Environment capacity maximization method and system for complex water area pollution outlet selection

Abstract

The present application belongs to the technical field of sewage treatment, and particularly relates to an environmental capacity maximization method and system for complex water area pollution outlet selection and demarcation, which comprises the following steps: obtaining multi-source original data including complex water area basic environmental data, pollution source data and constraint condition data, to obtain a standardized data set and a standardized constraint parameter library; constructing an initial digital twin containing a three-dimensional terrain submodel, a hydrodynamic submodel and a water quality diffusion submodel; inputting the standardized data set into an environmental capacity calculation large model to obtain the safe dilution degree of pollutants, and screening out the pollutant with the smallest safe dilution degree as the main control pollution factor; constructing a target function of total environmental capacity maximization according to the environmental capacity value of the water area and the main control pollution factor; determining the constraint quantization standard of the water area according to the standardized constraint parameter library; and screening out the optimal population meeting the constraint quantization standard as the optimal pollution outlet selection and demarcation scheme.

Description

Technical Field

[0001] This invention belongs to the field of wastewater treatment technology, specifically relating to a method and system for maximizing environmental capacity in the selection and delineation of sewage outlets in complex water areas. Background Technology

[0002] Complex water bodies, due to their diverse topography, complex water flow, uneven spatial and temporal distribution of water quality, and sensitive ecological environment, have become key and challenging areas for water environment management. The selection and delineation of sewage outlets is a core aspect of pollution control in complex water bodies. Its core objective is to maximize the utilization of the water body's environmental capacity and achieve a dynamic balance between pollutant discharge and the water body's carrying capacity, while meeting constraints such as water quality compliance, ecological protection, and flood control and drainage.

[0003] Currently, the existing technologies for selecting and delineating sewage outlets in complex water areas are mainly divided into empirical methods and traditional numerical simulation methods. Among them, empirical methods lack technical support, rely entirely on the past practical experience of technical personnel, lack standardized data processing and quantitative calculation processes, lack the ability to analyze multi-source data in complex water areas, cannot replicate the dynamic environmental characteristics of water areas, and the selection and delineation results are highly subjective, with extremely low accuracy and poor repeatability. Traditional numerical simulation methods use fixed static modeling techniques, do not build a real-time data-driven mechanism, lack adaptive adjustment capabilities in grid division, cannot adapt to dynamic changes such as complex water area topography, tides, and seasons, have weak multi-source heterogeneous data fusion capabilities, low computational efficiency and large errors, and are difficult to accurately simulate the diffusion patterns of pollutants. In summary, existing technologies for the selection and delineation of sewage outlets in complex water bodies cannot adapt to the diverse topography, complex water flow, and tidal / seasonal changes of complex water bodies. They are difficult to accurately replicate pollution under natural conditions, and the selection and delineation of sewage outlets do not conform to the natural characteristics of complex water bodies (such as water flow direction and natural distribution of ecologically sensitive areas). They do not follow the correlation between the location and number of sewage outlets and the natural water flow and topography of the water body, and cannot maximize the natural utilization rate of environmental capacity. Therefore, an improved method and system for maximizing environmental capacity for the selection and delineation of sewage outlets in complex water bodies has been designed. Summary of the Invention

[0004] To address the aforementioned shortcomings in the existing technology, this invention provides a method and system for maximizing environmental capacity in the selection and delineation of sewage outlets in complex water areas, thereby solving the problems mentioned in the background technology.

[0005] To solve the above-mentioned technical problems, the present invention adopts the following technical solution:

[0006] The present invention provides a method for maximizing the environmental capacity for the selection of sewage outlets in complex water areas, comprising the following steps: (1) acquiring multi-source raw data including basic environmental data of complex water areas, sewage source data and constraint data, performing federated homomorphic encryption distributed processing on the above multi-source raw data to obtain encrypted multi-source heterogeneous data; based on the encrypted multi-source heterogeneous data, performing preprocessing such as cleaning, outlier removal, normalization and tensor transformation in sequence to obtain a standardized dataset and a standardized constraint parameter library;

[0007] (2) Collect water area topographic image data and underwater topographic measurement data, and use BIM and GIS integrated modeling technology to construct an initial digital twin containing a three-dimensional topographic sub-model, a hydrodynamic sub-model, and a water quality diffusion sub-model;

[0008] (3) Input the standardized dataset into the environmental capacity calculation model to obtain the preliminary environmental capacity values ​​of each grid in the complex water area. Perform error calibration on the preliminary environmental capacity values ​​to obtain the environmental capacity values ​​of the water area. Perform safe dilution method calculation on the pollutant discharge concentration of the sewage outlet to obtain the safe dilution of the pollutants. Select the pollutant with the smallest safe dilution as the main control pollution factor.

[0009] (4) Based on the environmental capacity value of the water area and the main controlling pollution factor, construct the objective function for maximizing the total environmental capacity; determine the constraint quantification standard of the water area based on the standardized constraint parameter library; perform real number encoding and population initialization operations on the three parameters of sewage outlet location, number, and sewage discharge to obtain the initial population set; perform adaptive crossover mutation genetic algorithm to iterate on the initial population set to obtain the optimal population, and select the optimal population that meets the constraint quantification standard as the optimal sewage outlet selection scheme.

[0010] The present invention also provides an environmental capacity maximization system for the selection and delineation of sewage outlets in complex waterways, wherein the method described above includes:

[0011] Federated Data Acquisition Module: Used to collect basic environmental data, pollution source data, and constraint data of complex water areas. Based on the collected data, it performs federated homomorphic encryption processing to obtain encrypted data. Based on the encrypted data, it performs standardization preprocessing to obtain a standardized dataset and a standardized constraint parameter library.

[0012] Digital twin construction module: used to collect drone aerial photography data, underwater topography data and real-time dynamic data from sensors; based on the above topographic data, perform BIM+GIS fusion modeling operation to obtain an initial digital twin; based on the above real-time dynamic data from sensors, perform digital twin parameter synchronization update operation to obtain a dynamic digital twin;

[0013] Large Model Processing Module: Obtains the standardized dataset output by the federated data collection module; based on the standardized dataset, performs the construction and training operations of the large environmental capacity calculation model to obtain the large environmental capacity calculation model; inputs the standardized dataset into the qualified large model, performs capacity calculation and main control factor screening operations to obtain the environmental capacity value and the main control pollution factor;

[0014] Intelligent optimization module: It acquires the environmental capacity value, main control pollution factors, and standardized constraint parameter library output by the large model processing module and the federated data acquisition module; based on the above data, it performs objective function construction, constraint analysis, and adaptive crossover mutation genetic algorithm optimization operations to obtain the optimal sewage outlet selection scheme.

[0015] Compared with the prior art, the present invention has the following beneficial effects:

[0016] 1. By using a three-level federated data collection network and federated homomorphic encryption technology, distributed processing of multi-subject, multi-source data is achieved, ensuring that the data is "usable but not visible," effectively solving the problems of privacy leakage and heterogeneous fusion of multi-subject data; through a unified preprocessing process and differentiated processing strategies, standardized datasets and constraint parameter libraries are obtained, improving data adaptability and providing high-quality data support for subsequent calculations and optimizations, with data fusion efficiency improved by more than 40%.

[0017] 2. Scene Replication Level: A digital twin is constructed using BIM+GIS fusion technology. Combined with real-time sensor data, it achieves millisecond-level synchronization with the real water area, accurately replicating the dynamic changes of complex water areas and improving scene adaptability by more than 60%. Differentiated modeling accuracy is adapted according to different water types, balancing modeling efficiency and accuracy. This provides reliable support for scene simulation and effect prediction of sewage outlet selection schemes, reducing scheme design risks.

