Method and system for predicting pollutant drift and diffusion in estuary areas

Abstract

The present invention relates to the field of pollutant detection, and provides a method and system for predicting the drift and diffusion of pollutants in an estuary area, including: S1: obtaining pollutant concentration background field data, wind field background field data, and river monitoring data in the estuary area; S2: predicting the river monitoring data through deep learning to obtain pollutant concentration; S3: predicting the river monitoring data to obtain predicted flow; S4: inputting the predicted flow and wind field background field data into a wind-wave-current coupling model to obtain atmospheric drive and flow field drive; S5: inputting pollutant concentration background field data, flow field drive, atmospheric drive, and pollutant concentration into a three-dimensional pollutant transport model for prediction, and obtaining the prediction result of pollutant drift and diffusion in the estuary area. The present invention fully considers the efficiency and quality of the prediction, and combined with the wind-wave-current coupling model, can further improve the speed and accuracy of the three-dimensional pollutant transport model, and effectively improves the timeliness and accuracy of the numerical prediction of pollutant drift and diffusion in the estuary area.

Description

Technical Field

[0001] The present invention relates to the field of pollutant detection, and in particular to a method and system for predicting the drift and diffusion of pollutants in an estuary area. Background Art

[0002] In recent years, with the development of the economy, the pollution problem in estuaries has become increasingly prominent. The simulation of pollutant drift and diffusion in estuaries can provide theoretical guidance, technical reference and data support for estuary pollutant prediction, emergency response and environmental damage assessment, and plays a vital role in preventing and controlling estuary pollution. On the one hand, the traditional simulation of pollutant drift and diffusion in estuaries has a low prediction timeliness due to the huge amount of data. On the other hand, since the current in the marine environment is often accompanied by wind and waves during the propagation process, only considering wind, waves and currents without considering the mutual influence of the three will lead to limited accuracy of the prediction results of the three-dimensional transport model of pollutants.

[0003] The above contents are only used to assist in understanding the technical solution of the present invention and do not constitute an admission that the above contents are prior art. Summary of the invention

[0004] In order to solve the above technical problems, the present invention provides a method for predicting the drift and diffusion of pollutants in an estuary area, comprising:

[0005] S1: Obtain pollutant concentration background field data, wind field background field data and river monitoring data in the estuary area;

[0006] S2: predicting the river monitoring data through deep learning to obtain pollutant concentrations;

[0007] S3: predicting the river monitoring data by using EMD empirical mode decomposition method and LSTM neural network to obtain predicted flow;

[0008] S4: inputting the predicted flow and the wind field background data into a wind-wave-current coupling model to obtain atmospheric driving and flow field driving;

[0009] S5: Input the pollutant concentration background field data, the flow field drive, the atmospheric drive and the pollutant concentration into the pollutant three-dimensional transport model for prediction, and obtain the prediction result of pollutant drift and diffusion in the estuary area.

[0010] Preferably, in step S2, the pollutants in the pollutant concentration include: ammonia nitrogen, total nitrogen, total phosphorus, nitrate nitrogen, TOC, COD, nitrite, hexavalent chromium, copper and zinc.

[0011] Preferably, step S3 specifically comprises:

[0012] S31: decomposing the river monitoring data by the EMD empirical mode decomposition method to obtain data components; normalizing each of the data components to obtain corresponding normalized data components;

[0013] S32: Training each of the normalized data components through the LSTM neural network to obtain corresponding trained data components;

[0014] S33: Denormalize and reconstruct each of the trained data components to obtain the predicted flow.

[0015] Preferably, step S31 is specifically as follows:

[0016] S311: The river monitoring data at time t is represented as X(t), where t is the monitoring time;

[0017] Find all the maximum points of X(t), and fit all the maximum points to form the upper envelope through the cubic spline interpolation function;

[0018] Find all the minimum points of X(t), and fit all the minimum points to form the lower envelope through the cubic spline interpolation function;

[0019] The mean of the upper envelope and the lower envelope is denoted as ml;

[0020] S312: The expression of data component is as follows:

[0021] r t =X(t)-ml

[0022] Among them, r t is the data component at time t;

[0023] S313: normalize each of the data components to obtain a corresponding normalized data component; R t is the normalized data component at time t.

