Enterprise classification management method based on pollution discharge permission data evaluation index

Abstract

The invention discloses an enterprise classification management method based on a pollution discharge permission data evaluation index, and relates to the technical field of environment supervision, and the method comprises the following steps: S1, obtaining a real-time monitoring data flow and a pollution discharge permission execution report of an enterprise, and the real-time monitoring data flow collects pollutant concentration data at an hour frequency through an Internet of Things sensor. According to the enterprise classification management method based on the pollution discharge permission data evaluation index, the limitation of a traditional static data evaluation system is broken through by constructing a coupled dual-channel decision engine and a space-time self-adaptive evaluation model, fractal time sequence compression and process logic alignment are performed on monitoring data and declaration data, lossless fusion of multi-source heterogeneous data is realized, and the enterprise classification management method based on the pollution discharge permission data evaluation index is realized. Key management enterprises are highlighted through classification levels, resource intensified management and control are supervised in a centralized manner, and the intervention efficiency on high-pollution emission behaviors is effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of environmental supervision, and specifically to a method for classifying and managing enterprises based on pollution discharge permit data evaluation indicators. Background Art

[0002] In the field of environmental supervision, the method for classifying and managing enterprises based on pollution discharge permit data has become an important tool for implementing differentiated supervision; the pollutant emissions are evaluated standardly through preset fixed thresholds, and then the enterprise management levels are divided; however, this static and periodic data processing mode has significant defects: the data in the pollution discharge permit implementation report is essentially the phased summary result of the enterprise production activities, which can reflect the real emission behavior of the enterprise in a certain stage to a certain extent; during the actual production process of the enterprise, affected by factors such as the equipment operation status, process parameter fluctuations, and raw material quality changes, the pollutant emission concentration and total amount may fluctuate.

[0003] The existing technology cannot effectively integrate the high-frequency monitoring data collected in real time by Internet of Things sensors and the low-frequency declaration data, resulting in the classification evaluation model being in the state of "judging the new with the old" for a long time. The supervision department can only make management decisions based on the lagged historical data and cannot accurately identify the key management enterprises. In addition, the fixed threshold evaluation system ignores the real-time changes of external variables such as regional environmental capacity and meteorological and hydrological conditions, and the impact differences of the same emission intensity on the ecological environment under different spatio-temporal backgrounds are not quantitatively fed back, further weakening the scientificity and applicability of the classification results. Summary of the Invention

[0004] (1) Technical Problems to be Solved Aiming at the deficiencies of the existing technology, the present invention provides a method for classifying and managing enterprises based on pollution discharge permit data evaluation indicators, and solves the problem of the failure of the integration of real-time data and periodic management decisions in the existing technology.

[0005] (2) Technical Solutions To achieve the above object, the present invention is realized through the following technical solutions: A method for classifying and managing enterprises based on pollution discharge permit data evaluation indicators, comprising the following steps: S1: Obtain the real-time monitoring data stream of the enterprise and the pollution discharge permit implementation report. The real-time monitoring data stream collects pollutant concentration data through Internet of Things sensors at an hourly frequency, and the pollution discharge permit implementation report includes production process parameters and the enterprise environmental management ledger; S2: Input the real-time monitoring data stream and the pollutant discharge permit execution report into the coupled dual-channel decision engine. Use the spatio-temporal encoder-decoder module to perform fractal time series compression on the high-frequency monitoring data stream, extract the start-stop inflection points of the equipment and the curvature features of the emission trend. At the same time, reverse-derive the hidden emission nodes in the process chain based on the production log to generate enhanced declaration data, and use the process feature similarity matrix to achieve data fusion across time scales to generate an emission feature vector. The construction of the coupled dual-channel decision engine (DCDDE) architecture includes hardware deployment and software module interaction. Among them, the hardware deployment includes edge computing nodes and cloud servers. The edge computing nodes include deploying industrial gateways at the enterprise end, integrating Modbus and OPC UA protocol parsing modules to collect sensor data in real time with a sampling frequency of 1 Hz. The cloud server includes adopting a Kubernetes cluster architecture, configuring dedicated containers to process high-frequency data streams and low-frequency declaration data. The dedicated container for processing high-frequency data streams is an Apache Kafka message queue, and the low-frequency declaration data is a PostgreSQL database. The software module interaction includes a data preprocessing module and a process logic verification module. The data preprocessing module is used to perform sliding window normalization on the original monitoring data with a window length of 60 seconds and a step size of 5 seconds, and eliminate the sensor drift error. The process logic verification module is used to synchronize the work order execution status of the enterprise MES system every 10 minutes to verify the timestamp consistency between the production log and the declaration data. S3: Input the emission feature vector into the environmental capacity responsive evaluation model, access the meteorological grid forecast data and the hydrodynamic model to calculate the regional environmental capacity correction coefficient in real time. Combine the non-linear relationship matrix between process parameters and pollutant generation amounts constructed based on the knowledge graph, and the pollutant spatial influence weight function generated according to the hydrodynamic equation to calculate the enterprise pollution contribution degree weight. The correction of the hydrodynamic equation includes the corrected three-dimensional diffusion equation: ; where C is the pollutant concentration, t is the time, is the partial differential symbol, x, y, z are the spatial coordinates, V is the pollutant migration rate vector, K x , K y , K z It is updated every 6 hours according to satellite data, and the update rule is: , where K x´ is the corrected horizontal turbulent diffusion coefficient, K x is the original horizontal turbulent diffusion coefficient, θ is the real-time angle between the satellite-inverted diffusion direction and the dominant wind direction, is the satellite-inverted pollution diffusion vector field, is the maximum pollution diffusion rate in the region; S4: Input the preliminary classification results into a reverse process simulator and a pollution carrying capacity calculation model in sequence for double verification. The reverse process simulator reversely deduces the theoretical values of production process parameters through an LSTM neural network and conducts deviation analysis with the declared parameters. The pollution carrying capacity calculation model simulates the environmental capacity threshold using the Monte Carlo method, and the environmental capacity threshold is the ecological background; S5: Output the classification results after double verification, and update the spatial influence weight function parameters according to a preset update period based on the atmospheric pollution diffusion trajectory retrieved by satellite remote sensing.