[0018] 3. Capacity Calculation Level: A customized four-level environmental capacity calculation model is constructed, which combines federated distributed training and error calibration strategies to ensure that the environmental capacity calculation error is ≤2% and the model mean absolute error is ≤5%, improving the calculation accuracy by more than 30% and the calculation efficiency by more than 60%. The safe dilution method is used to accurately screen the main controlling pollution factors, providing targeted decision data for the selection and optimization of sewage outlets, and improving the targeting of optimization.

[0019] 4. Selection and optimization level: Adaptive crossover and mutation genetic algorithm is adopted to solve the problems of premature convergence and incomplete optimization of traditional algorithms. Combined with the actual constraints of the project, the global optimization of the sewage outlet selection scheme is realized, which improves the utilization rate of environmental capacity in complex waters by 20%~30% and the optimization efficiency by more than 25%. The optimized scheme has been verified for compliance with constraints, fits the actual project, and has strong feasibility and scalability.

[0020] 5. System Implementation: The four core modules work together and data flows bidirectionally, realizing full automation of the "data acquisition - scenario construction - capacity calculation - selection and optimization" process. No professional high-end technical personnel are required to operate it. It can be widely used in various complex water areas such as estuaries, bays, and lakes for the selection and delineation of sewage outlets, providing reliable technical support for the environmental protection and sewage outlet planning of complex water areas. Attached Figure Description

[0021] Figure 1 This is a flowchart of the method for maximizing environmental capacity in the selection and delineation of sewage outlets in complex water areas according to the present invention. Detailed Implementation

[0022] To enable those skilled in the art to better understand the present invention, the technical solution of the present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0023] The accompanying drawings are for illustrative purposes only and are schematic diagrams, not actual images. They should not be construed as limiting the scope of this application. To better illustrate the embodiments of the present invention, some parts in the drawings may be omitted, enlarged, or reduced, and do not represent the actual dimensions of the product. It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0024] Federationed Data Acquisition Network Construction: A multi-source heterogeneous data federation network is constructed, consisting of sensor terminal nodes, multi-entity data holding nodes, and a federation coordination node. Each node interacts with data via a federated homomorphic encryption protocol, employing 128-bit encryption to ensure data privacy and security during transmission and processing, achieving distributed processing where multi-entity data is "usable but not visible." Sensor terminal nodes include Doppler current meters, water quality sensors, weather stations, underwater topography instruments, and drones, used for real-time acquisition of basic environmental data from complex water areas. Multi-entity data holding nodes include administrative entities such as environmental protection, water resources, and maritime affairs, used to access cross-entity data on pollution sources, water area functional zoning, and ecological protection. The federation coordination node is used to uniformly schedule the distributed data processing of all nodes and output a federated standardized dataset.

[0025] Data Collection Scope and Types: The collected data corresponds one-to-one with the core influencing factors for the selection and delineation of sewage outlets in complex water areas, specifically including three categories:

[0026] ①Basic environmental data: water boundary coordinates (collection range covers the entire complex water area and a surrounding 50m buffer zone, with a collection density of one coordinate point per 10m; boundary inflection point coordinates are separately labeled and generated into a vector coordinate file in .shp format to meet subsequent import requirements of the digital twin 3D terrain sub-model); water depth distribution was collected using a single-beam underwater topography instrument; water flow velocity / direction; water quality monitoring data, with monitoring indicators clearly defined as six characteristic pollutants: COD, NH3-N, TP, BOD5, SS, and DO; topography and geomorphology, collected using a combination of UAV aerial photography and underwater topography instrument data, which were then fused using ArcGIS software. A 3D terrain data model is constructed; meteorological data collection indicators include four core indicators: precipitation, wind speed, wind direction, and temperature; tidal data (only applicable to complex waters such as estuaries / bays, with collection indicators including four indicators: tide level, tidal range, high tide duration, and low tide duration, used to calibrate the tidal parameters of the digital twin hydrodynamic sub-model; this type of data collection can be omitted for non-estuary / bay waters); all basic environmental data are uniformly converted to CSV format, with quantitative data retained to two decimal places and coordinate and angle data retained to one decimal place, and the data collection time range is clearly marked (no less than 3 months) to ensure the timeliness and representativeness of the data, which can be directly connected to the federated distributed data preprocessing stage.

[0027] ② Pollution source data: Pollution type (specific subdivisions: industrial pollution sources (chemical, printing and dyeing, food processing, etc.), domestic pollution sources (centralized urban domestic pollution, decentralized rural domestic pollution), agricultural non-point source pollution (fertilizer runoff from crop farming, wastewater from animal husbandry), each type has a unique code for easy model identification); pollution intensity (average daily pollution volume in m³ / d); pollutant emission concentration in mg / L, targeting 6 characteristic pollutants including COD, NH3-N, and TP; pollutant types (specific details: industrial pollution sources include COD, NH3-N, TP and corresponding industry characteristic pollutants; domestic pollution sources include COD, NH3-N, TP, and BOD5; agricultural non-point source pollution includes COD, NH3-N, TP, and nitrogen and phosphorus nutrients, each pollutant is labeled with CAS number and monitoring method); emission patterns (continuous emissions are labeled with 24-hour uninterrupted emission and time error; intermittent emissions are labeled with specific emission time period, duration and cycle).

[0028] ③ Constraint Data: 1. Water Quality Compliance Constraints: Based on the "Surface Water Environmental Quality Standard" GB3838-2002, quantitative limits are set for the core characteristic pollutants COD, NH3-N, and TP. The corresponding general limits for water bodies are specified as follows: COD ≤ 30 mg / L, NH3-N ≤ 1.5 mg / L, and TP ≤ 0.2 mg / L. These limits can be directly used as model constraint parameters; data exceeding these limits is considered non-compliant. 2. Distance Constraints for Discharge Outlets: The location of discharge outlets must be at least 100m from the core boundary of the ecologically sensitive area. Distance measurement is based on the straight-line distance between the GPS coordinates (accuracy 1m) of the outlet center and the GPS coordinates (accuracy 1m) of the sensitive area's core boundary. Distances less than this distance are considered invalid. The coordinates of the sensitive area's core boundary are provided and marked by the environmental protection department. Both types of constraints are converted into quantifiable and model-recognizable formats: water quality limit constraints are converted into numerical thresholds, and location distance constraints are converted into coordinate distance quantification standards. Both are entered into the constraint parameter library of the environmental capacity calculation model and optimization algorithm.

[0029] The standardized dataset consists of 6 dimensions, and its source and quantification standards are as follows: 1. Three core quantifiable features derived from basic environmental data (obtained through unified basic preprocessing + differential processing):

[0030] Dimension 1: Standardized water depth values ​​(Source: 1m×1m resolution water depth data collected by a single-beam underwater topography instrument, unit: m, after sound velocity correction, spatial interpolation completion and normalization processing, range: [0,1]).

[0031] Dimension 2: Standardized values ​​of water flow velocity in the water area (Source: flow velocity data collected by Doppler current meter once per hour, unit m / s, after time-series smoothing, outlier removal and normalization, range [0,1]);

[0032] Dimension 3: Standardized COD concentration (Source: COD data from water quality sensors and laboratory calibration once per day, unit mg / L, after refrigerated transport testing, linear interpolation, and normalization, range [0,1]).

[0033] 2. Three-dimensional core quantifiable features derived from pollution source data

[0034] Dimension 4: Standardized value of daily average sewage discharge (Source: sewage discharge data collected by flow meter once per hour, unit m³ / d, after arithmetic mean, linear interpolation and normalization, range [0,1]).

