Environment monitoring optimization system based on adaptive data driving

Abstract

The invention relates to the field of environment monitoring, in particular to the field of water area environment monitoring. An environment monitoring optimization system based on adaptive data driving generates a simulated water area environment according to a real water area environment by using a digital twinborn technology, and simulates pollutant diffusion at different positions. The system deploys an unmanned aerial vehicle and an unmanned ship to obtain water area environment data, and a deep learning network is trained by using the collected data to dynamically predict the diffusion trend of pollutants. According to the system, environment monitoring can be efficiently and adaptively realized, the monitoring efficiency and the coverage area are remarkably improved, and meanwhile, the monitoring cost is reduced.

Description

Technical Field

[0001] The present invention relates to the field of environmental monitoring, in particular to the field of water environment monitoring. Background Art

[0002] In the current field of environmental protection and watershed monitoring, traditional monitoring methods are limited by fixed monitoring stations and sampling strategies, resulting in limited spatial coverage, poor real-time performance, and high costs. Furthermore, traditional methods struggle to respond to sudden pollution incidents and lack the ability to dynamically predict pollutant diffusion trends. With the rapid development of image and signal processing technologies, drones and unmanned vessels are being widely used as intelligent devices in watershed monitoring, capable of acquiring large amounts of high-resolution data in real time. Digital twin technology, an emerging simulation and emulation technology, can train and test different strategies and models in a virtual environment that maps the actual watershed in real time, providing guidance for practical operations. Deep learning networks, by processing complex spatial and temporal data, are used to predict the distribution of pollutant concentrations over time. Summary of the Invention

[0003] In response to the shortcomings of the existing technology, the purpose of the present invention is to provide an environmental monitoring optimization system based on adaptive data-driven, which can obtain the distribution of pollution sources according to the situation where different waters are polluted by different pollution sources, rather than the blind search of traditional detection methods. By using deep learning models to train data in a simulated environment and further applying it to the real environment, it can solve a series of problems such as low monitoring efficiency, limited spatial range and high cost, and provide data warnings for future pollution situations.

