Urban internal lake water pollution treatment method and system

Abstract

The invention provides an urban internal lake water pollution treatment method and system. Real-time water quality data are collected from a lake; processing the collected data; using the prediction model to predict a water quality trend and identify a pollution risk; according to the analysis result of the machine learning algorithm and the water quality model, the decision support module generates a targeted treatment strategy; according to the treatment strategy provided by the decision support module, specific treatment measures are implemented; and the intelligent management and control module is used for monitoring and managing the whole treatment process in real time. All the modules are linked to construct a whole-process and intelligent treatment system, the treatment efficiency of urban internal lakes is effectively improved, and continuous improvement of water quality and long-acting restoration of an ecological system are achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of pollution control, in particular to a method and system for treating water pollution in an urban lake. BACKGROUND

[0002] As an important part of the urban ecosystem, the water quality safety of urban lakes directly affects the urban environmental quality and the health of residents. However, current urban lakes are generally facing high pollution load, insufficient water power, and fragile ecological system, etc. On the one hand, urban surface runoff, industrial wastewater and domestic sewage carry a large amount of nutrients such as nitrogen, phosphorus and organic matter into the lake through the lake inlet pipe or overflow port, leading to frequent eutrophication and black odor phenomenon in the water body. On the other hand, urban lakes are mostly weak flow or closed water bodies, with low self-purification capacity. Traditional treatment methods such as single dredging or water exchange have limited effect and high cost. In the prior art, water quality monitoring relies on single-point discrete data, lacks spatial and temporal correlation analysis, and the accuracy of pollution source tracing is insufficient. The treatment strategy is often based on experience, which is difficult to achieve precision and long-term effect. At the same time, the existing treatment system lacks a multi-technology coordination mechanism, and cannot dynamically integrate monitoring, prediction, decision-making and evaluation modules, resulting in insufficient treatment scheme and low cost-effectiveness ratio.

[0003] In addition, the traditional method for predicting water quality changes is mostly based on a simple statistical model, which cannot capture the nonlinear relationship between complex hydrological conditions and pollution processes, and does not fully consider the influence of uncertainty factors on the treatment effect, making it difficult to deal with sudden pollution events and long-term ecological restoration needs.

[0004] Therefore, it is urgent to develop a composite treatment method and system that integrates multi-dimensional monitoring, intelligent prediction, accurate source tracing and dynamic decision-making to improve the scientificity and effectiveness of urban lake treatment. SUMMARY

[0005] The main purpose of the present application is to provide a method and system for treating water pollution in an urban lake, which solves the problem that the traditional method for predicting water quality changes is mostly based on a simple statistical model, which cannot capture the nonlinear relationship between complex hydrological conditions and pollution processes, and does not fully consider the influence of uncertainty factors on the treatment effect, making it difficult to deal with sudden pollution events and long-term ecological restoration needs.

[0006] To solve the above technical problems, the technical solution adopted by the present application is: a method for treating water pollution in an urban lake, the method comprising: S1, collecting real-time water quality data from the lake; S2, processing the collected data; S3, using a prediction model to predict water quality trends and identify pollution risks; S4, according to the results of the analysis of the machine learning algorithm and the water quality model, the decision support module generates a targeted treatment strategy; S5, implementing specific governance measures according to the governance strategy provided by the decision support module; S6, using the intelligent management and control module to monitor and manage the entire governance process in real time.

[0007] In the preferred embodiment, real-time water quality data is collected using water quality detection sensors, nutrient salt sensors, and chlorophyll a sensors to collect water quality data. The data collected by the water quality detection sensor includes dissolved oxygen, pH value, turbidity, and water temperature data. The data collected by the nutrient salt sensor includes total phosphorus, total nitrogen, and ammonia nitrogen data. The data collected by the chlorophyll a sensor includes chlorophyll a concentration data. Meteorological data and hydrological data are also provided according to the weather website. Using wireless communication technology and wired communication technology, the monitored water quality data, meteorological data, and hydrological data are transmitted to the data processing module.

[0008] In the preferred embodiment, the data processing module processes the monitored data for sorting, cleaning, and fusion, and removes abnormal data and noise interference. A city lake water quality database is constructed to store historical data and real-time data.

[0009] In the preferred embodiment, in step S3, the prediction model includes a hybrid prediction framework that combines graph neural networks and hidden Markov models. The hybrid prediction framework includes a hybrid variational graph neural network and a non-homogeneous hidden Markov model, which combines a physically driven water quality dynamics model to achieve multi-scale water quality change prediction.

[0010] In the preferred embodiment, the improved spatiotemporal inverse distance weighting method is used to perform spatiotemporal interpolation on the water quality monitoring data, and the formula is where is the spatial weight, is the time decay factor, and the graph structure edge weight is constructed to include geographical distance and dynamic correlation coefficient , and the hydrodynamic driving features are extracted. The hybrid variational graph neural network is used for spatiotemporal feature learning, the multi-head attention mechanism is used to learn spatial dependencies, and the GRU layer is used to capture temporal features, and the variational inference layer is used to introduce hidden variables to handle uncertainty, and the objective function is optimized. The non-homogeneous hidden Markov model is used to construct the time-dependent transition matrix , the improved Baum-Welch algorithm is used to update the parameters, and the seasonal modulation function Implementing multi-scale prediction; Estimating the causal effect of pollution sources using dual machine learning , constructing a spatiotemporal risk index , and assigning pollution source contributions based on inverse distance weighting ; Jointly optimizing the model by fusing the loss function , evaluating the prediction performance using an improved Nash efficiency coefficient , and realizing short-term and long-term trend prediction of urban lake water quality, potential pollution risk area identification, and pollution source identification.

