Method and system for dynamic purification and regulation of river and lake water ecological pollution level

Abstract

The application provides a dynamic purification regulation and control method and system for river and lake water ecological pollution levels, relates to the technical field of intelligent control, and the method comprises the following steps: acquiring water level, flow rate, water quality parameters and river and lake topography multi-source monitoring data of a target water area, performing discrete degree analysis and filter normalization fusion processing to obtain an initial multi-dimensional state data set; inputting the water quality parameters into a pre-constructed pollution level judgment model for analysis to obtain a comprehensive pollution characterization index and a pollution level label, and determining an initial purification and restoration regulation and control scheme according to the pollution level label. Through multi-source data fusion modeling, hierarchical purification and ecological restoration collaborative regulation, residual feedback iterative correction, the application realizes accurate treatment, dynamic adaptation and closed-loop optimization of river and lake water ecological pollution, and improves the treatment efficiency.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology, and in particular to a method and system for dynamic purification and regulation of the ecological pollution level of rivers and lakes. Background Technology

[0002] Currently, technologies for treating river and lake water pollution have made some progress. For example, water quality monitoring equipment is deployed to obtain water level, flow velocity, and conventional water quality indicators, which are then combined with preset threshold rules to initiate aeration, filtration, or ecological restoration measures. However, in practical applications, existing methods still have room for improvement. For instance, river and lake pollution typically exhibits certain spatiotemporal dynamic changes; water level, flow velocity, and pollutant concentration may fluctuate with changes in upstream water flow, overflow from rainwater outlets, or human activities along the banks. Under such dynamic conditions, traditional technologies often rely on periodic sampling or data at fixed time intervals for decision-making. The resulting purification plans (such as aeration intensity and filter component activation rate) are sometimes statically set or can only be manually adjusted at limited levels. When pollution levels change significantly or suddenly, there may be a certain deviation between the initial control parameters and the actual needs. In addition, physical purification operations (such as aeration, filtration, and flow regulation) and ecological restoration measures (such as planting aquatic plants, administering microbial agents, and setting up ecological reefs) are sometimes implemented step by step or relatively independently in current practices. There is a lack of a collaborative adjustment mechanism based on real-time feedback between the two. If the initial treatment plan does not fully match the actual topography and pollutant distribution characteristics of the water area, it may be necessary to rely on on-site human intervention to correct subsequent operations.

[0003] For example, in a slow-flowing inland river in a plain city, initial monitoring showed that the river section was in a state of mild eutrophication (total phosphorus concentration approximately 0.12 mg / L, with a slight increase in chlorophyll a). Based on this, on-site management personnel initiated low-frequency intermittent aeration (aeration intensity set at 30% of the rated value) and planted a certain density of Vallisneria natans according to standard procedures, without simultaneously activating the dynamically adjustable filter components and flow control valves. Several days later, a short-term heavy rainfall occurred in the area, causing an overflow at a combined sewer overflow point along the riverbank. Some domestic sewage entered the river, raising the total phosphorus concentration to 0.65 mg / L and dissolved oxygen to 1.2 mg / L, thus raising the pollution level to moderate to severe. At this point, the original aeration control parameters remained at a low intensity level, the filter components were not activated, and the planting depth and microbial agent dosage remained at the values ​​set under the initial mild pollution conditions. Under on-site conditions, there was a certain delay in the manual response. The initial purification and ecological restoration parameters failed to adapt to the changed water quality in a timely manner, resulting in the pollutant concentration in some water areas remaining at a high level for several hours. The situation was gradually improved by increasing manual detection and manually adjusting equipment parameters. Summary of the Invention

[0004] This invention provides a dynamic purification and control method and system for the ecological pollution level of rivers and lakes, which can quickly respond to the spatiotemporal dynamic changes of water pollution and avoid the lag and subjective bias of human judgment.

[0005] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows: Firstly, a dynamic purification and control method for the pollution level of rivers and lakes, the method comprising: Step 1: Obtain multi-source monitoring data on water level, flow velocity, water quality parameters, and river and lake topography of the target water area. After dispersion analysis and filtering normalization fusion processing, an initial multidimensional state dataset is obtained. Step 2: Input the water quality parameters into the pre-constructed pollution level determination model for analysis to obtain the comprehensive pollution characterization index and pollution level label, and determine the initial purification and remediation control plan based on the pollution level label. Step 3: Deploy distributed purification units in the target water area according to the topographic features of the river and lake, analyze the initial purification and restoration control scheme, extract the aeration intensity control parameters, filter component opening rate control parameters and flow regulating valve opening control parameters, drive the distributed purification units to perform differentiated operations, and obtain a graded dynamic purification execution instruction set. Step 4: Receive the hierarchical dynamic purification execution instruction set and couple the river and lake topographic features and water quality parameters in the initial multidimensional state dataset. Dynamically calculate the weight of aquatic plant planting depth, the quantitative dosage ratio of microbial agents, and the coordinates of ecological reef layout to obtain the ecological restoration adaptation parameter set. Step 5: Collect time-series data of the treated water area after executing the hierarchical dynamic purification execution instruction set and the ecological restoration adaptation parameter set, obtain the residual feedback vector, and input the residual feedback vector back into the pollution level judgment model for iterative correction of the control parameters until the residual feedback vector converges to the preset compliance range, thus obtaining the final dynamic control instruction sequence.

[0006] Secondly, the dynamic purification and control system for the pollution levels of rivers and lakes includes: The acquisition module is used to acquire multi-source monitoring data on water level, flow velocity, water quality parameters and river and lake topography of the target water area. After discrete analysis and filtering normalization fusion processing, an initial multidimensional state dataset is obtained. The analysis module is used to input water quality parameters into a pre-built pollution level determination model for analysis, obtain a comprehensive pollution characterization index and pollution level label, and determine the initial purification and remediation control plan based on the pollution level label. The extraction module is used to deploy distributed purification units in the target water area according to the topographic features of rivers and lakes, analyze the initial purification and restoration control scheme, extract aeration intensity control parameters, filter component opening rate control parameters and flow regulating valve opening control parameters, drive the distributed purification units to perform differentiated operations, and obtain a graded dynamic purification execution instruction set. The calculation module is used to receive the hierarchical dynamic purification execution instruction set and couple the river and lake topographic features and water quality parameters in the initial multidimensional state dataset to dynamically calculate the weight of aquatic plant planting depth, the quantitative dosing ratio of microbial agents and the coordinates of ecological reef layout, and obtain the ecological restoration adaptation parameter set. The feedback module is used to collect time-series data of the treated water area after executing the hierarchical dynamic purification execution instruction set and the ecological restoration adaptation parameter set, obtain the residual feedback vector, and input the residual feedback vector back into the pollution level judgment model for iterative correction of the control parameters until the residual feedback vector converges to the preset compliance range, thus obtaining the final dynamic control instruction sequence.

[0007] The above-described solution of the present invention has at least the following beneficial effects: By analyzing the dispersion of multi-dimensional monitoring data (water level, flow velocity, water quality, and topography), filtering, noise reduction, and normalization fusion, a spatiotemporally unified initial multi-dimensional state dataset is constructed. This effectively eliminates abnormal interference, restores the true state of the water body, and overcomes the problems of incomplete single monitoring dimensions and insufficient data reliability. Based on the multi-dimensional data, a pollution level determination model is built, automatically outputting a comprehensive pollution characterization index and pollution level labels. This allows for rapid response to the spatiotemporal dynamic changes in water pollution, avoiding the lag and subjective bias of manual judgment, and achieving precise matching between purification and remediation plans and the actual pollution level. Distributed purification units are deployed in zones according to river and lake topographic features. Aeration, filtration, and flow regulation parameters are differentiated and dynamically executed according to pollution level and regional characteristics. This ensures that physical purification operations are highly adapted to the water topography and pollution distribution, avoiding resource waste caused by extensive operation and optimizing operational energy consumption while improving purification efficiency. By using the purification execution status as a constraint boundary and coupling topographic and water quality parameters to dynamically calculate ecological restoration parameters, precise adaptation of aquatic plant planting depth, microbial release ratio, and ecological reef placement location is achieved. This allows ecological restoration and physical purification to work synergistically, improving restoration survival rate and ecological stability. By collecting post-treatment time-series data to construct a residual feedback vector, the judgment model and control parameters are iteratively corrected in reverse, forming a complete closed loop of monitoring, judgment, control, restoration, feedback, and optimization. This allows the system to adapt to dynamic changes in the water body without frequent human intervention, enhancing the overall intelligence and automation level of the system. Attached Figure Description

[0008] Figure 1 This is a flowchart illustrating the dynamic purification and control method for the ecological pollution level of rivers and lakes provided in an embodiment of the present invention.

[0009] Figure 2 This is a schematic diagram of a dynamic purification and control system for the ecological pollution level of rivers and lakes provided in an embodiment of the present invention.

[0010] Figure 3A schematic diagram illustrating the internal feature coupling of the pollution level determination model and the calculation of the comprehensive pollution characterization index; Figure 4 A spatial contour map showing the suitability coefficient S for restoration; Figure 5 A heat map showing the spatial distribution of the ballast weight W at the planting depth of aquatic plants; Figure 6 This represents the changing trend of the five component values ​​of the one-dimensional residual feedback vector, with the dashed line indicating the threshold of ±0.05. Figure 7 A bar chart showing the concentration comparison of key water quality parameters (DO, NH3-N, TP) from T0 to T3 after the third iteration; Figure 8 The graph shows the downward trend of the average comprehensive pollution characterization index P over time (T0, T1, T2, T3) for the entire experimental section. Figure 9 A trend line graph showing how the system's energy consumption decreases as the number of iterations increases. Detailed Implementation

[0011] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0012] like Figure 1 As shown, embodiments of the present invention propose a dynamic purification and control method for the ecological pollution level of rivers and lakes, the method comprising the following steps: Step 1: Obtain multi-source monitoring data on water level, flow velocity, water quality parameters, and river and lake topography of the target water area. After dispersion analysis and filtering normalization fusion processing, an initial multidimensional state dataset is obtained. Step 2: Input the water quality parameters into the pre-constructed pollution level determination model for analysis to obtain the comprehensive pollution characterization index and pollution level label, and determine the initial purification and remediation control plan based on the pollution level label. Step 3: Deploy distributed purification units in the target water area according to the topographic features of the river and lake, analyze the initial purification and restoration control scheme, extract the aeration intensity control parameters, filter component opening rate control parameters and flow regulating valve opening control parameters, drive the distributed purification units to perform differentiated operations, and obtain a graded dynamic purification execution instruction set. Step 4: Receive the hierarchical dynamic purification execution instruction set and couple the river and lake topographic features and water quality parameters in the initial multidimensional state dataset. Dynamically calculate the weight of aquatic plant planting depth, the quantitative dosage ratio of microbial agents, and the coordinates of ecological reef layout to obtain the ecological restoration adaptation parameter set. Step 5: Collect time-series data of the treated water area after executing the hierarchical dynamic purification execution instruction set and the ecological restoration adaptation parameter set, obtain the residual feedback vector, and input the residual feedback vector back into the pollution level judgment model for iterative correction of the control parameters until the residual feedback vector converges to the preset compliance range, thus obtaining the final dynamic control instruction sequence.

[0013] In this embodiment of the invention, by analyzing the dispersion of multi-dimensional monitoring data of water level, flow velocity, water quality, and topography, filtering and denoising, and normalizing and fusing the data, a spatiotemporally unified initial multi-dimensional state dataset is constructed. This effectively eliminates abnormal interference, restores the true state of the water body, and overcomes the problems of incomplete single monitoring dimensions and insufficient data reliability. Based on the multi-dimensional data, a pollution level determination model is constructed, automatically outputting a comprehensive pollution characterization index and pollution level label. This model can quickly respond to the spatiotemporal dynamic changes of water pollution, avoiding the lag and subjective bias of manual judgment, and achieving accurate matching between purification and remediation plans and the actual pollution level. Distributed purification units are deployed in zones according to river and lake topographic features. Aeration, filtration, and flow regulation parameters are differentiated and dynamically executed according to pollution level and regional characteristics. This ensures that physical purification operations are highly adapted to the water topography and pollution distribution, avoiding resource waste caused by extensive operation, and optimizing operational energy consumption while improving purification efficiency. By using the purification execution status as a constraint boundary and coupling topographic and water quality parameters to dynamically calculate ecological restoration parameters, precise adaptation of aquatic plant planting depth, microbial release ratio, and ecological reef placement location is achieved. This allows ecological restoration and physical purification to work synergistically, improving restoration survival rate and ecological stability. By collecting post-treatment time-series data to construct a residual feedback vector, the judgment model and control parameters are iteratively corrected in reverse, forming a complete closed loop of monitoring, judgment, control, restoration, feedback, and optimization. This allows the system to adapt to dynamic changes in the water body without frequent human intervention, enhancing the overall intelligence and automation level of the system.