[0006] Preferably, the fractal time series compression includes: Conduct multifractal dimension analysis on the high-frequency monitoring data stream to extract key time windows with process relevance. Among them, the process of multifractal dimension analysis includes: Perform time series segmentation on the monitoring data stream with a window length of 5 minutes to cover a typical production cycle; Calculate the generalized Hurst exponent H(q) of each window, select the scaling range of q from -5 to +5, and generate the multifractal spectrum f(α); Screen the time windows with an f(α) spectral width greater than 0.25 as key process nodes, such as equipment start / stop and raw material switching.

[0007] Identify the persistence characteristics of the emission trend through wavelet transform and Hurst exponent calculation. During the wavelet transform and Hurst exponent calculation, use the Daubechies 4 wavelet basis to decompose the data stream into 5 layers, extract the detail coefficients d3 - d5 corresponding to the 2 - 8 minute cycle characteristics; calculate the Hurst exponent for the approximation coefficient a5, and when H is greater than 0.75, determine it as a persistent emission trend and automatically trigger a warning mark; Generate a compressed feature vector including the emission fluctuation amplitude, frequency, and the correlation weight of the equipment status.

[0008] Preferably, the construction of the process feature similarity matrix includes: Establish a multi-dimensional feature mapping relationship between real-time process parameters and declared data; Calculate the interpolation weight based on the correlation between the equipment operation status and the raw material feeding rate in the production log. The interpolation weight calculation includes: analyze the equipment start / stop records and raw material feeding amounts in the production log, and construct a time series relationship diagram of the process chain; for the time period with missing declared data, calculate the interpolation weight according to the following formula: ; where, w t : The interpolation weight at time t, n: The total number of equipment, i: The equipment index number, R i : The pollutant correlation degree of equipment i, S t，i: The operating status coefficient of device i at time t, t: time, R i is the pollutant correlation degree of device i, defined by the knowledge graph.

[0009] Match the process logic consistency of real-time data and historical declaration data through the cosine similarity algorithm.

[0010] Preferably, in the environmental capacity responsive evaluation model: The meteorological grid forecast data includes temperature, wind speed and atmospheric diffusion coefficient, and the hydrodynamic model includes basin runoff and water body self-purification rate; that is, meteorological-hydrological data fusion, including: The meteorological grid forecast data is accessed at a resolution of 1km×1km, and the real-time wind speed and temperature vertical gradient of the grid where the target enterprise is located are extracted; In the hydrodynamic model, the basin runoff is calculated using the Saint-Venant equation: ; where, : Partial differential symbol, t: time, Q: river channel flow, x: spatial coordinate, here referring to the longitudinal coordinate of the river channel, Q 2 / A: Momentum flux per unit wetted area, gA: Product of gravitational acceleration and wetted area, h: water depth, So: river channel bottom slope, Sf: friction slope drop, Q is the flow rate, A is the wetted area, h is the water depth.

[0011] The nonlinear relationship matrix optimizes the transfer function coefficients of the reactor temperature and pollutant generation amount through the stochastic gradient descent algorithm, and then conducts process-emission correlation modeling: First, optimize the transfer function to construct the nonlinear relationship between the reactor temperature T and the VOCs generation amount Q: ; where, G: Volatile organic compound generation amount, a, b, c, d: are all empirical coefficients, T: reactor temperature, ln(T): natural logarithm of temperature, e -d / T : Temperature-related exponential decay term; Initial coefficients a = 0.0023, b = 1.78, c = 5.6, d = 300; Then update the coefficients every 24 hours using the stochastic gradient descent method, where the learning rate η = 0.001, the batch size is 64, and the loss function is: ; where, L: Total loss value of the model, N: Total number of training samples, G i pred : Pollutant generation amount predicted for the i-th time, G i meas : Pollutant generation amount measured for the i-th time, λ: regularization coefficient, ‖θ‖2: L2 norm of the model parameters; λ = 0.01 to prevent overfitting.

[0012] Preferably, the verification process of the reverse process simulator includes: Input the real-time emission data into a pre-trained LSTM neural network to generate the theoretical values of the production process parameters. The LSTM neural network architecture includes an input layer, a hidden layer, and an output layer. The input layer is a 60-minute sliding window of real-time emission data, and the real-time emission data includes COD, NH3-N, and flow rate. The hidden layer consists of 3 layers of LSTM units, with 128 neurons in each layer and a dropout rate of 0.2. The output layer is the production process parameters, namely the reactor temperature, stirring rate, and raw material feeding rate. Among them, the training dataset in the LSTM network training parameters contains 100,000 matching samples of historical process parameters and emission data. Calculate the deviation value between the theoretical parameters and the enterprise declared parameters. When the deviation value exceeds the set threshold, associate the device operation log for fault tracing, that is, deviation analysis and tracing, including: calculating the root mean square error RMSE between the theoretical value and the declared value, and setting the threshold to 15% of the declared value range. When the RMSE exceeds the standard, extract the operation parameters in the 2 hours before the fault from the device log, such as the current fluctuation is greater than 10% and the temperature change is greater than 5°C / min, etc., and generate a verification report for recommended verification measures.