[0004] To achieve the above object, the present invention provides the following technical solution, comprising the following steps: An environmental monitoring and optimization system based on adaptive data drive, characterized by comprising the following steps: Step 1. Use digital twin technology to build a model of the water environment to be simulated. This model takes into account terrain, altitude, longitude and latitude of the water area, and water flow velocity, achieving dynamic mapping of the actual water environment to be simulated. Step 2. Set the square grid size and use it to segment the water environment model to be simulated. The center point of each segmented square grid within the simulated water area is used as a reference point. If a water flow direction exists, the water flow direction is determined. If no water flow direction exists, an arbitrary direction of the simulated water area is set as the water flow direction. Step 3. Consider a reference point as a pollution source, inject non-toxic chemicals as pollutants, and record the pollution source data, namely the amount of pollutants injected, the location of the pollution source, and the time of the pollution source injection. Then, use the unmanned boat to detect the pollutants at the reference points outside the pollution source for a period of time until the pollutant concentration at all detection points is considered to be zero. The detection data of any detection point x is expressed as (a x , b x , c x ), a x Indicates the position of the detection point x, which corresponds to a position point on the water environment model to be simulated. The water flow velocity and direction of the detection point x are obtained through the position correspondence. x represents the pollutant concentration at the detection point x, c x represents the detection time of detection point x; Step 4. Replace the reference point position corresponding to the pollution source and repeat step 3. The reference points serving as pollution sources are evenly distributed in the simulated water area. Step 5. Establish an adaptive data-driven environmental monitoring optimization model. This model uses a hybrid convolutional neural network (CNN) and a long short-term memory (LSTM) structure to comprehensively capture the spatial and temporal characteristics of the detection data. The network's input layer is divided into two branches: the spatial data processing branch uses the hybrid convolutional neural network (CNN) to process spatial data, dividing the environmental monitoring area into square grids of a set size. Each square grid contains data on the pollutant concentration, water flow velocity, and water flow direction at the current location, as well as the pollutant concentrations of the four adjacent square grids. This extracts spatial features and generates a spatial feature map. The time series data processing branch uses the LSTM to process time series data, including the diffusion of pollutant concentrations in each square grid within the current time period, to capture temporal dependencies. The network's middle layer uses a weighted average fusion strategy to fuse the features extracted by the two branches. Finally, the fully connected layer serves as the output layer, outputting the environmental pollutant concentration for the next time period and the pollution diffusion to adjacent grids. The prediction results are fed back to the network input layer. The iterative process continues until a preset termination condition is reached (the preset termination condition is either reaching the maximum number of iterations or the pollutant concentration in each square grid is considered zero). The final output is a prediction of the temporal and spatial variation of pollutant concentrations. Step 6. Use the data from the monitoring points obtained in Steps 3 and 4 to train the adaptive data-driven environmental monitoring optimization model established in Step 5. Before training, input the simulated water environment model data into the adaptive data-driven environmental monitoring optimization model to obtain the location, water velocity, and water flow direction corresponding to each square grid. Each sample data item represents the location, pollutant concentration, and detection time corresponding to a monitoring point. After data preprocessing, 80% of the sample data is used as training data, and 20% of the sample data is used as test data for training and testing. This results in a fully trained adaptive data-driven environmental monitoring optimization model. Step 7. When an abnormal situation occurs, use an unmanned boat to detect the water quality of the simulated water area. When it is found that the pollutant concentration at a certain reference point exceeds the standard, the position, water flow direction, speed, and pollutant concentration of the reference point and its surrounding reference points are detected, and the detection results are input into a well-trained adaptive data-driven environmental monitoring optimization model. The well-trained adaptive data-driven environmental monitoring optimization model outputs a prediction of the range of pollutant concentration changes over time and space, and the unmanned boat is used to detect and verify the prediction. The abnormal situations include the following: first, using drones to monitor the species and number of water birds in the simulated water area, when the species and number of water birds decrease due to unnatural reasons; second, when the pollutant concentration exceeds the standard at a water station set in the simulated water area; third, when notification of pollutant overturning in the simulated water area is received.

[0005] Furthermore, in step one, the dynamic mapping of the actual water environment of the water area to be simulated refers to reflecting the terrain, altitude, water longitude and latitude, and water flow speed of each point in the actual water environment of the water area to be simulated one by one on the water environment model to be simulated, and the terrain, altitude, water longitude and latitude, and water flow speed information of each point in the actual water environment of the water area to be simulated can be directly obtained through the water environment model to be simulated.

[0006] Furthermore, in step 2, the square grid size is set according to actual needs. The smaller the square grid size is set, the more accurate the result obtained by the adaptive data-driven environmental monitoring optimization model.

[0007] Furthermore, in step three, in a row or column of reference points in the same water flow direction, the pollutant concentration of the reference points appearing at the same detection time along the direction of decreasing pollutant concentration is deemed to be 0, and points outside the reference point along the direction of decreasing pollutant concentration are no longer used as detection points at the detection time; when the pollutant concentration is less than or equal to the zero set value, the pollutant concentration is deemed to be 0, and the zero set value is an artificial set value. When the pollutant concentration is less than or equal to the set value, the pollutant concentration can be deemed to be 0.

[0008] Furthermore, in step four, the reference points serving as pollution sources are evenly arranged in the water area to be simulated, which means that the positions of the reference points of all pollution sources are evenly distributed on the environmental model of the water area to be simulated, and there will not be a situation where a certain area has multiple reference points of pollution sources, while another area of the same size does not contain a reference point of a pollution source.