[0011] In the preferred embodiment, in step S4, the management strategy generated by the decision support module includes pollution source tracing and control scheme, ecological restoration scheme, cyclic purification scheme, and dredging scheme. The decision support module uses a causal reasoning network multi-objective optimization hybrid module to fuse causal discovery, multi-objective evolutionary algorithm, and dynamic Bayesian network, to realize intelligent generation and evaluation of the management strategy.

[0012] In the preferred embodiment, the method for the decision support module to realize management strategy generation and evaluation through the causal reasoning network and multi-objective optimization hybrid module is as follows: Calculate the causal effect value of the pollution source using the causal forest algorithm , generate a stepwise control strategy based on , where is the adjustment factor; Optimize the ecological restoration scheme using a multi-objective ant colony algorithm, taking the water quality comprehensive index improvement rate, total cost, and ecological diversity index as the objective function, and combining the pheromone update mechanism to search for Pareto optimal solutions; Build a four-layer network model based on dynamic Bayesian network to simulate the cyclic purification effect, predict water quality index changes through conditional probability table , and evaluate the short-term management effect using ; Distribute the dredging cost using an improved Shapley value algorithm , and determine the dredging priority in combination with ; where is the average distance from pollution source i to the coalition S, is the decay index, ensuring that nearby pollution sources bear higher costs; Establish a full life cycle cost model and an environmental benefit quantification formula to realize multi-scheme cost-benefit analysis and optimal strategy generation.

[0013] In the preferred embodiment, in step S5, the implemented management measures include: Pollution source tracing and control: Utilize water quality fingerprint analysis technology and isotope tracing technology to accurately locate pollution sources; Ecological restoration: Construct ecological revetment on the lake shore, plant strong purification ability of emergent plants, floating plants and submerged plants; Put filter-feeding fish and benthic animals into the lake to control the growth of algae; Circulating purification: Build bypass circulating purification facilities to introduce part of the lake water into the purification device; Use biological membrane method, activated carbon adsorption method, ozone oxidation method and other processes for deep treatment; The purified water is returned to the lake to improve the water quality of the lake; Ecological dredging: Regularly carry out ecological dredging of the lake bottom to remove the bottom mud rich in pollutants; The dredged bottom mud is treated as a resource to make ecological bricks and nutrient soil building materials.

[0014] In the preferred scheme, the pollution source positioning method of water quality fingerprint and isotope tracing comprises: Construct a water quality fingerprint feature matrix through space-time weighted principal component analysis , wherein the spatial weight is , and the time weight is ; Calculate the water quality fingerprint matching degree by using the improved Mahalanobis distance similarity, wherein the covariance matrix is estimated by the bootstrap method; Establish a Bayesian isotope mixing model to estimate the contribution proportion of each pollution source by Markov chain Monte Carlo sampling; Construct a multi-isotope mass balance equation , combined with the water dynamic transport equation to perform space-time tracing; Fuse water quality fingerprint evidence and isotope evidence by using the Dempster combination rule ; Determine the pollution source parameters by solving a nonlinear optimization problem , use Latin hypercube sampling to quantify uncertainty, and calculate the 95% confidence interval .

[0015] In the preferred scheme, the system comprises a real-time monitoring module, a data transmission module, a data processing module, an intelligent analysis module, a decision support module, a pollution control module and an intelligent control module; The real-time monitoring module monitors the water quality data of different areas of the inner lake in real time; The data transmission module is used to transmit the water quality data to the data processing module; The data processed by the data processing module is input to the intelligent analysis module; The intelligent analysis module deeply mines and analyzes the processed water quality data; A water quality change prediction model is established to predict the short-term and long-term change trends of the water quality of the urban inner lake, and identify potential pollution risk areas and pollution sources; The results of the analysis of the intelligent analysis module are used by the decision support module to generate targeted management strategies; According to the management strategies provided by the decision support module, the pollution management module implements specific management measures: The management measures include pollution source tracing and control, ecological restoration, cyclic purification, and ecological dredging; The intelligent management and control module monitors and manages the entire management process in real time; The staff can view the running status of each subsystem in real time through a computer or a mobile phone, receive alarm information, and perform remote operations; The intelligent management and control module has an automatic control function and automatically controls the operation of each device according to the management strategies; The water quality is continuously monitored, and the management measures are dynamically adjusted according to the real-time water quality conditions.

[0016] The present application provides a method and system for urban inner lake water pollution management, which realizes precise management through multi-dimensional technology integration: precise positioning of pollution sources using water quality fingerprints and isotope tracing, combined with intelligent model prediction of water quality change trends; a composite management system is constructed through multi-objective optimization and dynamic simulation to improve water flow and self-purification capacity; based on game theory, the cost is optimized and shared to realize scientific decision-making and long-term management of the management scheme, significantly improving the water quality of urban inner lakes, restoring the health of the water ecosystem, and enhancing the ecological resilience of the water body. BRIEF DESCRIPTION OF DRAWINGS

[0017] The present application will be further described below in conjunction with the drawings and examples: Figure 1 is a pollution management flowchart of the present application. DETAILED DESCRIPTION

[0018] Example 1 As shown in Figure 1 , a method for urban inner lake water pollution management, the method comprising: S1, collecting real-time water quality data from the lake; S2, processing the collected data; S3, using a prediction model to predict water quality trends and identify pollution risks; S4, according to the results of the analysis of the machine learning algorithm and the water quality model, the decision support module generates targeted management strategies; S5, according to the management strategies provided by the decision support module, specific management measures are implemented; S6, real-time monitoring and management of the whole treatment process is performed by using the intelligent management and control module.