[0024] Preferably, step S32 is specifically:

[0025] Inputting the normalized data component into the LSTM neural network, the LSTM neural network comprising: a forget gate, an input gate and an output gate;

[0026] The normalized data component R t Construct the weight matrices of the forget gate, input gate, and output gate;

[0027] The expression of the forget gate is as follows:

[0028] f t =σ(W f ·[h f(t-1),x ft ]+b f )

[0029] Where, t is the monitoring time; f t is the forget gate at time t; σ is the sigmoid function; W f is the weight matrix of the forget gate; weight matrix W f By matrix W fh and the matrix W fx spliced ​​together; h f(t-1) is the matrix W fh The corresponding output item at time t-1; x ft is the matrix W fx The corresponding output item at time t; b f is the bias term of the forget gate;

[0030] The expression of the input gate is as follows:

[0031] i t =σ(W c ·[h c(t-1) ,x ct ]+b c )

[0032] Among them, i t is the input gate at time t; W c is the weight matrix of the input gate; weight matrix W c By matrix W ch and the matrix W cx spliced ​​together; h c(t-1) is the matrix W ch The corresponding output item at time t-1; x ct is the matrix W cx The corresponding output item at time t; b c is the bias term of the input gate;

[0033] Input unit state at time t The expression is as follows:

[0034]

[0035] in, z is an arbitrary value;

[0036] The output unit state C at time t t The expression is as follows:

[0037]

[0038] Among them, C t-1 is the output unit state at time t-1; the symbol ⊙ represents element-wise multiplication;

[0039] The expression of the output gate is as follows:

[0040] O t =σ(W o ·[h o(t-1) ,x ot ]+b o )

[0041] Among them, O t is the output gate at time t, W o is the weight matrix of the input gate; weight matrix W o By matrix W oh and the matrix W ox spliced ​​together; h o(t-1) is the matrix W oh The corresponding output item at time t-1; x ot is the matrix W ox The corresponding output item at time t; b o is the bias term of the output gate;

[0042] The expression of the trained data component is as follows:

[0043] h t =O t ⊙tanh(C t )

[0044] Among them, h t Represents the trained data component at time t.

[0045] Preferably, step S4 is specifically:

[0046] S41: The wind-wave-current coupling model includes: an atmospheric model WRF, an ocean model FVCOM and a wave model SWAN; the wind field background field data is input into the atmospheric model WRF to obtain atmospheric driving;

[0047] S42: inputting the atmospheric drive into the ocean model FVCOM and the wave model SWAN, the ocean model FVCOM will transmit the flow velocity and water level data to the wave model SWAN in real time, and the wave model SWAN will transmit the wave height data, wave direction data, wavelength data and spectrum peak period to the ocean model FVCOM in real time;

[0048] S43: Using the predicted flow, the atmospheric drive, wave height data, wave direction data, wavelength data and spectrum peak period as inputs of the ocean model FVCOM to obtain the flow field drive.

[0049] A pollutant drift and diffusion prediction system for estuary areas, comprising:

[0050] A data acquisition module is used to obtain pollutant concentration background field data, wind field background field data and river monitoring data in the estuary area;

[0051] A pollutant concentration acquisition module, used to predict the river monitoring data through deep learning to obtain pollutant concentration;

[0052] A predicted flow acquisition module is used to predict the river monitoring data through the EMD empirical mode decomposition method and the LSTM neural network to obtain the predicted flow;

[0053] An atmospheric driving and flow field driving acquisition module is used to input the predicted flow and the wind field background field data into a wind-wave-current coupling model to obtain atmospheric driving and flow field driving;

[0054] The estuary pollutant drift and diffusion prediction module is used to input the pollutant concentration background field data, the flow field drive, the atmospheric drive and the pollutant concentration into the pollutant three-dimensional transport model for prediction, so as to obtain the estuary pollutant drift and diffusion prediction results.