[0009] Furthermore, in step 5, in the spatial data processing branch, the expression of the y data of any square grid is (a y , b y , d y , e y , b1 y , b2 y , b3 y , b4 y ), where a y Indicates the position of the square grid y, b y represents the pollutant concentration of the square grid y, d y represents the water velocity of square grid y, e y Indicates the direction of water flow in square grid y, b1 y represents the pollutant concentration of the left adjacent square grid of square grid y, b2 y represents the pollutant concentration of the right adjacent square grid of square grid y, b3 y represents the pollutant concentration of the previous adjacent square grid of square grid y, b4 y The pollutant concentration of the adjacent square grid after the square grid y is represented by b, and the pollutant concentration of the square grid y is represented by b. y Satisfy b y =f(a y , d y , e y , b1 y , b2 y , b3 y , b4 y ), f() is a function; in the time series data processing branch, the expression of any square grid y data is (b y , c y , b5 y ), c y represents the detection time of the square grid y, b5 y Indicates the previous adjacent detection time of the detection time of the square grid y. In the time series data processing branch, the time is the continuous time with the same interval.

[0010] In summary, the invention has the following beneficial effects: The present invention uses drones and unmanned boats equipped with high-precision filming equipment to monitor, evaluate and analyze biological components in water bodies. It can more comprehensively reflect environmental changes, especially monitor changes in biological communities, and help to detect environmental anomalies in a timely manner. The present invention uses digital twins to establish a simulated water environment, which can reduce the training cost of finding the optimal strategy in a real environment and improve the efficiency and feasibility of training. The model for predicting the scope of pollution established by the present invention adopts a weighted average strategy to fuse the features extracted by the hybrid convolutional neural network CNN and the long short-term memory network LSTM, takes the current position and pollutant concentration as input, and finally outputs the pollutant concentration and diffusion in the future. Compared with the comprehensive detection of traditional detection methods, it can achieve accurate prediction and monitoring of pollutant concentrations. The system can provide timely and accurate predictions of pollutant diffusion when anomalies occur, and guide corresponding monitoring and response measures. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 This is a schematic diagram of reference points after segmenting the simulated water environment model using a square grid of a set size in the present invention, where the small dots are reference points; Figure 2 It is a diagram of the implementation process of the present invention; Figure 3 This is a flowchart of deep learning training in the present invention. DETAILED DESCRIPTION

[0012] The present invention will be described in further detail below with reference to the accompanying drawings.

[0013] like Figures 1-3 As shown, the present invention discloses an adaptive data-driven environmental monitoring and optimization system that uses deep learning methods to predict pollutant distribution. The system first generates initial labels through a digital twin simulation environment. These labels are then used to further train and optimize the pollutant distribution prediction model, continuously improving prediction accuracy. Finally, the trained model is applied to actual waters. The system specifically includes the following steps: Step 1. Use digital twin technology to achieve dynamic mapping from the actual water area to the simulated water environment. Specifically, the actual environmental characteristics of each point in the real water area, such as terrain, altitude, longitude and latitude of the water area, and water flow speed, need to be accurately mapped to the simulated water environment model.

[0014] By using the fluid analysis CFD module in COMSOL Multiphysics software to establish a finite element model of the water area, combined with the three major fluid dynamics equations, the water environment of pollutant diffusion at different times is simulated. The finite element model is established according to the following steps: (1) Select the three-dimensional single-phase flow turbulence model in the simulation developer; (2) Set the component equations and properties of fluid flow, including density and dynamic viscosity, in the physical field interface. Select the mixture model to describe the physical field interface to simulate the mixing and interaction of different substances in the water area; (3) Use the geometric modeling tools provided by COMSOL to establish a geometric model of the water area in the software based on the simulated terrain data and longitude and latitude information. This model will serve as the basis for domain equations or boundary conditions; (4) Simulate and establish different terrain maps and select different pollution source locations. Repeat the above steps to obtain multiple sets of water environment simulation data under different conditions.

[0015] When a pollution source is put into use, the pollutant diffusion process is simulated according to the three conservation laws of fluid mechanics. Digital twin technology can simulate various conditions in the water area and provide prior knowledge for subsequent training of deep learning models.

