River network water pollution tracing method

Abstract

The application discloses a river network water pollution tracing method, comprising the following steps: firstly, acquiring river network data, inputting the river network data into MIKE software, constructing a water dynamic model of the river network and verifying the water dynamic model; then, dividing the whole river network into multiple river sections with water quality monitoring points as nodes, inputting the river network data corresponding to each river section into the verified water dynamic model to obtain simulation data of each river section; inputting the simulation data of each river section into a Gated Recurrent Unit (GRU) after pretreatment to construct a regression model of each river section; and tracing by using the regression model. The application simulates the diffusion of pollutants by using the MIKE software, supplements data under the condition that monitoring data is sparse, traces pollutants by combining the regression model, and uses hydrodynamic simulation data as the input of the regression model, so that the accuracy and density of the data set can be improved, and the accurate positioning of the pollution source and pollution source parameters is realized.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of river network water pollution tracing, and particularly relates to a river network water pollution tracing method. BACKGROUND

[0002] At present, sewage problem is still a major challenge to be solved in environmental pollution problems. If environmental pollution exceeds the environmental capacity, it will cause devastating impact on the local ecological circle, and along with the circulation of matter and energy, it will pose a great threat to people's life. Therefore, it is necessary to accurately locate the source of river network pollutants. If the river network pollutant tracing is simply relied on manual investigation, not only the work is doubled, but also the reliability and timeliness of the tracing result cannot be guaranteed. With the development of artificial intelligence technology, it is of great significance and value to trace the river network pollutants through big data analysis and artificial intelligence algorithm.

[0003] The patent No. CN202311274115.5 discloses a method for constructing a water pollution source fluorescence fingerprint database of an industrial park. The method can construct a pollutant fluorescence fingerprint library according to different fluorescence characteristic peaks of different pollutants, and can help accurately identify pollutants and trace corresponding factory emission sources.

[0004] However, although the invention has high tracing accuracy, it is costly to construct a fingerprint library containing a large number of pollutants, and the fingerprint library needs to be updated constantly with the continuous development of industrialization and the continuous change of raw materials used by factories in industrial parks, which requires a large amount of manpower, material resources and financial resources. Therefore, this method can only play a good role when the pollutants of the sewage enterprises in the tracing area have great differences, and it is difficult to accurately trace when multiple same type enterprises are concentrated in an industrial park.

[0005] Therefore, the application provides a river network water pollution tracing method based on MIKE software and a gated recurrent unit. SUMMARY

[0006] The application aims to overcome the defects of the prior art, and provides a river network water pollution tracing method. The MIKE software is combined with a gated recurrent unit and a regression model to trace the water pollution in reverse. Specifically, the MIKE software is used to simulate the concentration data of the monitoring point under any condition, and then the gated recurrent unit is used for model training to find the concentration correlation between the pollution source in any river section and the monitoring point.

[0007] To achieve the above object, the application provides the following technical scheme.

[0008] A river network water pollution tracing method comprises the following steps:

[0009] Step S1, acquiring river network data; the river network data includes river network geographic data, river network hydrological data, river channel patrol data, river network monitoring point monitoring data, etc.

[0010] Step S2, constructing a hydrodynamic model: input the river network data into the MIKE software to construct a hydrodynamic model of the whole river network.

[0011] Step S3, simulating pollutant diffusion and verifying the model: taking any monitoring point as a pollution source, setting the pollution discharge time, pollution concentration and concentration change interval of the pollution source and inputting the hydrodynamic model to obtain the simulation data of the water quality monitoring points around the pollution source, and verifying the hydrodynamic model with the real monitoring data corresponding to the monitoring points.

[0012] Step S4, constructing a regression model: dividing the whole river network into multiple river sections with monitoring points as nodes, then inputting the river network data corresponding to each river section into the hydrodynamic model to obtain the simulation data of each river section; after preprocessing the simulation data of each river section, inputting it into a gated recurrent unit to construct a regression model for each river section.