[0055] The present invention has the following beneficial effects:

[0056] Different from the traditional deep learning algorithm for data training, the present invention proposes to predict river monitoring data based on the EMD empirical mode decomposition method and LSTM neural network. This method fully considers the efficiency and quality of the prediction, and combined with the wind-wave-current coupling model, it can further improve the speed and accuracy of the three-dimensional pollutant transport model, and effectively improve the timeliness and accuracy of the numerical prediction of pollutant drift and diffusion in the estuary area. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 This is a flow chart of a method according to an embodiment of the present invention;

[0058] Figure 2 This is a system structure diagram of an embodiment of the present invention;

[0059] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0060] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0061] Reference Figure 1 The present invention provides a method for predicting the drift and diffusion of pollutants in an estuary area, comprising:

[0062] S1: Obtain pollutant concentration background field data, wind field background field data and river monitoring data in the estuary area;

[0063] Specifically, through large-scale observation surveys, background field data on pollutant concentrations and wind fields can be obtained, and through monitoring stations, river monitoring data in estuary areas can be obtained;

[0064] S2: Predicting the river monitoring data through deep learning to obtain pollutant concentration; specifically, the pollutant concentration is the predicted pollutant concentration within the next 48 hours;

[0065] S3: predicting the river monitoring data by using the EMD empirical mode decomposition method and the LSTM neural network to obtain a predicted flow; specifically, the predicted flow is the predicted flow within the next 48 hours;

[0066] S4: inputting the predicted flow and the wind field background data into a wind-wave-current coupling model to obtain atmospheric driving and flow field driving;

[0067] S5: inputting the pollutant concentration background field data, the flow field drive, the atmospheric drive and the pollutant concentration into a three-dimensional pollutant transport model for prediction, and obtaining a prediction result of pollutant drift and diffusion in the estuary area;

[0068] Specifically, the three-dimensional pollutant transport model takes into account the impact of the initial state and source sink in the simulation process of offshore pollutants, while retaining the convection and diffusion processes during the change of pollutants, and taking into account the impact of different types of pollutants. The model expression is as follows:

[0069]

[0070] Where t represents the monitoring time, x, y and z represent the spatial coordinates of the three-dimensional transport model of pollutants, p and q represent the numbers of the pollutant types, and C p represents the concentration of the pollutant No. p, u represents the velocity component in the east-west direction, v represents the velocity component in the north-south direction, w represents the velocity component in the vertical direction, and A H represents the diffusion coefficient in the horizontal direction, K H represents the diffusion coefficient in the vertical direction, θ p,q represents the impact of the qth pollutant on the pth pollutant, ∑ q θ p,q Represents the overall impact of other pollutants on pollutant No. p.

[0071] In this embodiment, in step S2, the pollutants of the pollutant concentration include: ammonia nitrogen, total nitrogen, total phosphorus, nitrate nitrogen, TOC, COD, nitrite, hexavalent chromium, copper and zinc;

[0072] Specifically, when other pollutants need to be detected, they can be added according to demand.

[0073] In this embodiment, step S3 is specifically as follows:

[0074] S31: decomposing the river monitoring data by the EMD empirical mode decomposition method to obtain data components; normalizing each of the data components to obtain corresponding normalized data components;

[0075] Specifically, the Empirical Mode Decomposition (EMD) method is based on the following assumptions: the data has at least two extreme values, a maximum value and a minimum value; the local time domain characteristics of the data are uniquely determined by the time scale between the extreme points; if the data has no extreme points but has inflection points, the extreme values ​​can be obtained by differentiating the data once or multiple times, and then the decomposition result can be obtained by integration;

[0076] S32: Training each of the normalized data components through the LSTM neural network to obtain corresponding trained data components;

[0077] S33: Denormalize and reconstruct each of the trained data components to obtain the predicted flow.

