Method and device for constructing salty tide upstream salinity forecasting model

Abstract

The invention discloses a method and a device for constructing a salt tide upstream salinity forecasting model, which are used for solving the technical problem that the existing salt tide upstream salinity forecasting technology cannot realize rapid forecasting. The method comprises the following steps: acquiring a hydro meteorology measured data set, and preprocessing the hydro meteorology measured data set to generate a target hydro meteorology measured data set; based on the target hydro-meteorological actual measurement data set, constructing a hydro-meteorological training data set, a hydro-meteorological test data set and a hydro-meteorological observation data set; performing model training on the initial long-short-term memory model, and determining an intermediate long-short-term memory model; obtaining a plurality of target long-short-term memory models by adopting a preset Shapril additive interpretation model and an intermediate long-short-term memory model according to the hydrometeorological test data set; and screening each target long-short-term memory model by adopting a hydro-meteorological test data set and a hydro-meteorological observation data set, and determining a salt tide upstream salinity forecasting model.

Description

Technical Field

[0001] The invention relates to the technical field of hydrology and water resources, and in particular to a method and device for constructing a saltwater tide upstream salinity forecasting model. Background Art

[0002] Saltwater tide is a natural phenomenon that occurs in estuaries, where seawater flows upstream due to the tide, causing the salinity of freshwater river mouths or waters near the mouths to increase. This phenomenon is caused by a variety of factors, including reduced river flow, tidal changes, wind direction and force, and rising sea levels. Especially during the dry season, the freshwater flow in rivers decreases, while the tide rises during high tides. In this case, seawater can easily flow back into the river, forming a saltwater tide.

[0003] The upstream saltwater tide will have a huge negative impact on the utilization of freshwater resources, industrial and agricultural production and aquatic habitat systems in the estuary area. According to the national water supply standard, the chloride content of drinking water sources should not exceed 250 mg / L. Long-term drinking of water with excessive chloride content will seriously endanger human health.

[0004] Most of the existing saltwater upstream salinity forecasting technologies are based on the establishment of hydrodynamic salinity numerical models based on the dynamic process of salinity change. The construction and parameter verification of this model requires a large amount of detailed information on boundary or initial conditions such as runoff, tides, wind, and riverbed topography. The process mechanism is complex and the calculation is time-consuming, making it impossible to achieve rapid salinity forecasting. Summary of the invention

[0005] The invention provides a method and device for constructing a saltwater upwelling salinity forecasting model, which are used to solve the technical problem that the existing saltwater upwelling salinity forecasting technology cannot achieve rapid forecasting.

[0006] A method for constructing a saltwater tide upstream salinity forecasting model provided by the first aspect of the present invention comprises:

[0007] Acquire a hydrological and meteorological measured data set, and preprocess the hydrological and meteorological measured data set to generate a target hydrological and meteorological measured data set;

[0008] Based on multiple preset lag times, multiple preset lead times, and the target hydrometeorological measured data set, construct a hydrometeorological training data set, a hydrometeorological test data set, and a hydrometeorological observation data set;

[0009] Using the hydrological and meteorological training data set and the hydrological and meteorological observation data set to perform model training on an initial long short-term memory model, and determine an intermediate long short-term memory model;

[0010] Using a preset Shapley additive interpretation model and the intermediate long short-term memory model, according to the hydrological and meteorological test data set, calculate the Shapley values ​​corresponding to the hydrological and meteorological test data subsets of multiple time series in the hydrological and meteorological test data set;

[0011] constructing a plurality of target long-short term memory models according to the plurality of time series hydrometeorological test data subsets, the hydrometeorological observation data set, and the Shapley values ​​corresponding to the plurality of time series hydrometeorological test data subsets;

[0012] The hydrological and meteorological test data set and the hydrological and meteorological observation data set are used to screen each of the target long-term and short-term memory models to determine a saltwater tide upstream salinity forecast model.

[0013] Optionally, the target hydrometeorological measured data set includes multiple salinity data sets and multiple water-gas coupling data sets; the step of constructing a hydrometeorological training data set, a hydrometeorological test data set and a hydrometeorological observation data set based on multiple preset lag times, multiple preset lead times and the target hydrometeorological measured data set includes:

[0014] Based on a plurality of preset lag times, performing lag processing on the plurality of salinity data sets, and outputting a plurality of lag salinity data sets corresponding to each of the preset lag times;

[0015] Calculating partial correlation coefficients associated with the plurality of lag-time salinity data sets corresponding to the respective preset lag times according to the plurality of lag-time salinity data sets corresponding to the respective preset lag times;

[0016] Based on multiple preset advance times, a salinity data set associated with a lagged salinity data set corresponding to a maximum partial correlation coefficient is used as a target salinity data set, and the target salinity data set is time-advanced to output a time-advanced salinity data set corresponding to each of the preset advance times;

[0017] Based on the plurality of preset advance times, time-advancing the plurality of water-gas coupling data sets, and outputting a plurality of initial time-advanced water-gas coupling data sets corresponding to the respective preset advance times;

[0018] Calculate the Pearson correlation coefficient of the multiple initial time leading water-vapor coupling data sets corresponding to each of the preset advance times according to the multiple salinity data sets and the multiple initial time leading water-vapor coupling data sets corresponding to each of the preset advance times;

[0019] Based on the plurality of preset advance times, time-advance the water-gas coupling data set associated with the initial time-advance water-gas coupling data set corresponding to the largest Pearson correlation coefficient, and output the target time-advance water-gas coupling data set corresponding to each preset advance time;

[0020] Performing data sorting and standardization on the plurality of target time-advance water-gas coupling data sets and the plurality of time-advance salinity data sets, and outputting a plurality of standardized time-advance water-gas coupling data sets and a plurality of standardized time-advance salinity data sets;

[0021] Dividing the plurality of the standardized time-advanced water-gas coupling data sets and the plurality of the standardized time-advanced salinity data sets to generate a hydrometeorological training data set and a hydrometeorological test data set;

[0022] The target salinity data set is used as a hydrological and meteorological observation data set.

[0023] Optionally, the using of the preset Shapley additive interpretation model and the intermediate long short-term memory model to calculate the Shapley values ​​corresponding to the hydrological and meteorological test data subsets of multiple time series in the hydrological and meteorological test data set according to the hydrological and meteorological test data set includes:

[0024] Using the intermediate long short-term memory model, outputting the salinity prediction value corresponding to each of the hydrological and meteorological test data subsets of the time series according to each of the hydrological and meteorological test data subsets of the time series;

[0025] The preset Shapley additive interpretation model is used to calculate the Shapley value corresponding to each of the time series' hydrometeorological test data subsets based on the salinity prediction value corresponding to each of the time series' hydrometeorological test data subsets.