[0016] Step 2. Set the appropriate square size and divide the simulated water environment into multiple square grids accordingly. The center point of each grid is used as a reference point. If the water flow direction is known, define it as the water flow direction; if there is no specific water flow direction, select any direction as the default water flow direction. The size of the square grid is set according to actual needs, such as Figure 1 As shown in the figure, it shows that the smaller the grid size, the more accurate the results of the environmental monitoring optimization model. However, it also means that there are more reference points to be processed, which increases the monitoring cost and computational burden.

[0017] Step 3. Obtain labels and prior knowledge for network training. Specifically, a non-toxic chemical substance is introduced as a pollutant at a known reference point. This point is considered a pollution source, and pollution source data is recorded, including the amount of pollutant introduced, the location of the pollution source, and the time of the pollution source introduction. Reference points outside the pollution source are used as detection points. An unmanned vessel is used to detect pollutants at each detection point over a sustained period of time until the pollutant concentration at all detection points is considered to be zero. The data for any detection point x is represented as (ax, bx, cx), where ax represents the location of detection point x, which corresponds to a point in the simulated water environment model. The water flow velocity and direction at detection point x are obtained through this positional correspondence; bx represents the pollutant concentration at detection point x; and cx represents the detection time at detection point x.

[0018] Within a row or column of reference points in the same water flow direction, detection is performed along the direction of decreasing pollutant concentration. If the pollutant concentration at a reference point is considered zero within the same detection period, all other points along the decreasing direction of that reference point will no longer be used as detection points for that period. When the pollutant concentration is less than or equal to the set value, the pollutant concentration is considered zero. The set value is a threshold set based on actual conditions; when the pollutant concentration is less than or equal to this value, it is considered zero.

[0019] Step 4. Replace the reference point location corresponding to the pollution source and repeat step 3 to ensure that the reference points for the pollution sources are evenly distributed throughout the simulated waters. Even distribution means that the reference points for all pollution sources are evenly distributed throughout the simulated waters. There should not be a situation where one area has multiple reference points for pollution sources while another area of the same size has none. The purpose of steps 3 and 4 is to obtain sample data. The more pollution source locations there are and the smaller the square grid size, the more accurate the data obtained, and the more accurately it reflects the spread of pollutants. However, the more square grids are cut, the more data needs to be collected.

[0020] In this example, a circular pond with a radius of 10 meters is used as an example. The pollution source is located at coordinates (0, 0). The pollutant is ink, the amount injected is 50 grams, and the injection time is 9:00 AM on a certain day. At 9:05 AM, 24 detection points are detected at the center of a square grid with a size of 1 meter x 1 meter, centered on the pollution source. The obtained detection point data is as follows: (1). The pollutant concentrations at coordinates (1,0), (-1,0), (1,1), (-1,-1), (1,-1), (-1,1), (0,1), and (0,-1) are all 30 mg / L, and the detection time is 9:05 am.

[0021] (2). The pollutant concentrations at coordinates (2,0), (-2,0), (2,1), (2,2), (2,-1), (2,-2), (-2,1), (-2,2), (-2,-1), (-2,-2), (0,2), (0,-2), (1,2), (1,-2), (-1,2), (-1,-2) are all 10 mg / L, and the detection time is 9:05 am.

[0022] If the pollutant concentration at a point other than 10 mg / L is less than 2 mg / L (zero setting value), the pollutant concentration is considered to be 0 and is not used as a detection point for that detection time.