[0078] In this embodiment, the EMD empirical mode decomposition method is specifically as follows:

[0079] S311: The river monitoring data at time t is represented as X(t), where t is the monitoring time;

[0080] Find all the maximum points of X(t), and fit all the maximum points to form the upper envelope through the cubic spline interpolation function;

[0081] Find all the minimum points of X(t), and fit all the minimum points to form the lower envelope through the cubic spline interpolation function;

[0082] The mean of the upper envelope and the lower envelope is denoted as ml;

[0083] S312: The expression of data component is as follows:

[0084] r t =X(t)-ml

[0085] Among them, r t is the data component at time t; if there are still negative local maximum values ​​and positive local minimum values, it is necessary to continue to loop the above steps S311-S312 until the requirements are met;

[0086] S313: normalize each of the data components to obtain a corresponding normalized data component; R t is the normalized data component at time t.

[0087] In this embodiment, the Long-Short Term Memory (LSTM) neural network is a special recurrent neural network (RNN);

[0088] Step S32 is specifically as follows:

[0089] Inputting the normalized data component into the LSTM neural network, the LSTM neural network comprising: a forget gate, an input gate and an output gate;

[0090] The normalized data component R t Construct the weight matrices of the forget gate, input gate, and output gate;

[0091] (1) Forget gate, which can help us delete certain data from memory;

[0092] (2) Input gate, which allows us to add some data to memory;

[0093] (3) Output gate, which can be used to generate and output predictions based on input and existing memory;

[0094] LSTM neural network is very suitable for classification, processing and prediction based on time series data. River monitoring data is a time series with strong correlation between the past and the present. The flow and pollutant concentration forecast for the next 48 hours belong to time series forecasting. Since traditional neural networks are not suitable for processing time series data, LSTM neural network will be used to predict the flow and pollutant concentration for the next 48 hours.

[0095] The gate is actually a fully connected layer, which can be expressed as:

[0096] g(x)=σ(Wx+b)

[0097] Where g is the gate, σ is the activation function (sigmoid function), x is the input vector, W is the weight matrix, and b is the bias term;

[0098] LSTM uses two gates to control the content of the cell state. One is the forget gate, which determines the cell state C at the previous moment. t-1 How many are retained to the current moment C t , the other is the input gate, which determines the input x of the network at the current time t t How much is saved to the cell state C t LSTM uses the output gate to control the cell state C t How many outputs are given to the current output value ht of LSTM?

[0099] The expression of the forget gate is as follows:

[0100] f t =σ(W f ·[h f(t-1) ,x ft ]+b f )

[0101] Where, t is the monitoring time; f t is the forget gate at time t; σ is the sigmoid function; W f is the weight matrix of the forget gate; weight matrix W f By matrix W fh and the matrix W fx spliced ​​together; h f(t-1) is the matrix W fh The corresponding output item at time t-1; x ft is the matrix W fx The corresponding output item at time t; b f is the bias term of the forget gate;

[0102] The expression of the input gate is as follows:

[0103] i t =σ(W c ·[h c(t-1) ,x ct ]+b c )

[0104] Among them, i t is the input gate at time t; W c is the weight matrix of the input gate; weight matrix W c By matrix W ch and the matrix W cx spliced ​​together; h c(t-1) is the matrix W ch The corresponding output item at time t-1; x ct is the matrix W cx The corresponding output item at time t; b c is the bias term of the input gate;

[0105] Input unit state at time t The expression is as follows:

[0106]

[0107] in, z is an arbitrary value;

[0108] The output unit state C at time t t The expression is as follows:

[0109]

[0110] Among them, C t-1 is the output unit state at time t-1; the symbol ⊙ represents element-wise multiplication;

[0111] The expression of the output gate is as follows:

[0112] O t =σ(W o ·[h o(t-1) ,x ot ]+b o )

[0113] Among them, O t is the output gate at time t, W o is the weight matrix of the input gate; weight matrix W o By matrix W oh and the matrix W ox spliced ​​together; h o(t-1) is the matrix W oh The corresponding output item at time t-1; x ot is the matrix W ox The corresponding output item at time t; b o is the bias term of the output gate;

[0114] The expression of the trained data component is as follows:

[0115] h t =O t ⊙tanh(C t )

[0116] Among them, h t Represents the trained data component at time t.

