Automatic monitoring and early warning method for water environment

Abstract

The invention provides a water environment automatic monitoring and early warning method, which comprises the following steps: constructing a water environment change state space model, and simulating a spatial and temporal distribution field of hydrodynamic force and water quality of river nodes; acquiring historical water quality pollution data of the drainage basin node relationship network, and calculating a time interval when upstream drainage basin node pollutants reach a current drainage basin node based on a water environment change state space model; taking the historical water quality pollution data of the upstream drainage basin node at the starting point of the time interval as input, taking the historical water quality pollution data of the current drainage basin node at the ending point of the time interval as output, and obtaining a trained feedforward deep neural network; and utilizing the trained feedforward deep neural network, taking the water environment data of the upstream drainage basin node and the current drainage basin node at the current moment as input, predicting the water environment data of the current drainage basin node at the next moment, and performing real-time early warning. According to the invention, the propagation process of pollutants in a water body can be simulated more accurately, and the precision and reliability of water quality prediction are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of environmental monitoring, and relates to a method for automatic monitoring and early warning of water environment. Background Art

[0002] With the acceleration of the global industrialization and urbanization processes, the problem of water pollution has become increasingly serious, becoming one of the primary environmental challenges threatening the ecosystem and human health. The transmission process of water pollutants is a complex dynamic system, affected by various factors such as hydrodynamics and pollutant diffusion. Traditional water quality models usually simulate based on simple mathematical formulas or empirical formulas, such as the classical Advection-Diffusion Equation, and are unable to accurately describe the complex interactions between basin nodes and their impact on water quality changes. Especially in large-scale basins, the transmission path and time delay of pollutants from upstream to downstream are important factors affecting water quality. Specifically, it is manifested in the following aspects:

[0003] 1. Model simplification and insufficient accuracy:

[0004] Traditional water quality models usually simulate based on simple mathematical formulas or empirical formulas, such as the classical Advection-Diffusion Equation. Although these models can provide certain reference values under certain specific conditions, in the complex and changing actual environment, it is often difficult to accurately describe the interactions between basin nodes and their impact on water quality changes. Specifically:

[0005] Based on linear assumptions: Many traditional models assume that the diffusion and degradation processes of pollutants are linear, ignoring the possible non-linear effects in the actual process. For example, the degradation rate of some organic pollutants in water may vary significantly with the change of concentration.

[0006] Ignoring spatio-temporal heterogeneity: Traditional models usually assume that water quality parameters are uniformly distributed in space and time, which is obviously unrealistic in large-scale basins. In fact, the water flow velocity, topography, and human activities in different regions will lead to spatial heterogeneity and temporal fluctuations of water quality parameters.

[0007] 2. Lack of comprehensive consideration from upstream to downstream:

[0008] In large-scale basins, the transmission path and time delay of pollutants from upstream to downstream are important factors affecting water quality. However, existing water quality models usually only focus on the water quality changes in local areas and fail to fully consider the complex interactions between upstream and downstream:

[0009] Unclear transmission path: Many traditional models ignore the transmission path of pollutants between different watershed nodes, resulting in the inability to accurately predict the time and concentration of pollutants reaching the downstream.

[0010] Time delay not considered: Due to the influence of factors such as water flow velocity and terrain, pollutants will experience different time delays during the process of transmission from the upstream to the downstream. Traditional models usually fail to fully consider this factor, resulting in inaccurate prediction results.

[0011] Therefore, how to provide a water environment automatic monitoring and early warning method that can improve the effectiveness of water environment monitoring is an urgent problem to be solved by those skilled in the art. Summary of the Invention

[0012] In view of this, the present invention proposes a water environment automatic monitoring and early warning method, which is based on the water environment change state space model to simulate the hydrodynamic and water quality spatio-temporal distribution fields of river nodes. On this basis, a feed-forward deep neural network is used to train the water quality data to predict the future water quality change trend and conduct real-time early warning. It can more accurately simulate the propagation process of pollutants in the water body, improving the accuracy and reliability of water quality prediction.