[0035] Dimension 5: Standardized value of comprehensive pollutant emission concentration (Source: weighted average of emission concentrations of three core pollutants: COD, NH3-N, and TP, unit: mg / L, normalized by sensor monitoring and laboratory calibration, range: [0,1]).

[0036] Dimension 6: Emission pattern characteristic value (Source: Continuous emissions are marked 1; intermittent emissions are converted according to "emission duration ratio × emission cycle uniformity coefficient", no unit, range [0,1], uniformity coefficient: 1 for emissions during fixed periods of the day, 0.5 for emissions during random periods, such as 0.8 × 1 = 0.8 for emissions during fixed 8 hours of the day).

[0037] Additional notes: All 6-dimensional features follow a unified preprocessing standard and can be restored to actual monitoring values ​​through an inverse normalization formula. The quantification logic is clear and repeatable. After preprocessing, the two types of constraint data form a separate standardized constraint parameter library, including water quality limits and location distance quantification standards, which are input into subsequent steps in conjunction with the 6-dimensional tensor dataset. 3. The dataset format is unified as tensor format, converted from the original CSV data. The quantified data retains two decimal places and can be directly connected to the input layer of the large environmental capacity calculation model, adapting to the needs of model training and inference.

[0038] Methods for constructing digital twins of complex water areas

[0039] This method is designed for the selection and design of sewage outlets in complex water areas (estuaries, bays, lakes, confluences of multiple tributaries, etc.). Combining the core objective of maximizing environmental capacity in this case, it integrates BIM+GIS, hydrodynamics, and water quality simulation technologies to construct a digital twin that is dynamically synchronized with the real complex water area. This digital twin supports the verification of sewage outlet selection schemes and the calculation of environmental capacity, solving the technical problems of poor adaptability and inability to replicate the dynamic characteristics of water areas in existing static models. It ensures that the construction process is repeatable and the results are verifiable. The specific steps are as follows:

[0040] (a) Data preparation

[0041] Based on the output of the federal data collection module in this case, all quantifiable data required for constructing complex water bodies are collected to ensure data accuracy and timeliness, specifically including:

[0042] Topographic data: The data was collected by combining drone aerial photography and single-beam underwater topography, covering the entire area of ​​complex waters and the surrounding 50m buffer zone. The water depth distribution resolution was no less than 1m×1m, and the boundary coordinate GPS accuracy was 1m. The data was generated as a shp format vector coordinate file and raw 3D topographic data, which were then spatially interpolated and error corrected for later use.

[0043] Hydrological dynamic data: including parameters such as water flow velocity, flow direction, and tides (for estuaries / bays). Tidal data must include tide level, tidal range, and duration of rise and fall. This can be omitted for non-tidal waters.

[0044] Water quality monitoring data: mainly includes concentration data of six characteristic pollutants such as COD, NH3-N, and TP, which are used for parameter calibration of subsequent water quality diffusion sub-models;

[0045] Auxiliary data: meteorological data (precipitation, wind speed, wind direction, temperature), water functional zone delineation data, and boundary coordinates of ecologically sensitive areas, used to simulate the constraints of sewage outlet selection.

[0046] By employing BIM+GIS fusion technology, accurate 3D terrain replication and geographic information association are achieved, supporting precise matching of water spatial location with candidate coordinates of sewage outlets. ArcGIS software is used for terrain data fusion and spatial analysis, BIM tools are used to construct 3D terrain models, and MATLAB and Python are used for solving sub-model equations and calibrating parameters. A data access interface is built to support real-time data synchronization from sensors, and edge computing technology is used to achieve rapid data processing and parameter updates, ensuring that the synchronization latency reaches the millisecond level.

[0047] (I) Construction of 3D Terrain Sub-model

[0048] The 3D terrain sub-model serves as the spatial foundation of the digital twin, replicating the static geographical features of complex water bodies and supporting the spatial positioning of hydrological and water quality sub-models. Specific steps include: Preprocessing: Importing topographic data from UAV aerial photography and water depth data collected by underwater topography instruments into ArcGIS software. Kriging interpolation is used to fill monitoring blind spots, eliminate data noise, and unify the coordinate system (WGS84 coordinate system) to ensure data consistency. Based on the preprocessed topographic and water depth data, a 3D terrain model is constructed using BIM tools. The modeling accuracy is adapted to the water body type: 10m×10m for estuaries / bays, and 50m×100m for lakes and confluences of multiple tributaries, accurately replicating water body boundaries, water depth distribution, topographic relief, and surrounding geomorphological features. Information such as water body functional zones, ecologically sensitive areas, and existing sewage outlets is labeled onto the 3D terrain model. This labeling information is linked to the constraint data of the federated data acquisition module, providing support for subsequent simulation of sewage outlet location constraints.

[0049] (II) Construction of Hydrodynamic Sub-model

[0050] The hydrodynamic sub-model is used to simulate the dynamic hydrological characteristics of complex water bodies, outputting water flow field data and providing core driving parameters for the water quality diffusion sub-model. This aligns with the pollutant diffusion simulation requirements of this case. Specific steps include: constructing a hydrodynamic sub-model based on the Saint-Venant equations; optimizing equation parameters to adapt to dynamic scenarios such as tides and seasonal changes, taking into account the water flow characteristics of complex water bodies; using real-time water flow velocity and tidal data acquired from the project's federated data acquisition module over the past three months to calibrate model parameters, adjusting key parameters such as roughness and drag coefficients to ensure that the error between the model's output water flow field data and the actual monitoring data is ≤5%; and building a time-series simulation function to simulate dynamic water flow changes at different times (intraday, seasonal, interannual), focusing on debugging the simulation accuracy of complex water flow areas such as tidal zones and tributary confluence areas to ensure accurate replication of water flow patterns and support for pollutant diffusion simulation under different discharge scenarios.

[0051] (III) Construction of Water Quality Diffusion Sub-model

[0052] The water quality diffusion sub-model is the core functional module of the digital twin in this case. It is used to simulate the distribution and range of pollutant diffusion concentration under different discharge conditions, supporting the verification of discharge outlet selection schemes and environmental capacity calculation. The specific steps are as follows: Based on the convection diffusion equation, the water flow field data output by the hydrodynamic sub-model is coupled to construct the water quality diffusion sub-model, clarifying the model's input (discharge volume, pollutant concentration, discharge location) and output (pollutant diffusion concentration distribution, diffusion range) parameters; combined with the discharge source data of this case, the types and emission patterns of pollutants from different types of discharge sources such as industrial, domestic, and agricultural non-point sources are entered, with a focus on setting diffusion parameters for six characteristic pollutants such as COD, NH3-N, and TP, matching the output requirements of the large environmental capacity calculation model; the actual monitored pollutant concentration data is compared with the model simulation data, and parameters such as the diffusion coefficient are adjusted to ensure that the average absolute error of pollutant diffusion concentration simulation is ≤5%; at the same time, combined with the discharge scenario in this case embodiment, the diffusion simulation accuracy of simultaneous discharge from multiple outlets is debugged to adapt to the verification requirements of schemes such as two-point discharge and multi-outlet optimization.

[0053] (iv) Twin fusion and visualization integration

[0054] The three sub-models of 3D topography, hydrodynamics, and water quality diffusion are integrated, and a data interaction interface is established between the sub-models. This enables the real-time driving of the water flow field data output by the hydrodynamic sub-model to drive the water quality diffusion sub-model. The water quality diffusion results are spatially correlated with the 3D topography model. A 3D visualization platform is built, integrating the output results of the three sub-models. Data interfaces are reserved for the large environmental capacity calculation model and the intelligent optimization module. This supports the import of environmental capacity values ​​and main control pollutants output by the large model, as well as the import of sewage outlet selection schemes output by the optimization algorithm, realizing the collaborative linkage between the twin and other modules.

