claim 2, characterized in that, Multi-source monitoring data from a unified data storage platform is input into a target prediction model built on a multi-task learning architecture. Joint calculations are performed through a shared monitoring data encoding layer to obtain predicted results for nitrogen oxide emissions, including: The natural gas consumption, steam consumption, water and electricity consumption, and production data of the power workshop, silk-making workshop, and rolling and packaging workshop in the unified data storage platform are input into the target prediction model built based on a multi-task learning architecture. Through the shared monitoring data encoding layer of the target prediction model, the input data is feature-encoded to obtain the corresponding production energy consumption and equipment operation monitoring data feature vector and shared process feature vector. The feature vectors of the production energy consumption and equipment operation monitoring data and the shared process feature vectors are input into the third dedicated feature extraction layer of the target prediction model for feature fusion, and the fused features are input into the third independent output layer of the target prediction model for calculation to obtain the predicted result of nitrogen oxide emissions in waste gas. The third dedicated feature extraction layer and the third independent output layer establish the correlation between production energy consumption and nitrogen oxide emissions based on multiple regression analysis.

6. The method according to claim 1, characterized in that, The collection of multi-source monitoring data during the tobacco processing process, and its categorized, encrypted storage on a unified data storage platform, includes: Online monitoring data, shift-level laboratory data, and batch-level production execution data were collected during the tobacco processing process to serve as multi-source monitoring data. The multi-source monitoring data is adaptively time-aligned and feature-weighted using an attention mechanism, and then classified, encrypted, and stored in a unified data storage platform.

7. The method according to claim 1, characterized in that, Generate and issue control instructions for pollution control processes that match the type of exceedance, including: Based on the predicted results of chemical oxygen demand (COD) water quality at the outlet, the predicted results of pollutant emissions per tank, and the predicted results of nitrogen oxide emissions in exhaust gas, an emission anomaly propagation diagram model is constructed. When any prediction result is determined to have a risk of exceeding the standard, the emission anomaly propagation model is used to trace back to the corresponding specific process unit and equipment operating status to locate the root cause of the emission anomaly. Based on the identified root causes, a comprehensive control scheme is generated that coordinates front-end production process optimization instructions with end-of-pipe pollution control equipment control instructions, and the emission change trend after the implementation of the comprehensive control scheme is pre-evaluated. Based on the results of the preliminary assessment, control instructions for pollution control processes that match the types of pollution exceeding the standards are determined and issued.

8. An electronic device, characterized in that, The electronic device includes: At least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores a computer program executable by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the tobacco processing environmental monitoring and prediction method based on multi-source data fusion as described in any one of claims 1-7.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that cause a processor to execute the method for environmental monitoring and prediction of tobacco processing based on multi-source data fusion as described in any one of claims 1-7.

10. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the method for environmental monitoring and prediction of tobacco processing based on multi-source data fusion according to any one of claims 1-7.

Images