[0117] In the specific implementation, in the process of calculating the trained data components, it is necessary to reversely calculate the error term δ value of each neuron of LSTM. The back propagation of the error term also includes two directions: one is the back propagation along time, and the other is to propagate the error term to the upper layer; according to the error term, the gradient of each error term weight is calculated;

[0118] At time t, the trained data component output by LSTM is h t , the error term δ at the corresponding moment t for:

[0119]

[0120] Where E is the error matrix, for the convenience of representation,

[0121]

[0122]

[0123]

[0124]

[0125]

[0126]

[0127]

[0128]

[0129] where δ f,t , δ i,t , δ o,t , respectively represent the error term of the forget gate at time t, the error term of the input gate at time t, the error term of the unit state at time t, and the error term of the output gate at time t;

[0130] To propagate the error term in reverse time is to calculate the error term δ at time t-1. t-1 :

[0131]

[0132] The formula for passing the error term forward to any time k is:

[0133]

[0134] j is the time count, and T represents its corresponding transposed matrix;

[0135] Pass the error to the previous layer:

[0136]

[0137] Among them, l is the number of layers, t is the monitoring time, f is the activation function, and f' is the derivative of the activation function;

[0138] The calculation formula of the weight gradient at time t is as follows:

[0139]

[0140] By adding together the gradients at each moment, we can get the final gradient of the weight.

[0141]

[0142]

[0143] Gradient calculation of the bias term at time t:

[0144]

[0145] By adding the gradients at each moment together, we can get the final gradient of the bias term.

[0146]

[0147] For W fx , W ix , W cx , W ox The weight gradient of , only needs to be calculated directly according to the corresponding error term:

[0148]

[0149] In this embodiment, step S4 is specifically as follows:

[0150] S41: The wind-wave-current coupling model includes: an atmospheric model WRF, an ocean model FVCOM and a wave model SWAN; the wind field background field data is input into the atmospheric model WRF to obtain atmospheric driving;

[0151] S42: inputting the atmospheric drive into the ocean model FVCOM and the wave model SWAN, the ocean model FVCOM will transmit the flow velocity and water level data to the wave model SWAN in real time, and the wave model SWAN will transmit the wave height data, wave direction data, wavelength data and spectrum peak period to the ocean model FVCOM in real time;

[0152] Specifically, the atmospheric model WRF provides atmospheric driving for the ocean model FVCOM and the wave model SWAN. The ocean model FVCOM and the wave model SWAN are directly coupled. The wave model SWAN provides wave height data, wave direction data, wavelength data and spectrum peak period to the ocean model FVCOM, and the ocean model FVCOM provides flow velocity and water level data to the wave model SWAN.

[0153] S43: Using the predicted flow, the atmospheric drive, wave height data, wave direction data, wavelength data and spectrum peak period as inputs of the ocean model FVCOM to obtain the flow field drive.

[0154] Compared with the prior art, the present invention has the following advantages: Different from the traditional deep learning algorithm for data training, the present invention proposes to predict river monitoring data based on the EMD empirical mode decomposition method and the LSTM neural network. This method fully considers the efficiency and quality of the prediction, and combined with the wind-wave-current coupling model, it can further improve the speed and accuracy of the three-dimensional pollutant transport model, and effectively improve the timeliness and accuracy of the numerical prediction of pollutant drift and diffusion in the estuary area.