[0055] The inputs and outputs of the digital twin for complex water areas are all quantifiable, and the inputs are all pre-processed quantifiable raw data, used only for twin modeling, dynamic driving, and accuracy calibration. They do not contain any simulation data and are specifically divided into three categories: Basic environmental raw data: water flow velocity / direction data collected by a single-beam underwater topography instrument (SDH-13D), daily average COD / NH3-N / TP concentration data from water quality sensors + laboratory calibration, as well as UAV aerial topographic data, meteorological data (wind speed, etc.), and tidal data (estuary / bay waters); Pollution source raw data: daily average pollution discharge (m³ / d), comprehensive pollutant discharge concentration (mg / L), and quantitative data related to discharge patterns collected by flow meters; Constraint-related data: GPS coordinates of the core boundary of the ecologically sensitive area, water quality limits for functional zones of the water area (COD≤30mg / L, etc., according to GB3838-2002), used for pollution outlet constraint verification.

[0056] The output (used only for visualization and verification of the solution, and not used as input to the large model) is specifically divided into two categories:

[0057] Dynamic simulation data (directly usable for validation): simulated water flow velocity / direction data, pollutant (COD / NH3-N / TP) diffusion concentration distribution data, pollutant diffusion range quantification data, and simulated water depth, water flow field stability characteristic values, etc.; Visualized output results: 3D terrain visualization model, real-time dynamic water flow simulation, and visualization of pollutant diffusion processes for different sewage outlet candidate schemes, which can intuitively show whether the sewage outlet site selection meets the constraints and whether the diffusion effect meets the standards. Input data all come from the preprocessed raw data of the federated data acquisition module. Output data is only used to verify the feasibility of sewage outlet selection schemes by connecting to the validation iteration module and the intelligent optimization module, and has no input correlation with the large-scale environmental capacity calculation model.

[0058] The construction, hierarchical connection relationship, training steps, and parameters of the four-level serial environmental capacity calculation model are as follows:

[0059] (31) Constructing the input layer: The input data is a 6-dimensional standardized dataset (3-dimensional basic environmental features: water depth, water flow velocity, COD concentration; 3-dimensional pollution source features: daily average pollution discharge, comprehensive pollutant discharge concentration, and emission pattern feature value). Based on the dimensions of the 6-dimensional standardized dataset, a linear fusion layer (containing 6 input neurons and 6 output neurons, with ReLU activation function) is constructed. A normalization function (min-max normalization, normalized to the [0,1] interval to eliminate the dimensional differences of data in different dimensions) is embedded in this linear fusion layer. The 6-dimensional standardized dataset is input into this linear fusion layer, and linear fusion (fusion weights are allocated according to feature importance: basic environmental feature weight 0.6, pollution source feature weight 0.4) and normalization operation are performed to obtain standardized fusion data (dimension: number of samples × 6), so that the output dimension of the input layer is fixed at 6. The output data of the input layer is directly transmitted to the input end of the federated pre-training basic layer to complete the initial adaptation and transmission of data.

[0060] (32) Constructing a federated pre-training base layer: The Transformer architecture is adopted, which includes a federated pre-training base layer with 16 encoder layers and 12 attention heads. The 16 encoder layers are set as follows: the first 8 layers are used to extract general features (adapting to the common features of various complex water areas), the last 8 layers are used to extract specific features (adapting to the individual features of different water areas such as estuaries, bays, and lakes), and the 12 attention heads are used to focus on key features in the 6-dimensional data (such as sewage discharge, water quality concentration, and other features directly related to environmental capacity). Each encoder layer includes a multi-head attention mechanism, layer normalization, and a FeedForward network. Standardized fused data is input into the federated pre-training base layer, and multi-feature deep association mining is performed (mining step size 1, feature association threshold 0.7) to obtain initial deep feature data. To balance the general performance and scene adaptability of the pre-trained model, the parameters of the first 8 encoder layers of the federated pre-training base layer are frozen (to retain general feature extraction capabilities), and parameter training is performed only on the last 8 encoder layers (learning rate 1e-4, training 100 rounds) to obtain optimized deep feature data. The output data of this layer is directly transmitted to the input of the scene adaptation fine-tuning layer to realize deep feature extraction and transmission.

[0061] (33) Constructing a scene adaptation fine-tuning layer: A scene adaptation fine-tuning layer consisting of 3 fully connected layers and 1 Dropout layer. The 3 fully connected layers are set as follows: the first layer is used for feature dimension transformation, the second layer is used for scene feature enhancement, and the third layer is used for output adaptation. The activation function is ReLU, and the dropout probability of the Dropout layer is set to 0.2. A scene adaptation factor is embedded in this layer (different factor values ​​are set according to the complex water body type (estuary, bay, lake), with estuary / bay set to 0.8 and lake set to 0.6). The optimized deep feature data is input into this scene adaptation fine-tuning layer, and scene adaptation transformation (feature weighted fusion, with scene adaptation factor as weight coefficient) and overfitting suppression operations are performed to obtain scene adaptation feature data. The output data of this layer is directly transmitted to the input end of the output layer to complete the scene adaptation of the features. The feature weighted fusion operation is implemented using a linear weighted summation formula, the specific formula is as follows: ,in The intermediate feature data after weighted fusion To optimize the k-th feature value in the deep feature data, The scene adaptation factor is set as follows (0.8 for bays / estuaries, 0.6 for lakes).

[0062] (34) Constructing the output layer: Input the scene adaptation feature data into the output layer, perform preliminary environmental capacity calculation, and obtain the preliminary environmental capacity value; In order to further improve the accuracy of the output results, link the adaptive multi-scale grid extrapolation algorithm and the safe dilution method (both methods are the core methods for calculating the environmental capacity of complex water areas, and their linkage can make up for the error of a single method), input the preliminary environmental capacity value into the linkage algorithm, perform calibration and correction calculation, obtain the accurate environmental capacity value, and complete the construction of the output layer; The output data of the output layer is the environmental capacity value of each grid in the complex water area. This output data is directly related to the accurate calculation of environmental capacity and the optimization of sewage outlet selection, forming a complete link of "model output - subsequent application", ensuring that the model output has clear scene application value.

[0063] (35) Model Training: Clearly define the specific operation procedures, parameter settings, and verification standards for training to ensure that technicians can repeat the training process. The mean squared error (MSE) is used as the loss function (the formula for calculating the loss function is...). ,in This is the theoretical environmental capacity value. To predict environmental capacity values ​​for the model, The formula is used to measure the numerical deviation of the sample size (β1=0.9, β2=0.999, ε=1e-8). The core training parameters are set as follows: learning rate is 1e-4 (experimentally verified, this learning rate can avoid the problem of non-convergence or slow convergence during training, balancing training efficiency and accuracy, and improving training efficiency by more than 60%), number of iterations is 300 rounds, and batch size is 32.

[0064] The calculation steps for the safe dilution method are as follows:

[0065] (41) Obtaining the baseline value of environmental quality Pre-defined water quality standard values ​​for functional water areas and the concentration of pollutants discharged from the sewage outlet. All three are units of mass concentration (mg / L), among which This represents the background concentration of pollutants in complex water bodies when they are not affected by sewage discharge. Based on the corresponding water functional zones determined according to the "Surface Water Environmental Quality Standard" (GB3838-2002), This is the actual discharge concentration from the sewage outlet, and it needs to be pre-treated to remove outliers before use.