[0023] Step 5. Establish an adaptive data-driven environmental monitoring optimization model. This model uses a hybrid convolutional neural network (CNN) and long short-term memory (LSTM) architecture to comprehensively capture the spatial and temporal characteristics of pollutant concentration diffusion in aquatic environments. The spatial data processing branch utilizes a convolutional neural network (CNN) to process spatial data within the environmental monitoring area. The input layer divides the water area into square grids of a set size. Each grid contains information about the pollutant concentration, water velocity, and flow direction at the current location, as well as the pollutant concentrations of its four neighboring grids. The CNN architecture includes multiple convolutional and pooling layers, as well as activation functions for nonlinear transformations. The convolution kernel size is typically 3x3 or 5x5 to extract local features. The pooling layer reduces the dimensionality of the feature map through max pooling or average pooling while retaining key information. During model training, the CNN uses a backpropagation algorithm to adjust the convolution kernel weights to maximize prediction accuracy and generalization. The resulting spatial feature map reflects the spatial distribution of pollutant concentrations and environmental factors within each square grid. The time series data processing branch uses a long short-term memory (LSTM) network to process the time series data of pollutant concentrations within each square grid. LSTMs consist of an input gate, a forget gate, an output gate, and memory cells, effectively capturing long-term temporal dependencies. Pollutant concentration data for each grid is fed into the LSTM network as a sequence of time steps. The input gate controls the influx of new information, the forget gate controls the loss of memory information from the previous state, and the output gate generates output based on the current input and memory cell states. The network's hidden and memory cell states are recursively updated and maintained over time steps, thereby capturing and learning the dynamic evolution of pollutant concentrations within the grid. LSTM parameters, including the hidden layer size, number of memory cells, and learning rate, are optimized using backpropagation and gradient descent algorithms to enhance the model's ability to model time series data.

[0024] The intermediate layer uses a weighted averaging fusion strategy to fuse the spatial feature maps processed by the CNN with the time series features processed by the LSTM. This strategy helps to comprehensively utilize both spatial and temporal information, enhancing the model's overall predictive performance. The fully connected layer, serving as the output layer, is responsible for mapping the fused features to the pollutant concentration at the next time step in the predicted environment and its diffusion within adjacent grid cells. The output is fed back to the network input layer, and the network undergoes multiple iterations until a preset termination condition is met, such as reaching the maximum number of iterations or the pollutant concentration within the grid reaching zero. The final output is a forecast of the temporal and spatial variation of pollutant concentration.

[0025] In the spatial data processing branch, the expression of any square grid y data is (a y , b y , d y , e y , b1y , b2 y , b3 y , b4 y ), where a y Indicates the position of the square grid y, b y represents the pollutant concentration of the square grid y, d y represents the water velocity of square grid y, e y Indicates the direction of water flow in square grid y, b1 y represents the pollutant concentration of the left adjacent square grid of square grid y, b2 y represents the pollutant concentration of the right adjacent square grid of square grid y, b3 y represents the pollutant concentration of the previous adjacent square grid of square grid y, b4 y The pollutant concentration of the adjacent square grid after the square grid y is represented by b, and the pollutant concentration of the square grid y is represented by b. y Satisfy b y =f(a y , d y , e y , b1 y , b2 y , b3 y , b4 y ), f() is a function; in the time series data processing branch, the expression of any square grid y data is (b y , c y , b5 y ), c y represents the detection time of the square grid y, b5 y Indicates the previous adjacent detection time of the detection time of the square grid y. In the time series data processing branch, the time is the continuous time with the same interval.

[0026] Step 6. Use the sample data obtained in Steps 3 and 4 to train the adaptive data-driven environmental monitoring optimization composite model established in Step 5. Before training, input the simulated water environment model data into the optimization model to determine the location, water velocity, and flow direction corresponding to each square grid. Each sample data point includes the location, pollutant concentration, and detection time. After data preprocessing, 80% of the sample data is used as training data, and 20% of the sample data is used as test data for both training and testing. Through these steps, a fully trained adaptive data-driven environmental monitoring optimization model is ultimately obtained.

[0027] Step 7. When an abnormal situation occurs, an unmanned vessel is used to monitor the water quality of the simulated waters. If a pollutant concentration at a reference point exceeds the permitted level, the location, flow direction, velocity, and pollutant concentration of that reference point and its surrounding reference points are monitored (reference points are selected along a decreasing direction until the pollutant concentration at each reference point reaches zero within the same monitoring time period). The test results are input into the trained adaptive data-driven environmental monitoring optimization model, which outputs a prediction of the temporal and spatial range of pollutant concentrations. The predictions can then be verified using an unmanned vessel.