[0013] Step S5, river network water pollution source tracing: when any monitoring point detects that the pollutant exceeds the standard, use the crossover and mutation operations in genetic algorithm to randomly change and recombine the pollution source parameters, and input them into the regression model corresponding to the river section where the monitoring point is located to obtain the prediction result of the current monitoring point's excessive pollutant concentration; taking the difference between the prediction result and the real excessive value as an evaluation index, repeating the above steps, and after multiple iterations, returning the optimal prediction result of the current monitoring point's excessive pollutant concentration, which corresponds to the source tracing result.

[0014] The pollution source parameters include the location of the pollution source, the start time of the pollution discharge, the duration of the pollution discharge, and the pollution concentration.

[0015] Preferably, the river network data includes river network geographic data, river network hydrological data, river channel patrol data, and river network water quality monitoring point monitoring data; specifically, the river network geographic data includes river channel latitude and longitude data and artificially measured river channel cross-section structure data, the river network hydrological data includes flow rate, flow direction, water level, etc. data of different river sections, the river channel patrol data includes background value data of four main pollutants in the river network collected by unmanned ships and unmanned aerial vehicles, and the river network water quality monitoring point monitoring data includes detection values of main pollutants at the river network monitoring point at regular intervals.

[0016] Preferably: use MIKE ZERO software, input river network two-dimensional geographic data (latitude and longitude, river direction, width, etc.), hydrological data, cross-section structure, etc. to build the hydrodynamic model of the whole river network and verify it; input specific pollution source characteristic parameters, use the hydrodynamic model to simulate the pollution influence degree, and can get the specific pollutant concentration data under any spatio-temporal condition and any accuracy, and use the background value data of the river network pollutants measured by the unmanned ship and the actual detection data of the monitoring points as the reference index for the correctness of the simulation and simulation, avoiding the influence of factors such as mechanical failure of the actual monitoring station and long monitoring interval on the accuracy of the trained model.

[0017] Preferably: divide the whole river network into multiple river sections with the water quality monitoring points in the river network as the division nodes. Due to the changes of river bed cross-section structure, water level and other factors, different parts of the river network show different flow and velocity, and the same pollution source also shows great difference in pollution in different parts of the river network. If the river network is not divided, a regression model is trained with the data of the whole river network, the model will inevitably try to meet the prediction accuracy of all river sections with different hydrological characteristics, which may eventually result in that only a few river sections have high accuracy or even the prediction accuracy of the whole river network is seriously insufficient.

[0018] Preferably: after dividing the river network into river sections, use the gated recurrent unit to input the simulation data and build independent regression models for each river section, and use the recurrent neural network to help find the corresponding relationship between the pollution source and the monitoring point value under different river section special conditions.

[0019] Preferably: the preprocessing in step S4 includes uniform time unit, removal of interval symbol, identification and removal of abnormal value.

[0020] Preferably: the identification and removal of abnormal value adopts two methods of Z-score (z-score) and IQR (interquartile range, four-part difference).

[0021] Preferably: introduce genetic algorithm, randomly change the pollution source parameters (position, discharge start time, discharge duration, discharge concentration, etc.) within a proper range, which can traverse the complete pollution source situation, avoid the local optimum of the gated recurrent unit, and effectively improve the tracing ability and accuracy of the pollution source.

[0022] Preferably: the simulation data of MIKE software needs to be verified by real detection data, and also needs to be analyzed visually. By calculating the difference between the background value under the same spatio-temporal condition and visualizing it, it can be determined whether the change is consistent with the actual river geographical and hydrological characteristics; in addition, it is also necessary to analyze whether the difference between the data of different monitoring points is obvious enough to be used as the input data set for model training without affecting the accuracy of the final model.

[0023] Technical effects and advantages of the present application:

[0024] 1、Compared with the prior art, the present application uses MIKE software to simulate the diffusion of pollutants, so as to supplement data under the condition of sparse monitoring data. The water dynamic model can be used to calculate the concentration of pollutants under complex conditions, and the real data returned by the sensor of the monitoring point is used as the verification of correctness, so as to overcome the problem of insufficient detection frequency and accuracy of the monitoring point in the real situation. The hydrodynamic simulation data is used as the input of the tracing model, so as to improve the accuracy and density of the data set and help the model to better find the corresponding relationship between the pollution source and the monitoring point data.