[0019] In the preferred embodiment, the real-time water quality data is collected by using a water quality detection sensor, a nutrient salt sensor, and a chlorophyll a sensor. The data collected by the water quality detection sensor includes dissolved oxygen, pH value, turbidity, and water temperature data. The data collected by the nutrient salt sensor includes total phosphorus, total nitrogen, and ammonia nitrogen data. The data collected by the chlorophyll a sensor includes chlorophyll a concentration data. Meteorological data and hydrological data are provided according to a meteorological webpage. The monitored water quality data, meteorological data, and hydrological data are transmitted to the data processing module by using wireless communication technology and wired communication technology.

[0020] The meteorological data and hydrological data provided according to the meteorological webpage provide basic data support for subsequent analysis. The monitored water quality data, meteorological data, and hydrological data are transmitted to the data processing module by using wireless communication technology and wired communication technology.

[0021] The monitoring sensor includes a water quality detection sensor, a nutrient salt sensor, and a chlorophyll a sensor. The data collected by the water quality detection sensor includes DO, pH, turbidity, and water temperature data, and is installed at the lake center, the lake inlet, and the broken stream location. The data collected by the nutrient salt sensor includes total phosphorus, total nitrogen, and ammonia nitrogen data, and is installed at the near-shore shallow water area and the downstream of the sewage outlet. The data collected by the chlorophyll a sensor includes chlorophyll a concentration data, and is installed at the backwater area of the inner lake.

[0022] In the preferred embodiment, the data processing module performs row arrangement, cleaning, and fusion on the monitored data, and removes abnormal data and noise interference. A city inner lake water quality database is constructed to store historical data and real-time data.

[0023] Embodiment 2 In combination with Embodiment 1, it is further illustrated that in step S3, the prediction model includes a hybrid prediction framework using a combination of a graph neural network and a hidden Markov model. The hybrid prediction framework includes a hybrid variational graph neural network and a non-homogeneous hidden Markov model, and realizes multi-scale water quality change prediction in combination with a physically driven water quality dynamics model.

[0024] In the preferred embodiment, the improved spatiotemporal inverse distance weighting method is used for spatiotemporal interpolation of the water quality monitoring data, and the formula is wherein is the spatial weight, is the time decay factor, the edge weight of the graph structure is constructed by combining geographical distance and dynamic correlation coefficient , and the hydrodynamic driving characteristics are extracted ; spatiotemporal feature learning is performed using a hybrid variational graph neural network, and the spatial dependence is learned through a multi-head attention mechanism , combined with a GRU layer to capture temporal features, and a variational inference layer The hidden variable is introduced to handle uncertainty, and the objective function is optimized ; The time-dependent transition matrix is constructed based on the non-homogeneous hidden Markov model , the parameters are updated by the improved Baum-Welch algorithm, and the seasonal modulation function is combined to achieve multi-scale prediction The dual machine learning is used to estimate the causal effect of pollution sources , the spatiotemporal risk index is constructed , and the pollution source contribution is allocated based on the inverse distance weighting ; The loss function is fused to optimize the model jointly, and the improved Nash efficiency coefficient is used to evaluate the prediction performance , realizing the prediction of short-term and long-term trends of water quality in urban lakes and the identification of potential pollution risk areas and pollution sources.

[0025] S3. Water quality change prediction and pollution source identification algorithm The improved spatiotemporal inverse distance weighting method (ST-IDW) is used to handle sensor missing values, and a time decay factor is introduced to enhance the weight of recent data: ; where: is the spatial weight, is the distance between monitoring points, is the distance power index is the time decay factor, is the time decay coefficient.

[0026] Graph structure construction and edge weight calculation: The spatiotemporal correlation graph is constructed , the nodes are monitoring sites, and the edge weight combines geographical distance and water quality correlation: ; where: is the Euclidean geographical distance, is the distance scaling parameter Dynamic correlation coefficient of water quality parameter for site i,j, calculated by sliding window: ; L is the window length, is the mean value in the window.

[0027] Physical driving feature extraction: Extract flow field and diffusion coefficient field from hydrodynamic model, construct water quality migration feature: ; Where: is the neighborhood of site i; is the flow velocity component from j to i, is the diffusion coefficient.

[0028] Hybrid variational graph neural network (HVGNN), spatial encoder (GAT layer): Learn the dynamic dependency between nodes through multi-head attention mechanism: ; Where: is the feature vector of node i, is the learnable weight matrix, is the attention vector; Multi-head attention aggregation: ; Temporal encoder (GRU layer); Capture the time evolution law of water quality parameters: ; Where: is the reset gate and update gate, is the candidate hidden state.

[0029] Variational inference layer: Introduce random latent variable to handle uncertainty, and realize it through reparameterization trick: ; Where: and are the mean and standard deviation learned through fully connected layer; Variational lower bound objective function: ; The first term is the reconstruction error, and the second term is the KL divergence regularization.