[0066] (42) Based on the above-obtained , , Three values, execute the formula The calculation operation yields the safe dilution of a single pollutant. ;

[0067] (43) Repeat steps (41)-(42) to calculate the safe dilution of all 6 types of characteristic pollutants. ;

[0068] (44) Based on the safe dilution of all characteristic pollutants obtained Perform numerical sorting and comparison operations to filter out safe dilution levels. The pollutant with the lowest value is identified as the controlling pollutant.

[0069] In step (5), the adaptive crossover mutation genetic algorithm serves as the core artificial intelligence algorithm:

[0070] (51) Obtain the environmental capacity value (one of the core input data of the algorithm, which directly determines the upper limit of the capacity of the sewage outlet selection, representing the maximum allowable discharge of the grid area per unit time). Based on this environmental capacity value, combined with the core requirement of "maximizing the total environmental capacity" for sewage outlet selection in complex water areas, perform the objective function construction operation to obtain the objective function for maximizing the total environmental capacity. The settings for each parameter and its relationship to the scene are as follows: The parameter vector for selecting sewage outlets includes three core parameters: location, number, and sewage discharge volume, all of which are key decision variables for selecting sewage outlets in complex water areas. For the first The total environmental capacity value of the grid where each candidate discharge outlet is located (the environmental capacity value output by the model is directly related to the scene characteristics of the specific water area grid and represents the maximum allowable daily discharge of that grid). The discharge intensity coefficient is set according to the water quality sensitivity of complex water areas, with 0.8 for ecologically sensitive areas and 0.9 for ordinary areas, in line with the environmental protection constraints of the scenario. The number of candidate sewage outlets (set based on the actual area of ​​complex water bodies and sewage discharge needs, with a maximum of 8, to suit the actual project). in: : The total environmental capacity value of the grid where the i-th candidate sewage outlet is located; : The types of characteristic pollutants that need to be controlled within this grid (e.g., if only COD is controlled, then m=1). The single-grid environmental capacity value of the j-th characteristic pollutant within the i-th grid is given by the core formula mentioned above. The calculations yielded the results (the parameters have the same meaning as above, and the subscript j represents the j-th pollutant). The results are derived from the large-scale model inference and calibration based on the successfully trained environmental capacity calculation.

[0071] (52) Obtain a standardized constraint parameter library. Based on this parameter library, and in conjunction with the engineering reality and environmental protection requirements of the selection of sewage outlets in complex water areas, perform constraint quantification analysis to clarify the specific constraint quantification standards and ensure that the algorithm optimization meets the actual needs of the scenario: the number of sewage outlets ≤ 8 (set according to the environmental carrying capacity of complex water areas and the distribution of sewage enterprises to avoid excessively dense sewage outlets leading to pollution exceeding standards), the sewage discharge of a single sewage outlet ≤ the total environmental capacity value of the corresponding grid × 0.95 (the sewage discharge of a single sewage outlet is calculated on a daily average basis, consistent with the measurement dimension of the environmental capacity value, and a 5% environmental capacity redundancy is reserved to conform to environmental protection constraints and avoid the sewage discharge exceeding the carrying capacity of the water area), and the straight-line distance between the sewage outlet and the core boundary of the ecologically sensitive area ≥ 100m.

[0072] (53) For the three selection parameters of sewage outlet location, quantity, and sewage discharge, a real number encoding operation is performed. The encoding rules are in line with the actual scenario and can be directly implemented: the sewage outlet location (latitude and longitude) is directly encoded as the corresponding real number (longitude range corresponds to [110,120]; latitude range corresponds to [20,30], which is adapted to the geographical distribution of complex waters in my country), the sewage outlet quantity is encoded as an integer from 1 to 8, and the sewage discharge volume is encoded as a real number from 0 to the corresponding grid environmental capacity value × 0.95. After the encoding is completed, a population initialization operation is performed. A random initialization algorithm is adopted, the population size is set to 100, and the initialization range strictly follows the constraint quantification standard to obtain the initial population set. Each individual in the initial population corresponds to a complete sewage outlet selection scheme (location + quantity + sewage discharge), and the individual parameters all meet the above constraint quantification standard.

[0073] (54) Based on the objective function and constraint quantification criteria, perform fitness value calculation operation on each population in the initial population set to obtain the fitness value of each population; the fitness value is calculated using linear normalization, and the calculation formula is as follows: ,in is the fitness value of the i-th population (range 0-100). Let be the objective function value of the i-th population. The minimum objective function value for all populations. The maximum objective function value for all populations is given; if a population does not meet the constraint quantization criteria, its fitness value is set to 0, and it is considered an invalid population that will not participate in subsequent iterations.

[0074] The fitness value is defined as a quantitative indicator used to evaluate the merits of each population (corresponding to a set of sewage outlet selection parameters: location, number, and sewage discharge). It is associated with the objective function of maximizing total environmental capacity and the quantitative standards of various constraints. A higher value indicates that the sewage outlet selection scheme is closer to the optimization objective and meets the constraints, while a lower value indicates that the scheme has a worse optimization effect or does not meet the constraints. Based on the objective function and the quantitative standards of constraints, the fitness value is calculated for each population in the initial population set to obtain the fitness value of each population.

[0075] Crossover probability (denoted as) 1. Definition: In a genetic algorithm, the probability of parameter crossover (exchanging some sewage outlet parameters) between two excellent populations (selection schemes) is used to retain the core parameters of the excellent scheme while generating new candidate schemes to meet the optimization requirements of sewage outlet selection. This is based on the average fitness value of the population (…). ) and the fitness value of a single population ( (2) Dynamic adjustment, the core principle is that the higher the fitness value (the better the solution), the lower the crossover probability (to avoid destroying excellent parameters); the lower the fitness value (the worse the solution), the higher the crossover probability (to increase the optimization space). in: (Basic crossover probability, adapted to the optimization requirements of a population size of 100). (Minimize crossover probability to avoid excessive crossover in superior populations); : The average fitness value of all effective populations; The highest fitness value among all effective populations.

[0076] Mutation probability (denoted as) 1. Definition: In genetic algorithms, the probability of random mutation of parameters (such as the location of sewage outlets and the amount of sewage discharged) of a single population is used to increase population diversity, avoid the algorithm from getting stuck in local optima, and ensure the comprehensiveness of the sewage outlet selection scheme. Similar to the crossover probability, it is dynamically adjusted based on the population fitness value. The core principle is that the higher the fitness value, the lower the mutation probability (protecting excellent schemes); the lower the fitness value, the higher the mutation probability (attempting to optimize inferior schemes). (2) Standard calculation formula: in: (Basic mutation probability, to avoid excessive mutation leading to population chaos); (Minimize mutation probability and maintain basic population diversity). , The meaning is the same as the crossover probability, ensuring parameter consistency.

[0077] (56) Based on the adjusted crossover and mutation probabilities, selection, crossover, and mutation operations are performed on the initial population. Each operation is tailored to the complex water area sewage outlet selection scenario to ensure accurate optimization direction: The selection operation adopts the roulette wheel selection method (the selection probability is positively correlated with the fitness value; the higher the fitness value, the greater the probability of being selected; the selection probability calculation formula is...). Each time, the top 50% of the best individuals are selected and retained. Crossover operation uses a single-point crossover method (crossover point is randomly selected, and the crossover position is at 1 / 2 of the code length. Crossover is performed on the sewage outlet selection parameter vector, retaining the core parameters of the best individuals, such as reasonable sewage outlet location and suitable sewage discharge volume). Mutation operation uses a random mutation method (the mutation range is controlled within the constraint quantification standard, the position mutation range is ≤0.000001°, the quantity mutation range is ±1, and the sewage discharge volume mutation range is ≤ the corresponding grid capacity × 0.05, to avoid individuals not conforming to the scene constraints after mutation, such as the sewage discharge volume not exceeding the corresponding grid capacity × 0.95 after mutation). The above operations are used to obtain a new generation of population (the population size is still 100). Each individual in the new generation of population has been verified by scene constraints, and the verification pass rate is ≥98%, ensuring that it meets the actual needs of sewage outlet selection in complex water areas.