[0013] Preferably, the capacity threshold simulation of the pollution carrying capacity calculation model includes: Generate the probability density function of the pollutant concentration distribution based on the current environmental capacity correction coefficient. Among them, for the fitting of the pollutant concentration distribution, use kernel density estimation KDE to generate the probability density function: ; Among them, : Probability density function of kernel density estimation, n: number of samples, o: bandwidth parameter, K(·): Gaussian kernel function, p: pollutant concentration value, O i : Concentration value of the i-th sample.

[0014] Randomly sample through the Monte Carlo method to simulate the classification boundaries under different meteorological conditions. Among them, the Monte Carlo simulation parameters include input variables, number of sampling times, and classification boundary determination. The input variables are wind speed, normally distributed, μ = measured value, σ = 1.5 m / s, and rainfall, Poisson distributed, λ = 0.2 mm / h. The number of sampling times is 10,000 times, and each simulation lasts for 6 hours. The classification boundary determination includes: when the pollutant concentration in the 95% confidence interval exceeds the GB 3838-2002 Class III water standard, trigger the limit adjustment. When the classification result exceeds the classification boundary, trigger the enterprise management category adjustment suggestion.

[0015] Preferably, the emission feature vector includes: The real-time pollution load intensity is calculated in real time according to the pollutant concentration and the discharge flow rate; The environmental capacity occupancy rate is generated based on the current meteorological conditions and the basin hydrological model. Among them, the calculation formula for the environmental capacity occupancy rate is as follows: ; where U is the environmental capacity occupancy rate, n is the total number of pollutant types, C i is the real-time concentration of pollutant i, W i is its spatial weight, C cap is the environmental benchmark capacity, and f meteo is the meteorological correction factor, with a range of 0.5 to 1.5; The process compliance index is calculated by weighting the deviation between the process parameters and the permitted values; The historical violation correlation degree is quantitatively evaluated by combining the enterprise's environmental penalty records in the past three years.

[0016] Among them, it also includes the enterprise management level determination rules, which include grading threshold adjustment and adjustment mechanism. Among them, during the grading threshold adjustment period, real-time pollution load intensity PLI grading is carried out. For those with a PLI interval less than 0.3, the management level is set as general management, and it can be regarded as a non-key supervision point and only spot-checked regularly; for those with a PLI interval between 0.3 and 0.7, the management level is set as simplified management, and the supervision measures need to be further strengthened, and the verification frequency should be more frequent; for those with a PLI interval greater than 0.7, the management level is set as key management, and the supervision measures are the strictest, and measures such as automatic monitoring equipment can be used for supervision.

[0017] The adjustment mechanism is: when the regional environmental capacity occupancy rate U is greater than 85%, the PLI threshold of all enterprises is lowered by 20%.

[0018] Real-time pollution load intensity PLI grading.

[0019] Preferably, step S4 further includes: when the difference between the dual verification results exceeds the preset tolerance, trigger the on-site monitoring equipment to start the water quality spectral analyzer and the waste gas on-line monitor for data review, that is, start the water quality spectral analyzer with a wavelength range of 200-800 nm and a resolution of 1 nm for instantaneous sampling; the waste gas on-line monitor synchronously collects the data of SO2 and NOx, and performs the Kolmogorov-Smirnov test with the system predicted value. If p is less than 0.01, it is determined as abnormal; according to the verification result, the enterprise classification level is optimized and adjusted, and a differentiated supervision strategy including key monitoring points and verification frequencies is generated; The control protocol of the water quality spectral analyzer includes a trigger instruction and sampling parameters. Among them, the trigger instruction is: when the dual verification difference value is greater than 0.25, a sampling instruction is sent through the MQTT protocol; Sampling parameters include spectral range, integration time, and data transmission format. Among them, the spectral range is 200 - 800 nm, that is, ultraviolet - visible light; the integration time is automatically adjusted, ranging from 100 ms to 10 s; the data transmission format is JSON containing wavelength - absorbance key - value pairs. Among them, the rules for generating key monitoring point instructions include: When the process compliance index is less than 0.7 and the historical violation correlation is greater than 0.6, the strictest supervision level is triggered. General management enterprises, that is, those with a compliance index greater than 0.9, generally use remote video verification and have the lowest on - site inspection frequency.

[0020] Preferably, the construction of the knowledge graph includes: Extract the process flow diagram and unstructured text description from the pollutant discharge permit report. Establish the associated edges between equipment nodes such as reaction vessels, pipelines, and purification devices and the pollutant emissions. Train the graph weights through historical accident data to identify high - level management process nodes and their hidden emission paths.

[0021] Preferably, the update of the spatial influence weight function includes: Analyze the atmospheric pollution diffusion trajectory and thermal map distribution in satellite remote sensing data. Among them, the satellite remote sensing data analysis process includes: Input: Aerosol Optical Depth (AOD) data of the Himawari - 8 satellite, with a resolution of 1 km; Processing: Use the dark pixel method to invert the PM2.5 spatial distribution and generate a pollution diffusion vector field ; Combine the data of ground monitoring stations to calibrate the pollutant migration rate parameters; Adjust the turbulent diffusion coefficient and sedimentation rate in the hydrodynamic equation according to the calibration results.

[0022] Among them, the update of the spatial influence weight function includes: Analyze the PM2.5 / COD concentration gradient distribution in the satellite remote sensing thermal map to identify the main pollution diffusion direction; adjust the turbulent diffusion coefficient (K x , K y ) in the hydrodynamic equation according to the diffusion direction. The update formula is: ; where θ is the angle between the diffusion direction and the dominant wind direction, and θ0 is the terrain correction angle.

