Rainfall runoff pollution prediction method based on spatiotemporal long short-term memory network and tree structure traversal

Abstract

The present invention provides a rainfall runoff pollution prediction method based on a spatiotemporal long short-term memory network and tree structure traversal, which belongs to a rainfall runoff prediction method of the rainfall runoff type. Prepare rainfall data, runoff data and ground elevation DEM data in the catchment area, select rain type, average rainfall intensity, peak rainfall intensity, rainfall duration, dry period before rain, atmospheric dust fall, PM10, vehicle flow, road surface material and functional area as influencing factors for rainfall runoff pollution prediction, select the average concentration of rainfall runoff pollution in each session and the initial flushing index as decision variables for rainfall runoff pollution prediction; divide the prediction area into basins according to the ground elevation of the catchment area, divide the catchment area into different sub-basins, and obtain the tree structure of the catchment area; construct a rainfall runoff pollution spatiotemporal long short-term memory network; train and test the rainfall runoff spatiotemporal long short-term memory network, and set the model parameters to ensure that the model can accurately predict the rainfall runoff in the catchment area.

Description

Technical Field

[0001] The invention relates to the prediction of rainfall runoff pollution in urban water environment, and in particular to a rainfall runoff pollution prediction method based on spatiotemporal long short-term memory network and tree structure traversal. Background Art

[0002] At present, the commonly used rainfall runoff pollution prediction methods can be divided into two categories according to the technical means adopted: process-driven hydrological models and data-driven models. Each type of method has its own advantages and disadvantages. The main difference between the two types of methods is that the data-driven model excludes the physical aspects of hydrological information and the complex process of rainfall runoff mechanism. Data-driven models are more flexible and can be analyzed using various structures and parameters, making them suitable for multi-disciplinary applications. For example, in terms of runoff, rainfall and water quality prediction, data-driven models are more efficient and targeted. Data-driven models are at the forefront of adaptation, learning, promotion and data processing. Compared with statistical models or physical models, data-driven models may have higher prediction efficiency and accuracy.

[0003] The basic principle of process-driven hydrological models is based on the theory of continuum mechanics. These models require comprehensive field information and are not suitable for applications in complex scenarios. These models require the definition of each sub-process and simulate the actual hydrological process in mathematical form, which can intuitively explain the relationship between input parameters and output variables. Obviously, when there is not enough data to establish the input-output connection, the advantages of process-driven hydrological models are stronger. However, process-driven hydrological models are prone to reduced prediction accuracy due to noise in the data. Noise in the data usually comes from routine maintenance, such as calibration work, data acquisition program errors, faulty data transmission, and other unknown sources. In addition, process-driven hydrological models are also usually limited by spatiotemporal complexity and are affected by model bias.

[0004] Therefore, data-driven methods have better application prospects in the field of rainfall-runoff prediction. In particular, the basic principles of physical models are based on the theory of continuum mechanics. These models require comprehensive field information and are not suitable for complex applications. The purpose is to define subprocesses and simulate actual hydrological processes in mathematical form. It uses a transparent model structure to illustrate the relationship between input parameters and output variables. It has advantages when there is not enough data to establish input-output connections. However, such process-driven hydrological models are prone to defects due to the presence of noise. Noise usually comes from routine maintenance such as calibration work, data acquisition program errors, faulty data transmission, and other unknown sources. Physical models are also limited by spatiotemporal complexity and are affected by model bias. Therefore, physical models are more susceptible to conceptual and systematic errors caused by the uncertainty of replication results. Sometimes, the difficulty of obtaining hydrological data hinders complete physical modeling. Other disadvantages of physical models are higher costs, locality and accuracy issues limited to process detection.

[0005] In recent years, deep learning methods have been widely used in many fields, especially in computer vision and natural language processing. Most deep learning methods have the ability to process long sequence data, such as recursive neural network, which is one of the most popular neural networks in the field of deep learning and has been proven to be an effective tool for processing sequence data for time series prediction. It is worth noting that due to its inherent gradient vanishing or exploding problem, the performance of traditional recursive neural network in sequence prediction has not been significantly improved. As an improved version of recursive neural network, long short-term memory network can effectively solve this problem. Long short-term memory network uses the time series information of rainfall data from various meteorological stations in the catchment area to predict the runoff data in the catchment area.

[0006] However, runoff pollution is not only related to the time dimension of rainfall data at each meteorological station (i.e., the contextual information of rainfall data), but also the runoff in the catchment area is related to the spatial dimension of rainfall data at each meteorological station (i.e., the spatial dependence of different rainfall data). Obviously, if only the time dimension information of rainfall data at each meteorological station is considered without considering the spatial dimension information of rainfall data at each meteorological station, some information may be missed, thus affecting the accuracy of rainfall runoff prediction. In addition, the rainfall data at each meteorological station at each moment is not necessarily completely reliable, that is, erroneous data needs to be excluded when predicting rainfall runoff. Summary of the invention

[0007] The technical problem to be solved by the present invention is to provide a rainfall runoff pollution prediction method based on spatiotemporal long short-term memory network and tree structure traversal, which is used for urban rainfall runoff pollution prediction.

[0008] In order to solve the above technical problems, the technical solution adopted by the present invention is: a rainfall runoff pollution prediction method based on spatiotemporal long short-term memory network and tree structure traversal, comprising the following steps:

[0009] Step 1: Prepare rainfall data of typical cities, location coordinates of each meteorological station and ground elevation DEM data, and check the quality of rainfall data;

[0010] Step 2: According to the location coordinates and ground elevation DEM data of each meteorological station in a typical city, the tree structure of the meteorological station is obtained;

[0011] Step 3: Build a spatial-temporal urban rainfall runoff pollution prediction model and use rainfall data to predict runoff pollution data;

[0012] Step 4: Train and test the urban spatiotemporal rainfall runoff pollution prediction model and set the parameters of the prediction model;

[0013] Step 5: Select rainfall data of typical cities, location coordinates of each meteorological station and ground elevation DEM data, use the urban spatiotemporal rainfall runoff pollution prediction model to predict urban rainfall runoff pollution, and evaluate the prediction performance of the model.

[0014] In a preferred embodiment, the operation steps of step 1 are as follows:

[0015] S1.1. Collect rainfall data of typical cities, location coordinates of meteorological stations and ground elevation DEM data, select rainfall type, average rainfall intensity, peak rainfall intensity, rainfall duration, dry period before rain, atmospheric dust fall, PM10, vehicle volume, pavement material and functional area as influencing factors for rainfall runoff pollution prediction, i.e. input, and select average concentration of rainfall runoff pollution EMC and initial flushing index FF30 as decision variables for rainfall runoff pollution prediction, i.e. output;

[0016] S1.2. After collecting typical urban rainfall data, check the quality of the rainfall data and use a sliding time window to replace missing data and erroneous data in the rainfall data.