[0030] Non-homogeneous Hidden Markov Model (NH-HMM), time-dependent transition matrix: State transition probability Dynamic over time, modeled as a function of external factors : ; Where: Contains time characteristics, weather factors, human activity indicators; Is a learnable weight vector, trained through historical data.

[0031] Improved Baum-Welch algorithm: introduce time window Local optimization within the window, avoid global computational complexity: ; Where: Observed sequence within the time window; Parameter update formula: ; Pi,j(t) is the probability of transitioning from state i to j at time t.

[0032] Multi-scale prediction Short-term prediction uses hourly resolution, long-term prediction introduces seasonal modulation function : ; Where: Pi(t) is the initial state distribution at time t; is the annual seasonal factor.

[0033] 4. Pollution source tracing and risk area identification Estimate the causal effect of pollution source x on water quality y using double machine learning (Double Machine Learning):

[0034] Where: Eliminate confounding factors by orthogonalization: ; And are estimated by random forest respectively. Combine prediction uncertainty and historical abnormal values to construct risk index : ; Where: And is the historical mean and standard deviation; is the predicted variance.

[0035] Pollution source localization algorithm: Pollution source contribution allocation based on inverse distance weighting: ; where: is the distance between monitoring point i and potential pollution source k; is the abnormal concentration change of monitoring point i; is the distance decay index (determined by cross-validation).

[0036] 5. Model ensemble and evaluation Physics-data fusion loss function: ; where: data-driven loss: ; Physics constraint loss: ; S is the source-sink term, calculated by the physics model.

[0037] Adaptive weight adjustment: Dynamic weight balancing short-term and long-term prediction errors: ; where: is the weight at time t, is the turning point, is the adjustment rate.

[0038] Model evaluation index: Introduce the improved Nash efficiency coefficient (Modified NSE): ; where: is the stability parameter, to prevent the denominator from being zero; more robust to outliers, suitable for water quality prediction scenarios.

[0039] Algorithm application process: 1. Data preparation: collect sensor data, hydrodynamic model output, meteorological data, etc., and perform spatio-temporal interpolation and feature engineering.

[0040] 2. Model training: Pre-train HVGNN, optimize variational lower bound; Train NH-HMM, estimate time-dependent transition matrix; Joint fine-tuning, minimize fusion loss function.

[0041] 3. Prediction and tracing: Short-term forecast: Combine real-time data with weather forecasts to generate hourly forecasts using the NH-HMM. Long-term prediction: Introducing seasonal factors to simulate monthly / quarterly water quality evolution; Risk identification: Calculate spatiotemporal risk heat maps and locate high-risk areas; Source analysis: quantify the causal effects and contributions of each pollution source.

[0042] 4. Result verification and update: Evaluate model performance through indicators such as MNSE; Model parameters are updated periodically with new data.

[0043] This algorithmic framework achieves high-precision water quality prediction and pollution source tracing by integrating physical mechanisms and data-driven methods, combining spatiotemporal modeling and uncertainty quantification, and is particularly suitable for complex water environment systems such as urban lakes.

[0044] Example 3 Further illustrating with reference to Example 1, in step S4, the governance strategies generated by the decision support module include a pollution source tracing and control plan, an ecological restoration plan, a recycling purification plan, and a dredging plan; The decision support module adopts a causal reasoning network multi-objective optimization hybrid module to integrate causal discovery, multi-objective evolutionary algorithm and dynamic Bayesian network to realize the intelligent generation and evaluation of governance strategies.

[0045] In the preferred solution, the decision support module uses a causal reasoning network and a multi-objective optimization hybrid module to realize governance strategy generation and evaluation as follows: Calculate the causal effect value of pollution sources using the causal forest algorithm ,based on Generate a step-by-step control strategy, where is the regulating factor; The ecological restoration plan is optimized through a multi-objective ant colony algorithm, with the water quality comprehensive index improvement rate, total cost, and ecological diversity index as the objective function, combined with the pheromone update mechanism Search for Pareto optimal solutions; Based on the dynamic Bayesian network, a four-layer network model is constructed to simulate the cyclic purification effect. Predict changes in water quality indicators and use Evaluate short-term governance effectiveness; Using the improved Shapley value algorithm Allocate desilting costs, combined with Determine dredging priorities; in, is the average distance from pollution source i to alliance S, To attenuate the index, ensure that neighboring pollution sources bear higher costs; Establish a full life cycle cost model Formula for quantifying environmental benefits , realizing multi-scheme cost-benefit analysis and optimal strategy generation.

[0046] S4. Governance Strategy Generation and Evaluation Algorithm Based on Causal Inference Network-Multi-Objective Optimization Hybrid Module 1. Quantification of causal effects of pollution sources and generation of control plans Step 1.1: Causal Forest Algorithm to Identify Key Pollution Sources Algorithm flow: 1. Using water quality indicators (such as total phosphorus (TP)) as outcome variables and pollution source characteristics (enterprise discharge, surface runoff, etc.) as input variables, a causal forest model was constructed. 2. Calculate the causal effect value (CATE) of each pollution source through double robust estimation (Double Robust Estimation), the formula is: ; in, To open the pollution source The water quality response, For the response at closure, confounding factors were eliminated by propensity score matching (PSM).