[0014] In a preferred embodiment of the present invention, step 1 includes: Step 100 involves real-time acquisition of water level time-series signals, flow velocity vector field data, water quality parameter solution concentration values, and river and lake topographic elevation point cloud data to form multi-source monitoring data. Specifically, this includes: firstly, clarifying the standards for dividing each monitoring area in the target water area to ensure clear and unambiguous area boundaries. The specific divisions are as follows: shallow water area is defined as water area with a depth ≤ 1.0m; deep water area is defined as water area with a depth > 2.0m; flow channel area is defined as water area with a flow velocity ≥ 0.5m / s, connecting the upstream and downstream of the target water area without significant stagnation; stagnant water area is defined as water area with a flow velocity < 0.1m / s, weak water exchange capacity, and easy accumulation of pollutants. The above area divisions are all based on the pre-set topographic survey data, and the boundaries of each area are accurately defined using WGS84 coordinate system coordinates.

[0015] Water level time-series signals are acquired through submersible water level sensors. The sensors are embedded in three or more monitoring sections of the target water area (the distance between sections is ≤50m, covering the upstream, midstream and downstream of the target water area). One sensor is deployed in each section for shallow water area, deep water area, flow channel area and stagnant water area. The sensor is embedded at a depth of 1 / 2 of the average water depth of the area. The acquisition frequency is set to 1 time / 10min, and the acquisition is continuous and uninterrupted to form a continuous water level time-series signal. The operating temperature is adapted to the ambient temperature of the target water area (0 to 60℃). Velocity vector field data were acquired using Doppler current meters. One Doppler current meter was deployed at each monitoring section, with the sampling direction consistent with the water flow direction (deviation ≤ 5°). The sampling frequency was set to 10 Hz. Simultaneously, high-definition image / video acquisition equipment (fixed resolution 1920×1080, fixed frame rate 28fps, lens focal length 8 to 12mm, night vision distance ≥ 50m) was used to acquire images of water flow. The motion feature points on the water surface were extracted using optical flow method. Combined with the velocity data acquired by the Doppler current meters, a complete velocity vector field was constructed to characterize the velocity magnitude and flow direction in different regions, ensuring the spatial correlation of the velocity data.

[0016] The solution concentration values ​​of water quality parameters are collected using a five-parameter integrated water quality sensor. One sensor is deployed at each monitoring section using an immersion method, with the sensor probe completely submerged in the water and at least 0.3m away from the bottom silt to avoid data deviation caused by probe contact with silt. The sensor focuses on collecting the solution concentration values ​​of five core water quality parameters: dissolved oxygen, ammonia nitrogen, total phosphorus, COD, and turbidity. The acquisition frequency is set to once every 30 minutes. During the acquisition process, the sensor's working status (including working voltage and probe cleanliness) is recorded in real time. The probe is cleaned and maintained every 24 hours to avoid data deviation caused by probe fouling. River and lake topographic elevation point cloud data were acquired through a combination of 3D laser scanning and ultrasonic topographic scanning. The 3D laser scanning equipment was installed at high points along the shore (installation height ≥ 8m) or on a drone (flight altitude 80 to 100m), with a scanning frequency of 15Hz, a laser wavelength of 1550nm, a ranging range of 0.5 to 100m, and an accuracy of ±1cm, completely covering the entire target water area. The ultrasonic topographic scanning equipment was mounted on a mobile monitoring vessel, which scanned at a constant speed (speed ≤ 2m / s) along a preset route, with an ultrasonic frequency of 200kHz, a measurement depth of 0 to 10m, and an accuracy of ±2cm. The data acquired by the two methods complemented each other, comprehensively capturing topographic information such as river and lake bottom elevation, slope contours, and underwater undulations, forming complete topographic elevation point cloud data (point cloud density ≥ 50 points / m²). 2 All data acquisition devices start up synchronously and work together. Each set of acquired data corresponds to a unique timestamp and spatial coordinates, ensuring that all types of data correspond one-to-one in time and space, forming multi-source monitoring data.

[0017] Step 101: Perform dispersion analysis on each type of multi-source monitoring data, calculate the local outlier deviation coefficient of each data point in its corresponding feature dimension, mark and remove abnormal data points whose local outlier deviation coefficient exceeds a preset dynamic threshold, and obtain the denoised monitoring data sequence, specifically including: The multi-source monitoring data collected in step 100 are divided into four categories according to data type: water level time series signal group, flow velocity vector field data group, water quality parameter solution concentration value group, and topographic elevation point cloud data group. Discreteness analysis is performed on each category of data. First, for a single type of monitoring data, 30 continuously collected data sets from that type of data are selected as the analysis sample set, and the mean and standard deviation of the sample set are calculated. Based on the mean and standard deviation, the local outlier coefficient for each data point is calculated as: Local outlier coefficient = |xi-μ| / σ, where xi is the value of a single data point, μ is the mean of the data sample set for that type, and σ is the standard deviation of the data sample set for that type. A preset dynamic threshold is set, which is differentiated according to different data types. The preset dynamic threshold for water level and flow velocity data is set to 3, and the preset dynamic threshold for water quality parameters and topographic data is set to 2.5. The calculated local outlier coefficient is compared with the corresponding preset dynamic threshold. If the local outlier coefficient of a data point exceeds the corresponding preset dynamic threshold, the data point is determined to be an outlier data point, marked, and removed from the original monitoring data. The remaining data after removing outlier data points are organized according to the collection time sequence to form a denoised monitoring data sequence.

[0018] Step 102: Filter out the high-frequency fluctuation components in the denoised monitoring data sequence to obtain smoothed monitoring data; perform range normalization on the smoothed monitoring data to obtain normalized water level, flow velocity, water quality parameters, and river and lake topography data; perform feature-level fusion of the normalized water level, flow velocity, water quality parameters, and river and lake topography data according to the same timestamp and spatial coordinates to construct an initial multidimensional state dataset. Each dimension of the initial multidimensional state dataset corresponds to a monitoring parameter type, and each data point contains a spatiotemporal label, specifically including: The denoised monitoring data sequence obtained in step 101 is smoothed by using a Gaussian filter algorithm to remove high-frequency fluctuation components, resulting in smoothed monitoring data. A 3×3 filter window is selected, and the weight of each data point within the filter window is calculated based on a Gaussian function. The weight calculation follows a Gaussian distribution, with data points closer to the center of the window having a higher weight and those farther away having a lower weight, i.e., w(x,y) = , where x and y are the coordinates of the data points within the filter window relative to the center of the window (with values ​​of -1, 0, and 1). The Gaussian filter coefficient is set to 0.5 to ensure that high-frequency fluctuations in the data are filtered out without losing the effective features of the data itself. Each data point is multiplied by its corresponding weight and then summed to obtain the smoothed result of that data point. All data points are filtered in sequence to obtain smoothed monitoring data.

[0019] Range normalization was performed on the smoothed monitoring data to uniformly map water level, flow velocity, water quality parameters, and river and lake topography data from different numerical ranges to the [0,1] interval. For each type of smoothed monitoring data, the maximum value x in that type of data was found. max and minimum value x min Normalize each data point in this type of data using the formula x. norm =(x-x min ) / (x max -x min ), where x norm Here, x represents the normalized data value, and x represents the original smoothed monitoring data point value. This calculation normalizes all data to the [0,1] interval, yielding normalized water level, flow velocity, water quality parameters, and river and lake topography data. Feature-level fusion is then performed on the normalized data to construct an initial multidimensional state dataset. During the fusion process, timestamps and spatial coordinates are used as the correlation benchmarks to match normalized water level time-series data, flow velocity vector field data, water quality parameter solution concentration values, and topographic elevation point cloud data one by one, ensuring that various parameter data under the same timestamp and spatial coordinates correspond and are correlated. Each spatial coordinate corresponds to a monitoring point in the target water area, and each timestamp corresponds to a synchronous acquisition time. The normalized water level data, normalized flow velocity data, normalized water quality parameter data, and normalized topographic data of each monitoring point at each acquisition time are treated as a data point. Each data point contains a corresponding spatiotemporal label, which consists of the timestamp of the acquisition time and the spatial coordinates of the acquisition point. All data points are organized in spatiotemporal order to form an initial multidimensional state dataset. Each dimension of this dataset corresponds to a monitoring parameter type, covering four major categories of parameters: water level, flow velocity, water quality, and topography, fully representing the initial state of the target water area.

[0020] This embodiment calculates local outlier deviation coefficients through discreteness analysis to identify and eliminate abnormal data points, effectively avoiding the impact of abnormal data caused by equipment failure, environmental interference, and other factors on subsequent processing. Gaussian filtering is then used to remove high-frequency fluctuation components, further improving data smoothness and ensuring that the monitoring data accurately reflects the actual conditions of the target water area. Range normalization unifies various data types into the same numerical range, solving the problem that parameters such as water level, flow velocity, water quality, and topography have different dimensions and large differences in numerical ranges, making direct fusion and analysis impossible. Spatiotemporal labels are used to associate various normalized data to construct an initial multidimensional state dataset, integrating scattered single-parameter data into comprehensive and systematic state data, achieving feature-level data fusion.

[0021] In a preferred embodiment of the present invention, step 2 includes: Step 200: Based on the water quality parameter dimension data and corresponding historical pollution level labels in the initial multidimensional state dataset, a pollution level determination model is constructed using a multi-layer neural network structure, specifically including: The input data and output labels for model training are clearly defined. The input data consists of water quality parameter dimension data from the initial multidimensional state dataset, specifically including normalized data for five core water quality parameters: dissolved oxygen, ammonia nitrogen, total phosphorus, COD, and turbidity. All input data undergoes preprocessing steps 100 to 102 to ensure that the data is anomaly-free, dimensionless, and within the [0,1] range. The output labels are the corresponding historical pollution level labels, which are divided into three categories: light pollution, moderate pollution, and heavy pollution. Light pollution refers to... Of the five core water quality parameters, only one exceeds the standard, and the exceedance of a single parameter is ≤50%. Specifically, dissolved oxygen ≥5 mg / L, ammonia nitrogen ≤1.0 mg / L, total phosphorus (lake / reservoir) ≤0.05 mg / L, and COD ≤20 mg / L. The exceedance range is calculated as (measured value - water standard value) / water standard value × 100%. Moderate pollution is defined as 2 to 3 of the five core water quality parameters exceeding the standard, with each exceeding the standard by ≤100%. Severe pollution is defined as ≥4 of the five core water quality parameters exceeding the standard, with any single parameter exceeding the standard by >100%. The three pollution levels are identified using numerical codes 0, 1, and 2, respectively, with 0 for light pollution, 1 for moderate pollution, and 2 for severe pollution.

[0022] The specific structure of the multi-layer neural network was determined, adopting a three-layer neural network architecture consisting of an input layer, hidden layers, and an output layer. The number of neurons in the input layer is consistent with the dimensions of the water quality parameters, set to 5, corresponding to the five water quality parameters: dissolved oxygen, ammonia nitrogen, total phosphorus, COD, and turbidity. Each input layer neuron is responsible for receiving normalized data for one water quality parameter. Two hidden layers are set, with the first hidden layer having 10 neurons and the second hidden layer having 5 neurons. The ReLU activation function is used in both hidden layers to enhance the nonlinear fitting ability of the model. This function can effectively solve the gradient vanishing problem and improve the model training efficiency and fitting accuracy. The number of neurons in the output layer is set to 3, corresponding to the three labels of light pollution, moderate pollution, and heavy pollution. The Softmax activation function is used in the output layer to transform the model output into a probability distribution.

[0023] The multi-layer neural network was trained and calibrated. The training dataset used water quality parameters and corresponding historical pollution level labels from the initial multidimensional state dataset. The historical pollution level labels were determined according to the aforementioned classification criteria. The training dataset was divided into a training set and a validation set in a 7:3 ratio. The training set was used to fit the model parameters, and the validation set was used to verify the model accuracy. During model training, the Adam optimization algorithm was used to minimize the loss function. The cross-entropy loss function was selected to measure the deviation between the model's predicted values ​​and the actual historical labels. The number of iterations was set to 1000, and the learning rate was set to 0.001. After each iteration, the prediction accuracy of the validation set was calculated. Training was stopped when the validation set accuracy did not improve for 50 consecutive iterations to avoid overfitting. After training, the model parameters were saved to obtain the trained pollution level judgment model. The model's judgment accuracy was not less than 98%, ensuring that the model could identify different pollution levels and that the judgment results were completely consistent with the preset pollution level classification criteria.