[0023] (III) Beneficial effects The present invention provides an enterprise classification management method based on pollutant discharge permit data evaluation indicators, having the following beneficial effects: (1). The enterprise classification management method based on the pollutant discharge permit data evaluation index breaks through the limitations of the traditional static data evaluation system by constructing a coupled dual-channel decision engine and a spatio-temporal adaptive evaluation model. It performs fractal time series compression and process logic alignment on the monitoring data and declaration data, realizes the lossless fusion of multi-source heterogeneous data, highlights key management enterprises through classification levels, concentrates supervision resources to strengthen control, and effectively improves the intervention efficiency for high-pollution emission behaviors. It introduces an environmental capacity correction mechanism and satellite remote sensing-driven spatial weight update to quantify the real-time impact of meteorological and hydrological conditions on pollution contribution, reducing the evaluation error during the dry season and wet season. The dual verification mechanism cross-verifies through reverse process simulation and Monte Carlo capacity boundary to ensure that the classification results have both production process rationality and ecological environment security, improving the accuracy and timeliness of environmental supervision.

[0024] (2). The enterprise classification management method based on the pollutant discharge permit data evaluation index realizes enterprise management assessment and hierarchical response through the intelligent generation of emission feature vectors and differentiated supervision strategies. The combination of satellite remote sensing parameter update and blockchain evidence storage technology constructs a sky-earth-enterprise collaborative supervision network, improving the supervision efficiency of key management enterprises and reducing environmental management costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 is a schematic flow chart of the whole invention; Figure 2 is a schematic system architecture diagram of the enterprise classification management method of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0026] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0027] Please refer to Figure 1 and Figure 2 , the present invention provides a technical solution: an enterprise classification management method based on the pollutant discharge permit data evaluation index, including the following steps: S1: Obtain the real-time monitoring data stream of the enterprise and the pollutant discharge permit execution report. The real-time monitoring data stream collects pollutant concentration data through Internet of Things sensors at an hourly frequency, and the pollutant discharge permit execution report includes production process parameters and the enterprise environmental management ledger; S2: Input the real-time monitoring data stream and the pollutant discharge permit execution report into the coupled dual-channel decision engine. Through the spatio-temporal encoder-decoder module, perform fractal time series compression on the high-frequency monitoring data stream, extract the inflection points of equipment start / stop and the curvature features of the emission trend. At the same time, reverse-derive the hidden emission nodes in the process chain based on the production log to generate enhanced declaration data, and use the process feature similarity matrix to achieve data fusion across time scales, generating an emission feature vector; Among them, the algorithm process of fractal time series compression includes time series segmentation, input the monitoring data stream {x t}, t = 1, 2,..., T; output the set of key time windows W = {w k |k = 1,..., K}; including the following steps: a: Calculate the Hölder exponent α(t) of the sliding window (length L = 300 seconds): ; where, W: sliding window length, S: standard deviation, R: range; R / S: rescaled range, α(t): Hölder exponent; b: Mark the windows with α(t) greater than 0.75 as candidate windows for equipment start / stop events.

[0028] The extraction of curvature features includes: Perform cubic spline interpolation on the candidate windows and calculate the curvature integral: ; C(w k ): the curvature integral value of window w k , te, ts: are the end and start times of the window respectively, : the second derivative of pollutant concentration, : the first derivative of pollutant concentration; Select the windows with C(w k ) greater than 0.15 as key process nodes.

[0029] S3: Input the emission feature vector into the environmental capacity responsive evaluation model, real-time access the meteorological grid forecast data and the hydrodynamic model to calculate the regional environmental capacity correction coefficient, combine the non-linear relationship matrix of process parameters and pollutant generation amounts constructed based on the knowledge graph, and the pollutant spatial impact weight function generated according to the hydrodynamic equation, and calculate the enterprise pollution contribution degree weight; S4: Input the preliminary classification results into the reverse process simulator and the pollutant assimilation capacity calculation model in sequence for double verification. The reverse process simulator reverse-derives the theoretical values of production process parameters through the LSTM neural network and performs deviation analysis with the declared parameters. The pollutant assimilation capacity calculation model uses the Monte Carlo method to simulate the environmental capacity threshold; S5: Output the doubly verified classification results and update the spatial impact weight function parameters every 6 hours based on the air pollution diffusion trajectory retrieved from satellite remote sensing.

[0030] Fractal time series compression includes: performing multifractal dimension analysis on high-frequency monitoring data streams to extract key time windows with process relevance; identifying the persistence characteristics of emission trends through wavelet transform and Hurst exponent calculation; generating a compressed feature vector containing the emission fluctuation amplitude, frequency, and the correlation weight of equipment status.

[0031] It should be further noted that in the specific implementation process, when performing fractal time series compression on high-frequency monitoring data streams, first divide the data stream into time windows with a length of 5 minutes, generate a multifractal spectrum by calculating the generalized Hurst exponent of each window, and select the windows with a spectrum width greater than 0.25 as key process nodes; then perform 5-layer decomposition using the Daubechies 4 wavelet basis, extract the detail coefficients in the period range of 2 - 8 minutes, and calculate the Hurst exponent for the approximation coefficients. When the exponent value exceeds 0.75, it is determined as a persistent emission trend; then perform cubic spline interpolation on the candidate windows, calculate the curvature characteristics by integrating along the time axis, and retain the windows with a curvature integral value greater than 0.15; finally, generate a compressed feature vector containing the emission fluctuation amplitude, frequency, and the correlation weight of equipment status, where the fluctuation amplitude is quantified by the ratio of the range to the mean within the window, the frequency is statistically analyzed by the zero-crossing rate, and the equipment correlation weight is assigned according to the matching degree between the process parameters and the equipment operation logs within the window.