[0047] 3. Select the pollution sources with the top 20% CATE absolute values ​​as priority control targets.

[0048] Step 1.2: Generate a step-by-step control strategy Algorithm rules: Design dynamic control intensity based on CATE value, the formula is: ; in, is the adjustment factor (the empirical value is 1.2) to ensure that the control intensity of high CATE pollution sources is ≥80%.

[0049] Output scheme: Enterprise pollution sources: Enterprises with CATE>0.1mg / L will be subject to pollution tax (tax rate = CATE × 50,000 yuan / unit pollution amount); Non-point source pollution: In areas where CATE>0.05mg / L, a rain garden should be built with an area = CATE×1000㎡.

[0050] 2. Ecological Restoration Optimization: Multi-Objective Ant Colony Algorithm (MOACO) Step 2.1: Strategy encoding and objective function Encoding method: Each ant represents an ecological restoration portfolio strategy, including: Submerged plant type (buckweed / fox tail / rotala, one-hot encoding); Planting density (1030 plants / m2, continuous variable); Release of benthic animal species (clams / mussels, proportional allocation).

[0051] Multi-objective function: ;, where, is the index weight, is the plant cost (yuan / plant), is the density, is the animal cost (yuan / individual), is the quantity, is the species weight.

[0052] Step 2.2: Pheromone update mechanism Global update: ; where, is the pheromone evaporation rate, is the optimal value of the th objective, determined by the Pareto frontier.

[0053] Local update: ; is the local update coefficient, which strengthens the current path pheromone.

[0054] 3. Design of cyclic purification scheme: Dynamic Bayesian Network (DBN) Step 3.1: DBN structure construction Train DBN parameters through historical data, such as the relationship between cycle flow and flow rate : ; where a, b are fitted by linear regression.

[0055] Step 3.2: Strategy effect simulation Short-term simulation (13 months): Input the cycle flow , and predict through DBN forward propagation: ; where, is the pollutant decay coefficient, is the lake volume, derived based on the first-order kinetic model.

[0056] Long-term simulation (15 years): Introducing environmental variable trend terms: ; The increase in dissolved oxygen brought by the circulation purification, is the stabilization time (years).

[0057] 4. Optimizing Dredging Solutions: Game Theory Cost Sharing Model Step 4.1: Sediment Risk Index (SRI) calculation Risk grading formula: ; in, Pollutants in sediments The concentration of is the standard value, is the weight (e.g. heavy metals have a higher weight than organic matter).

[0058] Step 4.2: Cost Allocation Algorithm Improved formula of Shapley value: Considering the spatial spillover effect, the distance attenuation factor is introduced: ; in, For pollution sources To the Alliance The average distance, is the attenuation exponent (taken as 1.5), ensuring that neighboring pollution sources bear higher costs.

[0059] Step 4.3: Prioritize dredging Comprehensive priority index: ; Combine sediment risk index (SRI) with pollution source contribution (Contribution) to ensure that high-risk and high-contribution areas are given priority for dredging.

[0060] 5. Cost-benefit analysis of multiple options Step 5.1: Life Cycle Cost Model Ecological restoration costs: ; is the discount rate, For the Annual maintenance cost.

[0061] Circulation purification cost: ; ; The depreciation cost of the equipment (yuan / m³), The energy consumption (kWh / m³).

[0062] Step 5.2: Quantification of environmental benefits Water quality improvement benefits: ; The unit area value of water quality improvement (yuan / ㎡) is determined through the contingent valuation method (CVM) survey.

[0063] Ecological benefits: ; The annual growth rate of the ecological index.

[0064] Algorithm application process 1. Causal analysis: Identify key pollution sources through causal forest and calculate CATE value to determine control priority; 2. Strategy generation: MOACO searches for ecological restoration Pareto solution set and combines DBN to simulate the effect of cyclic purification; 3. Cost allocation: Distribute the dredging cost based on the improved Shapley value to generate a priority list; 4. Comprehensive evaluation: Compare the cost-benefit ratio (BCR) of each scheme and output the optimal combination strategy.

[0065] This algorithm framework clearly defines pollution responsibility through causal reasoning, balances governance effect and cost through multi-objective optimization, and evaluates long-term impact through dynamic simulation, providing scientific and interpretable decision support for urban inner lake governance.

[0066] In the preferred scheme, the governance measures implemented in step S5 include: Pollution source tracing and control: Use water quality fingerprint analysis technology and isotope tracing technology to accurately locate pollution sources; Ecological restoration: Build ecological revetment on the lake shore and plant strong purifying ability of emergent plants, floating plants and submerged plants; Release filter-feeding fish and benthic animals into the lake to control algal growth; Cyclic purification: Build bypass cyclic purification facilities to introduce part of the lake water into purification devices; Use biological membrane method, activated carbon adsorption method, ozone oxidation method and other processes for advanced treatment; Purified water is returned to the lake to improve water quality; Ecological dredging: Regularly perform ecological dredging on the lake bottom to remove pollution-rich sediment; The dredged sediment is resourceized to make ecological bricks and nutrient soil building materials.

[0067] 2. Ecological restoration scheme Optimal plant combination based on MO-ACO search: Submerged plants (Vallisneria 60% + Stylidium 40%) are planted in the area with water depth of 1.2m, covering 70% of the area, and the cost-benefit ratio is improved to 1:3.2; DBN prediction: Total phosphorus decreased by 45% and algae density decreased by 60% after 6 months of implementation.