[0024] Step 201: The real-time water quality parameters in the initial multidimensional state dataset are used as input feature vectors and input into the pollution level determination model to calculate the feature coupling strength coefficient between each water quality parameter. Through iterative aggregation of features, a comprehensive pollution characterization index is obtained. The comprehensive pollution characterization index represents the overall deviation of the current water quality, specifically including: Real-time water quality parameters, including normalized data of five core parameters—dissolved oxygen, ammonia nitrogen, total phosphorus, COD, and turbidity—are extracted from the initial multidimensional state dataset and arranged in a fixed order to form an input feature vector. ,in For dissolved oxygen normalized data, For ammonia nitrogen normalized data, For total phosphorus normalized data, For COD normalized data, The turbidity data is normalized, and all data points are in the range [0,1]. The input feature vector is fed into the input layer of the pollution level determination model. The input layer neurons transmit the water quality parameter data to the first hidden layer. The first hidden layer neurons perform a linear weighted calculation on the input data, i.e. ,in For the first hidden layer The input value of each neuron. For the input layer The first neuron and the first hidden layer Connection weights between neurons It is the first Normalized input values ​​of the water quality parameters, For the first hidden layer Bias terms for each neuron, =1,2,...,5 (corresponding to 5 water quality parameters), =1,2,...,10 (corresponding to 10 neurons in the first hidden layer); The ReLU activation function is applied to... A non-linear transformation is performed to obtain the output value of the first hidden layer.

[0025] The output value of the first hidden layer The data is then passed to the second hidden layer, where neurons also undergo linear weighting operations. ,in For the second hidden layer The input value of each neuron. For the first hidden layer The first neuron and the second hidden layer Connection weights between neurons For the second hidden layer Bias terms for each neuron, =1,2,...,5 (corresponding to 5 neurons in the second hidden layer); then, a nonlinear transformation is performed using the ReLU activation function to obtain the output value of the second hidden layer. Based on the output value of the second hidden layer, the characteristic coupling strength coefficient between each water quality parameter is calculated. The characteristic coupling strength coefficient is used to characterize the degree of correlation between different water quality parameters. First, calculate the output value of each neuron in the second hidden layer. normalized value Then, the sum of the products of any two water quality parameters corresponding to the hidden layer output values ​​is calculated to obtain the feature coupling strength matrix, and the elements in the matrix are... That is, the first Water quality parameters and the first The characteristic coupling strength coefficient between the water quality parameters is calculated using the following formula: ,in , =1,2,...,5, and ≠ , , They are respectively the first , The normalized second-layer output value corresponding to each water quality parameter has a characteristic coupling strength coefficient ranging from [0,1]. The larger the coefficient, the stronger the correlation between the two water quality parameters.

[0026] The comprehensive pollution characterization index is obtained by fusing the input feature vector, feature coupling strength coefficient, and hidden layer output value. Specifically, the aggregation process involves first calculating a weighted sum of the input feature vectors, where the weights are the average values ​​of the corresponding feature coupling strength coefficients. ,in It is the comprehensive contribution value of the input water quality characteristics after being weighted by the coupling coefficient. It is the first Item and the The coupling strength coefficient of the water quality parameters; calculate the weighted sum of the output values ​​of the second hidden layer, where the weights are the connection weights of the output layer obtained from model training, i.e. ,in It is the corrected value of the output of the second hidden layer after being weighted by the output layer. For the second hidden layer The nth neuron and the output layer The connection weights between neurons are determined through model training and calibration, and their values ​​range from [0.15, 0.25]. For example... When =1 (corresponding to a slightly polluted label), the connection weight is 0.18. When the value is 2 (corresponding to a moderate pollution label), the connection weight is 0.20. When the value is 3 (corresponding to a heavily polluted label), the connection weight is 0.22. When the values ​​are 4 and 5 (corresponding to auxiliary feature outputs), the connection weights are all set to 0.16. This value is dynamically calibrated based on the accuracy of the validation set during model training to ensure that the weight values ​​are compatible with the accuracy of pollution level determination, so that the feature outputs corresponding to different pollution levels can accurately reflect the degree of water pollution. and Linear fusion is performed to obtain the comprehensive pollution characterization index. Where 0.6 and 0.4 are the fusion weights, The actual deviation of real-time water quality parameters has a more critical impact on the comprehensive pollution characterization index, so it is given a higher weight of 0.6 to ensure that the true state of real-time water quality parameters can dominate the index calculation. It is mainly used to assist in correcting index deviations and improve the stability and accuracy of the index. Therefore, it is given a weight of 0.4. The combination of the two makes the comprehensive pollution characterization index able to truly reflect the overall deviation of the current water quality. The value range of the comprehensive pollution characterization index is [0,1]. The larger the value, the greater the overall deviation of the current water quality and the more serious the pollution.

[0027] Step 202: Match the comprehensive pollution characterization index with multiple preset pollution level value ranges step by step to obtain the corresponding pollution level labels; based on the pollution level labels, retrieve the initial purification and remediation control scheme uniquely corresponding to the corresponding pollution level label from the pre-stored control scheme knowledge base. The initial purification and remediation control scheme includes aeration intensity benchmark value, filter component opening rate benchmark value, and flow regulating valve opening benchmark value, specifically including: A preset pollution level range is established, determined based on historical pollution time-series data of the target water area, the variation patterns of water quality parameters under different pollution levels, and data from previous remediation practices. Historical water quality monitoring data for the target water area over the past 3 to 5 years is collected, covering different seasons (spring, summer, autumn, and winter) and different hydrological conditions (high water, low water, and normal water) for measured values ​​of water quality parameters (measured values ​​of five core parameters: dissolved oxygen, ammonia nitrogen, total phosphorus, COD, and turbidity). Simultaneously, pollution level records for the corresponding time periods, determined according to the criteria in step 200, are also collected. The collected historical data is categorized into light, moderate, and heavy pollution levels. For each pollution level, corresponding historical water quality parameter data is extracted, and the comprehensive pollution characterization index for each set of historical water quality parameters is calculated using the calculation method in step 201. Statistical analysis is performed on the comprehensive pollution characterization index values ​​corresponding to each pollution level, calculating the mean, range, and natural distribution range (for each) of the indices. The comprehensive pollution characterization index is determined by combining the continuous range from the minimum to the maximum value of the index with the critical values ​​of the comprehensive pollution characterization index corresponding to the water quality reaching the treatment standard under different pollution levels in previous treatment practices (collecting measured water quality parameters at the time of water quality reaching the standard under each pollution level in previous treatment, obtaining the comprehensive pollution characterization index at the time of reaching the standard according to the calculation method in step 201, and taking the average value of the standard reaching index of the same pollution level as the critical value of that level). At the same time, referring to the water standards and pollution level classification standards specified in step 200, the statistically obtained natural distribution range is calibrated and adjusted, and abnormal data exceeding the critical value ±0.02 in the range are removed. At the same time, it is ensured that there is no overlap or omission between adjacent level ranges, and abnormal distribution data caused by extreme weather or equipment abnormalities are removed. Finally, the numerical range of the comprehensive pollution characterization index corresponding to the three pollution levels is determined.

[0028] The comprehensive pollution characterization index calculated in step 201 Compare each of the above preset intervals one by one, if If ∈[0.1,0.3), then it matches the lightly polluted label, encoded as 0; if If ∈[0.3,0.6), then it matches the moderate pollution label, encoded as 1; if If the value is ∈[0.6,1.0], then the heavily polluted label is matched, coded as 2. After matching, the corresponding pollution level label (mild, moderate, heavy) is output to the central control module, and the matching result is saved. A pre-stored control scheme knowledge base is constructed and stored in the storage unit of the central control module. The knowledge base contains three pollution level labels: mild (coded 0), moderate (coded 1), and heavy (coded 2). Each label uniquely corresponds to an initial purification and remediation control scheme. The core parameters of the control scheme are determined by calibration, specifically based on the complete historical treatment data of the target water area over the past 3 to 5 years. This includes treatment parameters such as aeration intensity, filter component opening rate, and flow regulating valve opening under three pollution levels, as well as core information such as treatment duration, water quality changes before and after treatment, and compliance status. Combined with the correlation analysis between water quality parameters and purification effect, that is, collecting the dissolved oxygen (ΔDO), ammonia nitrogen (ΔDO), and ammonia nitrogen (ΔDO) values ​​of the target water area under different pollution levels over the past 3 to 5 years, the values ​​of dissolved oxygen (ΔDO), ammonia nitrogen (ΔDO), and ammonia nitrogen (ΔDO) are collected. - The measured concentrations of five core parameters (ΔTP, ΔCOD, ΔTur) exceeding the standard were collected (exceeding standard concentration = measured value - water standard value specified in step 200). Simultaneously, the measured values ​​of three control parameters (Aeration intensity (A), Filter component opening rate (F), and Flow regulating valve opening degree (V)) were collected for the corresponding time period. The min-max normalization method was used to normalize the three control parameters, and the normalized values ​​of the three control parameters were taken in the range of [0,1].

[0029] The data included the water quality compliance rate and the rate of decrease in pollution concentration after treatment. Abnormal samples caused by equipment malfunctions or extreme weather were removed, and the selected samples were then... ≥1000 valid samples. Linear regression analysis was used to establish univariate linear regression models between the exceedance concentrations of the five core parameters and each control parameter. Taking ammonia nitrogen exceedance concentration and aeration intensity as an example, the model expression is: ,in It is the normalized setpoint for aeration intensity. For regression coefficients, The intercept is calculated using the least squares method. , ( For the first Histamine nitrogen concentration exceeded the standard. For the first Group aeration intensity, (These are the sample means of both); similarly, the remaining parameter combinations are calculated to establish 15 univariate linear regression models.

[0030] Through formula Calculate the Pearson correlation coefficient ( The core parameter exceeded the standard concentration. For control parameters, | (≥0.8 indicates a strong correlation). Strongly correlated models were selected and integrated into a quantitative correlation model, and regression coefficients were obtained through fitting. and correlation coefficient As a correlation coefficient, the matching rules are clearly quantified, for example... =0.5 (When ammonia nitrogen exceeds the standard by 0.2 mg / L, the aeration intensity increases by 0.1). =8 (the filter module opening rate increases by 0.08 when the total phosphorus exceeds the standard by 0.01 mg / L). Based on the topographic requirements of different areas of the target water body (shallow water depth ≤ 1.0 m, deep water depth > 2.0 m, flow velocity ≥ 0.5 m / s in the flow channel area, and flow velocity < 0.1 m / s in the stagnant water area), the control parameters output by the quantitative correlation model are iteratively calibrated. Simultaneously, combining the pollution level classification standards determined in step 200 and the comprehensive pollution characterization index range determined in step 201, the average values ​​of the control parameters that enable the water quality to reach the standard under the three pollution levels are calculated. These average values ​​are used as the benchmark values ​​for the corresponding pollution levels. For example, the benchmark value for aeration intensity corresponding to light pollution (code 0) is the average aeration intensity after topographic calibration among all valid samples within the light pollution range. The benchmark values ​​for filter module opening rate and flow regulating valve opening are calculated and determined using the same method. Ultimately, it is ensured that all three benchmark values ​​are within the [0,1] range and match the treatment requirements of the corresponding pollution level.

[0031] Based on the matched pollution level labels, the corresponding initial purification and remediation control schemes are retrieved from the control scheme knowledge base. The retrieval process uses keyword matching, with pollution level labels (mild, moderate, and severe) as keywords, to perform precise searches in the knowledge base, ensuring that the retrieved control schemes uniquely correspond to the pollution level labels. After the retrieval is completed, the aeration intensity benchmark value, filter component opening rate benchmark value, and flow regulating valve opening benchmark value are extracted from the scheme, and the three benchmark values ​​are transmitted to the instruction generation unit of the central control module.

[0032] This embodiment constructs a judgment model using a multi-layer neural network structure and trains it with historical pollution level labels. This captures the non-linear correlations between water quality parameters, avoiding subjective biases in manual judgments and the limitations of traditional linear judgment methods. Through feature iteration and aggregation by calculating the feature coupling strength coefficient, the correlation between various water quality parameters is fully considered, avoiding the one-sided influence of a single water quality parameter on the judgment result. This ensures that the comprehensive pollution characterization index can comprehensively reflect the overall deviation of the current water quality. By pre-setting a clear pollution level numerical range, each comprehensive pollution characterization index corresponds to a unique pollution level label. Furthermore, through precise retrieval from the control scheme knowledge base, the initial purification and remediation control scheme is highly adapted to the current pollution level, avoiding resource waste or insufficient treatment effects caused by extensive control methods.