[0028] Abnormal situations include the following: first, when the species and number of water birds in the simulated waters are monitored by drones, and the species and number of water birds decrease due to non-natural reasons; second, when the water station in the simulated waters detects that the pollutant concentration exceeds the standard; third, when a notification is received that pollutants have been overturned in the simulated waters.

[0029] To meet the project's requirements, both drones and unmanned boats were designed for water environment monitoring. Drones, such as the XAG P30, were equipped with high-performance batteries, a portable light source, a portable ICP emission spectrometer, and an LS1206B portable current velocity calculator. They were also equipped with a microcontroller, communication modules, a power control module, and fixed data acquisition devices to ensure long-term operation and data transmission. Meanwhile, the unmanned boats were designed to be appropriately sized and configured based on the specific situation. The water quality monitoring equipment, communication systems, and power supply systems were also configured to meet the requirements of the water monitoring mission. The battery and power supply modules were designed to meet the power requirements of the light source, spectrometer, and microcontroller to ensure long-term system operation.

[0030] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiment. All technical solutions based on the concept of the present invention are within the scope of protection of the present invention. It should be noted that for those skilled in the art, various improvements and modifications that do not depart from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. An environmental monitoring and optimization system based on adaptive data drive, characterized in that: The steps include: Step 1. Use digital twin technology to build a model of the water environment to be simulated. This model takes into account terrain, altitude, longitude and latitude of the water area, and water flow velocity, achieving dynamic mapping of the actual water environment to be simulated. Step 2. Set the square grid size and use it to segment the water environment model to be simulated. The center point of each segmented square grid within the simulated water area is used as a reference point. If a water flow direction exists, the water flow direction is determined. If no water flow direction exists, an arbitrary direction of the simulated water area is set as the water flow direction. Step 3. Consider a reference point as a pollution source, inject non-toxic chemicals as pollutants, and record the pollution source data, namely the amount of pollutants injected, the location of the pollution source, and the time of the pollution source injection. Then, use the unmanned boat to detect the pollutants at the reference points outside the pollution source for a period of time until the pollutant concentration at all detection points is considered to be zero. The detection data of any detection point x is expressed as (a x , b x , c x ), a x Indicates the position of the detection point x, which corresponds to a position point on the water environment model to be simulated. The water flow velocity and direction of the detection point x are obtained through the position correspondence. x represents the pollutant concentration at the detection point x, c x represents the detection time of detection point x; Step 4. Replace the reference point position corresponding to the pollution source and repeat step 3. The reference points serving as pollution sources are evenly distributed in the simulated water area. Step 5. Establish an adaptive data-driven environmental monitoring optimization model. This model uses a hybrid convolutional neural network (CNN) and long short-term memory (LSTM) structure to comprehensively capture the spatial and temporal characteristics of the detection data. The network's input layer is divided into two branches: the spatial data processing branch, which uses a hybrid convolutional neural network (CNN) to process spatial data and divides the environmental monitoring area into square grids of a set size. Each square grid data includes the pollutant concentration, water flow velocity, water flow direction, and the pollutant concentration of the four adjacent square grids at the current location, in order to extract spatial features and generate a spatial feature map; the time series data processing branch, which uses LSTM to process time series data, including the diffusion of pollutant concentrations in each square grid within the current time, to capture temporal dependencies. The middle layer of the network uses a weighted average fusion strategy to fuse the features extracted by the two branches. Finally, the fully connected layer serves as the output layer, outputting the concentration of environmental pollutants within the next timeframe and the extent of pollution diffusion to adjacent grids. The prediction results are fed back to the network input layer, and the network continues iteratively until a preset termination condition is reached. The preset termination condition is either reaching the maximum number of iterations or the pollutant concentration in each square grid is considered zero. The final output is a prediction of how pollutant concentrations change over time and space. Step 6. Use the data from the monitoring points obtained in Steps 3 and 4 to train the adaptive data-driven environmental monitoring optimization model established in Step 5. Before training, input the simulated water environment model data into the adaptive data-driven environmental monitoring optimization model to obtain the location, water velocity, and water flow direction corresponding to each square grid. Each sample data item represents the location, pollutant concentration, and detection time corresponding to a monitoring point. After data preprocessing, 80% of the sample data is used as training data, and 20% of the sample data is used as test data for training and testing. This results in a fully trained adaptive data-driven environmental monitoring optimization model. Step 7. When an abnormal situation occurs, use an unmanned boat to detect the water quality of the simulated water area. When it is found that the pollutant concentration at a certain reference point exceeds the standard, the position, water flow direction, speed, and pollutant concentration of the reference point and its surrounding reference points are detected, and the detection results are input into a well-trained adaptive data-driven environmental monitoring optimization model. The well-trained adaptive data-driven environmental monitoring optimization model outputs a prediction of the range of pollutant concentration changes over time and space, and the unmanned boat is used to detect and verify the prediction. The abnormal situations include the following: first, using drones to monitor the species and number of water birds in the simulated water area, when the species and number of water birds decrease due to unnatural reasons; second, when the pollutant concentration exceeds the standard at a water station set in the simulated water area; third, when notification of pollutant overturning in the simulated water area is received.