[0068] Cost analysis: Initial investment: 1.2 million yuan (seedlings + planting equipment); Long-term benefits: Carbon sink revenue increased by 200,000 yuan per year, and tourism revenue increased by 500,000 yuan per year.

[0069] 3. Circulation purification scheme MO-ACO optimized circulation path: 30m³ / s of water is introduced from the transit water source, preferentially flowing through highly polluted areas (such as broken streams), and the water replacement period is shortened from 90 days to 28 days; Energy cost reduced by 18% (through particle swarm optimization of pump set start-stop strategy).

[0070] Expected effect: Dissolved oxygen increased from 3mg / L to 5.5mg / L, and ammonia nitrogen decreased by 50%.

[0071] 4. Dredging scheme Dredging priority based on SRI index: SRI=3.2 in the center of the lake (severe pollution), dredging (cost 800,000 yuan / 10,000m³) is preferred, SRI=1.8 in the shore area, in-situ passivation (cost 300,000 yuan / 10,000m³) is adopted; After dredging, the release of nitrogen and phosphorus from the sediment is reduced by 70%.

[0072] Cost-benefit: Total dredging cost: 5 million yuan, saving 40% compared with full-lake mechanical dredging; Environmental benefits: Reduce the risk of water quality deterioration caused by internal pollution, save 1.5 million yuan in emergency treatment costs per year.

[0073] In the preferred scheme, the pollution source positioning method based on water quality fingerprint and isotope tracing includes: Construct water quality fingerprint feature matrix through spatiotemporal weighted principal component analysis , where the spatial weight is , and the temporal weight is ; Calculate the matching degree of water quality fingerprint using improved Mahalanobis distance similarity, where the covariance matrix is estimated by bootstrap method; Establish Bayesian isotope mixing model, estimate the contribution proportion of each pollution source by Markov chain Monte Carlo sampling; Construct multi-isotope mass balance equation , combined with hydrodynamic transport equation for spatiotemporal tracing; Using Dempster's composition rule Integrating water quality fingerprint evidence and isotope evidence ; By solving nonlinear optimization problems Determine pollution source parameters, quantify uncertainty using Latin hypercube sampling, and calculate 95% confidence intervals .

[0074] Water quality fingerprint analysis and isotope tracing pollution source location algorithm 1. Water quality fingerprint feature extraction and matching algorithm Step 1.1: Construction of multidimensional water quality feature vector Collect m water quality parameters (such as TP, TN, DO, COD, heavy metal concentration, etc.) from n monitoring points and construct the original feature matrix ; Apply spatiotemporal weighted principal component analysis (STWPCA) to extract features: ; in is the spatial weight matrix, is the time weight matrix, and the calculation formula is: ; is the attenuation coefficient, determined by cross-validation.

[0075] Step 1.2: Calculation of water quality fingerprint similarity Define the modified Mahalanobis distance similarity:

[0076] in is the feature covariance matrix, estimated by bootstrap method:

[0077] is the characteristic matrix of the b-th resampling.

[0078] Step 1.3: Pollution source matching and tracing Constructing a pollution feature library , each Corresponding to a type of pollution source Fuzzy C-means clustering (FCM) combined with evidence theory is used for multi-source matching: ; in For samples Belong to class The membership degree, The blur index is usually taken as 2.

[0079] 2. Isotope tracer source resolution algorithm Step 2.1: Multi-isotope mixing model For n isotopes, the mass balance equation is established: ; Where is the concentration of the mixed sample, is the contribution ratio of source i, is the characteristic concentration of source i, is the error term.

[0080] Step 2.2: Bayesian isotope mixing model (BIMM) Introduce prior information and uncertainty quantification: ; Where the posterior probability represents the distribution of the contribution ratio of each source under the data , which is estimated by Markov Chain Monte Carlo (MCMC) sampling.

[0081] Step 2.3: Spatio-temporal isotope tracing Combine the hydrodynamic model to construct the transport equation: ; Where is the isotope concentration, is the flow field, is the diffusion coefficient, is the source strength, is the Dirac function.

[0082] 3. Multi-source data fusion positioning algorithm Step 3.1: Evidence theory fusion framework Define water quality fingerprint evidence and isotope evidence , through Dempster combination rule: ; Where is the basic probability assignment function, $A,B,C$ is the proposition set.

[0083] Step 3.2: Spatio-temporal Bayesian network inference Construct a dynamic Bayesian network (DBN): ; Where is the state variable at time t, Control variables, Set of parent nodes of .

[0084] Step 3.3: Pollution source localization optimization model Establish a nonlinear optimization problem:

[0085] where is the pollution source parameter, is the observation value, is the prediction model, is the regularization term (such as total variation).

[0086] 4. Uncertainty quantification and verification Step 4.1: Monte Carlo error propagation Generate a set of parameters through Latin hypercube sampling (LHS)

[0087] Calculate the confidence interval of the localization result: ; where is the empirical quantile, is the pollution source location estimate.

[0088] Step 4.2: Cross-validation and model selection Use leave-one-out cross-validation (LOOCV): ; where is the prediction value excluding the ith sample.

[0089] Step 4.3: Traceability reliability evaluation Define the comprehensive reliability index: ; where each indicator is calculated through the confusion matrix and spatiotemporal consistency analysis.