[0032] The construction of the process feature similarity matrix includes: establishing a multi-dimensional feature mapping relationship between real-time process parameters and declared data; calculating the interpolation weight based on the correlation between the equipment operation status and the raw material feeding rate in the production log; matching the process logic consistency between real-time data and historical declared data through the cosine similarity algorithm. The process of the cosine similarity algorithm includes: converting real-time process parameters, i.e., temperature, pressure, and flow, and historical declared data into 128-dimensional feature vectors; calculating the cosine similarity between the real-time vector and the historical vector, and setting the threshold to 0.85. If it is lower than the threshold, interpolation enhancement is initiated.

[0033] It should be further noted that in the specific implementation process, when constructing the process feature similarity matrix, first, the equipment start-stop timestamps and raw material feeding acceleration rates recorded in the production log are parsed to establish an association map based on the process chain time sequence. For the missing period of declaration data, the interpolation weight is calculated based on the weighted calculation of the equipment operation state coefficient and pollutant correlation. The equipment operation state coefficient is adjusted according to the deviation degree of the real-time current fluctuation amplitude from the rated value, and the pollutant correlation is determined by the connection strength between the equipment node and the emission factor in the knowledge graph. Subsequently, the temperature, pressure, and flow parameters collected in real time and the historical declaration data are respectively converted into a standardized format containing 128-dimensional process feature vectors. The feature dimensions cover process parameter statistics, time sequence change rates, and covariance relationships with equipment states. The matching degree between the real-time vector and the historical vector is calculated through the cosine similarity algorithm. When the similarity is lower than the 0.85 threshold, interpolation enhancement based on the process chain time sequence relationship is initiated, and the declaration data of adjacent nodes are weighted and fused according to the proportion of equipment operation duration. Finally, enhanced declaration data with time continuity and process logic consistency is generated, achieving cross-scale alignment with real-time monitoring data.

[0034] In the environmental capacity responsive evaluation model: The meteorological gridded forecast data includes temperature, wind speed, and atmospheric diffusion coefficient, and the hydrodynamic model includes basin runoff and water body self-purification rate; The nonlinear relationship matrix optimizes the transfer function coefficients of the reactor temperature and pollutant generation amount through the stochastic gradient descent algorithm. It should be further noted that in the specific implementation process, when implementing the environmental capacity responsive evaluation model, the 1km×1km gridded forecast data released by the meteorological bureau is accessed in real time, and the temperature vertical gradient, wind speed at 10m height, and atmospheric diffusion coefficient of the grid where the target enterprise is located are extracted. At the same time, the Saint-Venant equation in the basin hydrodynamic model is called to calculate the current runoff and water body self-purification rate; A nonlinear relationship matrix between process parameters and pollutant generation amount is constructed based on the knowledge graph, and the transfer function coefficients are optimized through the stochastic gradient descent algorithm. Among them, the relationship function between the reactor temperature and the VOCs generation amount adopts a combination form of a cubic polynomial and an exponential term, and the initial coefficients are determined by fitting historical accident data. Parameter tuning is performed every 24 hours with a batch size of 64 and a learning rate of 0.001; The spatial influence weight function modifies the turbulent diffusion coefficient in the hydrodynamic equation based on the main direction of pollution diffusion inversely retrieved by real-time satellite remote sensing, combined with terrain elevation data. Specifically, the pollution vector field is generated by parsing the AOD data of the Himawari-8 satellite, and the horizontal diffusion coefficient K in the diffusion equation is updated every 6 hours. x and K y , and the correction formula is K x ´ = K x·exp(-(θ - θ0) / 30°), where θ is the angle between the remotely sensed inversion diffusion direction and the dominant wind direction, and θ0 is the terrain ruggedness correction angle; finally, the corrected environmental capacity coefficient, process emission transfer matrix, and spatial weight function are coupled in a multi-physical field to output the enterprise pollution contribution weight value, which is accurate to three decimal places and its confidence interval is verified through Monte Carlo simulation.

[0035] The verification process of the reverse process simulator includes: inputting real-time emission data into a pre-trained LSTM neural network to generate the theoretical values of production process parameters; calculating the deviation value between the theoretical parameters and the enterprise declared parameters, and when the deviation value exceeds the set threshold, associating the device operation log for fault traceability; generating a verification report of recommended verification measures. It should be further noted that in the specific implementation process, a deviation traceability rule library is set up, including current fluctuation association and abnormal raw material feeding. For current fluctuation association, if the theoretical reactor temperature deviation is greater than 8%, then retrieve the motor current data in the device log for the previous 2 hours, and the trigger rule is: IF ΔI is greater than 15% THEN fault type = motor overload, ΔI > 15%: the core determination threshold for the reverse process simulator to perform device fault traceability, ΔI: the real-time fluctuation amplitude of the motor current, and the calculation formula of ΔI: , I real : the real-time monitored motor current value, I rated : the rated working current of the motor; for abnormal raw material feeding, if the theoretical feeding rate deviation is greater than 20%, then associate the material weighing record, and the trigger rule: IF the standard deviation of the feeding amount is greater than 30% THEN the fault type is equal to abnormal feeding system.