[0013] To achieve the above object, the present invention adopts the following technical solutions:

[0014] The present invention discloses a water environment automatic monitoring and early warning method, including the following steps:

[0015] Determine the upstream watershed nodes of the current watershed node based on the watershed node relationship network; construct a water environment change state space model for simulating the hydrodynamic and water quality spatio-temporal distribution fields of river nodes;

[0016] Obtain the historical water quality pollution data of the watershed node relationship network, and calculate the time interval for pollutants from the upstream watershed nodes to reach the current watershed node based on the water environment change state space model; extract the historical water quality pollution data of the current watershed node within the time interval;

[0017] Construct a feed-forward deep neural network, using the historical water quality pollution data of the upstream watershed node at the starting point of the time interval as the input and the historical water quality pollution data of the current watershed node at the end point of the time interval as the output to obtain a trained feed-forward deep neural network;

[0018] Use the trained feed-forward deep neural network, with the water environment data of the upstream watershed node and the current watershed node at the current moment as the input, to predict the water environment data of the current watershed node at the next moment, and conduct real-time early warning according to the prediction result.

[0019] Preferably, the water environment change state space model is constructed based on the hydrodynamic equation and the water quality model equation, including the following steps:

[0020] The hydrodynamic equation is as follows:

[0021]

[0022] Where ρ represents the density of the basin fluid, u represents the velocity vector, t represents time, p represents pressure, v represents the kinematic viscosity, is the divergence operator, and g represents the gravitational acceleration vector;

[0023] Water quality model equation:

[0024]

[0025] Where C represents the pollutant concentration in the basin, D represents the diffusion coefficient matrix, and R ( C ) includes all source-sink terms and reaction terms.

[0026] Preferably, the upstream basin node is an upstream dynamic basin node and an upstream static basin node that have a direct connection relationship with the current basin node.

[0027] Preferably, the historical water quality pollution data is multi-source heterogeneous data, including online sensor data and laboratory sampling and analysis data, and further includes an assimilation step for the multi-source heterogeneous data:

[0028] According to the data formats of the multi-source heterogeneous data, perform structural parsing on the online sensor data and the laboratory sampling and analysis data respectively, and classify them according to the parsed data structures;

[0029] Extract key features for each type of data respectively, and screen all key features that belong to the same target object in all classifications to obtain a key feature set;

[0030] Establish the corresponding relationships of all key features in the key feature set, and generate a database table containing all associated key features;

[0031] Based on the association relationships in the database table, use a particle filter to fuse the key features of the same target object.

[0032] Preferably, the target object includes any one or a combination of the following: a preset geographical location, a preset water quality parameter, a preset type of pollutant or substance, a preset ecological environment index.

[0033] Preferably, the following steps are further included:

[0034] Based on the basin node relationship network, determine the direct upstream node and the indirect upstream node of the current river node;

[0035] Calculate the time intervals for pollutants from the directly upstream watershed node and the indirectly upstream watershed node to reach the current watershed node based on the water environment change state space model;

[0036] Using the historical water quality pollution data of the directly upstream watershed node at the start of the time interval and the current watershed node as input, and the historical water quality pollution data of the current watershed node at the end of the time interval as output, train to obtain a first feedforward deep neural network; using the historical water quality pollution data of the indirectly upstream watershed node, the directly upstream watershed node, and the current watershed node at the start of the time interval as input, and the historical water quality pollution data of the current watershed node at the end of the time interval as output, train to obtain a second feedforward deep neural network;

[0037] Input the water quality pollution data of the directly upstream watershed node and the current watershed node at time t - 1 into the first feedforward deep neural network to obtain the first water quality pollution data of the current watershed node at time t; input the water quality pollution data of the indirectly upstream watershed node, the directly upstream watershed node, and the current watershed node at time t - 1 into the second feedforward deep neural network to obtain the second water quality pollution data of the current watershed node at time t; extract the water quality pollution data of the current watershed node at time t, and calculate its relative errors with the first water quality pollution data and the second water quality pollution data respectively, and determine the optimal feedforward deep neural network according to the error results;

[0038] Using the optimal feedforward deep neural network, with the water environment data of the upstream watershed node and the current watershed node at the current moment as input, predict the water environment data of the current watershed node at the next moment, and conduct real-time early warning according to the prediction results.