[0025] 2、Compared with the prior art, the present application introduces a gating cycle unit for pollution source tracing. By finding the corresponding relationship between the concentration and time of the pollution source and the monitoring point, the concentration and the time of reaching the monitoring point are used to roughly judge the position of the pollution source. Even in the case of multiple similar enterprises gathering in the same industrial park, it is still applicable. And the whole tracing process is automatically completed by the system after the input of the pollution exceeding alarm information, so as to improve the work efficiency and the reliability of the tracing.

[0026] 3、The present application introduces a genetic algorithm, which randomly mutates and recombines the pollution source information and then re-enters the model. After continuous iteration, the best tracing result is obtained, which can effectively avoid the local optimal solution of the tracing model and affect the tracing result.

[0027] 4、The present application divides the whole river network into river sections with the monitoring point as the node. Due to the changes of river bed section structure, water level and other factors, different parts of the river network show different flow and flow rate. The same pollution source will also show great difference in pollution in different parts of the river network. If a regression model is trained with the data of the whole river network, the model will inevitably try to meet the prediction accuracy of all river sections with different hydrological characteristics, which may eventually result in high accuracy of only a few river sections and low prediction accuracy of the whole river network. The present application divides the river network to improve the accuracy of the regression model and finally accurately locates the pollution source. BRIEF DESCRIPTION OF DRAWINGS

[0028] Figure 1 The whole architecture diagram of a river network water pollution tracing method based on MIKE software and gating cycle unit is suitable for the present application.

[0029] Figure 2 The comparison data of the simulation results and the real results of the water dynamic model of the present application.

[0030] Figure 3The visualization data of the simulation result of the water power model of the application.

[0031] Figure 4 The training flowchart of the river section tracing model based on the gated recurrent unit of the application.

[0032] Figure 5 The pollution tracing step diagram based on the gated recurrent unit and the genetic algorithm of the application.

[0033] Figure 6 The tracing result of the method of the application. DETAILED DESCRIPTION

[0034] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by a person of ordinary skill in the art without creative work are within the protection scope of the application.

[0035] EMBODIMENT

[0036] In combination with Figure 1 The embodiment provides a river network water pollution tracing method based on MIKE software and a gated recurrent unit. The tracing method comprises the following steps.

[0037] Step 1: Data collection: the data to be collected comprises river network data of a target region, including river network geographic data, river network hydrological data, river channel patrol data and river network monitoring point monitoring data, specifically, longitude and latitude data of all river channels, riverbed cross-section structure data of all river channels at certain intervals, flow direction, flow rate and water level data of all river water bodies, complete river network pollutant background value data detected by an unmanned ship, and pollutant monitoring data of monitoring points in the river network (in the embodiment, the number of monitoring points is 20, and the types of pollutants are COD, TN, TP and NH3-N).

[0038] Step 2: Establishing a river network hydrodynamic model: extract the longitude and latitude data of all river channels, the riverbed cross-section structure data of all river channels at certain intervals, the flow direction, flow rate and water level data of all river water bodies, and input the data into MIKE software, and then use the software to establish a hydrodynamic model.