[0036] It should be further noted that in the specific implementation process, when implementing the verification of the reverse process simulator, the real-time collected pollutant concentration and flow data are normalized by a 60-minute sliding window and then input into a pre-trained three-layer LSTM neural network. The network hidden layer contains 128 neurons and a dropout rate of 0.2 is set. The output layer generates the theoretical values of the reactor temperature, stirring rate, and raw material feeding rate; calculate the root mean square error between the theoretical value and the declared parameter. When the temperature deviation exceeds ±8°C or the raw material feeding rate deviation is greater than 20%, automatically associate the current and vibration sensor data in the device operation log for the previous 2 hours, and extract the abnormal periods when the current fluctuation amplitude exceeds 15% or the temperature change rate is greater than 5°C / min; match the process nodes with the fault type library through the knowledge graph. If the deviation source is associated with the reactor cooling system, then trigger a motor overload warning. If it is associated with the feeding section, then generate a verification instruction for the material weighing record; finally, output a verification report containing the abnormal probability value, fault location path, and recommended disposal measures. Among them, the abnormal probability is generated by mapping the deviation value through the Sigmoid function, and the disposal measures include hierarchical response schemes such as on-site instrument calibration and equipment shutdown for maintenance.

[0037] The capacity threshold simulation of the pollution carrying capacity calculation model includes: generating a probability density function of pollutant concentration distribution based on the current environmental capacity correction factor; simulating the classification boundaries under different meteorological conditions by randomly sampling 10,000 times through the Monte Carlo method; when the classification result exceeds the classification boundary, triggering a proposal for adjusting the enterprise management category.

[0038] It should be further noted that in the specific implementation process, when verifying the pollution carrying capacity, a probability density function of pollutant concentration distribution is generated based on the environmental capacity correction factor, and the non-parametric fitting with a bandwidth of 0.5σ·n -1 / 5 is performed on the real-time monitoring data by using the kernel density estimation method, where σ is the sample standard deviation and n is the number of data points; when simulating through the Monte Carlo method, it is set that the wind speed follows a normal distribution with a mean of the measured value and a standard deviation of 1.5 m / s, the rainfall follows a Poisson distribution with λ equal to 0.2 mm / h, and 10,000 random samplings are used to generate a pollutant diffusion scenario with a 6-hour duration; each simulation uses the three-dimensional advection-diffusion equation to calculate the pollutant migration trajectory. When the concentration of the downstream section within the 95% confidence interval exceeds 1.2 times the limit value of the Class III water quality standard in GB 3838-2002, it is determined that the classification boundary is breached; after triggering the proposal for adjusting the management category, the system automatically generates a regulation plan including the maximum allowable emission, emission reduction ratio, and implementation time window, and synchronizes the new limit value to the enterprise sewage discharge monitoring terminal through the blockchain smart contract. At the same time, a collaborative production restriction assessment model for related enterprises within the basin is started to ensure that the total regional emissions do not exceed the pollution carrying capacity threshold.

[0039] The emission characteristic vector includes the real-time pollution load intensity, environmental capacity occupancy rate, process compliance index, and historical violation correlation degree; among them, the real-time pollution load intensity is calculated in real time according to the pollutant concentration and emission flow rate, the environmental capacity occupancy rate is generated based on the current meteorological conditions and the basin hydrological model, the process compliance index is calculated by weighting the deviation degree between the process parameters and the permitted values, and the historical violation correlation degree is quantitatively evaluated by combining the enterprise's environmental penalty records in the past three years.

[0040] It should be further noted that in the specific implementation process, when generating the emission feature vector, the real-time pollution load intensity is calculated by multiplying the instantaneous value of the pollutant concentration by the emission flow rate. Among them, the flow rate data is calibrated by an ultrasonic flowmeter and the invalid values during the equipment shutdown period are deducted; the environmental capacity occupancy rate is based on the atmospheric stability level in the meteorological grid data and the self-purification rate output by the basin hydrodynamic model, and the percentage of the current emission volume in the regional capacity is converted according to the pollutant diffusion half-life; the process compliance index uses the improved Euclidean distance algorithm to calculate the multi-dimensional deviation degree between the real-time process parameters and the pollutant discharge permit values. The weight of each dimension is determined by the enterprise management level of the process node in the knowledge graph, and the weight of the high-level management node is increased by 30%; the historical violation correlation degree analyzes the penalty records of the enterprise in the past three years through natural language processing technology, extracts features such as violation types, frequencies, and rectification effects, and uses a logistic regression model to output a standard value of 0-1. Among them, the weight coefficient of the violation records involving heavy metal emissions is doubled; finally, the four types of indicators are Min-Max standardized and then fused according to the weight ratio of 0.4:0.3:0.2:0.1 to generate an emission feature vector with a timestamp mark, which is updated every 5 minutes and the data integrity is ensured through hash verification.

[0041] Step S4 further includes: when the difference in the double verification results exceeds the preset tolerance, triggering the on-site monitoring equipment to start the water quality spectroanalyzer and the waste gas online monitor for data review; adjusting the enterprise classification level according to the review results, and generating a differentiated supervision strategy including key monitoring points and verification frequencies.

[0042] It should be further noted that in the specific implementation process, when triggering the double verification difference tolerance mechanism, when the comprehensive difference value between the theoretical parameter deviation output by the reverse process simulator and the verification result of the pollution carrying capacity exceeds the threshold of 0.25, a spectral scanning instruction is sent to the water quality spectroanalyzer at the target enterprise end through the MQTT protocol, and the wavelength range is set to 200-800nm, and the integration time automatically adapts to the light intensity from 100ms to 10s. At the same time, the sampling mode of the waste gas online monitor is started to capture SO2 and NO in real time xThe peak value of the concentration pulse; perform a Kolmogorov-Smirnov distribution consistency test on the on-site review data and the system prediction value. When the test statistic D is greater than 0.35 and the p-value is less than 0.01, it is determined as substantial data anomaly, and then call the blockchain evidence storage module to record the original spectral data fingerprint; adjust the enterprise classification level according to the anomaly confidence level. If the confidence level is greater than 90%, immediately upgrade the management level to senior management and generate a on-site verification plan including 3 key monitoring points, namely 50m upstream of the sewage outlet, the exhaust port of the reaction kettle, and the hazardous waste temporary storage area. The verification frequency is set to 2 full-index detections per day; at the same time, activate the basin collaborative supervision model, and issue a preliminary production restriction order to related enterprises with a process similarity greater than 0.8 in the same basin until the prediction error of the review data after being retrained by the LSTM-Attention hybrid model is restored within the threshold range.