[0155] The present invention provides a pollutant drift and diffusion prediction system in an estuary area, comprising:

[0156] A data acquisition module is used to obtain pollutant concentration background field data, wind field background field data and river monitoring data in the estuary area;

[0157] A pollutant concentration acquisition module, used to predict the river monitoring data through deep learning to obtain pollutant concentration;

[0158] A predicted flow acquisition module is used to predict the river monitoring data through the EMD empirical mode decomposition method and the LSTM neural network to obtain the predicted flow;

[0159] An atmospheric driving and flow field driving acquisition module is used to input the predicted flow and the wind field background field data into a wind-wave-current coupling model to obtain atmospheric driving and flow field driving;

[0160] The estuary pollutant drift and diffusion prediction module is used to input the pollutant concentration background field data, the flow field drive, the atmospheric drive and the pollutant concentration into the pollutant three-dimensional transport model for prediction, so as to obtain the estuary pollutant drift and diffusion prediction results.

[0161] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or system. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or system including the element.

[0162] The serial numbers of the embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments. In a unit claim that lists several means, several of these means may be embodied by the same hardware item. The use of the words first, second, and third, etc. does not indicate any order and these words may be interpreted as identifiers.

[0163] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for predicting the drift and diffusion of pollutants in an estuary area, characterized in that: include: S1: Obtain pollutant concentration background field data, wind field background field data and river monitoring data in the estuary area; S2: predicting the river monitoring data through deep learning to obtain pollutant concentrations; S3: predicting the river monitoring data by using EMD empirical mode decomposition method and LSTM neural network to obtain predicted flow; S4: inputting the predicted flow and the wind field background data into a wind-wave-current coupling model to obtain atmospheric driving and flow field driving; S5: inputting the pollutant concentration background field data, the flow field drive, the atmospheric drive and the pollutant concentration into a three-dimensional pollutant transport model for prediction, and obtaining a prediction result of pollutant drift and diffusion in the estuary area; Step S4 is specifically as follows: S41: The wind-wave-current coupling model includes: an atmospheric model WRF, an ocean model FVCOM and a wave model SWAN; the wind field background field data is input into the atmospheric model WRF to obtain atmospheric driving; S42: inputting the atmospheric drive into the ocean model FVCOM and the wave model SWAN, the ocean model FVCOM will transmit the flow velocity and water level data to the wave model SWAN in real time, and the wave model SWAN will transmit the wave height data, wave direction data, wavelength data and spectrum peak period to the ocean model FVCOM in real time; S43: using the predicted flow, the atmospheric driving, the wave height data, the wave direction data, the wavelength data and the spectrum peak period as inputs of the ocean model FVCOM to obtain the flow field driving; The expression of the three-dimensional pollutant transport model is: Where t represents the monitoring time, x, y and z represent the spatial coordinates of the three-dimensional transport model of pollutants, p and q represent the numbers of the pollutant types, and C p represents the concentration of the pollutant No. p, u represents the velocity component in the east-west direction, v represents the velocity component in the north-south direction, w represents the velocity component in the vertical direction, and A H represents the diffusion coefficient in the horizontal direction, K H represents the diffusion coefficient in the vertical direction, θ p,q represents the impact of the qth pollutant on the pth pollutant, ∑ q θ p,q Represents the overall impact of other pollutants on pollutant No. p.

2. The method for predicting the drift and diffusion of pollutants in estuary areas according to claim 1 is characterized in that: In step S2, the pollutants of the pollutant concentration include: ammonia nitrogen, total nitrogen, total phosphorus, nitrate nitrogen, TOC, COD, nitrite, hexavalent chromium, copper and zinc.

3. The method for predicting the drift and diffusion of pollutants in estuary areas according to claim 1 is characterized in that: Step S3 is specifically as follows: S31: decomposing the river monitoring data by the EMD empirical mode decomposition method to obtain data components; normalizing each of the data components to obtain corresponding normalized data components; S32: Training each of the normalized data components through the LSTM neural network to obtain corresponding trained data components; S33: Denormalize and reconstruct each of the trained data components to obtain the predicted flow.