[0090] Algorithm flow and implementation 1. Data collection: synchronously collect water quality routine parameters, isotope characteristic values, and hydro-meteorological data; 2. Feature extraction: apply STWPCA dimensionality reduction to construct a water quality fingerprint feature library; 3. Isotope modeling: establish a BIMM model and estimate source contribution proportion through MCMC sampling; 4. Evidence fusion: fuse multi-source evidence using the Dempster combination rule; 5. Localization optimization: solve the nonlinear optimization problem to determine the pollution source location and intensity; 6. Uncertainty analysis: quantify the reliability of results through Monte Carlo simulation; The algorithm framework realizes high-precision positioning of pollution sources through multi-evidence fusion and uncertainty quantification, and is particularly suitable for pollution tracing in complex urban water systems.

[0091] Embodiment 4 Further illustrated in combination with Embodiment 1, the system includes a real-time monitoring module, a data transmission module, a data processing module, an intelligent analysis module, a decision support module, a pollution control module, and an intelligent management and control module. The real-time monitoring module monitors water quality data in different areas of the inner lake in real time. The data transmission module is used to transmit water quality data to the data processing module. The data processed by the data processing module is input to the intelligent analysis module. The intelligent analysis module deeply mines and analyzes the processed water quality data. A water quality change prediction model is established to predict the short-term and long-term change trends of the water quality of the urban inner lake, and identify potential pollution risk areas and pollution sources. Based on the analysis results of the intelligent analysis module, the decision support module generates targeted control strategies. According to the control strategies provided by the decision support module, the pollution control module implements specific control measures: The control measures include pollution source tracing and control, ecological restoration, cyclic purification, and ecological dredging. The intelligent management and control module monitors and manages the entire control process in real time. The staff can view the running status of each subsystem in real time through a computer or a mobile phone, receive alarm information, and perform remote operations. The intelligent management and control module has an automatic control function and automatically controls the operation of each device according to the control strategies. The water quality is continuously monitored, and the control measures are dynamically adjusted according to the real-time water quality status.

[0092] Through the coordinated operation of multiple modules, the intelligent and precise water pollution control of urban inner lakes is realized. The cooperation of the real-time monitoring module and the data transmission module ensures the comprehensive and timely collection and transmission of water quality data, providing a reliable basis for subsequent treatment; the data processing and intelligent analysis module uses advanced algorithms to deeply mine data, accurately predicts water quality trends, identifies pollution risk areas and sources, and changes passive treatment to active prevention and control. The decision support module generates targeted treatment strategies based on the analysis results, covering pollution source control, ecological restoration and other aspects, ensuring that the scheme is scientific and reasonable. The pollution control module efficiently executes the strategy, and the intelligent control module monitors the entire treatment process in real time, accurately controls the equipment with automatic control functions, and dynamically optimizes the measures based on real-time water quality to ensure treatment effectiveness. Staff can monitor system operation, receive alarms and remotely operate through multiple terminals, improving management flexibility and efficiency. The linkage of various modules builds a full-process, intelligent treatment system, effectively improving the efficiency of urban inner lake treatment, and achieving sustainable improvement of water quality and long-term restoration of the ecological system.

[0093] The above embodiments are only preferred technical solutions of the present application, and should not be regarded as a limitation of the present application. The protection scope of the present application should be based on the technical solutions recited in the claims, including equivalent replacement solutions of the technical features recited in the claims. That is, equivalent replacement improvements within this scope are also within the protection scope of the present application.

Claims

1. A method for treating urban lake water pollution, characterized by: The method includes: S1. Collect real-time water quality data from lakes; S2, process the collected data; S3. Use predictive models to predict water quality trends and identify pollution risks; S4. Based on the analysis results of the machine learning algorithm and water quality model, the decision support module generates targeted governance strategies; S5. Implement specific governance measures based on the governance strategy provided by the decision support module; S6. Use the intelligent management and control module to monitor and manage the entire governance process in real time.

2. The method for treating urban lake water pollution according to claim 1, wherein: Collect real-time water quality data using water quality detection sensors, nutrient salt sensors, and chlorophyll a sensors; The data collected by the water quality detection sensor includes dissolved oxygen, pH value, turbidity, and water temperature data; The data collected by the nutrient sensor includes total phosphorus, total nitrogen, and ammonia nitrogen data; The data collected by the chlorophyll a sensor include chlorophyll a concentration data; It also provides meteorological data and hydrological data according to the weather webpage; The monitored water quality data, meteorological data and hydrological data are transmitted to the data processing module using wireless communication technology and wired communication technology.

3. A method for treating urban lake water pollution according to claim 2, characterized in that: The data processing module organizes, cleans and integrates the monitored data to remove abnormal data and noise interference; Build a water quality database for urban lakes to store historical and real-time data.

4. The method for treating urban lake water pollution according to claim 1, wherein: In step S3, the prediction model includes using a hybrid prediction framework combining a graph neural network and a hidden Markov model; The hybrid prediction framework includes a hybrid variational graphical neural network and a non-homogeneous hidden Markov model, combined with a physics-driven water quality dynamics model to achieve multi-scale water quality change prediction.