[0033] In a preferred embodiment of the present invention, step 3 includes: Step 300: Read river and lake topographic features from the initial multidimensional state dataset, wherein the river and lake topographic features include topographic elevation point cloud data; perform spatial clustering based on the topographic elevation point cloud data to identify shallow water areas, deep water areas, flow channels, and stagnant water areas of the target water body, specifically including: First, river and lake topographic features are read from the initial multidimensional state dataset. These features include topographic elevation point cloud data, water boundary data, and water flow velocity data. The topographic elevation point cloud data is acquired in-situ using a 3D laser scanning device, covering the entire target water area, with a point cloud density controlled at 50 points / ... Up to 80 points / The point cloud data contains the three-dimensional spatial coordinates (X, Y, Z) of each sampling point, where the Z value is the terrain elevation of the corresponding sampling point. Spatial clustering is performed based on the terrain elevation point cloud data using the density-based spatial clustering (DBSCAN) algorithm. This algorithm uses the spatial Euclidean distance and elevation difference between point clouds as core clustering features, setting a neighborhood radius of 2m and a minimum of 15 sample points within the neighborhood. This filters out point cloud clusters that meet the density and elevation continuity criteria. Density compliance means that the point cloud contains at least 15 valid sampling points within the set 2m neighborhood, and elevation continuity means that the elevation difference between adjacent point clouds does not exceed 0.5m. This ensures that the point cloud clusters form a continuous and complete terrain outline without obvious elevation discontinuities, excluding isolated anomalous point clouds (such as abnormal elevation points caused by floating objects or equipment shadows). During the clustering process, the actual water depth corresponding to each point cloud is calculated as the difference between the terrain elevation and the water surface elevation.

[0034] By combining clustering results, calculated actual water depth values, and flow velocity data from the initial multidimensional state dataset, the shallow water area, deep water area, flow channel area, and stagnant water area of ​​the target water body were identified. The specific identification criteria strictly adhered to the terrain adaptation requirements determined in previous communication: shallow water area: water depth ≤ 1.0m, slope ≤ 5°; deep water area: water depth > 2.0m, slope ≤ 15°; flow channel area: water depth 1.0m < h ≤ 2.0m, flow velocity ≥ 0.5m / s; stagnant water area: water depth 1.0m < actual water depth ≤ 2.0m, flow velocity < 0.1m / s. After identification, the spatial boundary coordinates, area, average water depth, average slope, and average flow velocity of the four types of areas were output, forming a complete water body zoning result.

[0035] Step 301: Deploy corresponding distributed purification units in each area. Each distributed purification unit includes an aeration device, a filter assembly, and a flow regulating valve. Use the initial purification and remediation control scheme as a high-level control command vector. Analyze the high-level control command vector to extract the aeration intensity benchmark value, the filter assembly opening rate benchmark value, and the flow regulating valve opening benchmark value. Use these three benchmark values ​​as the base values ​​for the aeration intensity control parameter, the filter assembly opening rate control parameter, and the flow regulating valve opening control parameter, respectively. Specifically, this includes: Based on the water area zoning results obtained in step 300, corresponding distributed purification units are deployed in the shallow water area, deep water area, flow channel area, and stagnant water area. Each distributed purification unit is supported by a float, the size of which is adapted to the water depth characteristics of each area. The draft of the float in the shallow water area is 0.6m to 0.8m, and the draft of the float in the deep water area, flow channel area, and stagnant water area is 0.8m to 1.2m. The spacing between adjacent distributed purification units is controlled at 10m to 20m to ensure that the purification range fully covers the target water area. Each distributed purification unit completely includes an aeration device, a filter component, and a flow regulating valve. The aeration device adopts a microporous aeration method, the filter component adopts quartz sand filter material, and the flow regulating valve adopts an electric regulating method. All three are connected to the central control module to receive control commands and provide feedback on the operating status.

[0036] The initial purification and remediation control scheme retrieved from the control scheme knowledge base constructed in step 202 is used as the high-level control command vector. This high-level control command vector is a three-dimensional normalized vector, including the aeration intensity benchmark value, the filter component opening rate benchmark value, and the flow regulating valve opening benchmark value. The values ​​of the three are all controlled within the range of [0,1]. For example, in the initial control scheme corresponding to light pollution (code 0), the aeration intensity benchmark value can be 0.3, the filter component opening rate benchmark value can be 0.4, and the flow regulating valve opening benchmark value can be 0.5; in the initial control scheme corresponding to moderate pollution (code 1), the three can be 0.5, 0.6, and 0.7 respectively; in the initial control scheme corresponding to heavy pollution (code 2), the three can be 0.8, 0.9, and 0.8 respectively. The instruction parsing unit of the central control module parses the high-level control instruction vector and extracts the above-mentioned aeration intensity benchmark value, filter component opening rate benchmark value, and flow regulating valve opening benchmark value through instruction decoding. These three benchmark values ​​are directly used as the base values ​​of the aeration intensity control parameter, filter component opening rate control parameter, and flow regulating valve opening control parameter, respectively.

[0037] Step 302: Based on the terrain features and pollution level labels of the areas where each distributed purification unit is located, the base values ​​of the three control parameters are differentially corrected to obtain the corrected control parameters. The differential correction of the base values ​​of the three control parameters includes: setting the aeration intensity control parameter for the shallow water area to 1.5 times the aeration intensity benchmark value and the filter assembly opening rate control parameter to 0.5 times the filter assembly opening rate benchmark value; setting the aeration intensity control parameter for the deep water area to 0.5 times the aeration intensity benchmark value and the flow regulating valve opening control parameter to 1.5 times the flow regulating valve opening benchmark value; setting the aeration intensity control parameter, filter assembly opening rate control parameter, and flow regulating valve opening control parameter for the flow channel area to their corresponding benchmark values; and setting the aeration intensity control parameter, filter assembly opening rate control parameter, and flow regulating valve opening control parameter for the stagnant water area to 1.2 times their corresponding benchmark values. Specifically, this includes: The central control module calls the water area zoning results from step 300 and the pollution level labels obtained from step 201. Combining the topographic features (average water depth, average flow velocity, and slope) of the area where each distributed purification unit is located with the pollution level labels of the corresponding area, it performs differential correction on the basic values ​​of the three control parameters extracted in step 301. All correction calculations are based on normalized parameters to ensure that the corrected parameters are still in the range of [0,1].

[0038] For distributed purification units deployed in shallow water areas (water depth ≤ 1.0m, slope ≤ 5°), the aeration intensity control parameter is corrected to 1.5 times the baseline aeration intensity value; the filter component opening rate control parameter is corrected to 0.5 times the baseline filter component opening rate value; the flow regulating valve opening control parameter remains unchanged. If the corrected aeration intensity control parameter exceeds the [0,1] range, the boundary value of the range is automatically taken (1 if it exceeds 1, 0 if it is below 0). This enhances water oxygenation and accelerates pollutant degradation while avoiding sediment disturbance caused by shallow water, preventing secondary diffusion of pollutants in the sediment. This aligns with the fragile and easily disturbed aquatic ecosystem in shallow water areas, ensuring purification effectiveness while avoiding secondary pollution. For distributed purification units deployed in deep water areas (water depth > 2.0m, slope ≤ 15°), the aeration intensity control parameter is corrected to 0.5 times the baseline aeration intensity value; the flow regulating valve opening rate control parameter is corrected to 1.5 times the baseline flow regulating valve opening rate value; the filter component opening rate control parameter remains unchanged. Similarly, if the corrected parameters exceed the [0,1] range, the corresponding boundary value will be automatically taken. This correction is mainly adapted to the characteristics of deep water bodies being thick and oxygen permeability being difficult, reducing the energy waste caused by ineffective aeration. At the same time, by increasing the flow rate regulation, it promotes water circulation, ensures that the bottom water body is fully purified, and avoids the aggravation of pollution caused by the lack of oxygen in the deep water body.

[0039] For distributed purification units deployed in flowing channel areas (1.0m < actual water depth ≤ 2.0m, flow velocity ≥ 0.5m / s), all three control parameters remain unchanged at their baseline values ​​without correction, aligning with the strong self-purification capacity of the water in these areas. Because the water in flowing channel areas has high fluidity and strong self-purification capacity, maintaining the baseline parameters is sufficient to meet purification needs, while avoiding energy waste caused by excessive regulation, thus balancing purification effectiveness and economy. For distributed purification units deployed in stagnant water areas (1.0m < actual water depth ≤ 2.0m, flow velocity < 0.1m / s), all three control parameters are corrected to 1.2 times their corresponding baseline values. If the corrected parameters exceed the [0,1] range, the corresponding boundary value is automatically applied. This correction primarily addresses the poor water flow and easy accumulation of pollutants in stagnant water areas. By increasing the intensity of various control parameters, aeration, filtration, and water circulation are enhanced to ensure timely degradation of pollutants and prevent pollution spread. After the correction is completed, the corrected control parameters for each distributed purification unit are obtained. Each parameter is a normalized value that is adapted to the terrain features and pollution level of the area.

[0040] Step 303: The modified control parameters are encapsulated into executable instructions for each distributed purification unit, driving each distributed purification unit to perform differentiated aeration, filtration, and flow regulation operations; according to the spatial location and execution priority of each distributed purification unit, all executable instructions are serialized and arranged to form a hierarchical dynamic purification execution instruction set, specifically including: The instruction generation unit of the central control module encapsulates the corrected control parameters obtained in step 302 according to the unique number of each distributed purification unit to form executable instructions for each distributed purification unit. Each executable instruction includes the unique unit number, unit spatial coordinates, aeration intensity setting value (corrected), filter component opening rate setting value (corrected), flow regulating valve opening setting value (corrected), execution priority, and execution duration. The instruction format strictly adapts to the communication protocol between the central control module and the distributed purification unit to ensure stable instruction transmission, error-free parsing, and transmission delay controlled within 500ms.

[0041] After encapsulation, all executable instructions are serialized and arranged according to the spatial location and execution priority of each distributed purification unit. The execution priority follows the principle of prioritizing pollution level and supplementing by regional characteristics. Specifically, the order is: distributed purification units in heavily polluted areas > distributed purification units in moderately polluted areas > distributed purification units in lightly polluted areas; under the same pollution level, the order is: stagnant water area > deep water area > shallow water area > flowing channel area; the spatial location is arranged in the order of upstream to downstream of the target water area and central area to edge area to avoid mutual interference between purification units in different areas and ensure maximum purification effect.

[0042] After serialization and arrangement, a hierarchical dynamic purification execution instruction set is formed. This instruction set includes all executable instructions, instruction execution order, instruction execution interval, instruction checksum, and exception handling plan. The checksum is used to ensure that no data is lost or tampered with during instruction transmission, and the exception handling plan is used to deal with emergencies such as instruction parsing failure and purification unit failure. After the instruction set is generated, it is issued by the central control module to each distributed purification unit, driving each distributed purification unit to perform differentiated aeration, filtration, and flow regulation operations according to its own executable instructions, thereby achieving precise zoning purification of the target water area.

[0043] This embodiment, based on elevation point cloud and density clustering, automatically divides shallow water areas, deep water areas, flow channel areas, and stagnant water areas. The high accuracy and complete coverage of these zoning provide a reliable spatial basis for the deployment of distributed purification units, avoiding blind deployment. Purification units are matched to water area zones to ensure that areas with different depths and flow velocities have corresponding purification capabilities, resulting in uniform purification coverage without blind spots and improving overall purification efficiency. Aeration, filtration, and flow parameters are quantitatively corrected based on water depth and flow velocity characteristics to avoid excessive disturbance in shallow water areas, oxygen deficiency at the bottom of deep water areas, and insufficient purification in stagnant water areas, ensuring precise matching of purification intensity to the local water environment. Through high-level command parsing, parameter correction, command encapsulation, and serialization, a standardized and directly executable command set is formed to drive precise operation of the purification units, achieving a high degree of automation and minimizing manual intervention. This avoids uniform high-intensity aeration and filtration across the entire area; aeration is reduced in deep water areas, while standard intensity is maintained in flow channel areas. Simultaneously, purification priority is increased in heavily polluted areas and stagnant water areas, shortening the treatment cycle and reducing operating costs.

[0044] In a preferred embodiment of the present invention, step 4 includes: Step 400: The execution states corresponding to the executable instructions of each distributed purification unit in the hierarchical dynamic purification execution instruction set are used as the dynamic constraint boundaries for ecological restoration; river and lake topographic features and water quality parameters are read from the initial multidimensional state dataset. The river and lake topographic features include water depth distribution data and sediment type classification data. The water quality parameters include ammonia nitrogen concentration, total phosphorus concentration, dissolved oxygen concentration, COD concentration, and turbidity, specifically including: The execution status corresponding to the executable instructions of each distributed purification unit in the hierarchical dynamic purification execution instruction set generated in step 303 is clearly defined as the dynamic constraint boundary for ecological restoration. This constraint boundary covers the operating parameter range of each purification unit (the set value of aeration intensity, filter component opening rate, and flow regulating valve opening), the scope of the work area, and the execution priority, ensuring that ecological restoration operations (planting aquatic plants, dispensing microbial agents, and setting up ecological reefs) do not conflict with the purification operations of the distributed purification units, while adapting to the dynamic adjustment of purification operations to avoid mutual interference.