[0043] The construction of the knowledge graph includes: extracting the process flow chart and unstructured text description in the pollution discharge permit and the implementation report; establishing the associated edges between equipment nodes such as reaction kettles, pipelines, and purification devices and the pollutant emissions; training the graph weights through historical accident data to identify the senior management process nodes and their hidden emission paths. It should be further noted that in the specific implementation process, when constructing the knowledge graph, first parse the process flow chart in the enterprise pollution discharge permit report through OCR technology to identify the topological connection relationship of equipment nodes such as reaction kettles, pipelines, and purification devices, and at the same time use the BERT model to extract the process parameter constraints and hidden danger descriptions in the unstructured text; when establishing the associated edges between equipment nodes and pollutant emissions, calculate the pollution contribution factors of each equipment based on historical monitoring data. Among them, the initial weight of the reaction kettle and VOCs emissions is set to 0.65, the weight of pipeline leakage and COD emissions is set to 0.23, and the contribution factors are corrected through a graph neural network; when training the graph using the environmental accident database in the past five years, add a weight coefficient of 0.3 to the nodes involving high-temperature and high-pressure processes, and use the GraphSAGE algorithm to iteratively optimize the edge weights. Each time, 500 groups of accident scenario data are sampled in batches, and the loss function includes the dual constraints of node classification error and edge weight prediction error; the finally generated visual knowledge graph includes a three-dimensional association network of equipment status - emission intensity - environment. The hidden emission path is detected by the random walk algorithm. When the cumulative value of the process path exceeds the threshold, it is automatically marked as a red warning node, and a contribution degree analysis report of key factors is output in combination with the LIME interpretability model.

[0044] The update of the spatial influence weight function includes: analyzing the atmospheric pollution diffusion trajectory and thermal map distribution in satellite remote sensing data, calibrating the pollutant migration rate parameters in combination with ground monitoring station data, and adjusting the turbulent diffusion coefficient and sedimentation rate in the hydrodynamic equation according to the calibration results. It should be further noted that in the specific implementation process, when updating the spatial influence weight function, the aerosol optical depth data of the Himawari-8 satellite is analyzed, the dark pixel method is used to invert the PM2.5 concentration gradient distribution, and a 5km×5km pollution diffusion thermal map centered on the enterprise is generated; combined with the hourly average data of the ground monitoring station, the pollutant migration rate parameters are calibrated through linear regression, and the calibration formula is V cal =0.87V sat +0.13V ground where V sat is the satellite inversion rate and V ground is the ground monitoring value; based on the calibrated diffusion direction and terrain elevation data, the turbulent diffusion coefficient K x and the sedimentation rate V s in the hydrodynamic equation are corrected. Among them, the horizontal diffusion coefficient is adjusted according to: K x ´=K x ·exp(-0.015·|θ-θ0|), where θ is the real-time angle between the satellite inversion diffusion direction and the dominant wind direction, and θ0 is the terrain roughness correction angle calculated according to the digital elevation model DEM; the sedimentation rate is calculated using the Stokes-Cunningham formula based on the PM2.5 particle size distribution data: where V s is the particulate matter sedimentation rate, g is the acceleration of gravity, μ is the air dynamic viscosity, d p is the median diameter of the particulate matter, ρ p is the density, and C c is the slip flow correction coefficient; the updated parameters are synchronized to the three-dimensional diffusion model every 6 hours to generate a spatial influence weight function surface, and the goodness of fit between the weight distribution and the measured data is verified through CUDA parallel computing. R² is greater than 0.92, and finally the adjusted spatial weight value of the pollution contribution degree is output, accurate to four decimal places.

[0045] By constructing a coupled dual-channel decision engine and a spatio-temporal adaptive evaluation model, the limitations of the traditional static data evaluation system are broken through. The monitoring data and declaration data are subjected to fractal time-series compression and process logic alignment to achieve lossless fusion of multi-source heterogeneous data. The list of key regulatory enterprises is clarified through the classification results, and high-frequency verification and dynamic limit control are implemented for highly polluting enterprises, significantly enhancing the targeting of regulatory measures. An environmental capacity correction mechanism and satellite remote sensing-driven spatial weight update are introduced to quantify the real-time impact of meteorological and hydrological conditions on pollution contribution, reducing the evaluation error during the dry season and wet season. The dual-verification mechanism cross-verifies through reverse process simulation and Monte Carlo capacity boundaries to ensure that the classification results have both production process rationality and ecological environment security, improving the accuracy and timeliness of environmental supervision.

[0046] Through the intelligent generation of emission feature vectors and differentiated regulatory strategies, enterprise management assessment and hierarchical response are realized. The combination of satellite remote sensing parameter update and blockchain certification technology constructs a sky-earth-enterprise collaborative supervision network, improving the supervision efficiency of key management enterprises and reducing environmental management costs.

[0047] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "comprising a..." does not exclude the presence of additional identical elements in the process, method, article or device comprising the element.