[0026] Optionally, constructing multiple target long-short term memory models according to multiple subsets of hydrometeorological test data of the time series, the hydrometeorological observation data set and the Shapley values ​​corresponding to multiple subsets of hydrometeorological test data of the time series includes:

[0027] Sorting the Shapley values ​​corresponding to the plurality of time series hydrological and meteorological test data subsets in descending order, selecting a preset number of time series hydrological and meteorological test data subsets as the first training data subset, and gradually accumulating the remaining time series hydrological and meteorological test data subsets into the first training data subset in order, to generate a plurality of second training data subsets;

[0028] The initial long short-term memory model is trained using the first training data subset, multiple second training data subsets, and the hydrological and meteorological observation data set to generate multiple target long short-term memory models.

[0029] Optionally, the using the hydrometeorological test data set and the hydrometeorological observation data set to screen each of the target long-short term memory models to determine a saltwater tide upstream salinity forecast model includes:

[0030] Using the hydrological and meteorological test data sets as inputs of the target long-short-term memory models, and outputting a plurality of salinity prediction values ​​corresponding to the target long-short-term memory models;

[0031] Using a plurality of salinity prediction values ​​corresponding to each of the target long and short-term memory models and the hydrological and meteorological observation data set to evaluate the model accuracy of each of the target long and short-term memory models, and outputting the Nash efficiency coefficient and root mean square error corresponding to each of the target long and short-term memory models;

[0032] The target long short-term memory model corresponding to the largest Nash efficiency coefficient and the smallest root mean square error is selected as the upstream salinity prediction model of saltwater tide.

[0033] Optionally, after the step of using the hydrometeorological test data set and the hydrometeorological observation data set to screen each of the target long-short term memory models to determine the saltwater tide upstream salinity forecast model, the method further comprises:

[0034] When the hydrological and meteorological measured data to be detected is received, the hydrological and meteorological measured data to be detected is input into the salt tide upstream salinity forecast model to perform salinity forecast, and an initial salinity forecast result is output;

[0035] The initial salinity forecast result is de-normalized to generate a target salinity forecast result.

[0036] A second aspect of the present invention provides a saltwater tide upstream salinity forecast model construction device, comprising:

[0037] An acquisition module is used to acquire a hydrological and meteorological measured data set, and preprocess the hydrological and meteorological measured data set to generate a target hydrological and meteorological measured data set;

[0038] Based on the module, it is used to construct a hydrometeorological training data set, a hydrometeorological test data set and a hydrometeorological observation data set based on multiple preset lag times, multiple preset lead times and the target hydrometeorological measured data set;

[0039] An adopting module is used to use the hydrological and meteorological training data set and the hydrological and meteorological observation data set to perform model training on the initial long short-term memory model to determine an intermediate long short-term memory model;

[0040] A calculation module, used to calculate the Shapley values ​​corresponding to the hydrometeorological test data subsets of multiple time series in the hydrometeorological test data set according to the hydrometeorological test data set by using a preset Shapley additive interpretation model and the intermediate long short-term memory model;

[0041] A construction module, for constructing a plurality of target long-short term memory models according to a plurality of said time series hydrometeorological test data subsets, said hydrometeorological observation data set and Shapley values ​​corresponding to a plurality of said time series hydrometeorological test data subsets;

[0042] The screening module is used to screen each of the target long-short term memory models using the hydrological and meteorological test data set and the hydrological and meteorological observation data set to determine the saltwater upstream salinity forecast model.

[0043] A third aspect of the present invention provides a computer device, comprising a memory and a processor, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the processor executes the steps of the method for constructing a saltwater upstream salinity forecast model as described in any one of the above items.

[0044] A fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed, implements the steps of the method for constructing a saltwater upstream salinity forecast model as described in any one of the above items.

[0045] A fifth aspect of the present invention provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium, and the computer program includes program instructions, wherein when the program instructions are executed by a computer, the computer executes the steps of the method for constructing a saltwater upstream salinity forecast model as described in any one of the above items.

[0046] It can be seen from the above technical solutions that the present invention has the following advantages:

[0047] The above technical scheme of the present invention provides a method for constructing a saltwater tide upstream salinity forecast model. First, a hydrometeorological measured data set is obtained, and the hydrometeorological measured data set is preprocessed to generate a target hydrometeorological measured data set; then, based on multiple preset lag times, multiple preset lead times, and the target hydrometeorological measured data set, a hydrometeorological training data set, a hydrometeorological test data set, and a hydrometeorological observation data set are constructed; the hydrometeorological training data set and the hydrometeorological observation data set are used to train an initial long-short-term memory model, and an intermediate long-short-term memory model is determined; a preset Shapley additive interpretation model and an intermediate long-short-term memory model are used to calculate the Shapley values ​​corresponding to the hydrometeorological test data subsets of multiple time series in the hydrometeorological test data set according to the hydrometeorological test data set; according to the hydrometeorological test data subsets of multiple time series, the hydrometeorological observation data set, and the intermediate long-short-term memory model, the Shapley values ​​corresponding to the hydrometeorological test data subsets are calculated according to the intermediate long-short-term memory model. The Shapley values ​​corresponding to the subsets of hydrometeorological test data of a time series are used to construct multiple target long-short-term memory models; finally, the hydrometeorological test data set and the hydrometeorological observation data set are used to screen the target long-short-term memory models to determine the saltwater upstream salinity forecast model; based on the above scheme, the acquired hydrometeorological measured data set is processed to obtain the hydrometeorological training data set, the hydrometeorological test data set and the hydrometeorological observation data set, and the preset Shapley additive interpretation model and the long-short-term memory model are combined to obtain the saltwater upstream salinity forecast model according to the hydrometeorological training data set, the hydrometeorological test data set and the hydrometeorological observation data set. The process does not require a large amount of detailed information on boundary or initial conditions such as runoff, tide, wind, riverbed topography, etc., and the obtained saltwater upstream salinity forecast model can be directly used to make the corresponding salinity forecast, which can reduce the calculation time and thus achieve rapid salinity forecast. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0049] Figure 1 A flow chart of the steps of a method for constructing a saltwater tide upstream salinity forecast model provided in Example 1 of the present invention;

[0050] Figure 2 A schematic diagram of partial autocorrelation coefficients of salinity series with different lag times provided in Example 1 of the present invention;

[0051] Figure 3A schematic diagram of the Pearson correlation coefficient between the tide level factor sequence, the runoff factor sequence, the wind speed and direction factor sequence and the salinity factor sequence at different lead times provided in the first embodiment of the present invention;

[0052] Figure 4 A basic structural diagram of a long short-term memory neural network neuron provided in Example 1 of the present invention;

[0053] Figure 5 A schematic diagram of the results of SHAP analysis based on the constructed long short-term memory model provided in the first embodiment of the present invention;

[0054] Figure 6 The accuracy changes of the model training set (a) and test set (b) with different numbers of features provided in Example 1 of the present invention;

[0055] Figure 7 A schematic diagram of the prediction results of model G provided in Example 1 of the present invention;

[0056] Figure 8 This is a structural block diagram of a device for building a saltwater upstream salinity forecast model provided in Example 2 of the present invention. DETAILED DESCRIPTION

[0057] The embodiment of the present invention provides a method and device for constructing a saltwater upwelling salinity forecast model, which is used to solve the technical problem that the existing saltwater upwelling salinity forecast technology cannot achieve rapid forecasting.