[0039] Preferably, the feedforward deep neural network includes: an input layer, and a plurality of hidden layers connected to the input layer in sequence, the hidden layer includes a fully connected layer and a dropout layer connected in sequence; the last hidden layer is connected to a single-node linear layer, and the output data of the single-node linear layer is input into the risk function.

[0040] Preferably, the risk function adopts the Cox proportional hazards model, and in the Cox proportional hazards model, the individual risk score and the baseline risk function are used to predict the probability of pollutant concentration exceeding the standard at the current watershed node at future time points.

[0041] Preferably, it further includes: inputting the water environment data of the upstream watershed node at the current moment into the water environment change state space model, calculating the time interval for pollutants from the upstream watershed node to reach the current watershed node, and conducting early warning at the same time.

[0042] Preferably, the historical water quality pollution data includes any combination of the following: dissolved oxygen DO, chemical oxygen demand COD, ammonia nitrogen NH3-N, total phosphorus TP, pH value, heavy metal concentration, pesticide residue, microbial index, suspended particulate matter.

[0043] As can be seen from the above technical solutions, compared with the prior art, the beneficial effects of the present invention include:

[0044] The present invention provides an efficient and accurate automatic monitoring and early warning solution for water environment. First, by analyzing data from different sources and using advanced data fusion technologies such as particle filters, the online sensor data and laboratory sampling analysis data are effectively integrated, improving the consistency and accuracy of the data. Second, the water environment change state space model constructed based on the hydrodynamic equation and water quality model equation can accurately simulate the hydrodynamic and water quality spatio-temporal distribution fields of river nodes, providing reliable data support for subsequent water quality prediction. Finally, by introducing a feedforward deep neural network, accurate prediction of future water quality change trends is achieved, and combined with a real-time early warning mechanism, the ability of water resource management and environmental protection is improved.

[0045] In summary, the present invention not only solves the problem of multi-source heterogeneous data fusion, but also improves the accuracy and reliability of water quality prediction. By combining low-frequency high-precision laboratory data and high-frequency low-precision online sensor data, the effective utilization and complementarity of the data are realized. At the same time, using advanced calculation models and technical means, the propagation process of pollutants in water can be simulated more accurately, providing strong support for water resource management and environmental protection. This solution integrating multiple advanced technologies is expected to become an important development direction in the future water quality monitoring field, promoting the water quality monitoring technology to a higher level. Description of the Drawings

[0046] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only the embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained according to the provided drawings;

[0047] Figure 1 It is a flowchart of the automatic monitoring and early warning method for water environment provided by the embodiment of the present invention;

[0048] Figure 2 It is a flowchart of assimilating the multi-source heterogeneous data provided by the embodiment of the present invention;

[0049] Figure 3 It is an architecture diagram of the feedforward deep neural network provided by the embodiment of the present invention. Detailed implementation mode

[0050] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0051] The embodiment of the present invention discloses an automatic monitoring and early warning method for water environment, including the following steps:

[0052] Based on the watershed node relationship network, determine the upstream watershed nodes of the current watershed node; construct a water environment change state space model for simulating the spatio-temporal distribution fields of the hydrodynamic and water quality of river nodes;

[0053] Obtain the historical water quality pollution data of the watershed node relationship network, and calculate the time interval for pollutants from the upstream watershed nodes to reach the current watershed node based on the water environment change state space model; extract the historical water quality pollution data of the current watershed node within the time interval;

[0054] Construct a feedforward deep neural network, using the historical water quality pollution data of the upstream watershed nodes at the starting point of the time interval as the input and the historical water quality pollution data of the current watershed node at the end point of the time interval as the output, to obtain the trained feedforward deep neural network;

[0055] Use the trained feedforward deep neural network, with the water environment data of the upstream watershed nodes and the current watershed node at the current moment as the input, predict the water environment data of the current watershed node at the next moment, and perform real-time early warning according to the prediction results.