[0048] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for classifying and managing enterprises based on evaluation indicators of pollutant discharge permit data, characterized in that It includes the following steps: S1: Obtain the real-time monitoring data stream of the enterprise and the pollutant discharge permit execution report. The real-time monitoring data stream collects pollutant concentration data at an hourly frequency through Internet of Things sensors. The pollutant discharge permit execution report includes production process parameters and the enterprise environmental management ledger; S2: Input the real-time monitoring data stream and the pollutant discharge permit execution report into the coupled dual-channel decision engine. Perform fractal time series compression on the high-frequency monitoring data stream through the spatio-temporal encoder-decoder module, extract the inflection points of equipment start-stop and the curvature features of emission trends. At the same time, reverse-derive the hidden emission nodes in the process chain based on production logs to generate enhanced declaration data, and use the process feature similarity matrix to achieve data fusion across time scales to generate emission feature vectors; S3: Input the emission feature vectors into the environmental capacity responsive evaluation model, access the meteorological grid forecast data and the hydrodynamic model in real time to calculate the regional environmental capacity correction coefficient, combine the non-linear relationship matrix between process parameters and pollutant generation amounts constructed based on the knowledge graph, and the pollutant spatial influence weight function generated according to the hydrodynamic equation to calculate the enterprise pollution contribution degree weight; S4: Input the preliminary classification results into the reverse process simulator and the pollution assimilation capacity calculation model in sequence for double verification. The reverse process simulator reverse-derives the theoretical values of production process parameters through the LSTM neural network and conducts deviation analysis with the declared parameters. The pollution assimilation capacity calculation model uses the Monte Carlo method to simulate the environmental capacity threshold; S5: Output the classification results verified twice, and update the parameters of the spatial influence weight function according to the preset update period based on the atmospheric pollution diffusion trajectory retrieved by satellite remote sensing.

2. The enterprise classification management method based on the evaluation index of pollutant discharge permit data according to claim 1, characterized in that: The fractal time series compression includes: Perform multifractal dimension analysis on the high-frequency monitoring data stream to extract key time windows with process relevance; Identify the persistence characteristics of the emission trend through wavelet transform and Hurst index calculation; Generate a compressed feature vector containing the emission fluctuation amplitude, frequency, and the associated weight of the equipment status.

3. The enterprise classification management method based on the pollution discharge permit data evaluation index according to claim 1, wherein: The construction of the process feature similarity matrix includes: Establish a multi-dimensional feature mapping relationship between real-time process parameters and declaration data; Calculate the interpolation weight based on the correlation between the equipment operation status and the raw material feeding acceleration rate in the production log; Match the process logic consistency between real-time data and historical declaration data through the cosine similarity algorithm.

4. The enterprise classification management method based on the evaluation index of pollutant discharge permit data according to claim 1, wherein: In the environmental capacity responsive evaluation model: The meteorological grid forecast data includes temperature, wind speed, and atmospheric diffusion coefficient. The hydrodynamic model includes the basin runoff and the water body self-purification rate; The non-linear relationship matrix optimizes the transfer function coefficient between the reactor temperature and the pollutant generation amount through the stochastic gradient descent algorithm.

5. The enterprise classification management method based on the pollution discharge permit data evaluation index according to claim 1, characterized in that: The verification process of the reverse process simulator includes: Input the real-time emission data into the pre-trained LSTM neural network to generate the theoretical values of production process parameters; Calculate the deviation value between the theoretical parameters and the enterprise declared parameters. When the deviation value exceeds the set threshold, associate the equipment operation log for fault tracing and generate a verification report recommending verification measures.

6. The enterprise classification management method based on the pollution discharge permit data evaluation index according to claim 1, wherein: The capacity threshold simulation of the pollution carrying capacity calculation model includes: generating a probability density function of pollutant concentration distribution based on the current environmental capacity correction coefficient, and randomly sampling through the Monte Carlo method to simulate the classification boundaries under different meteorological conditions, where the classification boundaries are management thresholds; when the classification result exceeds the classification boundaries, triggering suggestions for adjusting the enterprise management category.

7. A method for classifying and managing enterprises based on evaluation indicators of pollutant discharge permit data according to claim 1, characterized in that: The emission feature vector includes: The real-time pollution load intensity, which is calculated in real time according to the pollutant concentration and the emission flow rate; The environmental capacity occupancy rate, which is generated based on the current meteorological conditions and the basin hydrological model; The process compliance index, which is calculated by weighting the deviation degree between the process parameters and the permitted values; The historical violation correlation degree, which is quantitatively evaluated by combining the enterprise's environmental penalty records in the past three years.

8. The enterprise classification management method based on the evaluation index of pollutant discharge permit data according to claim 1, characterized in that: Step S4 also includes: when the difference in the double verification results exceeds the preset tolerance, triggering the on-site monitoring equipment to start the water quality spectroanalyzer and the waste gas on-line monitor for data review; adjusting the enterprise classification level according to the review results, and generating a differentiated supervision strategy including key monitoring points and verification frequencies.

9. The enterprise classification management method based on the pollution discharge permit data evaluation index according to claim 1 is characterized in that: The construction of the knowledge graph includes: Extracting the process flow diagrams and unstructured text descriptions in the pollutant discharge permit and the implementation report; Establishing the association edges between the nodes of the reaction kettle, pipeline, and purification device and the pollutant emission amount; Training the graph weights through historical accident data to identify the high-level management process nodes and their implicit emission paths.

10. A method for classifying and managing enterprises based on pollution discharge permit data evaluation indicators according to claim 1, characterized in that: The update of the spatial influence weight function includes: Analyzing the atmospheric pollution diffusion trajectory and the heat map distribution in the satellite remote sensing data; Calibrating the pollutant migration rate parameters by combining the data of the ground monitoring stations; Adjusting the turbulent diffusion coefficient and the sedimentation rate in the hydrodynamic equation according to the calibration results.

Images