[0058] In order to make the purpose, features and advantages of the present invention more obvious and easy to understand, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described below are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0059] See also Figure 1 , Figure 1 A flowchart of the steps of a method for constructing a saltwater upstream salinity forecast model provided in Example 1 of the present invention.

[0060] The present invention provides a method for constructing a saltwater tide upstream salinity forecasting model, comprising:

[0061] Step 101: Acquire a hydrological and meteorological measured data set, and preprocess the hydrological and meteorological measured data set to generate a target hydrological and meteorological measured data set.

[0062] It should be noted that, through the analysis of the influencing factors of the physical process of saltwater upwelling and data query, the hydrological and meteorological measured data sets in the dry season of the target area are collected. The hydrological and meteorological measured data sets include multiple salinity data sets and multiple water-gas coupling data sets that have not been preprocessed. The salinity data sets include multiple salinity data, and the water-gas coupling data sets include multiple tide level data sets, multiple flow data sets, and multiple wind speed and direction data sets.

[0063] Furthermore, the data preprocessing process includes data cleaning, such as removing duplicate data, correcting or deleting outliers, processing missing values, etc., to ensure the integrity and continuity of the data. In addition, if the scale of the collected data is inconsistent with the forecast scale, data integration must be performed first, such as aggregating hourly data into daily data.

[0064] Step 102: construct a hydrometeorological training dataset, a hydrometeorological test dataset, and a hydrometeorological observation dataset based on multiple preset lag times, multiple preset lead times, and a target hydrometeorological measured dataset.

[0065] Further, step 102 may include the following sub-steps S21-S29:

[0066] Step S21, based on multiple preset lag times, performing lag processing on multiple salinity data sets, and outputting multiple lag salinity data sets corresponding to each preset lag time;

[0067] Step S22, calculating the partial correlation coefficient associated with the multiple lag salinity data sets corresponding to the preset lag times according to the multiple lag salinity data sets corresponding to the preset lag times;

[0068] Step S23, based on multiple preset advance times, taking the salinity data set associated with the lagged salinity data set corresponding to the largest partial correlation coefficient as the target salinity data set, and time-advancing the target salinity data set, and outputting the time-advanced salinity data sets corresponding to each preset advance time;

[0069] It should be noted that the partial correlation coefficient between salinity series with different lags is calculated and visualized, which reflects the net correlation between two salinity series with different lags when one or more other variables are controlled. The salinity series corresponding to the lag with the higher (largest) partial correlation coefficient is selected as the input data series for model training.

[0070] Specifically, the multiple preset lag times are multiple different lag times that are set, and the multiple preset advance times are multiple different advance times that are set. Based on each lag time, each salinity data set is subjected to lag processing, and multiple lag salinity data sets corresponding to each preset lag time are output; for example, assuming that the preset lag time is a lag of 24 hours and a lag of 25 hours, and the salinity data set includes a salinity data set for January and a salinity data set for February, a salinity data set for January with a lag of 24 hours, a salinity data set for February with a lag of 24 hours, a salinity data set for January with a lag of 25 hours, and a salinity data set for February with a lag of 25 hours can be obtained.

[0071] Furthermore, among all the partial correlation coefficients associated with the lagged salinity data sets, the salinity data set associated with the lagged salinity data set corresponding to the largest partial correlation coefficient is selected. For example, if the lagged salinity data set corresponding to the largest partial correlation coefficient is a salinity data set in January that is lagged by 24 hours, then the salinity data set in January is used as the target salinity data set, and then the target salinity data set is time-advanced by setting a plurality of different advance times, and the time-advanced salinity data sets corresponding to each preset advance time are output. For example, assuming that the preset advance time is 24 hours in advance and 25 hours in advance, and the target salinity data set is a salinity data set in January, then the time-advanced salinity data set corresponding to the preset advance time may be a salinity data set in January that is 24 hours in advance and a salinity data set in January that is 25 hours in advance.

[0072] For example, see Figure 2 , assuming that the hydrological and meteorological measured data set is the measured hydrological and meteorological historical data collected during the dry season of Modaomen Waterway from 2020 to 2023, including the hourly average salinity data of Pinggang Station, the hourly tide data of Sanzao Station, the hourly flow data of Makou Station, and the wind direction and speed data. Among them, the wind speed and direction data of Macau, China are derived from the ERA5-Land hourly data from 1950 to present data set of the European Center for Medium-Term Weather Forecasts (ECMWF). The wind speed in the u and v directions at 10m in the data set is used as two candidate prediction factors. According to the hydrological and meteorological measured data set, the partial autocorrelation coefficients of the salinity series at different lags are calculated as follows: Figure 2 shown.

[0073] Step S24, based on multiple preset advance times, time-advance multiple water-gas coupling data sets, and output multiple initial time-advance water-gas coupling data sets corresponding to each preset advance time;

[0074] Step S25, calculating the Pearson correlation coefficient of the multiple initial time advance water-vapor coupling data sets corresponding to each preset advance time according to the multiple salinity data sets and the multiple initial time advance water-vapor coupling data sets corresponding to each preset advance time;

[0075] Step S26, based on multiple preset advance times, time-advance the water-gas coupling data set associated with the initial time-advance water-gas coupling data set corresponding to the largest Pearson correlation coefficient, and output the target time-advance water-gas coupling data set corresponding to each preset advance time;

[0076] It should be noted that for other collected data sequences except the salinity factor, the Pearson correlation coefficient between the tide factor sequence, runoff factor sequence, wind speed and direction factor sequence and salinity factor sequence at different lead times is first calculated. The calculation formula of the Pearson correlation coefficient can be expressed as:

[0077] ;

[0078] Among them, R is the Pearson correlation coefficient, which ranges from [-1,1] and indicates the degree of linear correlation. The absolute value close to 1 indicates that the greater the correlation, the better the fit; is the salinity data for the i-th hour; is the average value of the salinity series; is the value corresponding to the water-gas coupling data of the ith hour in the water-gas coupling data sequence with different preset advance times, where the water-gas coupling data includes tide level data, flow data, and wind speed and direction data; is the average value of the water-air coupling data series; n is the number of data contained in the data series.

[0079] Furthermore, for each prediction factor sequence, the exogenous driving item of the advance time series with the larger (maximum) correlation coefficient is taken as the sequence to be input to the model; for example, assuming that the preset advance time is 24 hours in advance and 25 hours in advance, the initial time advance water-gas coupling data set corresponding to the largest Pearson correlation coefficient is the water-gas coupling data set for January 24 hours in advance, then based on 24 hours in advance and 25 hours in advance, the water-gas coupling data set for January is time-advanced, and the tide level data set for January 24 hours in advance, the flow data for January 24 hours in advance, the wind speed and direction data set for January 24 hours in advance, the tide level data set for January 25 hours in advance, the flow data for January 25 hours in advance, and the wind speed and direction data set for January 25 hours in advance are output.