[0017] In a preferred solution, the operation steps of step 2 are as follows:

[0018] S2.1. Use the ground elevation (DEM data) of each area in the city to divide each area into small areas, divide the entire urban study area into individual small areas, and number each small area;

[0019] S2.2, according to the location coordinates of each meteorological station in each area of ​​the city, each meteorological station is divided into its corresponding small area, and the meteorological station is numbered according to the number of the small area where each meteorological station is located;

[0020] S2.3. Obtain a meteorological station structure tree based on the correlation between various meteorological stations in the city, determine each node of the meteorological station tree structure, select the meteorological station with the greatest impact on rainfall runoff pollution prediction as the root node of the meteorological station structure tree, and the remaining meteorological stations as leaf nodes.

[0021] In a preferred solution, the operation steps of step three are as follows:

[0022] S3.1. Construct an urban spatiotemporal rainfall runoff pollution prediction model based on spatiotemporal long short-term memory network. By inputting rainfall runoff data at various meteorological stations and monitoring points in the city and considering the spatial distribution of each meteorological station and monitoring point, the rainfall runoff pollution data of the city is obtained.

[0023] S3.2, according to the tree structure of each meteorological station and monitoring point, traverse from the root node of the tree structure, and input the rainfall runoff data at the root node and each leaf node of the tree structure into the rainfall runoff pollution spatiotemporal long short-term memory network in turn;

[0024] S3.3, after the rainfall runoff data at each meteorological station and monitoring point are input into the rainfall runoff pollution spatiotemporal long short-term memory network unit, a trust function is introduced to predict accurate and reliable rainfall runoff pollution data based on the available contextual information;

[0025] S3.4. At each time step, the features of the rainfall runoff data in each rainfall runoff pollution spatiotemporal long short-term memory network unit are fused, and the rainfall runoff pollution prediction data in the catchment area is obtained by using the feature-fused rainfall runoff data.

[0026] In a preferred embodiment, the operation steps of step S3.1 are as follows:

[0027] S3.1.1. Construct a long short-term memory network based on time series, and obtain the rainfall runoff pollution time series data of the city by inputting the rainfall data at various meteorological stations and monitoring points in the city;

[0028] S3.1.2. Input the rainfall data at each meteorological station or monitoring point in the city into the spatiotemporal long-term and short-term memory network unit corresponding to each meteorological station or monitoring point to obtain the spatiotemporal information of the rainfall data at each meteorological station or monitoring point in the city;

[0029] S3.1.3. After the rainfall runoff pollution prediction input data enters each rainfall runoff pollution spatiotemporal long-term memory network unit, each rainfall runoff pollution spatiotemporal long-term memory network unit processes the contextual information of the rainfall data in both time and space dimensions.

[0030] In a preferred embodiment, the operation steps of step S3.2 are as follows:

[0031] S3.2.1. Input the rainfall data of the meteorological station or monitoring point at the root node of the tree structure into the corresponding rainfall runoff pollution spatiotemporal long-term and short-term memory network unit;

[0032] S3.2.2, traverse the tree structure in depth-first order, and input the rainfall data of each leaf node at the meteorological station or monitoring point into the corresponding rainfall runoff pollution spatiotemporal long-short-term memory network unit in turn;

[0033] S3.2.3, after traversing to the leaf node of the tree structure, the tree structure is backtracked in reverse depth-first order, and the rainfall data of the meteorological station or monitoring point at each leaf node of the tree structure is input into the corresponding rainfall runoff pollution spatiotemporal long-short-term memory network unit in turn;

[0034] S3.2.4. After back-traversing to the root node of the tree structure, end the input of rainfall data at each meteorological station or monitoring point in the entire city at that time step, continue to input rainfall data at each meteorological station or monitoring point in the city at the next time step, and repeat steps S3.2.1 to S3.2.4.

[0035] In a preferred solution, the specific operation of step S3.3 is as follows:

[0036] S3.3.1. Add a new mechanism to each rainfall runoff pollution spatiotemporal long short-term memory network to predict the rainfall input data of the rainfall runoff pollution spatiotemporal long short-term memory network unit at each spatiotemporal step. The prediction of the rainfall input data of each rainfall runoff pollution spatiotemporal long short-term memory network unit is expressed as:

[0037]

[0038] Where M p Is to transfer data from Map to Affine transformation of h j,t-1 represents the hidden representation of the same weather station in the previous time step, h j-1,t represents the hidden representation of the neighboring weather stations at the current time step, and d represents the dimension of the rainfall-runoff pollution data;

[0039] S3.3.2. In each rainfall runoff pollution spatiotemporal long short-term memory network unit, the predicted rainfall data input value is compared with the actual rainfall data input value, and a trust gate is added to the network unit. The trust gate is trained using the error between the predicted value and the actual value of the input rainfall data. After activation, the trust gate can help the rainfall runoff pollution spatiotemporal long short-term memory network unit better decide when and how to remember or forget the content in the storage unit, using The vector in activates the trust gate, and the trust gate τ j,t The calculation of is as follows:

[0040]

[0041] τ j,t =G(p j,t -x′ j，t )

[0042] Where M x Is to transfer data from Affine transformation; x j,t Represents the input rainfall data; p j,t and x′ j,t They represent the state value and update value of rainfall runoff pollution data respectively;

[0043] The activation function G(·) is calculated as G(z)=exp(-λz 2 ), where λ is a parameter that controls the bandwidth of the Gaussian function, λ>0, if z has a large absolute value, a smaller response is generated; if z is close to zero, a larger response is generated; z represents the rainfall runoff pollution data at each moment.

[0044] After adding the trust gate to each rainfall runoff pollution spatiotemporal long short-term memory network unit, the status of each rainfall runoff pollution spatiotemporal long short-term memory network unit is updated as follows:

[0045]

[0046] In the formula, i j,t Represents the state value of the trust gate, u j,t Indicates the state value of the update gate, represents the rainfall runoff pollution data in the time dimension, c j-1,t Represents the rainfall runoff pollution data of the previous node (in space), represents the rainfall runoff pollution data in the spatial dimension, c j,t-1 Represents the rainfall runoff pollution data at the previous moment (in time);

[0047] If the input rainfall data x j,tis unreliable, the rainfall runoff pollution spatiotemporal long short-term memory network relies more on its historical information and tries to organize new rainfall data input at this time step; if the input rainfall data x j,t If it is credible, the storage unit of the rainfall runoff pollution spatiotemporal long short-term memory network is updated according to the input rainfall data.