4. The method for predicting the drift and diffusion of pollutants in estuary areas according to claim 3 is characterized in that: Step S31 is specifically as follows: S311: The river monitoring data at time t is represented as X(t), where t is the monitoring time; Find all the maximum points of X(t), and fit all the maximum points to form the upper envelope through the cubic spline interpolation function; Find all the minimum points of X(t), and fit all the minimum points to form the lower envelope through the cubic spline interpolation function; The mean of the upper envelope and the lower envelope is denoted as ml; S312: The expression of data component is as follows: r t =X(t)-ml Among them, r t is the data component at time t; S313: normalize each of the data components to obtain a corresponding normalized data component; R t is the normalized data component at time t.

5. The method for predicting the drift and diffusion of pollutants in estuary areas according to claim 3 is characterized in that: Step S32 is specifically as follows: Inputting the normalized data component into the LSTM neural network, the LSTM neural network comprising: a forget gate, an input gate and an output gate; The normalized data component R t Construct the weight matrices of the forget gate, input gate, and output gate; The expression of the forget gate is as follows: f t =σ(W f ·[h f(t-1) ,x ft ]+b f ) Where, t is the monitoring time; f t is the forget gate at time t; σ is the sigmoid function; W f is the weight matrix of the forget gate; weight matrix W f By matrix W fh and the matrix W fx spliced ​​together; h f(t-1) is the matrix W fh The corresponding output item at time t-1; x ft is the matrix W fx The corresponding output item at time t; b f is the bias term of the forget gate; The expression of the input gate is as follows: i t =σ(W c ·[h c(t-1) ,x ct ]+b c ) Among them, i t is the input gate at time t; W c is the weight matrix of the input gate; weight matrix W c By matrix W ch and the matrix W cx spliced ​​together; h c(t-1) is the matrix W ch The corresponding output item at time t-1; x ct is the matrix W cx The corresponding output item at time t; b c is the bias term of the input gate; Input unit state at time t The expression is as follows: in, z is an arbitrary value; The output unit state C at time t t The expression is as follows: Among them, C t-1 is the output unit state at time t-1; the symbol ⊙ represents element-wise multiplication; The expression of the output gate is as follows: The t =σ(W o ·[h o(t-1) ,x ot ]+b o ) Among them, O t is the output gate at time t, W o is the weight matrix of the input gate; weight matrix W o By matrix W oh and the matrix W ox spliced ​​together; h o(t-1) is the matrix W oh The corresponding output item at time t-1; x ot is the matrix W ox The corresponding output item at time t; b o is the bias term of the output gate; The expression of the trained data component is as follows: h t =O t ⊙tanh(C t ) Among them, h t Represents the trained data component at time t.

6. A system for predicting the drift and diffusion of pollutants in an estuary area, used to implement the method for predicting the drift and diffusion of pollutants in an estuary area according to any one of claims 1 to 5, characterized in that: include: A data acquisition module is used to obtain pollutant concentration background field data, wind field background field data and river monitoring data in the estuary area; A pollutant concentration acquisition module, used to predict the river monitoring data through deep learning to obtain pollutant concentration; A predicted flow acquisition module is used to predict the river monitoring data through the EMD empirical mode decomposition method and the LSTM neural network to obtain the predicted flow; An atmospheric driving and flow field driving acquisition module is used to input the predicted flow and the wind field background field data into a wind-wave-current coupling model to obtain atmospheric driving and flow field driving; The estuary pollutant drift and diffusion prediction module is used to input the pollutant concentration background field data, the flow field drive, the atmospheric drive and the pollutant concentration into the pollutant three-dimensional transport model for prediction, so as to obtain the estuary pollutant drift and diffusion prediction results.

Images