[0080] For example, see Figure 3, the Pearson correlation coefficients between the tide level factor sequence, runoff factor sequence, wind speed and direction factor sequence and salinity factor sequence at different lead times were calculated. The Pearson correlation coefficients of some characteristic variables are as follows: Figure 3 As shown in Table 1, taking the hourly scale forecast of 24-hour salinity as an example, the selected forecast factors (i.e., water-gas coupling data and salinity data of different time series) are shown in Table 1, where S is the hourly salinity of Pinggang Station (mg / L); F is the hourly flow of Makou (m 3 / s); T is the hourly tide level at Sanzao station (m); W1 is the wind speed in the u direction of Macao, China (m / s); W2 is the wind speed in the v direction of Macao, China (m / s); S_24 represents salinity data 24 hours ahead of the salinity forecast value (salinity data 24 hours ahead), S_25 represents salinity data 25 hours ahead of the salinity forecast value, T_24 represents tide level data 24 hours ahead of the tide level forecast value, F_24 represents flow rate data 24 hours ahead of the flow rate forecast value, W1_24 represents wind speed data in the u direction of Macao, China 24 hours ahead of the wind speed forecast value in the u direction of Macao, China, and W2_24 represents wind speed in the v direction of Macao, China 24 hours ahead of the wind speed forecast value in the v direction of Macao, China.

[0081]

[0082] Step S27, arranging and standardizing the multiple target time-advanced water-gas coupling data sets and the multiple time-advanced salinity data sets, and outputting multiple standardized time-advanced water-gas coupling data sets and multiple standardized time-advanced salinity data sets;

[0083] It should be noted that all the selected prediction factor sequences (target time-advanced water-gas coupling data set and time-advanced salinity data set) are organized into a two-dimensional data set, one dimension is the time series dimension, and the other dimension is the input feature type, and the input data is standardized; among them, all data need to be standardized before input to avoid the dimensional influence between the prediction factors.

[0084] Step S28, dividing the multiple standardized time-advanced water-gas coupling data sets and the multiple standardized time-advanced salinity data sets to generate a hydrometeorological training data set and a hydrometeorological test data set;

[0085] It should be noted that the hydrometeorological training dataset and the hydrometeorological test dataset are taken as a whole dataset, and divided into a training set and a test set according to the time series in a ratio of 6:4, namely, the hydrometeorological training dataset and the hydrometeorological test dataset; the hydrometeorological training dataset includes a plurality of salinity datasets, tidal level datasets, flow datasets, and wind speed and direction datasets of different time series used for the initial model training, and the hydrometeorological test dataset includes a plurality of salinity datasets, tidal level datasets, flow datasets, and wind speed and direction datasets of different time series used for shap value calculation, etc.; for example, assuming that the time series is 24 hours in advance and 25 hours in advance, the hydrometeorological training dataset includes a salinity dataset, a tidal level dataset, a flow dataset, and a wind speed and direction dataset in January 24 hours in advance, and a salinity dataset, a tidal level dataset, a flow dataset, and a wind speed and direction dataset in January 25 hours in advance.

[0086] Step S29: Use the target salinity data set as a hydrological and meteorological observation data set.

[0087] It should be noted that the data in the hydrometeorological observation dataset corresponds to the data in the hydrometeorological training dataset and the hydrometeorological test dataset. For example, assuming that the hydrometeorological training dataset includes the salinity dataset and tide level dataset of January 24 hours in advance, the hydrometeorological observation dataset also includes the salinity dataset and tide level dataset of January.

[0088] Step 103: Use the hydrological and meteorological training data set and the hydrological and meteorological observation data set to train the initial long short-term memory model to determine the intermediate long short-term memory model.

[0089] It should be noted that the Python library required for prediction is imported, the preprocessed data set is imported, and the data set is converted into a three-dimensional data set to obtain a three-dimensional data set (number of features, time series, number of samples).

[0090] For further information, see Figure 4 , build the LSTM model and set the corresponding model parameters. The LSTM (Long Short Term Memory) neural network model, also known as the long short-term memory model, is a special recurrent neural network (RNN) architecture proposed by Hochreiter and Schmidhuber in 1997. It can alleviate the gradient vanishing or gradient exploding problems encountered by standard RNN when processing long sequence data by selectively retaining and forgetting input memory, and can solve the dependency problem of different time series periods. Its basic structure is as follows Figure 4As shown. The LSTM memory unit contains three gates, namely forget, input, and output gates. The formula for its output implementation is as follows:

[0091] ;

[0092] ;

[0093] ;

[0094] ;

[0095] ;

[0096] ;

[0097] ;

[0098] ;

[0099] in, For the Gate of Oblivion; is a nonlinear activation function; is the first weight vector corresponding to the forget gate; is the input value of the first layer; is the second weight vector corresponding to the forget gate; is the activation vector of the block at time t-1; is the basis vector corresponding to the forget gate; is the input gate; is the first weight vector corresponding to the input gate; is the second weight vector corresponding to the input gate; is the basis vector corresponding to the input gate; is the candidate memory neuron state; is the tanh activation function; is the first weight matrix corresponding to the candidate memory unit state; is the second weight matrix corresponding to the candidate memory unit state; is the bias vector of the candidate memory cell state; is the neuron state at time t; is the neuron state at time t-1; is the output gate; is the first weight matrix corresponding to the output gate; is the second weight matrix corresponding to the output gate; is the bias vector of the output gate; is the activation vector of the block at time t; is the output value of the second layer; is the weight matrix of the hidden state; is the bias vector of the hidden state.

[0100] Furthermore, the model accuracy is preliminarily verified based on the hydrological and meteorological observation dataset, and the model hyperparameters are adjusted. If the model accuracy is poor, it is necessary to continue to adjust the model parameters and retrain until the model accuracy reaches a certain standard.

[0101] In this embodiment, the above-mentioned initially selected prediction factor data sets (hydrological and meteorological training data sets and hydrological and meteorological observation data sets) are used to construct an LSTM model, and the corresponding model parameters are set. The model is trained by combining different parameters to construct a suitable model.

[0102] Step 104: Using a preset Shapley additive interpretation model and an intermediate long short-term memory model, the Shapley values ​​corresponding to the hydrometeorological test data subsets of multiple time series in the hydrometeorological test data set are calculated according to the hydrometeorological test data set.

[0103] The hydrological and meteorological test data subsets of multiple time series include salinity data sets of multiple time series, tide level data sets of multiple time series, flow data sets of multiple time series, and wind speed and direction data sets of multiple time series.

[0104] Further, step 104 may include the following sub-steps:

[0105] Step S41, using the intermediate long short-term memory model to output the salinity prediction value corresponding to the hydrological and meteorological test data subset of each time series according to the hydrological and meteorological test data subset of each time series;

[0106] Step S42: using a preset Shapley additive interpretation model to calculate the Shapley value corresponding to the hydrometeorological test data subset of each time series according to the salinity prediction value corresponding to the hydrometeorological test data subset of each time series.