[0048] In a preferred embodiment, the step S3.4 is performed as follows:

[0049] S3.4.1. In a single rainfall runoff pollution spatiotemporal long-term memory network, the temporal and spatial information of rainfall data at each meteorological station in the city and monitoring data at monitoring points are integrated to obtain the rainfall runoff pollution prediction value of the single rainfall runoff pollution spatiotemporal long-term memory network;

[0050] At each time step, the spatiotemporal rainfall data of each meteorological station in the city and the spatiotemporal monitoring data at each monitoring point are input into the rainfall runoff spatiotemporal long short-term memory network unit. represents the spatiotemporal rainfall data of the jth meteorological station or monitoring point in the city at the tth time step, where represents the rainfall information of the jth meteorological station or monitoring point in the city at the tth time step, The location information of the jth meteorological station or monitoring point in the catchment area at the tth time step, and The inputs are fed into the corresponding rainfall runoff pollution spatiotemporal long short-term memory network units respectively;

[0051] In each rainfall runoff pollution spatiotemporal long short-term memory network unit, the input gate About time and space The forget gate and trust gate And process spatiotemporal rainfall data and The two sets of heterogeneous modal characteristics of the temporal and spatial rainfall data and The unit representation Put together to establish the multimodal context information of the two sets of modal features. After inputting the spatiotemporal rainfall data, the output of each rainfall runoff pollution spatiotemporal long short-term memory network is calculated according to the multimodal context information of the two sets of modal features, and the output gate o shared by the two sets of features is used. j,t control.

[0052] S3.4.2. Use the rainfall data of each rainfall runoff pollution spatiotemporal LSTM network unit feature fusion to predict the rainfall runoff pollution data of the city at that time step.

[0053] In a preferred solution, the operation steps of step 4 are as follows:

[0054] S4.1. Determine the learnable parameters and hyperparameters of the spatiotemporal long short-term memory network for rainfall-runoff pollution;

[0055] S4.2. Determine the rainfall runoff pollution prediction performance evaluation index of the rainfall runoff pollution spatiotemporal long short-term memory network to ensure that the rainfall runoff pollution spatiotemporal long short-term memory network can be effectively trained to obtain an urban rainfall runoff pollution prediction model that meets the accuracy requirements;

[0056] S4.3. Input the rainfall runoff data from each meteorological station and monitoring point in the catchment area into the rainfall runoff pollution spatiotemporal long-term and short-term memory network, train and test the rainfall runoff pollution spatiotemporal long-term and short-term memory network, and ensure that the average concentration EMC of rainfall runoff pollution and the initial flushing index FF30 can accurately reflect the pollution caused by the rainfall in the city.

[0057] In a preferred solution, in step S4.1, the learning parameter refers to the runoff pollution data value, and the hyperparameter refers to the likelihood function value;

[0058] In step S4.2, the rainfall runoff pollution spatiotemporal long short-term memory network uses a softmax layer to predict the rainfall runoff pollution prediction output data, and the objective function is as follows:

[0059]

[0060] In the formula, Represents the negative log-likelihood loss, which is used to represent the contaminated prediction value at each spatiotemporal step The difference between the actual value of rainfall runoff pollution, J represents the number of meteorological stations in the catchment area, and T represents the total time of rainfall data of each meteorological station in the catchment area;

[0061] In the step S4.3, real observation data of k rainfall runoffs in the city are selected to train the proposed rainfall runoff pollution long short-term memory network, k likelihood function values ​​are obtained through k trainings, and the hyperparameter base that produces the highest median likelihood function value among all hyperparameter combinations is selected as the hyperparameter of the rainfall runoff pollution long short-term memory network.

[0062] The present invention provides a rainfall runoff pollution prediction method based on spatiotemporal long short-term memory network and tree structure traversal, constructs a rainfall runoff pollution spatiotemporal long short-term memory network, models the dynamic and dependency relationship of rainfall data in the spatiotemporal domain, and the rainfall data of each meteorological station at each moment can receive context information from the rainfall data stored at the previous moment, and can also receive context information from the rainfall data at the adjacent meteorological station at the same moment. At the same time, a meteorological station structure tree traversal method is proposed, starting from the root node of the tree structure, and sequentially inputting the rainfall data at the root node and each leaf node of the tree structure into the rainfall runoff spatiotemporal long short-term memory network. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] The present invention will be further described below in conjunction with the accompanying drawings and embodiments:

[0064] Figure 1 This is a flow chart of the rainfall runoff pollution prediction method based on spatiotemporal long short-term memory network and tree structure traversal;

[0065] Figure 2 It is a schematic diagram of the structure tree of various meteorological stations and monitoring points in the catchment area;

[0066] Figure 3 Schematic diagram of a single neuron in the spatiotemporal long short-term memory network of rainfall runoff pollution;

[0067] Figure 4 This is a schematic diagram of the spatiotemporal long-short-term memory network of rainfall runoff pollution;

[0068] Figure 5 A schematic diagram of the structure tree traversal for each meteorological station and monitoring point in the city;

[0069] Figure 6 It is a schematic diagram of the spatiotemporal long-short-term memory network of rainfall-runoff pollution with a deep tree structure;

[0070] Figure 7 Trust gates for spatiotemporal long short-term memory networks for rainfall runoff pollution;

[0071] Figure 8 It is the internal feature fusion unit of the spatiotemporal long short-term memory network of rainfall runoff pollution;

[0072] Fig. 9 This is a comparison chart of the EMC predicted value and the actual value obtained using the urban spatiotemporal rainfall runoff pollution prediction model;

[0073] Fig.10 This is a comparison chart of the FF30 predicted value and the actual value obtained using the urban spatiotemporal rainfall runoff pollution prediction model;

[0074] Fig.11Comparison chart of EMC prediction and true value obtained by using long short-term memory network of rainfall runoff pollution;

[0075] Fig.12 This is a comparison chart between the FF30 predicted value and the actual value obtained using the long short-term memory network for rainfall runoff pollution. DETAILED DESCRIPTION

[0076] Combination Figures 1 to 12 The specific embodiments of the present invention are described in further detail.

[0077] like Figure 1 As shown in the figure, a rainfall runoff pollution prediction method based on spatiotemporal long short-term memory network and tree structure traversal includes five parts: (1) preparation of hydrological and meteorological data in the city; (2) obtaining the tree structure of meteorological stations and monitoring points in the city; (3) spatiotemporal long short-term memory of rainfall runoff pollution; (4) spatiotemporal long short-term memory network training of rainfall runoff pollution; (5) comparison and evaluation of rainfall runoff pollution prediction performance.