[0056] In one embodiment, it is necessary to construct a watershed node relationship network, which can clearly represent the connectivity between each watershed node. The specific steps are as follows:

[0057] Data collection: Obtain basic data such as the terrain and hydrology of the watershed from a Geographic Information System (GIS), including river network diagrams, watershed boundaries, etc.

[0058] Node definition: Define each monitoring site or key location as a watershed node, and establish connections according to their upstream and downstream relationships. For example, assume that A, B, and C are three consecutive monitoring sites, and A is upstream of B, and B is upstream of C, then A can be defined as the direct upstream node, B as the current node, and C as the downstream node. The following is an example:

[0059] Suppose there are 5 monitoring points A, B, C, D, and E in the basin. Among them, A and B are the direct upstream nodes of C, while D and E are the direct upstream nodes of A and B respectively. Then the upstream nodes of C include A, B, D, and E, where A and B are the direct upstream nodes, and D and E are the indirect upstream nodes.

[0060] In one embodiment, a state space model of water environment change is constructed based on the hydrodynamic equation and the water quality model equation, including the following steps:

[0061] The hydrodynamic equation is:

[0062]

[0063] In the formula, ρ represents the density of the basin fluid, u represents the velocity vector, t represents time, p represents pressure, v represents the kinematic viscosity, is the divergence operator, and g represents the gravitational acceleration vector;

[0064] The water quality model equation:

[0065]

[0066] In the formula, C represents the concentration of basin pollutants, D represents the diffusion coefficient matrix, and R ( C ) includes all source-sink terms and reaction terms, such as biodegradation, photolysis, sedimentation, etc.

[0067] In order to simulate the hydrodynamic and water quality spatio-temporal distribution fields of river nodes, numerical solution methods are needed to solve the above equations. Commonly used numerical methods include the finite difference method, the finite element method, and the finite volume method. A specific implementation example is provided below:

[0068] To simulate the hydrodynamic and water quality spatio-temporal distribution fields of a certain river node, the specific steps are as follows:

[0069] Data preparation: Collect the topographic data, hydrological data (such as flow rate, water level), meteorological data (such as rainfall, wind speed), and historical water quality data (such as dissolved oxygen DO, chemical oxygen demand COD, ammonia nitrogen NH3-N, etc.) of this river node.

[0070] Grid division: Divide the river node and its upstream and downstream areas into several grid cells, and determine the appropriate grid size and time step.

[0071] Model setting: Set the initial conditions (such as initial flow velocity, initial pollutant concentration) and boundary conditions (such as inlet flow rate, outlet pressure, pollutant input amount).

[0072] Simulation operation: Run the numerical model to gradually calculate the fluid density, velocity, and pollutant concentration at different times and spatial positions.

[0073] Result analysis: According to the simulation results, spatial distribution maps of fluid density, velocity, and pollutant concentration are plotted, and their changing trends are analyzed. For example, the process of pollutant propagation from the upstream to the downstream can be observed, and the water quality changes within different time periods can be evaluated.

[0074] In one embodiment, the upstream basin nodes are the upstream dynamic basin nodes and upstream static basin nodes that have a direct connection relationship with the current basin node. The dynamic basin nodes include river basins, river channels, etc., and the static basin nodes include lakes, reservoirs, etc.