[0107] It should be noted that the shap library (preset Shapley additive explanatory model) in Python is called to calculate the shap value of the hydrometeorological test data subsets of multiple time series, that is, to calculate the shap value corresponding to each hydrometeorological test data subset of different time series and different data types. For example, assuming that the hydrometeorological test data subsets of multiple time series include salinity data sets 24 hours in advance, salinity data sets 25 hours in advance, and flow data sets 24 hours in advance, the shap value corresponding to the salinity data sets 24 hours in advance, the shap value corresponding to the salinity data sets 25 hours in advance, and the shap value corresponding to the flow data sets 24 hours in advance can be calculated. The SHAP (SHapley Additive ex Planations) model was proposed by Lundberg et al. in 2017. The SHAP method can calculate the marginal contribution of features to the model prediction results, that is, the SHAP value (Shapley value). It borrows the concept of Shapley value in game theory and can be applied to various types of machine learning models, including decision trees, random forests, gradient boosting machines, deep learning models, etc. The SHAP value can be used to better understand how features affect the model's prediction results and the extent of the impact, thereby improving the transparency and credibility of the model.

[0108] Furthermore, the calculation formula of the Shapley value can be expressed as:

[0109] ;

[0110] in, is the Shapley value corresponding to the feature subset of the ith feature, which represents the salinity data set, tidal data set, flow data set, or wind speed and direction data set of any time series input to the intermediate long short-term memory model. When it is greater than 0, it means that the feature improves the prediction value and plays a positive role. Otherwise, it means that the feature reduces the prediction value and plays a negative role. S is the feature subset that does not contain the ith feature, that is, the feature subset that does not contain the salinity data set, tidal data set, flow data set, or wind speed and direction data set of any time series input to the intermediate long short-term memory model. N is the set of all features, that is, the data set composed of the hydrological and meteorological test data subsets of all time series. The salinity prediction value of feature subset S plus feature subset i; It is the salinity prediction value output by the intermediate long short-term memory model when the input is the feature subset S.

[0111] Furthermore, the SHAP model decomposes the model's prediction results into the contribution of each feature. Each feature has a corresponding SHAP value, which indicates the impact of the feature on the model's prediction results. For each feature, the SHAP value is obtained by calculating the marginal contribution of the feature in all possible feature combinations.

[0112] For example, see Figure 5 , call the shap library in Python to calculate the shap value of the test set of the long short-term memory model. The results of the SHAP analysis of the initially constructed long short-term memory model are as follows Figure 5 As shown, for the convenience of representation, the present invention names the salinity prediction factors as S1, S2, S3, S4, S5, and other variables are deduced by analogy, that is, Si represents salinity data of different time series, Fi represents flow data of different time series, Ti represents tide data of different time series, W1i represents wind speed in the direction of wind speed u of different time series, and W2i represents wind speed in the direction of wind speed v of different time series. As can be seen from Figure 5, in the prediction process of the model, the feature importance ranking of S1, S2, and F1 is at the top, that is, the salinity series (salinity data set) 24 hours in advance and 25 hours in advance play a greater role in the model prediction process, which is consistent with the research results of predecessors who used physical models to explore the physical mechanism of saltwater tide traceability.

[0113] Step 105: construct multiple target long short-term memory models based on the hydrometeorological test data subsets of multiple time series, the hydrometeorological observation data set, and the Shapley values ​​corresponding to the hydrometeorological test data subsets of multiple time series.

[0114] Further, step 105 may include the following sub-steps S51-S52:

[0115] S51, sorting the Shapley values ​​corresponding to the hydrological and meteorological test data subsets of multiple time series in descending order, selecting a preset number of hydrological and meteorological test data subsets of the time series as the first training data subset, and gradually accumulating the hydrological and meteorological test data subsets of the remaining time series into the first training data subset in order, to generate multiple second training data subsets;

[0116] S52: Use the first training data subset, multiple second training data subsets and the hydrological and meteorological observation data set to perform model training on the initial long short-term memory model to generate multiple target long short-term memory models.

[0117] It should be noted that according to the calculation results of the shap value, the features included in the input data set of the model are adjusted, the predictor sequence with the feature importance ranked first is input first, the number of input features, that is, the preset number generally starts from 3 and increases one by one, and models containing different numbers and types of features are established and trained respectively. For example, assuming that after the Shapley values ​​corresponding to the hydrological and meteorological test data subsets of all time series are sorted in descending order, the salinity sequence 24 hours in advance ranks first, the salinity sequence 25 hours in advance ranks second, the salinity sequence 26 hours in advance ranks third, the tidal level sequence 24 hours in advance ranks fourth, and the tidal level sequence 25 hours in advance ranks fifth, then the salinity sequence 24 hours in advance, the salinity sequence 25 hours in advance, and the salinity sequence 26 hours in advance are used to construct the first training data subset, and the tidal level sequence 24 hours in advance and the tidal level sequence 25 hours in advance are gradually accumulated to the first training data subset, that is, the first training data subset and the tidal level sequence 24 hours in advance are used to construct data, and the first training data subset, the tidal level sequence 24 hours in advance, and the tidal level sequence 25 hours in advance are used to construct data, and two second training data subsets are obtained. Then, the first training data subset is used to train the initial long short-term memory model, and the two second training data subsets are used to train the initial long short-term memory model respectively, and three target long short-term memory models are output.

[0118] Exemplarily, according to the shap calculation result, the features included in the input data set of the model are adjusted, the predictor sequence with the feature importance ranking at the front is prioritized, the number of input features increases one by one from 3, and models containing different numbers of types of features are established and trained respectively, named A, B, C, D, E, F, G, H, I, and J. The feature factors input to each model are shown in Table 2.

[0119]

[0120] Step 106: Use the hydrometeorological test data set and the hydrometeorological observation data set to screen the target long-term and short-term memory models to determine the saltwater upstream salinity forecast model.

[0121] Further, step 106 may include the following sub-steps:

[0122] S61, using the hydrological and meteorological test data set as the input of each target long-short-term memory model, and outputting a plurality of salinity prediction values ​​corresponding to each target long-short-term memory model;

[0123] S62, using multiple salinity prediction values ​​and hydrological and meteorological observation data sets corresponding to each target long and short-term memory model to evaluate the model accuracy of each target long and short-term memory model, and outputting the Nash efficiency coefficient and root mean square error corresponding to each target long and short-term memory model;

[0124] S63. Select the target long-short term memory model corresponding to the largest Nash efficiency coefficient and the smallest root mean square error as the saltwater upstream salinity prediction model.

[0125] It should be noted that the model with different characteristics (target long-short term memory model) established based on the above steps uses the Nash efficiency coefficient (NSE) and root mean square error (RMSE) to evaluate the model salinity forecast. The calculation methods of NSE and RMSE are as follows:

[0126] ;

[0127] ;

[0128] Among them, NSE is the Nash efficiency coefficient, and the value range of NSE is (-∞,1]. When NSE=1, it means that the model forecast is completely consistent; when NSE=0, it means that the explanatory power of the forecast result is the same as the mean of the observed value; when NSE<0, it means that the model forecast effect is lower than the mean forecast method; i is the predicted value of salinity at the ith time; P i is the i-th target salinity data (salinity observation value); RMSE is the root mean square error, which indicates the difference between the predicted value and the observed value. When the difference increases, it gradually increases from 0 to +∞; is the average value of the target salinity data (the average value of the salinity observations); n is the total number of samples.