[0078] Step 1. First, a city is selected as an example to elaborate on the method proposed in the present invention. The city has 6 meteorological stations, 11 monitoring points and 1 hydrological station. Prepare the city's rainfall data, the location coordinates of the meteorological station and the ground elevation DEM data, select the rain type, average rainfall intensity, peak rainfall intensity, rainfall duration, dry period before rain, atmospheric dust fall, PM10, vehicle flow, pavement material, functional area, average concentration of rainfall runoff pollution EMC, initial flushing index FF30 and ground elevation DEM data, and check the quality of rainfall data and runoff pollution data, and use a sliding time window to replace the missing data and erroneous data in the collected data to ensure that the input data and output data quality of the rainfall runoff pollution prediction model meet the prediction requirements.

[0079] The city area is divided into sub-areas according to the ground elevation of the city, and the study area is divided into different sub-areas, such as Figure 2 As shown, the study area can be divided into two small areas, gray and white.

[0080] Step 2: Secondly, according to the ground elevation (DEM data) of the study area and the location coordinate information of each meteorological station and monitoring point in the catchment area, the tree structure of each meteorological station and monitoring station in the catchment area is obtained.

[0081] The study area is divided into small areas using the ground elevation (DEM data) in each small area. The study area is divided into individual small areas and each small area is numbered, such as Figure 2 As shown, the meteorological stations and monitoring points are numbered 1-16.

[0082] According to the location coordinates of each meteorological station and monitoring point in the study area, each meteorological station and monitoring point is divided into its corresponding small area, and the meteorological station and monitoring point are numbered according to the number of the small area where each meteorological station and monitoring point is located, such as Figure 2 As shown, the meteorological stations and monitoring points are numbered 1-16.

[0083] The urban study area is converted into a tree structure according to the correlation between each small area, and each node of the tree structure is determined.

[0084] Step 3. Then, a recursive neural network is used to construct an urban spatiotemporal rainfall runoff pollution prediction model. Each spatiotemporal long short-term memory network unit corresponds to each meteorological station and monitoring point in the study area, and receives the hidden representation of the rainfall runoff pollution prediction input data of the meteorological station or monitoring point from the previous time step, as well as the hidden representation of the rainfall runoff pollution prediction input data at the adjacent meteorological station or monitoring point in the study area at the current time step. Figure 3 Schematic diagram of the spatiotemporal long-short-term memory network of rainfall runoff pollution.

[0085] Each spatiotemporal LSTM network unit corresponds to a meteorological station or monitoring point in the study area, and receives the hidden representation of the rainfall-runoff pollution prediction input data of the meteorological station or monitoring point itself from the previous time step, as well as the hidden representation of the rainfall-runoff pollution prediction input data of the adjacent meteorological stations or monitoring points in the study area at the current time step.

[0086] The rainfall runoff pollution prediction input data at each meteorological station or monitoring point in the study area is input into the spatiotemporal long-short-term memory network unit corresponding to each meteorological station or monitoring point to obtain the spatiotemporal information of the rainfall runoff data at each meteorological station or monitoring point in the study area.

[0087] The input data of rainfall runoff pollution prediction for each meteorological station or monitoring point in the study area is x j,t Represents, where j∈{1,…,J}, t∈{1,…,T}, j represents the jth meteorological station or monitoring point in the study area, t represents the rainfall runoff pollution prediction input data of each meteorological station or monitoring point in the study area at time t, J represents the total number of meteorological stations or monitoring points in the study area, and T represents the total time of rainfall runoff pollution prediction input data at each meteorological station or monitoring point in the study area.

[0088] The input data of each rainfall runoff pollution spatiotemporal long short-term memory network unit includes: j,t 、h j-1,t and h j,t-1 , x j,t represents the rainfall runoff pollution prediction input data at the meteorological station or monitoring point corresponding to the current time step, h j-1,t represents the hidden representation of the neighboring weather stations or monitoring points at the current time step, hj,t-1 Represents the hidden representation of the same weather station or monitoring point in the previous time step.

[0089] After the rainfall runoff pollution prediction input data enters each rainfall runoff pollution spatiotemporal long short-term memory network unit, each rainfall runoff pollution spatiotemporal long short-term memory network unit processes the context information of the rainfall runoff pollution prediction input data in both time and space dimensions.

[0090] Each rainfall runoff pollution spatiotemporal long short-term memory network unit has two forget gates and Forget Gate The context information of the rainfall runoff pollution prediction input data at each meteorological station or monitoring point in the study area is processed in the time dimension, and the forget gate In the spatial dimension, the context information of the rainfall runoff pollution prediction input data at each meteorological station or monitoring point in the study area is processed. The transition equation of the rainfall runoff pollution spatiotemporal long short-term memory network unit is expressed as:

[0091]

[0092]

[0093] h j,t =o j,t ⊙tanh(c j,t )

[0094] like Figure 4 As shown, according to the tree structure of each meteorological station or monitoring point, the traversal starts from the root node of the tree structure, and the rainfall runoff pollution prediction input data at the root node and each leaf node of the tree structure are input into the rainfall runoff pollution spatiotemporal long short-term memory network in turn.

[0095] The rainfall runoff pollution prediction input data of the meteorological station or monitoring point at the root node of the tree structure is input into its corresponding rainfall runoff pollution spatiotemporal long short-term memory network unit.

[0096] The tree structure is traversed in depth-first order, and the rainfall runoff pollution prediction input data of the meteorological station or monitoring point at each leaf node is input into the corresponding rainfall runoff pollution spatiotemporal long-short-term memory network unit in turn.

[0097] After traversing to the leaf node of the tree structure, the tree structure is backtracked in reverse depth-first order, and the rainfall runoff pollution prediction input data of the meteorological station or monitoring point at each leaf node of the tree structure is input into the corresponding rainfall runoff pollution spatiotemporal long-short-term memory network unit in turn.

[0098] After traversing back to the root node of the tree structure, the input of the rainfall runoff pollution prediction input data at each meteorological station or monitoring point in the entire study area at this time step is completed. Continue to input the rainfall runoff pollution prediction input data at each meteorological station or monitoring point in the study area at the next time step, and repeat the above traversal steps.

[0099] After the rainfall runoff pollution prediction input data at each meteorological station or monitoring point is input into the rainfall runoff pollution spatiotemporal long short-term memory network unit, the trust function is introduced to predict accurate and reliable rainfall runoff pollution prediction input data based on the available context information, such as Figure 5 shown.