[0045] River and lake topographic features and water quality parameters were read from the initial multidimensional state dataset. The river and lake topographic features specifically included water depth distribution data and sediment type classification data. The water quality parameters specifically included ammonia nitrogen concentration, total phosphorus concentration, dissolved oxygen concentration, COD concentration, and turbidity. Water depth distribution data was jointly collected by a 3D laser scanning device and a water level monitoring device, corresponding to the actual water depth at each spatial sampling point in the target water area. Sediment type classification data was obtained through underwater sampling and was divided into four categories based on sediment properties: clay, sandy clay, sand, and gravel, corresponding to codes 1, 2, 3, and 4, respectively. These codes serve only as classification identifiers. All water quality parameters were collected in-situ by distributed water quality monitoring nodes at a sampling frequency of once per hour. The collected data, after outlier removal (excluding data caused by equipment malfunctions or extreme weather), served as the basis for subsequent calculations.

[0046] Step 401: Spatial grid coupling is performed on the water depth distribution data and sediment type classification data in the river and lake topography features with the ammonia nitrogen concentration, total phosphorus concentration, dissolved oxygen concentration, COD concentration, and turbidity in the water quality parameters. The target water area is divided into multiple grid cells according to a fixed side length. For each grid cell, the average water depth value, sediment type code value, average ammonia nitrogen concentration, average total phosphorus concentration, average dissolved oxygen concentration, average COD concentration, and average turbidity value within the corresponding grid cell are obtained. The main flow path segment set is extracted from the flow direction vector field in the flow velocity vector field data. The main flow path segment set is formed by connecting the points with the maximum flow velocity in the flow direction vector field in sequence. For each grid cell, the geometric center point of the corresponding grid cell is taken, and the shortest distance from the geometric center point to each line segment in the main flow path segment set is calculated. The minimum value among all shortest distances is taken as the vertical distance from the corresponding grid cell to the main flow path. The normalized distance factor is calculated by ratio of the vertical distance from the corresponding grid cell to the main flow path to half the maximum width of the target water area. Based on the average water depth, sediment type code value, average ammonia nitrogen concentration, average total phosphorus concentration, average dissolved oxygen concentration, average COD concentration, average turbidity, and normalized distance factor, the remediation suitability coefficient of each grid cell is calculated using a weighted summation method, specifically including: The water depth distribution data and sediment type classification data from river and lake topography are spatially coupled with water quality parameters such as ammonia nitrogen concentration, total phosphorus concentration, dissolved oxygen concentration, COD concentration, and turbidity. A square grid is used, dividing the target water area into multiple grid units with fixed side lengths of 5m × 5m. Each grid unit corresponds to a unique spatial coordinate range, ensuring no overlap or blind spots. After grid division, for each grid unit, the average and identifier values ​​of various basic parameters within that grid unit are extracted and calculated. Specifically, the arithmetic mean of water depth data from all sampling points within the grid unit is calculated as the average water depth value for that grid unit. The sediment type with the highest proportion within the grid unit is extracted, and its corresponding code value is used as the sediment type code value for that grid unit. Similarly, the arithmetic mean of ammonia nitrogen concentration, total phosphorus concentration, dissolved oxygen concentration, COD concentration, and turbidity data from all water quality monitoring points within the grid unit is calculated and used as the average ammonia nitrogen concentration, total phosphorus concentration, dissolved oxygen concentration, COD concentration, and turbidity concentration for that grid unit.

[0047] From the velocity vector field data in the initial multidimensional state dataset, the flow direction vector field is extracted. That is, for each spatial sampling point in the velocity vector field data, the corresponding flow direction component (i.e., the direction parameter of the velocity vector) is extracted. This direction parameter is obtained by calculating the azimuth angle of the velocity vector in the two-dimensional plane. Let the components of the velocity vector in the x-axis and y-axis directions of each spatial sampling point be u (x-axis component) and v (y-axis component), respectively. Taking the positive x-axis direction as the reference (corresponding to due east), the azimuth angle is calculated using the arctangent function. The calculated azimuth angle is the direction parameter of the velocity vector at that sampling point, with a value range of [0, 2π), representing the actual flow direction of the water. Data with directional anomalies (sampling points with directional deviations exceeding ±10°) caused by equipment errors and sampling interference are removed. The water flow direction components of the remaining valid sampling points are then organized in spatial coordinate order to form a continuous and complete water flow direction vector field. This allows for the acquisition of the main water flow path segment set. The extraction method for the main water flow path segment set involves filtering all points with maximum flow velocity in the water flow direction vector field and connecting these points sequentially according to the water flow direction to form a continuous segment set. This segment set covers all major water flow channels in the target water area. For each grid cell, its geometric center point is first determined. The coordinates of the geometric center point are obtained by calculating the arithmetic mean of the coordinates of the four vertices of the grid cell. The shortest distance from this geometric center point to each segment in the main water flow path segment set is calculated using the formula for the shortest distance from a point to a segment.

[0048] After calculating the shortest distance from the geometric center point to each line segment, the minimum of all shortest distances is taken as the vertical distance from the grid cell to the main flow path. This vertical distance is then compared to half the maximum width of the target water area to obtain the normalized distance factor, which ranges from [0,1]. Combining the average water depth, sediment type code, average ammonia nitrogen concentration, average total phosphorus concentration, average dissolved oxygen concentration, average COD concentration, average turbidity, and normalized distance factor for each grid cell, a weighted summation method is used to calculate the remediation suitability coefficient for that grid cell. Before calculation, all parameters must be normalized to ensure that all parameter values ​​are within the [0,1] interval. When normalizing the average water depth, subtract the minimum water depth of the target water area from the average water depth of the grid cell, and then divide by the difference between the maximum and minimum water depths of the target water area. When normalizing the concentrations of ammonia nitrogen, total phosphorus, COD, and turbidity, subtract the average concentration of the grid cell from the historical maximum concentration of the corresponding parameter, and then divide by the difference between the historical maximum and minimum concentrations of the parameter. When normalizing the dissolved oxygen concentration, subtract the historical minimum dissolved oxygen concentration from the average dissolved oxygen concentration of the grid cell, and then divide by the difference between the historical maximum and minimum dissolved oxygen concentrations. The substrate type is coded according to 1 to 4 (clay = 1, sandy clay = 2, sandy = 3, gravel = 4). When normalizing, subtract 1 from the substrate code of the grid cell, and then divide by 3 (because the code range is 1 to 4, the difference is 3).

[0049] Set the weights for each normalized parameter, where the normalized value of the average water depth is... Weight 0.15, normalized value of substrate type Weight 0.10, normalized ammonia nitrogen concentration Weight 0.20, normalized total phosphorus concentration Weight 0.20, normalized dissolved oxygen value Weight 0.15, normalized COD concentration Weight 0.10, turbidity normalized value Weight 0.05, normalized distance factor Weight 0.05, sum of all weights is 1. Repair Suitability Coefficient , The value ranges from [0,1]. The larger the value, the more suitable the grid cell is for ecological restoration operations.

[0050] Step 402: Based on the ecological restoration suitability coefficient of each grid unit, dynamically calculate the planting depth weight of aquatic plants in the proposed planting area. The planting depth weight represents the weight value of the burial depth of plant roots in the bottom sediment, specifically including: The definition of planting depth weighting is clarified. Planting depth weighting represents the weight value of the burial depth of plant roots in the sediment, with a value range of [0.3, 0.8]. A larger weight value indicates that the plant roots need to be buried deeper in the sediment, which is suitable for areas with stronger sediment stability and deeper water; a smaller weight value indicates that the plant roots need to be buried shallower, which is suitable for areas with loose sediment and shallower water. The remediation suitability coefficient of each grid cell calculated in step 401 is obtained. Combined with the normalized value of the average water depth of the grid cell Normalized values ​​of substrate type Dynamic calculations of planting depth weight distribution are performed, and the calculation formula is as follows: ; in The weight is calculated based on the planting depth, with 0.3 being the minimum weight for planting depth (suitable for shallow water areas and loose substrate), and 0.5 being the weight adjustment coefficient.

[0051] During the calculation process, it is necessary to ensure the weight distribution at the planting depth. Within the range of [0.3, 0.8], if the calculation result exceeds this range, the corresponding boundary value is automatically taken (0.8 if it exceeds 0.8, and 0.3 if it is below 0.3). This calculation logic conforms to the growth pattern of aquatic plants. In areas with high suitability, deeper water, and more stable bottom sediment, the weight of plant root burial depth is increased to improve plant survival rate. Conversely, the weight of burial depth is reduced to adapt to the growth environment of shallow water and loose bottom sediment.

[0052] Step 403: Obtain the remediation suitability coefficient for each grid cell, as well as the corresponding ammonia nitrogen concentration, total phosphorus concentration, and COD concentration; multiply the ammonia nitrogen concentration, total phosphorus concentration, and COD concentration, and then take the square root to obtain the comprehensive pollution index for the corresponding grid cell; multiply the comprehensive pollution index by the remediation suitability coefficient for the corresponding grid cell to obtain the basic dosage ratio of the microbial agent; obtain the average water depth value from the water depth distribution data of the corresponding grid cell, and obtain the quantitative dosage ratio of the microbial agent based on the basic dosage ratio, the corresponding average water depth value, and the arithmetic mean of the average water depth values ​​of all grid cells in the target water area, specifically including: Obtain the remediation suitability coefficient for each grid cell calculated in step 401, as well as the average ammonia nitrogen concentration, average total phosphorus concentration, and average COD concentration for that grid cell (all valid data read and filtered in step 400). Calculate the comprehensive pollution index for that grid cell by multiplying the ammonia nitrogen concentration, total phosphorus concentration, and COD concentration and then taking the square root. The higher the comprehensive pollution index, the more severe the pollution level of the grid cell, and the more microbial agents are required. Calculate the basic application ratio of the microbial agents. The comprehensive pollution index Repair suitability coefficient of corresponding grid cell Perform multiplication and normalize the result to ensure the base deployment ratio is within the [0,1] interval. The specific calculation formula is as follows: ,in To find the maximum value of the comprehensive pollution index across all grid cells in the target water area, the method is to iterate through all grid cells. The values ​​were selected, and this normalization process can avoid deviations in the dosage ratio caused by differences in the dimensions of pollution concentration.

[0053] Calculate the quantitative dosage ratio of microbial agents Obtain the average water depth value from the water depth distribution data of the corresponding grid cell. Simultaneously, the arithmetic mean of the average water depth values ​​of all grid cells in the target water area is calculated. The formula for calculating the quantitative distribution ratio is as follows: The deeper the grid cell, the larger the water volume per unit area. Microbial agents of the same concentration diffuse and dilute faster in deep water. If the dosage ratio is the same as in shallow water, the agent concentration will be too low to achieve the desired pollutant degradation. Conversely, in shallow water, the smaller water volume results in slower agent diffusion and dilution; a higher dosage ratio would waste the agent and potentially disrupt the aquatic microbial balance. By dynamically adjusting the agent dosage ratio based on the water volume differences within the grid cells, the system ensures that the agent concentration in deep water meets the pollutant degradation requirements while avoiding waste in shallow water. Furthermore, by combining the comprehensive pollution index (pollution level) and the remediation suitability coefficient (ecological compatibility), the system ensures that the agent dosage is precisely matched to the actual pollution status and ecological environment of each grid cell, achieving efficient pollutant degradation and assisting in purification operations. After calculation, the quantitative dosage ratio is ensured to be within the [0,1] range. If it exceeds this range, the corresponding boundary value is automatically used as the final dosage basis for that grid cell.

[0054] Step 404: Based on the water flow direction vector field, velocity scalar distribution data, and existing obstacle spatial coordinates in the velocity vector field data, dynamically calculate the layout coordinates of the ecological reefs. This ensures the ecological reefs are located at points where the water flow direction changes or in areas where the velocity value in the velocity scalar distribution data is below a preset velocity threshold. The layout coordinates of each ecological reef include its two-dimensional planar position and bottom elevation. Combine the calculated planting depth weighting, microbial agent quantitative application ratio, and ecological reef layout coordinates into a data set. Define this data set as the ecological restoration adaptation parameter set, specifically including: First, flow velocity vector field data is read from the initial multidimensional state dataset. This data includes the previously extracted water flow direction vector field and complete flow velocity scalar distribution data. At the same time, the spatial coordinates of existing obstacles in the target water area are read. Existing obstacles include bridge piers, shore structures, underwater pipelines, shoals and reefs, etc. Their spatial coordinates are obtained by combining underwater sonar surveys and on-site verification. The two-dimensional plane position (x, y) and bottom elevation of each obstacle are marked to form an obstacle coordinate list. Based on the above data, the layout coordinates of the ecological reef are dynamically calculated. The layout coordinates must simultaneously meet three requirements: adapting to water circulation, ensuring reef stability, and avoiding obstacles. Specifically, the conditions are: first, it must be located at a point where the water flow direction changes; second, it must be located in an area where the flow velocity value in the flow velocity scalar distribution data is lower than a preset flow velocity threshold. The preset flow velocity threshold is set at 0.2 m / s. If the flow velocity is too high, the reef will be subjected to excessive water flow impact, making it prone to displacement and overturning, and unable to play a stable role in the long term; if the flow velocity is too low, it will not be able to disturb the water through the reef, making it difficult to achieve the effects of strengthening water circulation and assisting in pollutant degradation. Therefore, it is sufficient to meet either of the two conditions. At the same time, it is necessary to avoid the area corresponding to the spatial coordinates of existing obstacles and the surrounding 1m range to further ensure the safety and stability of the reef layout.