[0129] For example, see Figure 6-Figure 7 , using NSE and RMSE to evaluate the model prediction accuracy, the model accuracy changes with different numbers of features are obtained as follows Figure 6 As shown in Figure 3, according to the calculation results of each model accuracy evaluation index, the model with the largest NSE and the smallest RMSE is selected as the optimal model. Based on the analysis of Table 3, it can be seen that when the number of model input features is 9, the model accuracy is the best, that is, model G is the optimal model. The prediction results of the model are visualized as follows: Figure 7 As shown, the NSE of the training set and the test set can reach 0.764 and 0.738 respectively. Compared with the model of the initial prediction factor, the accuracy is improved and can basically meet the prediction needs. Among them, the prediction results and the accuracy evaluation of model G show that the method of the present invention for upstream prediction of salt tide based on interpretable machine learning is combined with the LSTM model and the SHAP model for refined prediction of salinity, which not only improves the problem of too many initial prediction factors and the troublesome application, improves the rationality of the model, but also further improves the prediction accuracy of the model. Furthermore, the current optimal model can be saved, and when the latest data of the study area is collected, efficient prediction of future salinity can be made.

[0130]

[0131] Optionally, after the step of screening each target long-short term memory model using the hydrometeorological test data set and the hydrometeorological observation data set to determine the saltwater tide upstream salinity forecast model, the method includes:

[0132] When the hydrological and meteorological measured data to be detected are received, the hydrological and meteorological measured data to be detected are input into the salt tide upstream salinity forecast model for salinity forecasting, and the initial salinity forecast result is output;

[0133] The initial salinity forecast results are denormalized to generate the target salinity forecast results.

[0134] It should be noted that the best model (salinity forecast model for upstream saltwater tide) is saved in pb format or h5 format. When necessary, the forecast factor with the required lag is directly input to call the model (salinity forecast model for upstream saltwater tide) to make upstream saltwater tide forecast for the corresponding forecast period. Each time the upstream saltwater tide salinity forecast model is used for prediction, the prediction result data needs to be de-standardized.

[0135] As a comparison of technical effects, it can be combined with existing technologies for reference. Salt tide upstream is a natural hydrological phenomenon, which is mainly caused by factors such as reduced river flow, tidal changes, wind direction and wind force, and rising sea levels. The freshwater flow in rivers decreases during the dry season, and the rising tide during high tides makes it easy for seawater to flow back into the river, forming a salt tide. In addition, when the wind force and direction are consistent with the tidal direction, the advancement of the salt tide will be aggravated. The rise in sea levels caused by global warming has also expanded the scope of influence of salt tides. In general, the upstream of salt tides will have a huge negative impact on the utilization of freshwater resources, industrial and agricultural production, and aquatic habitat systems in estuary areas. Long-term drinking of water with excessive chloride content will seriously endanger human health.

[0136] Recurrent Neural Network (RNN) is a deep learning model suitable for processing sequence data. RNN can keep the memory of previous information when processing the sequence and capture the dynamic characteristics in the time series, which is very suitable for time series prediction tasks. The LSTM model effectively solves the gradient vanishing problem of traditional RNN by introducing a gating mechanism, and captures the long-distance dependency in the sequence data with fewer parameters and higher training efficiency. It has been widely used in the forecasting field. However, simulation research on saltwater upstream salinity forecast based on LSTM model is relatively rare. LSTM model is mostly used for runoff forecast in hydrology.

[0137] In view of the above problems, the present invention proposes a method for predicting the time series of saltwater tides by using a long short-term memory network (LSTM), so as to realize the prediction of saltwater tides at multiple time scales. The introduction of the SHAP model allows us to intuitively understand the contribution of the input features in the model and the importance ranking of the features. Combined with the physical mechanism of the saltwater tide, the rationality of the forecast model can be further verified. Therefore, combining the LSTM model and the SHAP model to quickly and effectively predict the saltwater tides in the estuary area during the dry season is of great significance for ensuring the water resources security and ecological environment protection in coastal areas.

[0138] Specifically, the present invention first collects and preprocesses the data of the study area, and preliminarily selects the prediction factors suitable for the lead time based on the partial autocorrelation analysis and the Pearson correlation coefficient, and uses the preliminarily selected prediction factors to build an LSTM (Long Short-Term Memory Network) model, and further uses the SHAP model to analyze the established model, sorts according to the importance characteristics, and screens the prediction factors of the input model again, thereby building an optimal salinity prediction model to achieve the best prediction effect. The present invention can effectively and finely predict the salinity value of the saltwater tide in the estuary area during the dry season, and at the same time, the introduction of the SHAP model improves the interpretability of the machine learning black box model, thereby providing a scientific basis for the precise prevention and control of the saltwater tide in the estuary area.

[0139] In an embodiment of the present invention, the present invention provides a method for constructing a saltwater tide upstream salinity forecast model. First, a hydrometeorological measured data set is obtained, and the hydrometeorological measured data set is preprocessed to generate a target hydrometeorological measured data set; then, based on multiple preset lag times, multiple preset lead times, and the target hydrometeorological measured data set, a hydrometeorological training data set, a hydrometeorological test data set, and a hydrometeorological observation data set are constructed; the hydrometeorological training data set and the hydrometeorological observation data set are used to train an initial long-short-term memory model to determine an intermediate long-short-term memory model; a preset Shapley additive interpretation model and an intermediate long-short-term memory model are used to calculate the Shapley values ​​corresponding to the hydrometeorological test data subsets of multiple time series in the hydrometeorological test data set according to the hydrometeorological test data set; according to the hydrometeorological test data subsets of multiple time series, the hydrometeorological observation data set, and the multiple time series The Shapley value corresponding to the subset of the hydrometeorological test data of the sequence is used to construct multiple target long-short-term memory models; finally, the hydrometeorological test data set and the hydrometeorological observation data set are used to screen the target long-short-term memory models to determine the saltwater upstream salinity forecast model; based on the above scheme, the acquired hydrometeorological measured data set is processed to obtain the hydrometeorological training data set, the hydrometeorological test data set and the hydrometeorological observation data set, and the preset Shapley additive interpretation model and the long-short-term memory model are combined to obtain the process of the saltwater upstream salinity forecast model according to the hydrometeorological training data set, the hydrometeorological test data set and the hydrometeorological observation data set. The detailed information of the boundary or initial conditions such as a large amount of runoff, tide, wind, riverbed topography, etc. is not required. The obtained saltwater upstream salinity forecast model can be directly used to make the corresponding salinity forecast, which can reduce the calculation time and simplify the process mechanism, thereby realizing rapid salinity forecasting.