[0100] like Figure 6-8 As shown in the figure, affected by the monitoring and collection errors of the rainfall runoff pollution prediction input data, the rainfall runoff pollution prediction input data at each meteorological station or monitoring point in the study area may have certain errors. In order to improve the accuracy of rainfall runoff pollution prediction in the study area, the present invention adds an additional gate to each rainfall runoff pollution spatiotemporal long short-term memory network unit, so as to estimate the reliability of the rainfall runoff pollution prediction input data based on the output data derived from the available context information of the rainfall runoff pollution prediction input data at each meteorological station or monitoring point in the study area at each spatiotemporal step. In each spatiotemporal step, the context information h is used to generate the rainfall runoff pollution prediction input data. j-1,t and h j,t-1 To predict rainfall runoff pollution prediction input data x j,t , as follows:

[0101] A new mechanism is added to each rainfall runoff pollution spatiotemporal long short-term memory network to predict the rainfall runoff pollution prediction input data of the rainfall runoff pollution spatiotemporal long short-term memory network unit at each spatiotemporal step. The prediction of the rainfall runoff pollution prediction input data input to each rainfall runoff pollution spatiotemporal long short-term memory network unit is expressed as:

[0102]

[0103] Where M p Is to transfer data from Map to , so the dimension of is d. Note that the context information of each spatiotemporal step contains not only the representation of the previous time step, but also the hidden state of the previous spatial step. This shows that the long-term context information of the same weather station in the previous time step and other weather stations in the current time step has been seamlessly merged. Therefore, the prediction function can predict reasonable prediction values ​​of the input data of rainfall runoff pollution prediction.

[0104] In each rainfall runoff pollution spatiotemporal long short-term memory network unit, the predicted rainfall runoff pollution input data is compared with the actual rainfall runoff pollution prediction input data, and a trust gate is added to the network unit. The trust gate is trained using the error between the predicted and actual values ​​of rainfall runoff pollution. After activation, the trust gate can help the rainfall runoff pollution spatiotemporal long short-term memory network unit better decide when and how to remember or forget the contents of the storage unit. The vector in activates the trust gate, and the trust gate τ j,t The calculation of is as follows:

[0105]

[0106] τ j,t =G(p j,t -x′ j，t )

[0107] In the formula, is an affine transformation, and the activation function G(·) is calculated as G(z)=exp(-λz 2 ), where λ is a parameter that controls the bandwidth of the Gaussian function (λ>0). If z has a large absolute value, a smaller response is produced; if z is close to zero, a larger response is produced.

[0108] After adding the trust gate to each rainfall runoff pollution spatiotemporal long short-term memory network unit, the status of each rainfall runoff pollution spatiotemporal long short-term memory network unit is updated as follows:

[0109]

[0110] In the formula, i j,t Represents the state value of the trust gate, u j,t Indicates the state value of the update gate, represents the rainfall runoff pollution data in the time dimension, c j-1,t Represents the rainfall runoff pollution data of the previous node (in space), represents the rainfall runoff pollution data in the spatial dimension, c j,t-1 Represents the rainfall runoff pollution data at the previous moment (in time).

[0111] If the input rainfall runoff pollution prediction input data x j,t is not credible, then the rainfall runoff pollution spatiotemporal long short-term memory network relies more on its historical information and tries to organize new rainfall runoff pollution prediction input data input at this time step; if the input rainfall runoff pollution prediction input data x j,t If it is credible, the storage unit of the rainfall runoff pollution spatiotemporal long short-term memory network is updated according to the input rainfall runoff pollution prediction input data.

[0112] At each time step, the features of the rainfall runoff pollution prediction input data in each rainfall runoff pollution spatiotemporal long short-term memory network unit are fused, and the rainfall runoff pollution prediction data in the study area is predicted using the feature-fused rainfall runoff pollution prediction input data.

[0113] In a single rainfall runoff pollution spatiotemporal long-short-term memory network, the temporal and spatial information of the rainfall runoff pollution prediction input data at each meteorological station or monitoring point in the study area are integrated to obtain the rainfall runoff pollution prediction value of the single rainfall runoff pollution spatiotemporal long-short-term memory network.

[0114] At each time step, the rainfall-runoff pollution prediction input data at each meteorological station or monitoring point in the study area is input into the rainfall-runoff pollution spatiotemporal long-term memory network unit. represents the rainfall runoff pollution prediction input data of the jth meteorological station or monitoring point in the study area at the tth time step, where represents the input data of rainfall runoff pollution prediction at the jth meteorological station or monitoring point in the study area at the tth time step, The location information (longitude, latitude, elevation and structure tree number) of the jth meteorological station or monitoring point in the study area at the tth time step, and The inputs are fed into their corresponding rainfall-runoff pollution spatiotemporal long short-term memory network units.

[0115] In each rainfall runoff pollution spatiotemporal long short-term memory network unit, the input gate About time and space The forget gate and trust gate And process the spatial and temporal rainfall runoff pollution prediction input data and The spatial-temporal rainfall runoff pollution prediction input data and The unit representation After inputting the spatiotemporal rainfall runoff pollution prediction input data, the output of each rainfall runoff pollution spatiotemporal long short-term memory network is calculated based on the multimodal context information of the two sets of modal features, and the output gate (o j,t )control.

[0116] Prediction of spatial and temporal rainfall-runoff pollution with input data For example, when inputting the spatial-temporal rainfall runoff pollution prediction input data With the help of The corresponding input gate, forget gate, and trust gate can more effectively learn how to update its current unit state Specifically, for the new spatiotemporal rainfall-runoff pollution prediction input data The spatial-temporal long short-term memory network unit of rainfall runoff pollution needs to consider not only the spatial-temporal rainfall runoff pollution prediction input data when predicting. The context information of the spatial and temporal rainfall runoff pollution prediction input data also needs to be considered. Therefore, the trust gate Evaluating Spatiotemporal Rainfall-Runoff Pollution Forecast Input Data Greater reliability capabilities.

[0117] The formula of the rainfall runoff pollution spatiotemporal long short-term memory network with integrated multimodal feature fusion is as follows:

[0118]

[0119]

[0120]

[0121]

[0122]

[0123]

[0124]

[0125] The rainfall runoff pollution prediction input data after the fusion of the characteristics of each rainfall runoff pollution spatiotemporal long short-term memory network unit is used to predict the rainfall runoff pollution prediction output data in the study area at that time step.

[0126] Step 4. Then, the urban spatiotemporal rainfall runoff pollution prediction model is trained and tested, and the model parameters are set to ensure that the model can accurately simulate and predict rainfall runoff pollution in the study area.