[0075] In one embodiment, traditional water quality monitoring methods mainly rely on laboratory sampling and analysis. Although this method can provide high-precision data, due to its high cost, low frequency, and long time cycle required to process samples, it is difficult to meet the needs of real-time monitoring. In addition, laboratory sampling usually can only cover limited monitoring points and cannot comprehensively reflect the water quality status of the entire basin. With the development of sensor technology and the Internet of Things (IoT), online water quality monitoring systems have gradually become popular. These systems can achieve high-frequency real-time monitoring of important water quality parameters such as dissolved oxygen (DO), chemical oxygen demand (COD), ammonia nitrogen (NH3-N), total phosphorus (TP), and pH value through various sensors deployed in the water body. However, although online sensors provide a large amount of real-time data, their measurement accuracy is relatively low and is easily affected by the external environment, resulting in large data fluctuations. Therefore, historical water quality pollution data is multi-source heterogeneous data, including online sensor data and laboratory sampling analysis data, as well as the assimilation steps for multi-source heterogeneous data:

[0076] According to the data formats of the multi-source heterogeneous data, the online sensor data and laboratory sampling analysis data are respectively subjected to structural parsing, and classification is performed according to the parsed data structures.

[0077] Key feature extraction is performed on each type of data respectively, and all key features belonging to the same target object in all classifications are screened to obtain a key feature set.

[0078] The corresponding relationships of all key features in the key feature set are established to generate a database table containing all related key features.

[0079] Based on the association relationships in the database table, a particle filter is used to fuse all key features of the same target object.

[0080] In this embodiment, the specific steps of applying a particle filter for data fusion are as follows:

[0081] Initialization: Generate a set of weighted particles based on the association relationships in the database tables in the key feature set. For example, assuming two sets of data (online sensor data and laboratory sampling and analysis data), a set of particles can be generated for each target object.

[0082] Prediction phase: Predict the state of the particles at the next moment according to the dynamic models of the system (such as hydrodynamic equations and water quality model equations). For example, use the hydrodynamic equations to predict the changes in water flow velocity and pressure, and use the water quality model equations to predict the changes in pollutant concentrations.

[0083] Update phase: Update the weights of the particles according to the new observation data (such as newly collected online sensor data or laboratory sampling and analysis data). For example, calculate the similarity between each particle and the observation data and adjust the weights of the particles.

[0084] Resampling: Resample according to the weights of the particles to generate a new set of particles. For example, delete the particles with lower weights and copy the particles with higher weights to ensure that the particle set is closer to the real state.

[0085] Output the fusion result: Calculate the fused water quality data according to the particle set after resampling.

[0086] In one embodiment, the target object includes any one or a combination of the following: a preset geographical location, a preset water quality parameter, a preset type of pollutant or substance, and a preset ecological environment index.

[0087] In one embodiment, the following steps are further included:

[0088] Determine the direct upstream node and the indirect upstream node of the current river node based on the basin node relationship network;

[0089] Calculate the time interval for pollutants from the direct upstream basin node and the indirect upstream basin node to reach the current basin node based on the water environment change state space model;

[0090] Use the historical water quality pollution data of the direct upstream basin node at the starting point of the time interval and the current basin node as the input, and the historical water quality pollution data of the current basin node at the end point of the time interval as the output to train and obtain a first feedforward deep neural network; use the historical water quality pollution data of the indirect upstream basin node, the direct upstream basin node, and the current basin node at the starting point of the time interval as the input, and the historical water quality pollution data of the current basin node at the end point of the time interval as the output to train and obtain a second feedforward deep neural network;

[0091] Input the water quality pollution data of the direct upstream watershed node and the current watershed node at time t-1 into the first feedforward deep neural network to obtain the first water quality pollution data of the current watershed node at time t; input the water quality pollution data of the indirect upstream watershed node, the direct upstream watershed node and the current watershed node at time t-1 into the second feedforward deep neural network to obtain the second water quality pollution data of the current watershed node at time t; extract the water quality pollution data of the current watershed node at time t, and calculate the relative errors between it and the first water quality pollution data and the second water quality pollution data respectively, and determine the optimal feedforward deep neural network according to the error results.

[0092] Use the optimal feedforward deep neural network, with the water environment data of the upstream watershed node and the current watershed node at the current moment as the input, predict the water environment data of the current watershed node at the next moment, and conduct real-time early warning according to the prediction results.