[0078] (57) Repeat steps (54)-(56), calculate the optimal fitness value and its rate of change after each iteration, until the rate of change of the optimal fitness value is ≤1% for 15 consecutive iterations, and stop the iteration operation;

[0079] (58) Based on the optimal population obtained after stopping the iteration (corresponding to multiple better sewage outlet selection schemes), combined with the engineering reality and environmental constraints of sewage outlet selection in complex water areas, the constraint compliance verification operation is performed. The verification content completely corresponds to the constraint quantification standard in step (52). The number of sewage outlets, sewage discharge, and location of each population are verified one by one to see if they meet the requirements (among which the sewage discharge verification is in line with the measurement dimension of the environmental capacity value, and the verification of the daily average sewage discharge is compliant). The population that meets all constraint quantification standards is selected to obtain the optimal sewage outlet selection scheme. The output data of this optimal scheme directly corresponds to the actual application scenario of sewage outlet selection in complex water areas, and can be directly used for engineering implementation, realizing the closed loop of "algorithm optimization-scenario application", and clarifying the intrinsic relationship between algorithm output and scenario requirements.

[0080] This invention also provides a system for maximizing the environmental capacity of sewage outlet selection in complex water areas to implement the above method, comprising four core modules, the execution steps of each module being as follows:

[0081] (1) Federated data acquisition module: Deploy Doppler current meter, water quality sensor, underwater topography instrument, UAV and other acquisition equipment to collect basic environmental data, sewage source data and constraint data of complex water area; Based on the above-collected data, perform federated homomorphic encryption processing operation to obtain encrypted data; Based on the encrypted data, perform standardization preprocessing operation to obtain 6-dimensional standardized dataset and standardized constraint parameter library; Output the above-mentioned 6-dimensional standardized dataset to the large model processing module, and output the standardized constraint parameter library to the intelligent optimization module;

[0082] (2) Digital Twin Construction Module: Collect UAV aerial photography data, underwater topography data and real-time dynamic data from sensors; Based on the above topographic data, perform BIM+GIS fusion modeling operation to obtain an initial digital twin; Based on the above real-time dynamic data from sensors, perform digital twin parameter synchronization update operation to obtain a dynamic digital twin; Output the simulation data of the dynamic digital twin to the large model processing module and the intelligent optimization module.

[0083] (3) Large model processing module: Obtain the 6-dimensional standardized dataset output by the federated data collection module; Based on the 6-dimensional standardized dataset, perform the construction and training operation of the environmental capacity calculation large model to obtain the environmental capacity calculation large model; Input the 6-dimensional standardized dataset into the qualified large model, perform capacity calculation and main control factor screening operation to obtain the environmental capacity value and the main control pollution factor; Output the above two data to the intelligent optimization module;

[0084] (4) Intelligent optimization module: Obtain the environmental capacity value, main control pollution factor and standardized constraint parameter library output by the large model processing module; based on the above data, perform objective function construction, constraint analysis and adaptive crossover mutation genetic algorithm optimization operation to obtain the optimal sewage outlet selection scheme; and execute the output operation of the optimal sewage outlet selection scheme.

[0085] Furthermore, the sensor terminal nodes of the federated data acquisition module include Doppler current meters, water quality sensors, underwater topography instruments, and drones, which collect basic data such as water depth, flow velocity, water quality concentration, and topography of complex water areas. Based on the collected basic data, data transmission and encryption processing operations are performed to obtain encrypted basic data. The large model processing module has a built-in error calibration function to obtain deviation data generated during environmental capacity calculation, and performs model parameter adjustment operations based on the deviation data to obtain a calibrated large model. The intelligent optimization module obtains scene parameters for different complex water areas, and performs parameter adjustment operations using an adaptive crossover mutation genetic algorithm based on the scene parameters to adapt to the delineation requirements of different complex water areas.

[0086] Example 1: A method for maximizing the environmental capacity of sewage outlet selection in complex water areas

[0087] (1) Federalized data collection and preprocessing

[0088] Basic environmental data (water depth, water flow velocity, COD concentration, NH3-N concentration, etc.), pollution source data (daily average discharge volume, pollutant discharge concentration, discharge time period, etc. of surrounding polluting enterprises), and constraint data (water quality limits of the bay, location and range of ecologically sensitive areas, etc.) were collected for the bay. Based on the collected multi-source raw data, a three-level federated data acquisition network was constructed, consisting of sensor terminal nodes composed of Doppler current meters and water quality sensors, multi-entity data holding nodes composed of environmental protection, water resources, and maritime departments, and federated coordination nodes composed of data scheduling servers. Based on this three-level federated data acquisition network, federated homomorphic encryption distributed processing was performed on the multi-source raw data to obtain encrypted multi-source heterogeneous data. Based on the encrypted multi-source heterogeneous data, cleaning (removing missing values ​​and anomalies) was performed sequentially. A unified preprocessing workflow was adopted, including dynamic data, outlier removal (using the 3σ criterion to remove outliers), normalization (normalizing the data to the [0,1] interval), and tensor transformation. Time-series smoothing was performed on dynamic data such as water flow velocity and water quality concentration, and emission pattern conversion was performed on the discharge data of intermittently discharging enterprises. This resulted in a 6-dimensional standardized dataset consisting of 3-dimensional basic environmental derived features of water depth, water flow velocity, and COD concentration, and 3-dimensional pollution source derived features of daily average discharge volume, comprehensive pollutant discharge concentration, and emission pattern characteristic value. Based on the original data of the constraint conditions, a quantitative assignment operation was performed to obtain a standardized constraint parameter library that includes the water quality limits of the bay (COD≤20mg / L, NH3-N≤1.5mg / L) and the straight-line distance of the discharge outlet from the core boundary of the ecologically sensitive area ≥100m.

[0089] (2) Construction of digital twins for complex water areas

[0090] Aerial topographic images of the bay taken by drones and measured underwater topographic data collected by underwater topography instruments were collected. Based on these two types of topographic data, BIM and GIS fusion modeling technology was used to construct an initial digital twin containing a 3D topographic sub-model, a hydrodynamic sub-model, and a water quality diffusion sub-model. A 10m×10m precision modeling operation was performed for the bay scene. Real-time dynamic data of the bay (water flow velocity, flow direction, COD concentration, NH3-N concentration) were collected from sensors at a frequency of once per hour. Based on the hourly collected water dynamic data, the parameters of the hydrodynamic sub-model and the water quality diffusion sub-model of the initial digital twin were synchronously updated to obtain a dynamic digital twin that is synchronized with the bay at the millisecond level. This was used to simulate the pollutant diffusion effect under different sewage discharge schemes.

[0091] (3) Construction and training of large-scale environmental capacity calculation model

[0092] Based on the dimensions and features of this 6-dimensional standardized dataset, a large-scale environmental capacity calculation model with a four-level cascaded structure is constructed:

[0093] (31) Constructing the input layer: Construct a linear fusion layer, embed a normalization function, input the 6-dimensional standardized dataset into the linear fusion layer, perform linear fusion and normalization operations to obtain standardized fusion data, so that the output dimension of the input layer is fixed at 6;

[0094] (32) Constructing a federated pre-training base layer: Using the Transformer architecture, a federated pre-training base layer containing 16 encoder layers and 12 attention heads is constructed. Standardized fused data is input into this layer, and multi-feature deep association mining operation is performed to obtain initial deep feature data. The parameters of the first 8 encoder layers are frozen, and only the last 8 encoder layers are trained to obtain optimized deep feature data.