[0055] The identification of points where the water flow direction changes is determined by traversing all valid sampling points in the water flow direction vector field and calculating the angle (azimuth difference) between adjacent sampling points. An angle greater than 30° is considered a point of change. In such areas, the water flow disturbance is moderate, which can enhance water mixing and assist in the diffusion of pollutants. Low-velocity areas are identified by traversing all sampling points in the velocity scalar distribution data and selecting all sampling points with velocities ≤ 0.2 m / s. These areas have poor water flow and are prone to pollutant accumulation. Deploying ecological reefs in these areas can disturb the water body, promote water circulation, and solve the problem of pollutant accumulation. After identification, suitable sampling points for ecological reef deployment are selected from these two types of areas according to the principle of uniform distribution. The distance between adjacent reefs is controlled between 15 and 20 m to ensure that the reef's effective range fully covers the key restoration area of ​​the target water body. This distance is determined in conjunction with the effective radius of the reef (assuming 9 m) to ensure effective connection and no overlap between the effective ranges of adjacent reefs. The final selected sampling points are the coordinates for the deployment of the ecological reefs.

[0056] Each ecological reef's layout coordinates include a two-dimensional planar position and a bottom elevation. The two-dimensional planar position refers to the coordinates (x, y) of the selected sampling points, corresponding to the specific location of the target water area. The bottom elevation is based on the topographic elevation of the sampling point, minus 0.5m. All elevation data are calculated in meters. The purpose of this calculation logic is to ensure that the bottom of the ecological reef is embedded in the bottom mud by 0.5m. This not only improves the stability of the reef and prevents it from being washed away by water flow, but also avoids the bottom of the reef being quickly buried by sediment, ensuring the attachment of microorganisms on the reef surface and water circulation, thus fully leveraging the reef's auxiliary purification function. After the ecological reef layout coordinates are calculated, the planting depth weight of each grid unit obtained from the dynamic calculation in step 402, the quantitative application ratio of microbial agents calculated in step 403, and the ecological reef layout coordinates obtained in this step are integrated into a complete data set according to the grid unit correspondence. This data set is defined as the ecological restoration adaptation parameter set. Each set of ecological restoration adaptation parameters uniquely corresponds to a grid cell in the target water area. For grid cells without reef deployment, the reef deployment coordinates are marked as "none," while all other parameters are fully retained. This parameter set can be used to guide on-site ecological restoration operations, clarifying the aquatic plant planting depth requirements, microbial agent dosage, and ecological reef deployment locations for each grid cell, thus achieving coordinated adaptation between ecological restoration and distributed purification unit purification operations.

[0057] This embodiment uses the execution status of distributed purification units as dynamic constraint boundaries to avoid interference between ecological restoration and purification operations. Based on spatial grid coupling and quantitative calculation, the topographic and water quality characteristics of each grid unit are obtained. The calculated restoration suitability coefficient provides precise quantitative basis for subsequent aquatic plant planting, microbial agent application, and reef deployment, avoiding blind restoration and improving the targeting and effectiveness of ecological restoration. The planting depth and weight distribution, as well as the proportion of microbial agents, are determined based on specific parameter calculations, conforming to the actual environmental characteristics of each grid unit, ensuring the survival rate of aquatic plants and the degradation efficiency of microbial agents, while avoiding resource waste. The ecological reef deployment coordinates are combined with dynamic calculations of flow velocity characteristics and obstacle distribution, ensuring the stability of the reef deployment, strengthening water circulation, assisting pollutant degradation, and working synergistically with aquatic plants and microbial agents to form a closed-loop governance model of purification and restoration, improving the long-term effectiveness of aquatic ecological restoration. It achieves differentiated and precise control of ecological restoration, with different grid units adapted to different restoration parameters, avoiding the inefficiency caused by uniform restoration across the entire area.

[0058] In a preferred embodiment of the present invention, step 5 includes: Step 500: Collect time-series data of the treated water area again according to a fixed sampling period. The time-series data of the treated water area includes water level time-series signals, flow velocity vector field data, water quality parameter solution concentration values, and river and lake topographic elevation point cloud data. Perform discrete analysis and filtering normalization fusion processing on the collected time-series data of the treated water area to obtain a feedback multidimensional state dataset, specifically including: Time-series data of the treated water area were collected according to a fixed 24-hour sampling cycle, which balances the timeliness of dynamic changes in the water environment with monitoring energy consumption control. The collection scope, parameter types, and initial multidimensional state datasets were kept consistent to ensure data comparability. Collected parameters included water level time-series signals, flow velocity vector field data, water quality parameter solution concentration values, and river and lake topographic elevation point cloud data. The water level time-series signals were collected by water level monitoring sensors at a sampling frequency of once every 10 minutes, recording water level values ​​(unit: meters). The flow velocity vector field data were collected by underwater flow velocity monitoring equipment at a sampling point density of 50 points / m². 2 Up to 80 points / m 2 The methods for extracting the water flow direction vector, identifying points of change in water flow direction, and identifying low-velocity areas are consistent with step 404. Water quality parameters, including dissolved oxygen, ammonia nitrogen, total phosphorus, COD, and turbidity, are collected by distributed water quality monitoring nodes at a sampling frequency of once per hour (unit: mg / L). River and lake topographic elevation point cloud data are collected by a 3D laser scanning device, covering the entire target water area. All collected data are filtered to remove abnormal data caused by equipment failure or extreme weather. After collection, the time-series data of the treated water area are processed according to the discreteness analysis, filtering, and normalization fusion processing flow described in step 101 to obtain a feedback multidimensional state dataset.

[0059] Step 501: Extract water quality parameter dimension data from the feedback multidimensional state dataset, compare it feature-by-feature with the water quality parameter dimension data in the initial multidimensional state dataset, calculate the real-time deviation value of each water quality parameter feature, arrange the real-time deviation values ​​of all water quality parameter features in a preset order, and construct a one-dimensional residual feedback vector, specifically including: Water quality parameter dimensional data were extracted from the feedback multidimensional state dataset, including normalized values ​​of five core parameters: dissolved oxygen, ammonia nitrogen, total phosphorus, COD, and turbidity. These five parameters in the feedback dataset were compared feature-by-feature with their corresponding values ​​at the same spatial coordinates and time (or sampling period) in the initial multidimensional state dataset. For each water quality parameter feature, a real-time deviation value (feedback value - initial value) was calculated. The real-time deviation value ranged from -1 to 1, with positive values ​​indicating that the parameter concentration was higher than the initial value (treatment not meeting expectations or pollution worsening), negative values ​​indicating that the concentration was lower than the initial value (treatment better than expected), and zero values ​​indicating consistency with the initial state. Based on the priority of river and lake water pollution treatment (dissolved oxygen reflects the water body's self-purification capacity, ammonia nitrogen and total phosphorus are the core causes of eutrophication, COD characterizes the degree of organic pollution, and turbidity affects water transparency and ecological restoration effectiveness), the five real-time deviation values ​​were arranged in the order of dissolved oxygen, ammonia nitrogen, total phosphorus, COD, and turbidity to construct a one-dimensional residual feedback vector. This vector quantifies the differences through the specific numerical values ​​of each component (normalized to [-1, 1]). The larger the absolute value of the component, the more significant the difference in the corresponding parameter before and after treatment. Positive components indicate increased pollution, while negative components indicate reduced pollution. Overall, it reflects the overall deviation of water quality and the specific differences of each pollutant.

[0060] Step 502: The one-dimensional residual feedback vector is input back into the pollution level determination model to iteratively correct the connection weights within the model and the boundary thresholds of multiple pollution level numerical ranges. After each correction, the pollution level determination model outputs a new comprehensive pollution characterization index and a new pollution level label, specifically including: First, the one-dimensional residual feedback vector constructed in step 501 is used for two purposes. One is to backpropagate it to the pollution level determination model to correct the connection weights between neurons in each layer of the model (including the weights from the input layer to the first hidden layer, the weights from the first hidden layer to the second hidden layer, and the weights from the second hidden layer to the output layer), thereby indirectly changing the feature coupling strength coefficient (the feature coupling strength coefficient is dynamically calculated by the output of the second hidden layer of the model and is not used as a separate correction variable). The other is to correct the boundary threshold of the pollution level numerical range (this threshold is not included in the neural network model, but is used as a post-processing mapping rule from the comprehensive pollution characterization index to the pollution level label).

[0061] The iterative correction process uses the gradient descent method, with the iteration step size set to . =0.01, the iteration termination condition is the change in the model connection weights. and the change in boundary threshold All are less than 0.001. Define the residual loss function: ;in It is the first of the one-dimensional residual feedback vectors. One portion, , These are the parameters before the correction. , These are the corrected parameters.

[0062] Calculate the loss function separately model connection weights partial derivatives and boundary thresholds partial derivatives The boundary thresholds include the endpoint values ​​of the light pollution range [0.1, 0.3], the moderate pollution range [0.3, 0.6], and the heavy pollution range [0.6, 1.0]. =1, 2, 3 correspond to the lower or upper boundaries of the three intervals, respectively.

[0063] The model connection weights are updated as follows: The boundary threshold is updated as follows: After each update, the new model connection weights are substituted into the pollution level determination model, and forward propagation is performed again to calculate the new feature coupling strength coefficients. Simultaneously, the new comprehensive pollution characterization index is matched with the updated boundary thresholds to obtain new pollution level labels (mild, moderate, severe). The connection weights, boundary thresholds, comprehensive pollution characterization index, and pollution level labels are recorded after each iteration. This iterative correction process is repeated until the changes in model connection weights and boundary thresholds are both less than 0.001, ensuring model convergence. After convergence, the pollution level determination model outputs pollution level labels that accurately reflect the actual pollution status of the water area currently being treated.

[0064] Step 503: Update the aeration intensity benchmark value, filter component opening rate benchmark value, and flow regulating valve opening benchmark value in the initial purification and remediation control scheme according to the new pollution level label, and re-obtain the updated hierarchical dynamic purification execution instruction set based on the updated benchmark values; update the ecological restoration adaptation parameter set based on the updated hierarchical dynamic purification execution instruction set and the river and lake topographic features and water quality parameters in the feedback multidimensional state dataset; repeat the update process, comparing the newly constructed one-dimensional residual feedback vector with the preset compliance interval after each iteration, until the absolute value of each component of the one-dimensional residual feedback vector is less than the threshold corresponding to the preset compliance interval; repackage the aeration intensity benchmark value, filter component opening rate benchmark value, and flow regulating valve opening benchmark value in the finally converged updated initial purification and remediation control scheme into an executable instruction sequence, and define the executable instruction sequence as the final dynamic control instruction sequence, specifically including: Based on the pollution level label revised in step 502, update the three benchmark values ​​of aeration intensity, filter component opening rate, and flow regulating valve opening in the initial purification and remediation control scheme of step 202. The update logic is the same as in step 202: first, retrieve the corresponding benchmark values ​​from the control scheme knowledge base; then, combining the components of the one-dimensional residual feedback vector, calculate the comprehensive deviation after directional correction. = ,in , , , , These are the real-time deviation values ​​(feedback value - initial value) for dissolved oxygen, ammonia nitrogen, total phosphorus, COD, and turbidity, respectively. According to the formula... Fine-tune the baseline value and limit it. Within the [0,1] interval. When the actual pollution worsens compared to the initial state ( Negative, (positive) When the value is positive, the benchmark value is increased to strengthen regulation; when pollution decreases, If the value is negative, the baseline value is reduced to avoid overtreatment. Following steps 301 to 303, the instructions are re-parsed, parameters are corrected, and serialized instructions are encapsulated to generate an updated hierarchical dynamic purification execution instruction set adapted to the current pollution status. Simultaneously, combining the updated hierarchical dynamic purification execution instruction set with the terrain features and water quality parameters (dissolved oxygen, ammonia nitrogen, total phosphorus, COD, turbidity) in the feedback multidimensional state dataset, the ecological restoration adaptation parameter set is updated according to steps 400 to 404. This includes updating the restoration suitability coefficient and planting depth weighting according to the new water quality parameters, updating the quantitative dosing ratio of microbial agents according to the new comprehensive pollution index (based on ammonia nitrogen, total phosphorus, and COD), and recalculating the ecological reef layout coordinates according to the new flow velocity vector field data.