[0093] It can be understood that due to the phenomena of biodegradation, photolysis and sedimentation among watershed nodes, the influence of different watersheds by upstream nodes may be different. Moreover, the possible pollution sources of watershed nodes in different geographical locations may be different, resulting in differences in the decomposition of corresponding pollutants. Therefore, it is necessary to consider the influence of the current watershed node by the direct upstream watershed node and the indirect upstream watershed node, and then select a more appropriate feedforward deep neural network according to different actual situations.

[0094] In one embodiment, the feedforward deep neural network includes: an input layer, and a plurality of hidden layers sequentially connected to the input layer, and the hidden layer includes a fully connected layer and a dropout layer sequentially connected; the last hidden layer is connected to a single-node linear layer, and the output data of the single-node linear layer is input to the risk function.

[0095] In one embodiment, the risk function adopts the Cox proportional hazards model, and in the Cox proportional hazards model, the individual risk score and the baseline risk function are used to predict the probability of pollutant concentration exceeding the standard at the current watershed node at a future time point.

[0096] In one embodiment, it further includes: inputting the water environment data of the upstream watershed node at the current moment into the water environment change state space model, calculating the time interval for pollutants in the upstream watershed node to reach the current watershed node, and conducting early warning at the same time.

[0097] In one embodiment, the historical water quality pollution data includes any combination of the following: dissolved oxygen DO, chemical oxygen demand COD, ammonia nitrogen NH3-N, total phosphorus TP, pH value belonging to on-line sensor data, heavy metal concentration, pesticide residue, microbial index, suspended particulate matter belonging to laboratory sampling analysis data.

[0098] Through the above detailed implementation details and examples, it can be seen that the present invention provides an efficient and accurate automatic water environment monitoring and early warning method, which can significantly improve the accuracy and reliability of water quality prediction and provide strong support for water resource management and environmental protection.

[0099] The above has introduced in detail the automatic water environment monitoring and early warning method provided by the present invention. In this embodiment, specific examples are used to elaborate on the principle and implementation manner of the present invention. The description of the above embodiments is only used to help understand the method and its core idea of the present invention; at the same time, for those of ordinary skill in the art, according to the idea of the present invention, there will be changes in the specific implementation manner and application scope. In summary, the content of this specification should not be construed as a limitation to the present invention.

[0100] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present invention. Various modifications to these embodiments will be obvious to those skilled in the art. The general principles defined in this embodiment can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to these embodiments shown in this embodiment, but will conform to the widest scope consistent with the principles and novel features disclosed in this embodiment.

Claims

1. A method for automatic monitoring and early warning of water environment, characterized in that: The steps include: Determine the upstream basin node of the current basin node based on the basin node relationship network; construct a water environment change state space model to simulate the spatiotemporal distribution field of the hydrodynamics and water quality of the river node; Obtain the historical water quality pollution data of the river basin node relationship network, and calculate the time interval for pollutants from upstream river basin nodes to reach the current river basin node based on the water environment change state space model; extract the historical water quality pollution data of the current river basin node within the time interval; Constructing a feedforward deep neural network, taking the historical water quality pollution data of the upstream river basin node at the starting point of the time interval as input and the historical water quality pollution data of the current river basin node at the end point of the time interval as output, to obtain a trained feedforward deep neural network; The trained feedforward deep neural network is used to predict the water environment data of the current river basin node at the next moment by taking the water environment data of the upstream river basin node and the current river basin node at the current moment as input, and a real-time warning is issued based on the prediction results.

2. A method for automatic monitoring and early warning of water environment according to claim 1, characterized in that: The water environment change state space model is constructed based on the hydrodynamic equation and the water quality model equation, including the following steps: The hydrodynamic equation is: In the formula, ρ represents the density of the fluid in the flow field, u represents the velocity vector, t represents time, p represents pressure, and v represents kinematic viscosity. is the divergence operator, g represents the gravitational acceleration vector; Water quality model equation: In the formula, C represents the pollutant concentration in the watershed, D represents the diffusion coefficient matrix, and R ( C ) Contains all source, sink and reaction terms.