2. The adaptive data-driven environmental monitoring and optimization system according to claim 1, characterized in that: In step one, the dynamic mapping of the actual water environment of the water area to be simulated refers to reflecting the terrain, altitude, water longitude and latitude, and water flow speed of each point in the actual water environment of the water area to be simulated on the water environment model to be simulated. The terrain, altitude, water longitude and latitude, and water flow speed information of each point in the actual water environment of the water area to be simulated can be directly obtained through the water environment model to be simulated.

3. The adaptive data-driven environmental monitoring and optimization system according to claim 1, characterized in that: In step 2, the square grid size is set according to actual needs. The smaller the square grid size is set, the more accurate the results obtained by the adaptive data-driven environmental monitoring optimization model.

4. The adaptive data-driven environmental monitoring and optimization system according to claim 1, characterized in that: In step three, in a row or column of reference points in the same water flow direction, the pollutant concentration of the reference points appearing at the same detection time along the direction of decreasing pollutant concentration is deemed to be 0, and points outside the reference point along the direction of decreasing pollutant concentration are no longer used as detection points at the detection time; when the pollutant concentration is less than or equal to the zero set value, the pollutant concentration is deemed to be 0, and the zero set value is an artificial set value. When the pollutant concentration is less than or equal to the set value, the pollutant concentration can be deemed to be 0.

5. The adaptive data-driven environmental monitoring and optimization system according to claim 1, characterized in that: In step 4, the reference points serving as pollution sources are evenly arranged in the water area to be simulated, which means that the positions of the reference points of all pollution sources are evenly distributed on the environmental model of the water area to be simulated, and there will not be a situation where a certain area has multiple reference points of pollution sources, while another area of the same size does not contain a reference point of a pollution source.

6. The adaptive data-driven environmental monitoring and optimization system according to claim 1, characterized in that: In step 5, in the spatial data processing branch, the expression of the y data of any square grid is (a y , b y , d y , e y , b1 y , b2 y , b3 y , b4 y ), where a y Indicates the position of the square grid y, b y represents the pollutant concentration of the square grid y, d y represents the water velocity of square grid y, e y Indicates the direction of water flow in square grid y, b1 y represents the pollutant concentration of the left adjacent square grid of square grid y, b2 y represents the pollutant concentration of the right adjacent square grid of square grid y, b3 y represents the pollutant concentration of the previous adjacent square grid of square grid y, b4 y The pollutant concentration of the adjacent square grid after the square grid y is represented by b, and the pollutant concentration of the square grid y is represented by b. y Satisfy b y =f(a y , d y , e y , b1 y , b2 y , b3 y , b4 y ), f() is a function; in the time series data processing branch, the expression of any square grid y data is (b y , c y , b5 y ), c y represents the detection time of the square grid y, b5 y Indicates the previous adjacent detection time of the detection time of the square grid y. In the time series data processing branch, the time is the continuous time with the same interval.

Images