[0127] Determine the learnable parameters and hyperparameters of the urban spatiotemporal rainfall runoff pollution prediction model so as to train and test the urban spatiotemporal rainfall runoff pollution prediction model and ensure that the urban spatiotemporal rainfall runoff pollution prediction model can accurately output the rainfall runoff pollution prediction value in the study area. The learning parameters refer to the runoff pollution data values, and the hyperparameters refer to the likelihood function values.

[0128] The number of rainfall runoff pollution spatiotemporal long-term memory network units is determined according to the number of meteorological stations and monitoring points J in the study area. Each meteorological station or monitoring point in the study area corresponds to a rainfall runoff pollution spatiotemporal long-term memory network unit; the hidden state length of each rainfall runoff pollution spatiotemporal long-term memory network unit is set to 48. In order to avoid overfitting, the exit rate of each rainfall runoff pollution spatiotemporal long-term memory network is set to 0.05, the learning rate is set to 0.01, and the number of samples of each batch of spatiotemporal rainfall runoff pollution prediction input data is set to 128. The rainfall runoff pollution spatiotemporal long-term memory network uses a softmax layer to predict the rainfall runoff pollution prediction output data, and the objective function is as follows:

[0129]

[0130] In the formula, Represents the negative log likelihood loss, which is used to represent the rainfall runoff pollution prediction value at each time and space step The difference between the actual value of rainfall runoff pollution.

[0131] The real observation data of k rainfall runoff pollution in the study area are selected to train the proposed rainfall runoff pollution spatiotemporal long short-term memory network, and k likelihood function values ​​are obtained through k training. Choose the hyperparameter combination that produces the highest likelihood function value among all The median hyperparameter basis is used as the hyperparameter of the rainfall-runoff pollution spatiotemporal long short-term memory network.

[0132] Determine the rainfall runoff pollution prediction performance evaluation indicators of the rainfall runoff pollution spatiotemporal long short-term memory network to ensure that the urban spatiotemporal rainfall runoff pollution prediction model can be effectively trained to obtain a rainfall runoff pollution prediction model that meets the accuracy requirements.

[0133] The rainfall data of each meteorological station in the study area and the monitoring data of each monitoring point are input into the rainfall runoff pollution spatiotemporal long-term memory network, and the rainfall runoff pollution spatiotemporal long-term memory network is trained and tested to ensure that the output data of the rainfall runoff pollution spatiotemporal long-term memory network (i.e., the average concentration of rainfall runoff pollution EMC and the initial flushing index FF30) can accurately reflect the actual rainfall runoff pollution in the study area. Fig. 9 and Fig.10 As shown in the figure, the predicted average concentration of rainfall runoff pollution EMC and initial flushing index FF30 are closer to the actual observed data.

[0134] Step 5. Finally, select the real rainfall runoff pollution observation data in the remaining cities to evaluate the prediction performance of the urban spatiotemporal rainfall runoff pollution prediction model. The model prediction performance evaluation includes comparison with other benchmark models and changing the input method of rainfall runoff pollution prediction input data.

[0135] First, a long short-term memory network for rainfall runoff pollution is constructed as a benchmark model for urban rainfall runoff pollution prediction, and the rainfall runoff pollution prediction performance of the spatiotemporal long short-term memory network for rainfall runoff pollution is compared and evaluated. A long short-term memory network for rainfall runoff pollution is constructed as a benchmark network for rainfall runoff pollution prediction, and the rainfall data of each meteorological station in the study area and the monitoring data at each monitoring point are used to predict the rainfall runoff pollution prediction output data in the study area. The dimension of the unit of the long short-term memory network for rainfall runoff pollution is the number J of meteorological stations and monitoring points in the study area. The hidden state length of each unit of the long short-term memory network for rainfall runoff pollution is set to 48. In order to avoid overfitting, the exit rate of each rainfall runoff pollution spatiotemporal long short-term memory network is set to 0.05, the learning rate is set to 0.01, and the number of samples of each batch of spatiotemporal rainfall pollution runoff input data is set to 128. The spatiotemporal long short-term memory network for rainfall runoff pollution uses a softmax layer to predict the runoff value, and the objective function is as follows:

[0136]

[0137] In the formula, Represents the negative log likelihood loss, which is used to represent the rainfall runoff pollution prediction value at each time and space step The difference between the actual value of rainfall runoff pollution.

[0138] Where, represents the negative logarithm of the likelihood loss, which is used to represent the difference between the predicted value of rainfall runoff pollution and the true value of rainfall runoff pollution at each spatiotemporal step.

[0139] The proposed rainfall runoff pollution long short-term memory network is trained by selecting k real observation data of rainfall runoff pollution in the study area. K likelihood function values ​​are obtained through k trainings. The hyperparameter base that produces the highest median likelihood function value among all hyperparameter combinations is selected as the hyperparameter of the rainfall runoff pollution long short-term memory network. Fig.11 and 12 As shown in the figure, the predicted average concentration of rainfall runoff pollution EMC and initial flushing index FF30 are closer to the actual observed data.

[0140] W real rainfall runoff pollution prediction input data in the study area were selected and input into the urban spatiotemporal rainfall runoff pollution prediction model and the rainfall runoff pollution long short-term memory network to predict the rainfall runoff pollution prediction value in the study area. After the rainfall runoff pollution prediction value in the study area was predicted, the W real rainfall runoff pollution prediction data in the study area were used to calculate the likelihood loss values ​​of the rainfall runoff pollution spatiotemporal long short-term memory network and the rainfall runoff pollution long short-term memory network to evaluate the prediction performance of the urban spatiotemporal rainfall runoff pollution prediction model.

[0141] Secondly, in order to further evaluate and analyze the rainfall runoff pollution prediction performance of the proposed method, different methods were used to input the rainfall runoff pollution prediction data in the study area into the urban spatiotemporal rainfall runoff pollution prediction model to obtain the rainfall runoff pollution prediction value in the study area. According to the model prediction results, the prediction performance of different input methods of rainfall runoff pollution prediction input data was evaluated and analyzed.