[0039] Step three: simulation data generation: due to the long detection interval (4 hours / time) and incomplete detection data (less than 20% of the monitoring points for four pollutants are detected and the detection equipment is not malfunctioning), the pollutant parameters of specific conditions are substituted into the hydrodynamic model in this embodiment to simulate the simulation data, which overcomes the problem of insufficient real data; in this embodiment, any one monitoring point is taken as a hypothetical pollution source, the simulation parameters are: the discharge amount is 500 m 3 / d, the discharge time is 8:00-11:00 every day, the pollution source concentration is 0-20000 mg / L, and the simulation concentration data density is 200 mg / L (the simulation parameters are set according to the nature of the discharge enterprises in the study area, the relationship between water flow and time, and the concentration density required for the modeling of the tracing model, which can be changed according to the actual situation), the above data are input into the hydrodynamic model to obtain the water quality pollution simulation data of four monitoring points in different branches, which are used to establish the tracing model combined with Figure 2 , the real monitoring data is used to verify the accuracy of the simulation data set, in the actual situation, the simulation data error value of the river network water quality monitoring point is less than 15%; combined with Figure 3 , four river channels with large differences in hydrological characteristics are selected, and the corresponding relationship between the simulation concentration and time of the water quality monitoring points in the river channels is visualized and analyzed, the results show that the simulation data of each river channel conforms to the actual corresponding river channel geography and hydrological characteristics, and the data between different monitoring points have obvious differences, which can be used as the input data set for subsequent model training.

[0040] Step four: establishment of the tracing model: before establishing the model, the river network needs to be divided, the water quality monitoring points in the river network are taken as the division nodes in this embodiment, and the whole river network is divided into multiple independent river sections, due to the changes of river bed section structure, water level and other factors, different parts of the river network show different flow and velocity, a unique regression model is trained for each river section, which can prevent the model from sacrificing part of the river section to meet the overall river network prediction accuracy. In this embodiment, a gated recurrent unit is selected to construct the tracing model, combined with Figure 4 , the training process of the tracing model of each river section is: the water quality simulation data of the river channel after division is subjected to time unit unification, interval symbol removal, abnormal value identification and removal by Z-score and IQR, etc. Pretreatment operation, the data set after pretreatment is divided into training set (80%) and test set (20%), after the operation is completed, the gated recurrent unit is substituted, the independent regression model of each river channel is constructed, the model effect is good, then the model is saved, if the effect is not good, the model parameters are modified to obtain good simulation accuracy.

[0041] The formula of Z-score is:

[0042]

[0043] Wherein, X represents the original data, represents the average number of original data, and sigma represents the standard deviation of original data.

[0044] The formula of IQR is:

[0045]

[0046] Wherein, Q1 represents the lower quartile, that is, the 25th number of all values arranged from small to large in the sample, and Q3 represents the upper quartile, that is, the 75th number of all values arranged from small to large in the sample.

[0047] The core formula of the gated recurrent unit is:

[0048]

[0049] Wherein, z t , r t are the outputs of the update gate and the reset gate respectively, x t is the input at time t, represents the candidate hidden state at time t, h t-1 , h t represent the hidden state at the corresponding time, [] represents the connection of two vectors, W z , W r represent the weight matrix, represents the sigmoid function, represents the tanh activation function, represents the weight matrix, represents the product of matrices.

[0050] Step five: start of the traceability model: combined with Figure 5 , the pollutant traceability steps are as follows:

[0051] (1) The monitoring point detects that the pollutant concentration in the place exceeds the standard, returns information, and triggers the traceability model.

[0052] (2) The traceability model randomly generates a pollution source parameter. In this embodiment, the pollution source parameter includes four items: pollution source position, pollution discharge start time (0:00~8:00), pollution discharge duration (1~4 hours), and pollution discharge concentration (0~20000 mg / L). The pollution source position is randomly selected in the interval between the over-standard monitoring point and the monitoring point closest to it in the upstream different branch of the river, and the remaining three parameters are determined according to the nature of the pollution discharge enterprise, the relationship between water flow and time, and the simulation density of the simulation data.

[0053] (3) The pollution source parameter generated in step (2) is identified and substituted into the regression model of the river section to make a prediction, and a quantitative value of the correctness of the source tracing is returned by subtracting the actual detected value of the excessive pollutant from the predicted value of the excessive pollutant corresponding to the pollution source parameter. If the difference is too large, it proves that the pollution source is quite different from the actual pollution source.

[0054] (4) The pollution source parameter generated in step (2) is randomly changed and recombined by using the crossover and mutation operations in the genetic algorithm, and the change and recombination range is consistent with the parameter range in step (2).