[0140] See also Figure 8 , Figure 8 A structural block diagram of a saltwater upstream salinity forecast model construction device provided in the second embodiment of the present invention

[0141] The present invention provides a saltwater tide upstream salinity forecast model construction device, comprising:

[0142] The acquisition module 801 is used to acquire a hydrological and meteorological measured data set, and pre-process the hydrological and meteorological measured data set to generate a target hydrological and meteorological measured data set;

[0143] Based on module 802, it is used to construct a hydrometeorological training data set, a hydrometeorological test data set and a hydrometeorological observation data set based on multiple preset lag times, multiple preset lead times and a target hydrometeorological measured data set;

[0144] Adopting module 803, for using the hydrological and meteorological training data set and the hydrological and meteorological observation data set to perform model training on the initial long short-term memory model, and determine the intermediate long short-term memory model;

[0145] A calculation module 804 is used to calculate the Shapley values ​​corresponding to the hydrometeorological test data subsets of multiple time series in the hydrometeorological test data set according to the hydrometeorological test data set by using a preset Shapley additive interpretation model and an intermediate long short-term memory model;

[0146] A construction module 805 is used to construct multiple target long short-term memory models according to the hydrological and meteorological test data subsets of multiple time series, the hydrological and meteorological observation data set and the Shapley values ​​corresponding to the hydrological and meteorological test data subsets of multiple time series;

[0147] The screening module 806 is used to screen each target long-term and short-term memory model using the hydrological and meteorological test data set and the hydrological and meteorological observation data set to determine the saltwater upstream salinity forecast model.

[0148] Furthermore, the target hydrological and meteorological measured data set includes multiple salinity data sets and multiple water-gas coupling data sets; based on module 802, it is specifically used for:

[0149] Based on multiple preset lag times, multiple salinity data sets are subjected to lag processing, and multiple lag salinity data sets corresponding to each preset lag time are output;

[0150] According to the multiple lag salinity data sets corresponding to the respective preset lag times, calculating the partial correlation coefficient associated with the multiple lag salinity data sets corresponding to the respective preset lag times;

[0151] Based on multiple preset advance times, the salinity data set associated with the lagged salinity data set corresponding to the largest partial correlation coefficient is used as the target salinity data set, and the target salinity data set is time-advanced to output the time-advanced salinity data sets corresponding to each preset advance time;

[0152] Based on multiple preset advance times, multiple water-gas coupling data sets are time-advanced, and multiple initial time-advanced water-gas coupling data sets corresponding to each preset advance time are output;

[0153] Calculate the Pearson correlation coefficient of the multiple initial time advance water-gas coupling data sets corresponding to each preset advance time according to the multiple salinity data sets and the multiple initial time advance water-gas coupling data sets corresponding to each preset advance time;

[0154] Based on multiple preset advance times, the water-gas coupling data set associated with the initial time-advance water-gas coupling data set corresponding to the largest Pearson correlation coefficient is time-advanced, and the target time-advance water-gas coupling data set corresponding to each preset advance time is output;

[0155] Perform data sorting and standardization on multiple target time-advance water-air coupling data sets and multiple time-advance salinity data sets, and output multiple standardized time-advance water-air coupling data sets and multiple standardized time-advance salinity data sets;

[0156] Divide multiple standardized time-advanced water-air coupling data sets and multiple standardized time-advanced salinity data sets to generate hydrometeorological training data sets and hydrometeorological test data sets;

[0157] The target salinity dataset is used as the hydrological and meteorological observation dataset.

[0158] Furthermore, the calculation module 804 is specifically configured to:

[0159] The intermediate long short-term memory model is used to output the salinity prediction value corresponding to the hydrological and meteorological test data subset of each time series according to the hydrological and meteorological test data subset of each time series;

[0160] The preset Shapley additive interpretation model is used to calculate the Shapley value corresponding to the subset of hydrometeorological test data of each time series based on the salinity prediction value corresponding to the subset of hydrometeorological test data of each time series.

[0161] Further, the construction module 805 is specifically used for:

[0162] Sort the Shapley values ​​corresponding to the hydrological and meteorological test data subsets of multiple time series in descending order, select a preset number of hydrological and meteorological test data subsets of the time series as the first training data subset, and gradually add the hydrological and meteorological test data subsets of the remaining time series to the first training data subset in order to generate multiple second training data subsets;

[0163] The initial long short-term memory model is trained using a first training data subset, multiple second training data subsets, and a hydrological and meteorological observation data set to generate multiple target long short-term memory models.

[0164] Furthermore, the screening module 806 is specifically used for:

[0165] The hydrological and meteorological test data sets are used as the input of each target long and short-term memory model, and multiple salinity prediction values ​​corresponding to each target long and short-term memory model are output;

[0166] The model accuracy of each target long and short-term memory model is evaluated using multiple salinity prediction values ​​and hydrological and meteorological observation data sets corresponding to each target long and short-term memory model, and the Nash efficiency coefficient and root mean square error corresponding to each target long and short-term memory model are output;

[0167] The target long short-term memory model corresponding to the largest Nash efficiency coefficient and the smallest root mean square error is selected as the upstream salinity prediction model of saltwater tide.

[0168] In an optional embodiment, the device further comprises:

[0169] The first module is used for inputting the hydrological and meteorological measured data to be detected into the salt tide upstream salinity forecast model to perform salinity forecast and output the initial salinity forecast result when receiving the hydrological and meteorological measured data to be detected;

[0170] The second module is used to de-standardize the initial salinity forecast results to generate target salinity forecast results.

[0171] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described devices and modules can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0172] An embodiment of the present invention also provides a computer device, including a memory and a processor, wherein a computer program is stored in the memory; when the computer program is executed by the processor, the processor executes the steps of the method for constructing a saltwater upstream salinity forecast model as described in any of the above embodiments.

[0173] An embodiment of the present invention further provides a computer-readable storage medium having a computer program / instruction stored thereon. When the computer program / instruction is executed by a processor, the steps of the method for constructing a saltwater upstream salinity forecast model as described in any of the above embodiments are implemented.

[0174] An embodiment of the present invention further provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the method for constructing a saltwater upstream salinity forecast model as described in any of the above embodiments.

[0175] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0176] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0177] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for constructing a saltwater tide upstream salinity forecast model, characterized in that: include: Acquire a hydrological and meteorological measured data set, and preprocess the hydrological and meteorological measured data set to generate a target hydrological and meteorological measured data set; Based on multiple preset lag times, multiple preset lead times, and the target hydrometeorological measured data set, construct a hydrometeorological training data set, a hydrometeorological test data set, and a hydrometeorological observation data set; Using the hydrological and meteorological training data set and the hydrological and meteorological observation data set to perform model training on an initial long short-term memory model, and determine an intermediate long short-term memory model; Using a preset Shapley additive interpretation model and the intermediate long short-term memory model, according to the hydrological and meteorological test data set, calculate the Shapley values ​​corresponding to the hydrological and meteorological test data subsets of multiple time series in the hydrological and meteorological test data set; constructing a plurality of target long-short term memory models according to the plurality of time series hydrometeorological test data subsets, the hydrometeorological observation data set, and the Shapley values ​​corresponding to the plurality of time series hydrometeorological test data subsets; The hydrological and meteorological test data set and the hydrological and meteorological observation data set are used to screen each of the target long-term and short-term memory models to determine a saltwater tide upstream salinity forecast model.