[0142] The way of inputting the spatiotemporal rainfall data of each meteorological station and the monitoring data at each monitoring point in the study area into the urban spatiotemporal rainfall runoff pollution prediction model is changed to further evaluate and analyze the prediction performance of the urban spatiotemporal rainfall runoff pollution prediction model. The input methods of rainfall runoff pollution prediction input data mainly include the following: ① Input the rainfall data of each meteorological station and the monitoring data of the monitoring point into the urban spatiotemporal rainfall runoff pollution prediction model in a simple chain sequence, and remove the trust gate in each rainfall runoff pollution spatiotemporal long-term memory network unit; ② Input the rainfall data of each meteorological station and the monitoring data of the monitoring point into the urban spatiotemporal rainfall runoff pollution prediction model in a simple chain sequence, and retain the trust gate in each rainfall runoff pollution spatiotemporal long-term memory network unit; ③ Input the spatiotemporal rainfall data of each meteorological station and the monitoring data of the monitoring point into the urban spatiotemporal rainfall runoff pollution prediction model in a structure tree traversal manner, and remove the trust gate in each rainfall runoff pollution spatiotemporal long-term memory network unit. The urban spatiotemporal rainfall runoff pollution prediction model is trained and tested using real observation data of k rainfall runoff pollution in the study area to ensure that the predicted values ​​of the spatiotemporal long short-term memory network units of rainfall runoff pollution can reflect the actual rainfall runoff pollution situation.

[0143] W real rainfall runoff pollution data in the study area were selected, and the rainfall runoff pollution prediction input data were input into the urban spatiotemporal rainfall runoff pollution prediction model in three different ways to predict the rainfall runoff pollution prediction value in the study area. After the rainfall runoff pollution prediction value in the study area was predicted, the likelihood function loss values ​​of the three different ways were calculated using the W real rainfall runoff pollution data in the study area, and compared with the likelihood loss value of the rainfall runoff pollution spatiotemporal long-short-term memory network, the rainfall runoff prediction performance of the rainfall runoff pollution spatiotemporal long-short-term memory network was evaluated and analyzed.

Claims

1. A rainfall runoff pollution prediction method based on spatiotemporal long short-term memory network and tree structure traversal, characterized in that: The steps include: Step 1: Prepare rainfall data of typical cities, location coordinates of each meteorological station and ground elevation DEM data, and check the quality of rainfall data; Step 2: According to the location coordinates and ground elevation DEM data of each meteorological station in a typical city, the tree structure of the meteorological station is obtained; Step 3: Construct an urban spatiotemporal rainfall runoff pollution prediction model and use rainfall data to predict runoff pollution data. The operation steps are as follows: S3.

1. Construct an urban spatiotemporal rainfall runoff pollution prediction model based on spatiotemporal long short-term memory network. By inputting rainfall runoff data at various meteorological stations and monitoring points in the city and considering the spatial distribution of each meteorological station and monitoring point, the rainfall runoff pollution data of the city is obtained. S3.2, according to the tree structure of each meteorological station and monitoring point, traverse from the root node of the tree structure, and input the rainfall runoff data at the root node and each leaf node of the tree structure into the rainfall runoff pollution spatiotemporal long short-term memory network in turn; S3.

3. After the rainfall runoff data at each meteorological station and monitoring point are input into the rainfall runoff pollution spatiotemporal long short-term memory network unit, the trust function is introduced to predict accurate and reliable rainfall runoff pollution data based on the available context information. The specific operations are as follows: S3.3.

1. A new mechanism is added to each rainfall runoff pollution spatiotemporal long short-term memory network to predict the rainfall input data of the rainfall runoff pollution spatiotemporal long short-term memory network unit at each spatiotemporal step. The prediction of the rainfall input data of each rainfall runoff pollution spatiotemporal long short-term memory network unit is expressed as: ； In the formula, Is to transfer data from Map to The affine transformation of represents the hidden representation of the same weather station in the previous time step, represents the hidden representation of the neighboring weather stations at the current time step, Represents the dimensionality of rainfall runoff pollution data; S3.3.

2. In each rainfall runoff pollution spatiotemporal long short-term memory network unit, the predicted rainfall data input value is compared with the actual rainfall data input value, and a trust gate is added to the network unit. The trust gate is trained using the error between the predicted value and the actual value of the input rainfall data. After activation, the trust gate can help the rainfall runoff pollution spatiotemporal long short-term memory network unit better decide when and how to remember or forget the content in the storage unit, using The vector in activates the trust gate, which The calculation of is as follows: ； ； In the formula, Is to transfer data from Affine transformation; Represents the input rainfall data; and The state value and update value of rainfall runoff pollution data respectively; Activation Function is calculated as Element-wise operation, where is the parameter that controls the bandwidth of the Gaussian function, >0, if z has a larger absolute value, a smaller response is produced; If z is close to zero, a larger response is produced; Represents the rainfall runoff pollution data at each moment; After adding the trust gate to each rainfall runoff pollution spatiotemporal long short-term memory network unit, the status of each rainfall runoff pollution spatiotemporal long short-term memory network unit is updated as follows: ； In the formula, Indicates the state value of the trust gate, Indicates the state value of the update gate, Represents rainfall runoff pollution data in the time dimension, Represents the rainfall runoff pollution data of the previous node, represents the rainfall runoff pollution data in the spatial dimension, Represents the rainfall runoff pollution data at the previous moment; If the input rainfall data If the input rainfall data is not credible, the rainfall runoff pollution spatiotemporal long short-term memory network relies more on its historical information and tries to organize new rainfall data input at this time step; if the input rainfall data is If it is credible, the storage unit of the rainfall runoff pollution spatiotemporal long short-term memory network is updated according to the input rainfall data; S3.

4. At each time step, feature fusion is performed on the rainfall runoff data in each rainfall runoff pollution spatiotemporal long short-term memory network unit, and the rainfall runoff pollution prediction data in the catchment area is obtained by using the rainfall runoff data after feature fusion; Step 4: Train and test the urban spatiotemporal rainfall runoff pollution prediction model and set the parameters of the prediction model; Step 5: Select rainfall data of typical cities, location coordinates of each meteorological station and ground elevation DEM data, use the urban spatiotemporal rainfall runoff pollution prediction model to predict urban rainfall runoff pollution, and evaluate the prediction performance of the model.

2. According to claim 1, a rainfall runoff pollution prediction method based on spatiotemporal long short-term memory network and tree structure traversal is characterized in that: The operation steps of step one are as follows: S1.

1. Collect rainfall data of typical cities, location coordinates of meteorological stations and ground elevation DEM data, select rainfall type, average rainfall intensity, peak rainfall intensity, rainfall duration, dry period before rain, atmospheric dust fall, PM10, vehicle volume, pavement material and functional area as influencing factors for rainfall runoff pollution prediction, i.e. input, and select average concentration of rainfall runoff pollution EMC and initial flushing index FF30 as decision variables for rainfall runoff pollution prediction, i.e. output; S1.

2. After collecting typical urban rainfall data, check the quality of the rainfall data and use a sliding time window to replace missing and erroneous data in the rainfall data.