[0065] The process involves repeated model correction (step 502), baseline value update, hierarchical dynamic purification execution instruction set update, and ecological restoration adaptation parameter set update. After each iteration, the newly constructed one-dimensional residual feedback vector is compared with the preset compliance interval [-0.05, 0.05] until the absolute value of all components is less than 0.05 (the governance compliance threshold). After stopping the iteration, the three finally converged baseline values ​​(aeration intensity, filter component opening rate, and flow regulating valve opening) are encapsulated into an executable instruction sequence (format consistent with step 303, including unit number, parameter setting value, and execution priority). This sequence is defined as the final dynamic control instruction sequence, used to guide purification and ecological restoration operations in the long term, ensuring long-term stable compliance of the aquatic ecosystem.

[0066] This embodiment accurately captures dynamic changes in water quality, topography, and flow velocity by collecting and processing time-series data at fixed intervals. It constructs residual feedback vectors to quantify governance deviations, avoiding rigid control schemes and ensuring adaptation to the dynamic changes in the river and lake water environment. Through iterative correction of the pollution level determination model, the internal coupling strength coefficient and boundary thresholds are optimized to address the disconnect between the model and the actual governance state, ensuring accurate pollution level determination. Synchronous updates of purification control and ecological restoration parameters are achieved, ensuring that parameters such as aeration intensity, microbial agent dosage, and reef deployment are always precisely matched with the current pollution level and water environment, avoiding blind control. The final generated dynamic control command sequence can be directly used for on-site operation guidance without excessive manual intervention, achieving a high degree of automation. It also considers the synergy between purification and restoration, reducing operating costs and achieving long-term stable compliance of the aquatic ecological environment.

[0067] like Figure 2 As shown, embodiments of the present invention also provide a dynamic purification and control system for the ecological pollution levels of rivers and lakes, including: The acquisition module is used to acquire multi-source monitoring data on water level, flow velocity, water quality parameters and river and lake topography of the target water area. After discrete analysis and filtering normalization fusion processing, an initial multidimensional state dataset is obtained. The analysis module is used to input water quality parameters into a pre-built pollution level determination model for analysis, obtain a comprehensive pollution characterization index and pollution level label, and determine the initial purification and remediation control plan based on the pollution level label. The extraction module is used to deploy distributed purification units in the target water area according to the topographic features of rivers and lakes, analyze the initial purification and restoration control scheme, extract aeration intensity control parameters, filter component opening rate control parameters and flow regulating valve opening control parameters, drive the distributed purification units to perform differentiated operations, and obtain a graded dynamic purification execution instruction set. The calculation module is used to receive the hierarchical dynamic purification execution instruction set and couple the river and lake topographic features and water quality parameters in the initial multidimensional state dataset to dynamically calculate the weight of aquatic plant planting depth, the quantitative dosing ratio of microbial agents and the coordinates of ecological reef layout, and obtain the ecological restoration adaptation parameter set. The feedback module is used to collect time-series data of the treated water area after executing the hierarchical dynamic purification execution instruction set and the ecological restoration adaptation parameter set, obtain the residual feedback vector, and input the residual feedback vector back into the pollution level judgment model for iterative correction of the control parameters until the residual feedback vector converges to the preset compliance range, thus obtaining the final dynamic control instruction sequence.

[0068] It should be noted that this system is a system corresponding to the above method. All implementation methods in the above method embodiments are applicable to this embodiment and can achieve the same technical effect.

[0069] Experimental example: This experiment uses a 500-meter-long, 30-meter-wide, slow-flowing section of the Qinghe River, an inland river in a plain city (hereinafter referred to as the Qinghe experimental section), as the demonstration area. Historically, this section has been affected by intermittent discharges of domestic sewage and non-point source pollution, resulting in water quality fluctuations between light and moderate pollution. The experimental objective is to achieve rapid, adaptive treatment and ecological restoration of moderate to severe pollution caused by a simulated short-term heavy rainfall event (leading to overflow of a combined sewer overflow and the discharge of domestic sewage into the river) by deploying the system and method described in this invention.

[0070] Experimental procedures and data simulation analysis: Step 1: Acquisition of multi-source monitoring data and construction of initial state dataset Three monitoring sections (upstream S1, midstream S2, and downstream S3) were set up in the Qinghe experimental section. Sensors were deployed at each section according to the shallow water area, deep water area, flow channel area, and stagnant water area. The water level sensor (accuracy ±0.5cm) collected data once every 10 minutes; the Doppler current meter (10Hz) combined with a high-definition camera acquired the flow velocity vector field; the five-parameter water quality sensor (dissolved oxygen DO, ammonia nitrogen NH3, total phosphorus TP, COD, and turbidity NTU) collected data once every 30 minutes; and the terrain was scanned by UAV-borne lidar and survey vessel-borne sonar to generate a high-density point cloud (greater than 50 points / m2).

[0071] Monitoring data were collected at the initial moment after rainfall (T0). First, dispersion analysis was performed. For example, an abnormally high value was found in a turbidity sensor at section S2 due to temporary obstruction (local outlier coefficient 3.5 > threshold 2.5), and this sensor was removed. Subsequently, Gaussian filtering (σ=0.5) was used to smooth the data, and range normalization was performed. The normalized water level, flow velocity, water quality, and topographic data (elevation) at all spatial coordinate points at time T0 were spatiotemporally aligned and fused to form an initial multidimensional state dataset. This dataset forms the basis for all subsequent analyses.

[0072] Step 2: Pollution level determination and initial control plan generation The water quality parameter dimension data (five normalized values) from the initial multidimensional state dataset at time T0 are input into the pre-trained pollution level determination model. The model calculates the feature coupling strength coefficient (e.g., the coupling coefficient between NH3-N and TP reaches 0.85), and outputs a comprehensive pollution characterization index P=0.42 through iterative aggregation. This index is matched with preset intervals: [0.1, 0.3) for light pollution, [0.3, 0.6) for moderate pollution, and [0.6, 1.0] for severe pollution. P=0.42∈[0.3, 0.6), therefore the current pollution level is determined to be moderate pollution.

[0073] Based on the moderate pollution label, the corresponding control scheme was retrieved from the control scheme knowledge base to obtain the baseline values ​​for the initial purification and remediation control scheme: aeration intensity baseline value Abase=0.5, filter component opening rate baseline value Fbase=0.6, and flow control valve opening baseline value Vbase=0.7 (all normalized values). Figure 3 This diagram illustrates the internal feature coupling and comprehensive pollution characterization index calculation within the pollution level determination model. It shows the transmission process after water quality parameters are input into the model, highlighting the feature coupling strength matrix calculated in the hidden layer, and the final aggregation into the comprehensive pollution characterization index P=0.42, indicating that it falls into the moderate pollution range.

[0074] Step 3: Deployment of distributed purification units and generation of hierarchical dynamic purification instructions Based on the topographic elevation point cloud data in the initial dataset, the DBSCAN clustering algorithm (neighborhood radius 2m, minimum number of points 15) was used, combined with water depth and flow velocity, to divide the experimental section into four regions: shallow water zone (water depth ≤ 1.0m), deep water zone (> 2.0m), flow channel zone (flow velocity ≥ 0.5m / s), and stagnant water zone (flow velocity < 0.1m / s). A total of 12 distributed purification units (each unit includes an aerator, filter components, and regulating valves) were deployed in each zone.

[0075] The initial scheme is analyzed to obtain Abase, Fbase, and Vbase. Based on the region where the cell is located, differential correction rules are applied: Shallow water unit: A=1.5×0.5=0.75, F=0.5×0.6=0.3, V=0.7.

[0076] Deep water zone unit: A=0.5×0.5=0.25, F=0.6, V=1.5×0.7=1.0 (upper limit truncated to 1.0).

[0077] Flow channel area unit: A=0.5, F=0.6, V=0.7.

[0078] Perched water zone unit: A=1.2×0.5=0.6, F=1.2×0.6=0.72, V=1.2×0.7=0.84.

[0079] The modified parameters are encapsulated and serialized according to unit number, spatial location and execution priority (stagnant water area > deep water area > shallow water area > flow channel area) to generate a hierarchical dynamic purification execution instruction set, which is then sent to each purification unit for execution.

[0080] Step 4: Dynamic calculation of ecological restoration adaptation parameter set The experimental section was divided into 5m × 5m grid cells. The average water depth, substrate type, water quality parameters (NH3-N, TP, DO, COD, NTU), and normalized distance factor to the main flow path within each grid were coupled. A remediation suitability coefficient S (between 0 and 1, with higher values ​​indicating better remediation suitability) for each grid was calculated through weighted summation. The calculated S values ​​showed a spatially uneven distribution, with lower S values ​​(0.3-0.5) in stagnant water areas and nearshore deep water areas, and higher S values ​​(0.6-0.8) in flow channels and some shallow water areas. Based on the velocity vector field, areas with flow velocities below 0.2m / s and points of change in flow direction were identified. Existing obstacles were avoided, and coordinates were calculated and deployed in areas with moderate S values ​​(0.4-0.6) to ensure reef stability and promote water circulation. Finally, a coefficient for each grid (W, K) was generated. quant and the selected set of ecological restoration adaptation parameters for the grid reef coordinates, among which, Figure 4 A spatial contour map showing the suitability coefficient S for restoration; Figure 5 A heat map showing the spatial distribution of the ballast weight W at the planting depth of aquatic plants.

[0081] Step 5: Residual feedback and iterative correction to dynamic equilibrium Twenty-four hours after the purification and remediation measures were implemented (at time T1), time-series data of the entire area were collected again. After the same preprocessing, a feedback multidimensional state dataset was obtained. Water quality parameters at time T1 were extracted and compared feature by feature with the initial values ​​at time T0, and the residuals were calculated. Arranged in order, a one-dimensional residual feedback vector R=[+0.12, -0.23, -0.10, -0.18, -0.25] (after normalization) was constructed.

[0082] The residual feedback vector R is input inversely into the pollution level determination model. Using gradient descent, the model's internal connection weights are fine-tuned (e.g., reducing the weight contribution of NH3-N characteristics) and the pollution level interval boundary thresholds are corrected (e.g., the moderate pollution interval is fine-tuned to [0.28, 0.58]). The water quality at time T1 is reassessed using the corrected model, yielding a new comprehensive pollution characterization index P1 = 0.35, indicating a moderate pollution level, but closer to the light pollution boundary. Next, the baseline values ​​of the control scheme are updated using the new pollution level and residual direction (overall negative, indicating reduced pollution). Steps 3 and 4 are then repeated to generate new purification instructions and remediation parameters.

[0083] The monitoring-judgment-regulation-repair-feedback-correction process iterates every 24 hours. Simulation data shows that after three iterations (approximately 72 hours), the absolute values ​​of all components of the residual feedback vector R are less than the preset threshold of 0.05, indicating that the system is stabilizing. At this point, the pollution level is updated to light pollution, and various control parameters are dynamically adjusted accordingly, forming a final dynamic control command sequence adapted to the current water quality state. The system then enters a low-energy maintenance purification and ecological conservation mode. Figure 6 This represents the changing trend of the five component values ​​of the one-dimensional residual feedback vector, with the dashed line indicating the threshold of ±0.05.

[0084] Figure 7 A bar chart showing the concentration comparison of key water quality parameters (DO, NH3-N, TP) from T0 to T3 after the third iteration; Figure 8 The line represents the downward trend of the average comprehensive pollution characterization index P over time (T0, T1, T2, T3) for the entire experimental section. Figure 9 The trend line represents the decrease in system energy consumption (characterized by total aeration power) as the number of iterations increases, reflecting the energy-saving effect after dynamic optimization.

[0085] This experimental example, through simulating specific scenarios and data, fully demonstrates the entire process of the method described in this application, from multi-source data fusion, intelligent pollution judgment, zone-specific differentiated purification, precise ecological restoration to residual feedback closed-loop optimization. The results show that the system can effectively respond to sudden pollution, rapidly reduce pollution load through the synergy of physical purification and ecological restoration, and automatically adjust strategies using a feedback mechanism, ultimately achieving stable water quality compliance and dynamic optimization of the treatment process, thus verifying the method's advanced nature, adaptability, and engineering application potential.

[0086] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A dynamic purification and control method for the pollution level of rivers and lakes, characterized in that, The method includes: Step 1: Obtain multi-source monitoring data on water level, flow velocity, water quality parameters, and river and lake topography of the target water area. After dispersion analysis and filtering normalization fusion processing, an initial multidimensional state dataset is obtained. Step 2: Input the water quality parameters into the pre-constructed pollution level determination model for analysis to obtain the comprehensive pollution characterization index and pollution level label, and determine the initial purification and remediation control plan based on the pollution level label. Step 3: Deploy distributed purification units in the target water area according to the topographic features of the river and lake, analyze the initial purification and restoration control scheme, extract the aeration intensity control parameters, filter component opening rate control parameters and flow regulating valve opening control parameters, drive the distributed purification units to perform differentiated operations, and obtain a graded dynamic purification execution instruction set. Step 4: Receive the hierarchical dynamic purification execution instruction set and couple the river and lake topographic features and water quality parameters in the initial multidimensional state dataset. Dynamically calculate the weight of aquatic plant planting depth, the quantitative dosage ratio of microbial agents, and the coordinates of ecological reef layout to obtain the ecological restoration adaptation parameter set. Step 5: Collect time-series data of the treated water area after executing the hierarchical dynamic purification execution instruction set and the ecological restoration adaptation parameter set, obtain the residual feedback vector, and input the residual feedback vector back into the pollution level judgment model for iterative correction of the control parameters until the residual feedback vector converges to the preset compliance range, thus obtaining the final dynamic control instruction sequence.