3. A method for automatic monitoring and early warning of water environment according to claim 1, characterized in that: The upstream watershed node is an upstream dynamic watershed node and an upstream static watershed node that are directly connected to the current watershed node.

4. The method for automatic monitoring and early warning of water environment according to claim 1, characterized in that: The historical water pollution data is multi-source heterogeneous data, including online sensor data and laboratory sampling and analysis data, and also includes the assimilation step of the multi-source heterogeneous data: According to the data format of multi-source heterogeneous data, the online sensor data and the laboratory sampling and analysis data are respectively structurally analyzed, and classified according to the analyzed data structure; Extract key features for each type of data, and filter all key features belonging to the same target object in all categories to obtain a key feature set; Establishing the corresponding relationship between all key features in the key feature set, and generating a database table containing all related key features; Based on the association relationship in the database table, the particle filter is used to fuse all the key features of the same target object.

5. A method for automatic monitoring and early warning of water environment according to claim 4, characterized in that: The target object includes any one or more combinations of the following: a preset geographical location, a preset water quality parameter, a preset type of pollutant or substance, and a preset ecological environment indicator.

6. The method for automatic monitoring and early warning of water environment according to claim 1, characterized in that: The following steps are also included: Determine the direct upstream nodes and indirect upstream nodes of the current river node based on the watershed node relationship network; Calculate the time interval for pollutants from the direct upstream river basin node and the indirect upstream river basin node to reach the current river basin node based on the water environment change state space model; Taking the historical water quality pollution data of the direct upstream river basin node and the current river basin node at the starting point of the time interval as input, and the historical water quality pollution data of the current river basin node at the end point of the time interval as output, a first feedforward deep neural network is obtained by training; taking the historical water quality pollution data of the indirect upstream river basin node, the direct upstream river basin node and the current river basin node at the starting point of the time interval as input, and the historical water quality pollution data of the current river basin node at the end point of the time interval as output, a second feedforward deep neural network is obtained by training; Input the water quality pollution data of the direct upstream river basin node and the current river basin node at time t-1 into the first feedforward deep neural network to obtain the first water quality pollution data of the current river basin node at time t; input the water quality pollution data of the indirect upstream river basin node, the direct upstream river basin node and the current river basin node at time t-1 into the second feedforward deep neural network to obtain the second water quality pollution data of the current river basin node at time t; extract the water quality pollution data of the current river basin node at time t, and calculate the relative error between it and the first water quality pollution data and the second water quality pollution data, and determine the optimal feedforward deep neural network according to the error results; The optimal feedforward deep neural network is used to take the water environment data of the upstream river basin node and the current river basin node at the current moment as input to predict the water environment data of the current river basin node at the next moment, and to issue a real-time warning based on the prediction results.

7. A method for automatic monitoring and early warning of water environment according to claim 1 or 6, characterized in that: The feedforward deep neural network includes: an input layer, and multiple hidden layers connected to the input layer in sequence, the hidden layers include a fully connected layer and a dropout layer connected in sequence; the last hidden layer is connected to a single-node linear layer, and the output data of the single-node linear layer is input into the risk function.

8. The method for automatic monitoring and early warning of water environment according to claim 7, characterized in that: The risk function adopts the Cox proportional risk model, in which the individual risk score and the benchmark risk function are used to predict the probability of the pollutant concentration exceeding the standard at the current watershed node at a future time point.

9. The method for automatic monitoring and early warning of water environment according to claim 1, characterized in that: Also includes: The water environment data of the upstream river basin node at the current moment is input into the water environment change state space model, the time interval for the pollutants of the upstream river basin node to reach the current river basin node is calculated, and an early warning is issued at the same time.

10. The method for automatic monitoring and early warning of water environment according to claim 1, characterized in that: The historical water pollution data include any combination of the following: dissolved oxygen DO, chemical oxygen demand COD, ammonia nitrogen NH3-N, total phosphorus TP, pH value, heavy metal concentration, pesticide residues, microbial indicators, and suspended particulate matter.

Images