3. The method for predicting rainfall runoff pollution based on spatiotemporal long short-term memory network and tree structure traversal according to claim 1 is characterized in that: The operation steps of step 2 are as follows: S2.

1. Use the ground elevation of each area in the city to divide each area into small areas, divide the entire urban study area into individual small areas, and number each small area; S2.2, according to the location coordinates of each meteorological station in each area of ​​the city, each meteorological station is divided into its corresponding small area, and the meteorological station is numbered according to the number of the small area where each meteorological station is located; S2.

3. Obtain a meteorological station structure tree based on the correlation between various meteorological stations in the city, determine each node of the meteorological station tree structure, select the meteorological station with the greatest impact on rainfall runoff pollution prediction as the root node of the meteorological station structure tree, and the remaining meteorological stations as leaf nodes.

4. The method for predicting rainfall runoff pollution based on spatiotemporal long short-term memory network and tree structure traversal according to claim 1 is characterized in that: The operation steps of step S3.1 are as follows: S3.1.

1. Construct a long short-term memory network based on time series, and obtain the rainfall runoff pollution time series data of the city by inputting the rainfall data at various meteorological stations and monitoring points in the city; S3.1.

2. Input the rainfall data at each meteorological station or monitoring point in the city into the spatiotemporal long-term and short-term memory network unit corresponding to each meteorological station or monitoring point to obtain the spatiotemporal information of the rainfall data at each meteorological station or monitoring point in the city; S3.1.

3. After the rainfall runoff pollution prediction input data enters each rainfall runoff pollution spatiotemporal long-term memory network unit, each rainfall runoff pollution spatiotemporal long-term memory network unit processes the contextual information of the rainfall data in both time and space dimensions.

5. The method for predicting rainfall runoff pollution based on spatiotemporal long short-term memory network and tree structure traversal according to claim 1 is characterized in that: The operation steps of step S3.2 are as follows: S3.2.

1. Input the rainfall data of the meteorological station or monitoring point at the root node of the tree structure into the corresponding rainfall runoff pollution spatiotemporal long-term and short-term memory network unit; S3.2.2, traverse the tree structure in depth-first order, and input the rainfall data of each leaf node at the meteorological station or monitoring point into the corresponding rainfall runoff pollution spatiotemporal long-short-term memory network unit in turn; S3.2.3, after traversing to the leaf node of the tree structure, the tree structure is backtracked in reverse depth-first order, and the rainfall data of the meteorological station or monitoring point at each leaf node of the tree structure is input into the corresponding rainfall runoff pollution spatiotemporal long-short-term memory network unit in turn; S3.2.

4. After back-traversing to the root node of the tree structure, end the input of rainfall data at each meteorological station or monitoring point in the entire city at that time step, continue to input rainfall data at each meteorological station or monitoring point in the city at the next time step, and repeat steps S3.2.1~S3.2.

4.

6. The method for predicting rainfall runoff pollution based on spatiotemporal long short-term memory network and tree structure traversal according to claim 1 is characterized in that: The operation of step S3.4 is as follows: S3.4.

1. In a single rainfall runoff pollution spatiotemporal long-term memory network, the temporal and spatial information of rainfall data at each meteorological station in the city and monitoring data at monitoring points are integrated to obtain the rainfall runoff pollution prediction value of the single rainfall runoff pollution spatiotemporal long-term memory network; At each time step, the spatiotemporal rainfall data of each meteorological station in the city and the spatiotemporal monitoring data at each monitoring point are input into the rainfall runoff spatiotemporal long short-term memory network unit. Indicates the city The weather station or monitoring point is Spatiotemporal rainfall data of time steps, ,in, Indicates the city The weather station or monitoring point is The rainfall information of time steps, within the catchment area The weather station or monitoring point is The position information of each time step, and The inputs are fed into the corresponding rainfall runoff pollution spatiotemporal long short-term memory network units respectively; In each rainfall runoff pollution spatiotemporal long short-term memory network unit, the input gate About time and space The forget gate and trust gate And process spatial and temporal rainfall data and The two sets of heterogeneous modal characteristics of the temporal and spatial rainfall data and The unit representation Put together to establish the multimodal context information of the two sets of modal features. After inputting the spatiotemporal rainfall data, the output of each rainfall runoff pollution spatiotemporal long short-term memory network is calculated according to the multimodal context information of the two sets of modal features, and the output gate shared by the two sets of features is used. control; S3.4.

2. Use the rainfall data of each rainfall runoff pollution spatiotemporal LSTM network unit feature fusion to predict the rainfall runoff pollution data of the city at that time step.

7. The method for predicting rainfall runoff pollution based on spatiotemporal long short-term memory network and tree structure traversal according to claim 1 is characterized in that: The operation steps of step 4 are as follows: S4.

1. Determine the learnable parameters and hyperparameters of the spatiotemporal long short-term memory network for rainfall-runoff pollution; S4.

2. Determine the rainfall runoff pollution prediction performance evaluation index of the rainfall runoff pollution spatiotemporal long short-term memory network to ensure that the rainfall runoff pollution spatiotemporal long short-term memory network can be effectively trained to obtain an urban rainfall runoff pollution prediction model that meets the accuracy requirements; S4.

3. Input the rainfall runoff data from each meteorological station and monitoring point in the catchment area into the rainfall runoff pollution spatiotemporal long-term and short-term memory network, train and test the rainfall runoff pollution spatiotemporal long-term and short-term memory network, and ensure that the average concentration EMC of rainfall runoff pollution and the initial flushing index FF30 can accurately reflect the pollution caused by the rainfall in the city.

8. The method for predicting rainfall runoff pollution based on spatiotemporal long short-term memory network and tree structure traversal according to claim 7 is characterized in that: In step S4.1, the learning parameter refers to the runoff pollution data value, and the hyperparameter refers to the likelihood function value; In step S4.2, the rainfall runoff pollution spatiotemporal long short-term memory network uses a softmax layer to predict the rainfall runoff pollution prediction output data, and the objective function is as follows: ； In the formula, Represents the negative log-likelihood loss, which is used to represent the contaminated prediction value at each spatiotemporal step The difference between the actual value of rainfall runoff pollution and Indicates the number of weather stations within the catchment area It represents the total time of rainfall data of each weather station in the catchment area; In the step S4.3, real observation data of k rainfall runoffs in the city are selected to train the proposed rainfall runoff pollution long short-term memory network, k likelihood function values ​​are obtained through k trainings, and the hyperparameter base that produces the highest median likelihood function value among all hyperparameter combinations is selected as the hyperparameter of the rainfall runoff pollution long short-term memory network.

Images