2. The dynamic purification and control method for the ecological pollution level of rivers and lakes according to claim 1, characterized in that, Step 1 includes: Real-time acquisition of water level time-series signals, flow velocity vector field data, water quality parameter solution concentration values, and river and lake topographic elevation point cloud data forms multi-source monitoring data; Discreteness analysis is performed on each type of multi-source monitoring data. The local outlier deviation coefficient of each data point in the corresponding feature dimension is calculated. Abnormal data points whose local outlier deviation coefficient exceeds the preset dynamic threshold are marked and removed to obtain the noise-reduced monitoring data sequence. High-frequency fluctuation components in the noise-reduced monitoring data sequence are filtered out to obtain smoothed monitoring data; the range is normalized on the smoothed monitoring data to obtain normalized water level, flow velocity, water quality parameters and river and lake topography data. Normalized water level, flow velocity, water quality parameters, and river and lake topography data are fused at the feature level according to the same timestamp and spatial coordinates to construct an initial multidimensional state dataset. Each dimension of the initial multidimensional state dataset corresponds to a monitoring parameter type, and each data point contains a spatiotemporal label.

3. The dynamic purification and control method for the ecological pollution level of rivers and lakes according to claim 2, characterized in that, Step 2 includes: Based on the water quality parameter dimension data and corresponding historical pollution level labels in the initial multidimensional state dataset, a pollution level determination model is constructed using a multi-layer neural network structure. The real-time water quality parameters in the initial multidimensional state dataset are used as input feature vectors and input into the pollution level determination model to calculate the feature coupling strength coefficient between each water quality parameter. Through feature iteration and aggregation, a comprehensive pollution characterization index is obtained, which represents the overall deviation of the current water quality. The comprehensive pollution characterization index is matched step by step with multiple preset pollution level value ranges to obtain the corresponding pollution level labels. Based on the pollution level labels, the initial purification and remediation control scheme that uniquely corresponds to the corresponding pollution level label is retrieved from the pre-stored control scheme knowledge base. The initial purification and remediation control scheme includes the aeration intensity benchmark value, the filter component opening rate benchmark value, and the flow regulating valve opening degree benchmark value.

4. The dynamic purification and control method for the ecological pollution level of rivers and lakes according to claim 3, characterized in that, Step 3 includes: River and lake topographic features are read from the initial multidimensional state dataset, which includes topographic elevation point cloud data; spatial clustering is performed based on the topographic elevation point cloud data to identify shallow water areas, deep water areas, flow channels, and stagnant water areas of the target water area. A corresponding distributed purification unit is deployed in each area. Each distributed purification unit includes an aeration device, a filter component, and a flow regulating valve. The initial purification and repair control scheme is used as a high-level control command vector. The high-level control command vector is parsed to extract the aeration intensity benchmark value, the filter component opening rate benchmark value, and the flow regulating valve opening benchmark value. These three benchmark values ​​are used as the basic values ​​of the aeration intensity control parameter, the filter component opening rate control parameter, and the flow regulating valve opening control parameter, respectively. Based on the terrain features and pollution level labels of the areas where each distributed purification unit is located, the base values ​​of the three control parameters are differentiated and corrected to obtain the corrected control parameters. The modified control parameters are encapsulated into executable instructions for each distributed purification unit, driving each distributed purification unit to perform differentiated aeration, filtration, and flow regulation operations; according to the spatial location and execution priority of each distributed purification unit, all executable instructions are serialized and arranged to form a hierarchical dynamic purification execution instruction set.

5. The dynamic purification and control method for the ecological pollution level of rivers and lakes according to claim 4, characterized in that, The differential correction of the base values ​​of the three control parameters includes: Set the aeration intensity control parameter for the shallow water area to 1.5 times the aeration intensity benchmark value and set the filter module opening rate control parameter to 0.5 times the filter module opening rate benchmark value. Set the aeration intensity control parameter for the deep water area to 0.5 times the aeration intensity reference value and set the flow regulating valve opening control parameter to 1.5 times the flow regulating valve opening reference value; Set the aeration intensity control parameters, filter component opening rate control parameters, and flow regulating valve opening control parameters of the flow channel area to their corresponding baseline values. The aeration intensity control parameters, filter component opening rate control parameters, and flow regulating valve opening control parameters of the stagnant water zone are all set to 1.2 times the corresponding benchmark values.

6. The dynamic purification and control method for the ecological pollution level of rivers and lakes according to claim 5, characterized in that, Step 4 includes: The execution state corresponding to the executable instructions of each distributed purification unit in the hierarchical dynamic purification execution instruction set is used as the dynamic constraint boundary for ecological restoration; river and lake topographic features and water quality parameters are read from the initial multidimensional state dataset. The river and lake topographic features include water depth distribution data and sediment type classification data. The water quality parameters include ammonia nitrogen concentration, total phosphorus concentration, dissolved oxygen concentration, COD concentration and turbidity. The topographic features of rivers and lakes are spatially gridded and coupled with water quality parameters, and the restoration suitability coefficient is calculated for each grid cell. Based on the ecological restoration suitability coefficient of each grid unit, the planting depth weight of aquatic plants in the proposed planting area is dynamically calculated. The planting depth weight represents the weight value of the burial depth of plant roots in the bottom sediment. Based on the ammonia nitrogen concentration, total phosphorus concentration and COD concentration in the water quality parameters, the quantitative dosage ratio of microbial agents is dynamically calculated. The quantitative dosage ratio represents the mass of microbial agents required per unit volume of water. Based on the water flow direction vector field, velocity scalar distribution data, and existing obstacle spatial coordinates in the velocity vector field data, the layout coordinates of the ecological reefs are dynamically calculated so that the layout coordinates of the ecological reefs are located at the point where the water flow direction changes or in the region where the velocity value in the velocity scalar distribution data is lower than the preset velocity threshold. The layout coordinates of each ecological reef include the two-dimensional plane position and bottom elevation. The calculated planting depth weight, the quantitative application ratio of microbial agents, and the layout coordinates of the ecological reefs are combined into a data set, which is defined as the ecological restoration adaptation parameter set.

7. The dynamic purification and control method for the ecological pollution level of rivers and lakes according to claim 6, characterized in that, The topographic features of rivers and lakes are spatially gridded and coupled with water quality parameters. For each grid cell, a remediation suitability coefficient is calculated, including: The water depth distribution data and sediment type classification data in the river and lake topography features are spatially coupled with the ammonia nitrogen concentration, total phosphorus concentration, dissolved oxygen concentration, COD concentration and turbidity in the water quality parameters, and the target water area is divided into multiple grid units according to a fixed side length. For each grid cell, the average water depth, sediment type code, average ammonia nitrogen concentration, average total phosphorus concentration, average dissolved oxygen concentration, average COD concentration, and average turbidity are obtained respectively. The main flow path segment set is extracted from the flow direction vector field in the velocity vector field data. The main flow path segment set is formed by connecting the points with the maximum velocity in the flow direction vector field in sequence. For each grid cell, take the geometric center point of the corresponding grid cell, calculate the shortest distance from the geometric center point to each line segment in the main flow path segment set, and take the minimum value among all shortest distances as the vertical distance from the corresponding grid cell to the main flow path. The normalized distance factor is obtained by calculating the ratio of the vertical distance from the corresponding grid cell to the main flow path to half the maximum width of the target water area. The remediation suitability coefficient of each grid cell is calculated by weighted summation based on the average water depth, bottom sediment type code value, average ammonia nitrogen concentration, average total phosphorus concentration, average dissolved oxygen concentration, average COD concentration, average turbidity, and normalized distance factor.

8. The method for dynamic purification and control of river and lake water ecological pollution levels according to claim 7, characterized in that, Based on the ammonia nitrogen concentration, total phosphorus concentration, and COD concentration in the water quality parameters, the quantitative dosage ratio of microbial agents is dynamically calculated. This quantitative dosage ratio represents the required mass of microbial agents per unit volume of water, including: Obtain the remediation suitability coefficient for each grid cell, as well as the corresponding ammonia nitrogen concentration, total phosphorus concentration, and COD concentration for that grid cell; multiply the ammonia nitrogen concentration, total phosphorus concentration, and COD concentration together and take the square root to obtain the comprehensive pollution index for the corresponding grid cell. The basic dosage ratio of microbial agents is obtained by multiplying the comprehensive pollution index with the remediation suitability coefficient of the corresponding grid unit; the average water depth value in the water depth distribution data of the corresponding grid unit is obtained; and the quantitative dosage ratio of microbial agents is obtained based on the basic dosage ratio, the corresponding average water depth value and the arithmetic mean of the average water depth values ​​of all grid units in the target water area.

9. The method for dynamic purification and control of river and lake water ecological pollution levels according to claim 8, characterized in that, Step 5 includes: Time-series data of the treated water area are collected again according to a fixed sampling period. The time-series data of the treated water area includes water level time-series signals, flow velocity vector field data, water quality parameter solution concentration values, and river and lake topographic elevation point cloud data. Discreteness analysis and filtering normalization fusion processing are performed on the collected time-series data of the treated water area to obtain a feedback multidimensional state dataset. Water quality parameter dimension data is extracted from the feedback multidimensional state dataset and compared feature by feature with the water quality parameter dimension data in the initial multidimensional state dataset. The real-time deviation value of each water quality parameter feature is calculated, and the real-time deviation values ​​of all water quality parameter features are arranged in a preset order to construct a one-dimensional residual feedback vector. The one-dimensional residual feedback vector is input in reverse into the pollution level determination model to iteratively correct the connection weights and boundary thresholds of multiple pollution level numerical ranges within the pollution level determination model. After each correction, the pollution level determination model outputs a new comprehensive pollution characterization index and a new pollution level label. The aeration intensity benchmark value, filter component opening rate benchmark value, and flow control valve opening benchmark value in the initial purification and remediation control scheme are updated according to the new pollution level label. An updated hierarchical dynamic purification execution instruction set is then obtained based on the updated benchmark values. The ecological restoration adaptation parameter set is updated based on the updated hierarchical dynamic purification execution instruction set and the river and lake topographic features and water quality parameters in the feedback multidimensional state dataset. The update process is repeated, and after each iteration, the newly constructed one-dimensional residual feedback vector is compared with the preset compliance interval until the absolute value of each component of the one-dimensional residual feedback vector is less than the threshold corresponding to the preset compliance interval. The aeration intensity benchmark value, filter component opening rate benchmark value, and flow control valve opening benchmark value in the finally converged initial purification and remediation control scheme are repackaged into an executable instruction sequence, which is defined as the final dynamic control instruction sequence.

10. A dynamic purification and control system for the pollution level of river and lake water ecosystems, wherein the system implements the method as described in any one of claims 1 to 9, characterized in that, include: The acquisition module is used to acquire multi-source monitoring data on water level, flow velocity, water quality parameters and river and lake topography of the target water area. After discrete analysis and filtering normalization fusion processing, an initial multidimensional state dataset is obtained. The analysis module is used to input water quality parameters into a pre-built pollution level determination model for analysis, obtain a comprehensive pollution characterization index and pollution level label, and determine the initial purification and remediation control plan based on the pollution level label. The extraction module is used to deploy distributed purification units in the target water area according to the topographic features of rivers and lakes, analyze the initial purification and restoration control scheme, extract aeration intensity control parameters, filter component opening rate control parameters and flow regulating valve opening control parameters, drive the distributed purification units to perform differentiated operations, and obtain a graded dynamic purification execution instruction set. The calculation module is used to receive the hierarchical dynamic purification execution instruction set and couple the river and lake topographic features and water quality parameters in the initial multidimensional state dataset to dynamically calculate the weight of aquatic plant planting depth, the quantitative dosing ratio of microbial agents and the coordinates of ecological reef layout, and obtain the ecological restoration adaptation parameter set. The feedback module is used to collect time-series data of the treated water area after executing the hierarchical dynamic purification execution instruction set and the ecological restoration adaptation parameter set, obtain the residual feedback vector, and input the residual feedback vector back into the pollution level judgment model for iterative correction of the control parameters until the residual feedback vector converges to the preset compliance range, thus obtaining the final dynamic control instruction sequence.

Images