5. The method for treating urban lake water pollution according to claim 4, wherein: The improved space-time inverse distance weighting method is used to perform space-time interpolation on water quality monitoring data. The formula is: ,in is the spatial weight, As the time decay factor, a graph structure edge weight is constructed that includes geographical distance and dynamic correlation coefficient , and extract the hydrodynamic driving characteristics ; Using hybrid variational graph neural network for spatiotemporal feature learning, through multi-head attention mechanism Learning spatial dependencies, combining GRU layers to capture temporal features, and using variational inference layers Introducing hidden variables to deal with uncertainty and optimize the objective function ; Constructing time-dependent transfer matrix based on nonhomogeneous hidden Markov model , update parameters by the improved Baum-Welch algorithm, combined with the seasonal modulation function Achieve multi-scale prediction; Estimating the causal effects of pollution sources using dual machine learning , constructing a spatiotemporal risk index , and allocate the contribution of pollution sources based on inverse distance weight ; By fusion loss function Joint optimization model using improved Nash efficiency coefficient Evaluate prediction performance to predict short-term and long-term trends in urban lake water quality and identify potential pollution risk areas and pollution sources.

6. The method for treating urban lake water pollution according to claim 1, wherein: Step S4 In the decision support module, the governance strategies generated include pollution source tracing and control scheme, ecological restoration scheme, recycling purification scheme and dredging scheme; The decision support module adopts a causal reasoning network multi-objective optimization hybrid module to integrate causal discovery, multi-objective evolutionary algorithm and dynamic Bayesian network to realize the intelligent generation and evaluation of governance strategies.

7. The method for treating urban lake water pollution according to claim 6, wherein: The decision support module uses a causal reasoning network and a multi-objective optimization hybrid module to achieve governance strategy generation and evaluation: Calculate the causal effect value of pollution sources using the causal forest algorithm ,based on Generate a step-by-step control strategy, where is the regulating factor; The ecological restoration plan is optimized through a multi-objective ant colony algorithm, with the water quality comprehensive index improvement rate, total cost, and ecological diversity index as the objective function, combined with the pheromone update mechanism Search for Pareto optimal solutions; Based on the dynamic Bayesian network, a four-layer network model is constructed to simulate the cyclic purification effect. Predict changes in water quality indicators and use Evaluate short-term governance effectiveness; Using the improved Shapley value algorithm Allocate desilting costs, combined with Determine dredging priorities; in, is the average distance from pollution source i to alliance S, To attenuate the index, ensure that neighboring pollution sources bear higher costs; Establish a full life cycle cost model Formula for quantifying environmental benefits , realizing multi-scheme cost-benefit analysis and optimal strategy generation.

8. The method for treating urban lake water pollution according to claim 1, wherein: In step S5, the implemented governance measures include: Pollution source tracing and control: using water quality fingerprint analysis technology and isotope tracing technology to accurately locate pollution sources; Ecological restoration: construct ecological slope protection on the lakeshore and plant emergent plants, floating plants and submerged plants with strong purification capabilities; Stocking the lake with filter-feeding fish and benthic animals to control algae growth; Circulation purification: Build a bypass circulation purification facility to divert part of the lake water into the purification device; Use biofilm method, activated carbon adsorption method, ozone oxidation method and other processes for deep treatment; The purified water flows back into the lake, improving its water quality; Ecological desilting: Regular ecological desilting of the lake bottom to remove pollutant-rich sediment; The bottom mud after dredging is processed for resource utilization and made into ecological bricks and nutrient soil building materials.

9. The method for treating urban lake water pollution according to claim 8, wherein: Water quality fingerprinting and isotope tracing pollution source location methods include: Constructing water quality fingerprint feature matrix through spatiotemporal weighted principal component analysis , where the spatial weight , time weight ; The improved Mahalanobis distance similarity was used to calculate the water quality fingerprint matching, where the covariance matrix was estimated by the bootstrap method; A Bayesian isotope mixing model was established, and the contribution ratio of each pollution source was estimated through Markov chain Monte Carlo sampling. Constructing a multi-isotope mass balance equation , combined with the hydrodynamic transmission equation Conduct spatiotemporal tracing; Using Dempster's composition rule Integrating water quality fingerprint evidence and isotope evidence ; By solving nonlinear optimization problems Determine pollution source parameters, quantify uncertainty using Latin hypercube sampling, and calculate 95% confidence intervals .

10. The urban lake water pollution control system according to any one of claims 1 to 9, characterized in that: The system includes real-time monitoring module, data transmission module, data processing module, intelligent analysis module, decision support module, pollution control module and intelligent management and control module; The real-time monitoring module monitors the water quality data of different areas of Neihu in real time; The data transmission module is used to transmit water quality data to the data processing module; The data processed by the data processing module is input into the intelligent analysis module; The intelligent analysis module conducts in-depth mining and analysis based on the processed water quality data; Establish a water quality change prediction model to predict short-term and long-term trends in urban lake water quality; identify potential pollution risk areas and pollution sources; Based on the analysis results of the intelligent analysis module, the decision support module generates targeted governance strategies; According to the control strategy provided by the decision support module, the pollution control module implements specific control measures: Control measures include pollution source tracing and control, ecological restoration, recycling purification, and ecological dredging; The intelligent management and control module monitors and manages the entire governance process in real time; Staff can view the operating status of each subsystem in real time through computers or mobile phones, receive alarm information and perform remote operations; The intelligent management and control module has an automated control function, which automatically controls the operation of each device according to the management strategy; Continuously monitor changes in water quality and dynamically adjust treatment measures based on real-time water quality conditions.

Images