[0095] (33) Constructing a scene adaptation fine-tuning layer: Construct a scene adaptation fine-tuning layer containing 3 fully connected layers and 1 Dropout layer. Set the dropout probability of the Dropout layer to 0.2 and embed the bay scene adaptation factor (value 0.8, to adapt to the dynamic features of bay tides). Input the optimized deep feature data into this layer and perform scene adaptation transformation (feature weighted fusion, scene adaptation factor is the weight coefficient) and overfitting suppression operation to obtain scene adaptation feature data (dimension: number of samples × 64).

[0096] (34) Constructing the output layer: Input the scene adaptation feature data into the output layer, perform preliminary environmental capacity calculation, and obtain the preliminary environmental capacity value (representing the maximum allowable daily emission in the grid area, which fits the continuous sewage discharge scenario); link the adaptive multi-scale grid extrapolation algorithm and the safe dilution method to perform calibration and correction calculation on the preliminary environmental capacity value, obtain the environmental capacity value, and complete the construction of the output layer;

[0097] (35) Model training: Using mean squared error (MSE) as the loss function and Adam as the optimizer, the learning rate is set to 1e-4, the number of iterations is 300 rounds, and the batch size is 32. The 6-dimensional standardized dataset is divided into training set and validation set in a 7:3 ratio. The training set is input into the above four-level structure and federated distributed training is performed. After each round of training, the validation set is used to perform accuracy verification. Training is stopped when the mean absolute error of the model is ≤5%, and the large model for environmental capacity calculation is obtained.

[0098] (4) Accurate calculation of environmental capacity and screening of main control factors

[0099] A 6-dimensional standardized dataset is input into a large-scale model for calculating qualified environmental capacity. Model inference is performed (inference batch size = 32, inference latency ≤ 500ms) to obtain preliminary environmental capacity values ​​for each grid cell in the bay. Based on these preliminary values, an adaptive multi-scale grid inference algorithm is executed. For the dynamic region (tidal range ≥ 1m) significantly affected by tides, the grid is refined (to 5m × 5m accuracy), while for the static region (tidal range < 1m), the grid is simplified (to 20m × 20m accuracy). Bilinear interpolation is then used to perform interpolation calibration on the preliminary environmental capacity values ​​(interpolation error ≤ 1%), yielding the background environmental quality value and the calibrated environmental capacity value for the bay. Based on the calibrated environmental capacity value, an error calibration function is executed (the error calibration function is...). ,in For the calibrated environmental capacity value, The theoretical environmental capacity value is used to determine whether its deviation from the theoretical environmental capacity value exceeds 2%. If the deviation of a certain grid is 3%, the deviation data is fed back to the scene adaptation fine-tuning layer. The parameters of the fully connected layer of this layer are adjusted (the learning rate is adjusted to 1e-5, and the fine-tuning rounds are 10). Then, the model inference and calibration operations are re-executed to obtain an environmental capacity value with an error ≤2%. This environmental capacity value, the water quality limit of the bay (COD ≤ 20 mg / L), and the collected COD emission concentration of the sewage outlet (80 mg / L) are obtained, and the safe dilution method is performed.

[0100] (41) Obtaining the baseline value of environmental quality Water quality standard values ​​for functional zones of water areas Pollutant emission concentration at sewage outlets ;

[0101] (42) Execute the formula To obtain the safe dilution of COD ;

[0102] (43) Repeat the above steps to calculate the safe dilution rates of the other 5 characteristic pollutants, such as NH3-N and TP, which are 0.23, 0.31, etc., respectively;

[0103] (44) Compare the safe dilution of each pollutant and select the COD with the smallest safe dilution as the main controlling pollutant.

[0104] (5) Intelligent optimization of sewage outlet selection

[0105] Obtain the environmental capacity value (representing the maximum allowable daily emissions for each grid) and the main controlling pollutant (COD). Based on these two data points, construct an objective function to maximize the total environmental capacity. (in , The analysis yields the following constraint quantification standards: number of sewage outlets ≤ 8, sewage discharge from a single outlet ≤ total environmental capacity value of the corresponding grid × 0.95 (single outlet sewage discharge is calculated on a daily average basis, consistent with the environmental capacity value measurement dimension), and straight-line distance between the sewage outlet and the core boundary of the ecologically sensitive area ≥ 100m. Based on the objective function and constraint quantification standards, an adaptive crossover mutation genetic algorithm is executed.

[0106] (51) Perform real number encoding on the three parameters of sewage outlet location (latitude and longitude), number (1-8), and sewage discharge (≤ corresponding grid capacity × 0.95), perform population initialization operation, and obtain the initial population set (population size is 100).

[0107] (52) Based on the objective function and constraint quantification criteria, perform fitness value calculation operation on each population in the initial population set to obtain the fitness value of each population;

[0108] (53) Adjust crossover and mutation probabilities: When fitness value ≥ 80, crossover probability = 0.6 and mutation probability = 0.03 (at this time the population is better, reduce the crossover and mutation probability and retain excellent individuals); when fitness value < 80, crossover probability = 0.8 and mutation probability = 0.05 (at this time the population has a large optimization space, increase the crossover and mutation probability and increase population diversity); the probability adjustment adopts a linear switching method to avoid mutation leading to population instability.

[0109] (54) Based on the adjusted probabilities, perform selection, crossover, and mutation operations on the initial population to obtain a new generation of population;

[0110] (55) Repeat the fitness calculation, probability adjustment, selection, crossover and mutation operations until the rate of change of the optimal fitness value is ≤1% for 15 consecutive rounds, then stop the iteration (the number of iterations is 280 rounds).

[0111] (56) Perform constraint compliance verification on the optimal population after stopping iteration, screen out the population that meets all constraint quantification standards, and obtain the optimal sewage outlet selection scheme for the bay (6 sewage outlets, all avoiding ecologically sensitive areas, the average daily sewage discharge of a single sewage outlet is ≤ the corresponding grid capacity × 0.95, which matches the environmental capacity value measurement dimension, and the total environmental capacity utilization rate reaches 92%).

[0112] The above are merely embodiments of the present invention. The circuits, electronic components, and modules involved are all prior art, fully achievable by those skilled in the art, and require no further explanation. The scope of protection in this application does not involve improvements to the software and methods. Commonly known structures and characteristics in the solutions are not described in detail here. Those skilled in the art are aware of all common technical knowledge in the field prior to the application date or priority date, are aware of all prior art in that field, and have the ability to apply conventional experimental methods prior to that date. Those skilled in the art can, under the guidance of this application, improve and implement this solution in combination with their own capabilities. Some typical known structures or methods should not be obstacles for those skilled in the art to implement this application. It should be noted that those skilled in the art can make several modifications and improvements without departing from the structure of the present invention. These should also be considered within the scope of protection of the present invention, and will not affect the effectiveness of the implementation of the present invention or the practicality of the application.