2. The method for constructing a saltwater tide upstream salinity forecasting model according to claim 1, characterized in that: The target hydrometeorological measured data set includes multiple salinity data sets and multiple water-gas coupling data sets; the hydrometeorological training data set, the hydrometeorological test data set and the hydrometeorological observation data set are constructed based on multiple preset lag times, multiple preset lead times and the target hydrometeorological measured data set, including: Based on a plurality of preset lag times, performing lag processing on the plurality of salinity data sets, and outputting a plurality of lag salinity data sets corresponding to each of the preset lag times; According to the multiple lag salinity data sets corresponding to the respective preset lag times, calculating the partial correlation coefficient associated with the multiple lag salinity data sets corresponding to the respective preset lag times; Based on multiple preset advance times, a salinity data set associated with a lagged salinity data set corresponding to a maximum partial correlation coefficient is used as a target salinity data set, and the target salinity data set is time-advanced to output a time-advanced salinity data set corresponding to each of the preset advance times; Based on the plurality of preset advance times, the plurality of water-gas coupling data sets are time-advanced, and a plurality of initial time-advanced water-gas coupling data sets corresponding to the respective preset advance times are output; Calculate the Pearson correlation coefficient of the multiple initial time leading water-vapor coupling data sets corresponding to each of the preset advance times according to the multiple salinity data sets and the multiple initial time leading water-vapor coupling data sets corresponding to each of the preset advance times; Based on the plurality of preset advance times, time-advance the water-gas coupling data set associated with the initial time-advance water-gas coupling data set corresponding to the largest Pearson correlation coefficient, and output the target time-advance water-gas coupling data set corresponding to each preset advance time; Performing data sorting and standardization on the plurality of target time-advance water-gas coupling data sets and the plurality of time-advance salinity data sets, and outputting a plurality of standardized time-advance water-gas coupling data sets and a plurality of standardized time-advance salinity data sets; Dividing the plurality of the standardized time-advanced water-gas coupling data sets and the plurality of the standardized time-advanced salinity data sets to generate a hydrometeorological training data set and a hydrometeorological test data set; The target salinity data set is used as a hydrological and meteorological observation data set.

3. The method for constructing a saltwater tide upstream salinity forecasting model according to claim 1, wherein: The method of using the preset Shapley additive interpretation model and the intermediate long short-term memory model to calculate the Shapley values ​​corresponding to the hydrological and meteorological test data subsets of multiple time series in the hydrological and meteorological test data set according to the hydrological and meteorological test data set includes: Using the intermediate long short-term memory model, outputting the salinity prediction value corresponding to each of the hydrological and meteorological test data subsets of the time series according to each of the hydrological and meteorological test data subsets of the time series; The preset Shapley additive interpretation model is used to calculate the Shapley value corresponding to each of the time series' hydrometeorological test data subsets based on the salinity prediction value corresponding to each of the time series' hydrometeorological test data subsets.

4. The method for constructing a saltwater tide upstream salinity forecasting model according to claim 2, wherein: The method of constructing multiple target long-short term memory models according to the multiple time series hydrometeorological test data subsets, the hydrometeorological observation data set and the Shapley values ​​corresponding to the multiple time series hydrometeorological test data subsets includes: Sorting the Shapley values ​​corresponding to the plurality of time series hydrological and meteorological test data subsets in descending order, selecting a preset number of time series hydrological and meteorological test data subsets as the first training data subset, and gradually accumulating the remaining time series hydrological and meteorological test data subsets into the first training data subset in order, to generate a plurality of second training data subsets; The initial long short-term memory model is trained using the first training data subset, multiple second training data subsets, and the hydrological and meteorological observation data set to generate multiple target long short-term memory models.

5. The method for constructing a saltwater tide upstream salinity forecast model according to claim 1, wherein: The method of using the hydrological and meteorological test data set and the hydrological and meteorological observation data set to screen each of the target long-term and short-term memory models to determine a saltwater tide upstream salinity forecast model includes: Using the hydrological and meteorological test data sets as inputs of the target long-short-term memory models, and outputting a plurality of salinity prediction values ​​corresponding to the target long-short-term memory models; Using a plurality of salinity prediction values ​​corresponding to each of the target long and short-term memory models and the hydrological and meteorological observation data set to evaluate the model accuracy of each of the target long and short-term memory models, and outputting the Nash efficiency coefficient and root mean square error corresponding to each of the target long and short-term memory models; The target long short-term memory model corresponding to the largest Nash efficiency coefficient and the smallest root mean square error is selected as the upstream salinity prediction model of saltwater tide.

6. The method for constructing a saltwater tide upstream salinity forecasting model according to claim 1, characterized in that: After the step of using the hydrological and meteorological test data set and the hydrological and meteorological observation data set to screen each of the target long-short term memory models to determine the saltwater tide upstream salinity forecast model, the method includes: When the hydrological and meteorological measured data to be detected is received, the hydrological and meteorological measured data to be detected is input into the salt tide upstream salinity forecast model to perform salinity forecast, and an initial salinity forecast result is output; The initial salinity forecast result is de-normalized to generate a target salinity forecast result.

7. A saltwater tide upstream salinity forecast model construction device, characterized in that: include: An acquisition module is used to acquire a hydrological and meteorological measured data set, and preprocess the hydrological and meteorological measured data set to generate a target hydrological and meteorological measured data set; Based on the module, it is used to construct a hydrometeorological training data set, a hydrometeorological test data set and a hydrometeorological observation data set based on multiple preset lag times, multiple preset lead times and the target hydrometeorological measured data set; An adopting module is used to use the hydrological and meteorological training data set and the hydrological and meteorological observation data set to perform model training on the initial long short-term memory model to determine an intermediate long short-term memory model; A calculation module, used to calculate the Shapley values ​​corresponding to the hydrometeorological test data subsets of multiple time series in the hydrometeorological test data set according to the hydrometeorological test data set by using a preset Shapley additive interpretation model and the intermediate long short-term memory model; A construction module, for constructing a plurality of target long-short term memory models according to a plurality of said time series hydrometeorological test data subsets, said hydrometeorological observation data set and Shapley values ​​corresponding to a plurality of said time series hydrometeorological test data subsets; The screening module is used to screen each of the target long-short term memory models using the hydrological and meteorological test data set and the hydrological and meteorological observation data set to determine the saltwater upstream salinity forecast model.

8. A computer device, characterized in that: It comprises a memory and a processor, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the processor executes the steps of the method for constructing a saltwater upstream salinity forecast model as described in any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed, the method for constructing a saltwater upstream salinity forecast model as described in any one of claims 1 to 6 is implemented.

10. A computer program product, characterized in that The computer program product includes a computer program stored on a non-transitory computer-readable storage medium, and the computer program includes program instructions, wherein when the program instructions are executed by a computer, the computer executes the method for constructing a saltwater upstream salinity forecast model as described in any one of claims 1-6.

Images