[0055] (5) The new pollution source parameter after mutation in step (4) is returned to step (3).

[0056] (6) An iteration number is set, which is 50 in this embodiment, and can be modified according to the actual situation and the accuracy of source tracing. In the continuous iteration, steps (3) to (5) are repeatedly performed, and finally a minimum difference between the predicted value and the true value is returned. The pollution source parameter corresponding to the predicted value is the result of the pollution source tracing this time, and the source tracing result is finally returned and visualized, and the result is shown in Figure 6 The accuracy of source tracing under different conditions has differences. In this embodiment, the source tracing error of each parameter is about 5%.

[0057] The whole step is automatically completed by the system, which improves the work efficiency and the accuracy of source tracing.

[0058] The above embodiment is the preferred implementation of the present application, but the implementation of the present application is not limited to the above embodiment. Any modification, substitution, combination, simplification made without departing from the spirit and principles of the present application shall be an equivalent replacement, which is included in the protection scope of the present application.

Claims

1. A method for tracing the source of water pollution in a river network, characterized in that: The method comprises the following steps: Step S1, obtaining river network data; Step S2, constructing a hydrodynamic model: inputting the river network data into the MIKE software to construct a hydrodynamic model of the river network; Step S3, verifying the hydrodynamic model: taking any water quality monitoring point in the river network as a pollution source, setting the pollution source's discharge duration, discharge concentration, and concentration change interval and inputting them into the hydrodynamic model, obtaining simulated data from the water quality monitoring points around the pollution source, and verifying the hydrodynamic model using the actual monitoring data corresponding to the water quality monitoring points around the pollution source; Step S4, constructing a regression model: the entire river network is divided into multiple river sections using water quality monitoring points as nodes, and the river network data corresponding to each river section is then input into the verified hydrodynamic model to obtain simulated data for each river section; the simulated data of each river section is pre-processed and then input into the gated recurrent unit to construct a regression model for each river section; Step S5, tracing the source of water pollution in the river network: When any water quality monitoring point detects that the pollutants exceed the standard, the crossover and mutation operations in the genetic algorithm are used to change and reorganize the pollution source parameters, and then input them into the regression model corresponding to the river section where the water quality monitoring point is located to obtain the predicted result of the pollutant concentration exceeding the standard at the current water quality monitoring point; the difference between the predicted result and the actual exceeding value is used as the evaluation index, and the above steps are repeated. After multiple iterations, the optimal predicted result of the pollutant concentration exceeding the standard at the current monitoring point is returned, and the pollution source parameter corresponding to the optimal predicted result is the tracing result.

2. The method for tracing the source of river water pollution according to claim 1, characterized in that: The river network data includes river network geographic data, river network hydrological data, river channel inspection data, and river network water quality monitoring point monitoring data.

3. The method for tracing the source of river water pollution according to claim 1, characterized in that: The preprocessing in step S4 includes unifying time units, removing separators, and identifying and removing outliers.

4. The method for tracing the source of river water pollution according to claim 1, characterized in that: The pollution source parameters include the location of the pollution source, the start time of pollution source emission, the duration of emission, and the emission concentration.

5. A river network water pollution tracing system implementing the method according to any one of claims 1 to 4, characterized in that: Includes the following modules: Data acquisition module, used to obtain river network data; A hydrodynamic model building module is used to build a hydrodynamic model of a river network based on river network data; The regression model building module is used to divide the river network into multiple river sections, input the river network data corresponding to the river sections into the hydrodynamic model, and build the regression model of each river section after the gated circulation unit; The water pollution source tracing module uses regression models to predict pollution source parameters.

6. An electronic device, characterized in that: The method comprises a processor and a memory, wherein the memory stores machine-executable instructions that can be executed by the processor, and the processor executes the machine-executable instructions to implement the method according to any one of claims 1 to 4.

7. A machine-readable storage medium, characterized in that The machine-readable storage medium stores machine-executable instructions. When the machine-executable instructions are called and executed by a processor, the machine-executable instructions prompt the processor to implement the method according to any one of claims 1 to 4.

Images