Claims

1. A method for maximizing environmental capacity in the selection and delineation of sewage outlets in complex waterways, characterized in that: Includes the following steps: (1) Obtain multi-source raw data including basic environmental data of complex water areas, sewage source data and constraint data, perform federated homomorphic encryption distributed processing on the above multi-source raw data to obtain encrypted multi-source heterogeneous data; based on the encrypted multi-source heterogeneous data, perform preprocessing in sequence such as cleaning, outlier removal, normalization and tensor transformation to obtain standardized dataset and standardized constraint parameter library; (2) Collect water area topographic image data and underwater topographic measurement data, and use BIM and GIS integrated modeling technology to construct an initial digital twin containing a three-dimensional topographic sub-model, a hydrodynamic sub-model, and a water quality diffusion sub-model; (3) Input the standardized dataset into the environmental capacity calculation model to obtain the preliminary environmental capacity values ​​of each grid in the complex water area. Perform error calibration on the preliminary environmental capacity values ​​to obtain the environmental capacity values ​​of the water area. Calculate the safe dilution rate of pollutants discharged from the sewage outlet using the safe dilution rate method to obtain the safe dilution rate of the pollutants. Select the pollutant with the smallest safe dilution rate as the main controlling pollution factor. The calculation formula for the safe dilution rate method is as follows: ,in This represents the environmental quality baseline value. The preset water quality standard values ​​for the functional zones of the water area, The concentration of pollutants discharged from the sewage outlet. , , All three are units of mass concentration. For the safe dilution of a single contaminant; (4) Based on the environmental capacity value of the water area and the main controlling pollution factor, construct the objective function for maximizing the total environmental capacity; determine the constraint quantification standard of the water area based on the standardized constraint parameter library; perform real number encoding and population initialization operations on the three parameters of sewage outlet location, number, and sewage discharge to obtain the initial population set; perform adaptive crossover mutation genetic algorithm to iterate on the initial population set to obtain the optimal population, and select the optimal population that meets the constraint quantification standard as the optimal sewage outlet selection scheme; The execution steps of the adaptive crossover mutation genetic algorithm are as follows: Based on the environmental capacity value, an objective function construction operation is performed to obtain the objective function that maximizes the total environmental capacity. ,in Select and define parameter vectors for sewage outlets. For the first The total environmental capacity value of the grid where each candidate sewage outlet is located. The discharge intensity coefficient is the pollution intensity coefficient. The number of candidate sewage outlets; Based on the standardized constraint parameter library, perform constraint quantification and parsing operations to clarify the constraint quantification standards; For the three selection parameters of sewage outlet location, number, and sewage discharge volume, perform real number encoding operation, and then perform population initialization operation to obtain the initial population set; Based on the objective function of maximizing total environmental capacity and the constraint quantification standard, the fitness value calculation operation is performed on each population in the initial population set to obtain the fitness value of each population. Based on the calculated fitness values ​​of various populations, crossover and mutation probability adjustment operations are performed. When the fitness value is ≥80, the crossover probability is set to 0.6 and the mutation probability is set to 0.03; when the fitness value is <80, the crossover probability is set to 0.8 and the mutation probability is set to 0.

05. Based on the adjusted crossover and mutation probabilities, selection, crossover, and mutation operations are performed on the initial population to obtain a new generation of population. Repeatedly perform fitness calculation, probability adjustment, selection, crossover and mutation operations. After each iteration, calculate the optimal fitness value and its rate of change until the rate of change of the optimal fitness value is ≤1% for 15 consecutive iterations, then stop the iteration operation. Based on the optimal population obtained after stopping the iteration, perform constraint compliance verification operation to select the population that meets all constraint quantification standards and obtain the optimal sewage outlet selection scheme.

2. The method for maximizing environmental capacity for selecting and delineating sewage outlets in complex waterways as described in claim 1, characterized in that: The standardized dataset includes standardized values ​​for water depth, water flow velocity, COD concentration, daily average sewage discharge, comprehensive pollutant discharge concentration, and emission pattern characteristics.

3. The method for maximizing environmental capacity in the selection and delineation of sewage outlets in complex waterways as described in claim 1, characterized in that: The initial digital twin includes; A three-dimensional terrain sub-model used to replicate the static geographical features of complex waterways; Hydrodynamic sub-model for simulating the dynamic hydrological characteristics of complex water bodies; A water quality diffusion sub-model used to simulate the diffusion process of pollutants under different sewage discharge conditions.

4. The method for maximizing environmental capacity in the selection and delineation of sewage outlets in complex waterways as described in claim 1, characterized in that: The environmental capacity value is the maximum allowable discharge of a certain type of pollutant within a complex water area grid per unit time, provided that the water quality standards and environmental protection constraints of the water functional zone are met, in tons / day.

5. The method for maximizing environmental capacity in the selection and delineation of sewage outlets in complex waterways as described in claim 1, characterized in that: The large-scale environmental capacity calculation model includes: Input layer: Input a standardized dataset, construct a linear fusion layer with embedded normalization functions, perform linear fusion and normalization operations to obtain standardized fused data, and transmit the standardized fused data to the federated pre-trained base layer; Federated pre-trained base layer: A Transformer architecture is used to build an encoder with 16 layers and 12 attention heads. Initial deep feature data is obtained based on standardized fused data. Based on the initial deep feature data, the parameters of the first 8 encoder layers are frozen, and only the last 8 encoder layers are trained to obtain optimized deep feature data. Scene adaptation fine-tuning layer: Based on the optimized deep feature data, a scene adaptation fine-tuning layer containing 3 fully connected layers and 1 Dropout layer is constructed and a scene adaptation factor is embedded. Scene adaptation transformation and overfitting suppression operations are performed to obtain scene adaptation feature data. Output layer: Input scene adaptation feature data into the output layer and perform preliminary environmental capacity calculation to obtain a preliminary environmental capacity value; based on the preliminary environmental capacity value, perform calibration and correction calculations in conjunction with the adaptive multi-scale grid extrapolation algorithm and the safety dilution method to obtain the environmental capacity value.

6. The method for maximizing environmental capacity for selecting and delineating sewage outlets in complex waterways as described in claim 5, characterized in that: The formula for calculating the environmental capacity is as follows: ,in This represents the single-grid environment capacity value. For scene adaptation factors, For the volume of the grid water area, The water quality standard values ​​for functional zones of water areas This represents the background value for environmental quality. For time, This refers to the concentration of pollutants emitted from the sewage outlet.

7. The method for maximizing environmental capacity for selecting and delineating sewage outlets in complex waterways as described in claim 6, characterized in that: The formula for the training loss function of the large-scale environmental capacity calculation model is as follows: ,in This is the theoretical value of environmental capacity. To predict environmental capacity values ​​for the model, This represents the number of samples.

8. A system for maximizing environmental capacity in the selection and delineation of sewage outlets in complex waterways, characterized in that, by performing the method as described in any one of claims 1-7: include: Federated Data Acquisition Module: Used to collect basic environmental data, pollution source data, and constraint data of complex water areas. Based on the collected data, it performs federated homomorphic encryption processing to obtain encrypted data. Based on the encrypted data, it performs standardization preprocessing to obtain a standardized dataset and a standardized constraint parameter library. Digital twin construction module: used to collect drone aerial photography data, underwater topography data and real-time dynamic data from sensors; based on the above topographic data, perform BIM+GIS fusion modeling operation to obtain an initial digital twin; based on the above real-time dynamic data from sensors, perform digital twin parameter synchronization update operation to obtain a dynamic digital twin; Large Model Processing Module: Obtains the standardized dataset output by the federated data collection module; based on the standardized dataset, performs the construction and training operations of the large environmental capacity calculation model to obtain the large environmental capacity calculation model; inputs the standardized dataset into the trained large model to perform capacity calculation and main control factor screening operations to obtain the environmental capacity value and the main control pollution factor; Intelligent optimization module: It acquires the environmental capacity value, main control pollution factors, and standardized constraint parameter library output by the large model processing module and the federated data acquisition module; based on the above data, it performs objective function construction, constraint analysis, and adaptive crossover mutation genetic algorithm optimization operations to obtain the optimal sewage outlet selection scheme.

Images