Method and system for setting turbidity initial field of water turbidity forecasting model

Abstract

The invention provides a turbidity initial field setting method and system for a water turbidity forecasting model, and the method comprises the following steps: S1, collecting multi-source data, and carrying out the preprocessing of the data; s2, feature extraction and feature selection; s3, training a machine learning model; s4, using the trained machine learning model to obtain a turbidity initial field forecast result; s5, performing spatial interpolation by using actually measured turbidity data of the monitoring points to obtain a turbidity initial field interpolation result; s6, fusing the turbidity initial field forecast result and the turbidity initial field interpolation result to obtain a fused turbidity initial field; s7, performing physical constraint and smoothing processing on the fusion turbidity initial field; s8, quantitatively analyzing the total uncertainty of the fusion turbidity initial field; and S9, correcting and fusing the turbidity initial field, and finally obtaining a reasonable turbidity initial field setting scheme of the water turbidity forecasting model. According to the method, the turbidity initial field setting precision of the water turbidity forecasting model is improved.

Description

Technical Field

[0001] The invention relates to the technical field of water environment prediction, in particular to a method and system for setting a turbidity initial field of a water body turbidity prediction model. Background Art

[0002] Water turbidity is one of the important indicators of water quality, and its accurate prediction is of great significance to water environment protection and water resources management. Traditional water turbidity prediction methods mainly rely on mechanism models and statistical models, but they have limitations when dealing with complex environmental variables.

[0003] With the development of machine learning technology, the water turbidity prediction model based on machine learning has shown stronger prediction ability. However, the prediction accuracy of water turbidity depends not only on the water turbidity prediction model itself, but more importantly on the accurate setting of the turbidity initial field.

[0004] At present, the main methods for setting the turbidity initial field include: (1) directly using monitoring data: simple but easily affected by local outliers; (2) statistical interpolation methods: such as the inverse distance weighted method, which is simple to calculate but has limited accuracy; (3) geostatistical methods: such as Kriging interpolation, which considers spatial correlation but is complex to calculate; (4) numerical simulation methods: based on physical models, but greatly affected by initial conditions and boundary conditions. These methods have their own advantages and disadvantages, and it is difficult to take into account both accuracy and efficiency at the same time. Therefore, it is of great significance to develop a turbidity initial field setting method that can comprehensively utilize multi-source data and combine the advantages of machine learning. Summary of the invention

[0005] The present invention proposes a turbidity initial field setting method and system for a water body turbidity prediction model, so as to solve the technical problem that the existing turbidity initial field setting method is difficult to take into account both accuracy and efficiency.

[0006] One aspect of the present invention is to provide a method for setting an initial turbidity field of a water turbidity prediction model, the method comprising the following steps:

[0007] S1. Collect hydrological, water quality and meteorological data from multiple sources and pre-process the data;

[0008] S2, extracting features from the preprocessed data and selecting feature subsets from all features;

[0009] S3, using the feature subset screened in step S2 as input variables and water turbidity as the target variable to train the machine learning model;

[0010] S4, using the machine learning model trained in step S3 to obtain the turbidity initial field prediction result of the water turbidity prediction model;

[0011] S5, using the measured turbidity data of the monitoring points to perform spatial interpolation to obtain the turbidity initial field interpolation result of the water turbidity prediction model;

[0012] S6, fusing the turbidity initial field prediction result and the turbidity initial field interpolation result to obtain a fused turbidity initial field;

[0013] S7, physically constraining and smoothing the fused turbidity initial field;

[0014] S8, quantitatively analyzing the overall uncertainty of the fused turbidity initial field and setting the overall uncertainty threshold;

[0015] S9. Use the overall uncertainty threshold to correct the fused turbidity initial field; use the corrected fused turbidity initial field as the turbidity initial field of the water turbidity prediction model.

[0016] In a preferred embodiment, in step S1, preprocessing the collected data includes: data screening and cleaning, and removing noise and outliers.

[0017] In a preferred embodiment, in step S4, the turbidity initial field prediction result of the water turbidity prediction model is obtained by the following method:

[0018] Get all grid nodes of water turbidity prediction model;

[0019] For each grid node, extract the data required for the feature subset screened in step S2 from the mechanism model simulation results and the meteorological forecast data as the input data of the machine learning model of each grid node;

[0020] Input data of the machine learning model of each grid node into the machine learning model, and output the turbidity prediction result of each grid node of the water turbidity prediction model;

[0021] The turbidity initial field prediction result of the water turbidity prediction model is obtained by using the turbidity prediction result of each grid node of the water turbidity prediction model.

[0022] In a preferred embodiment, in step S5, the turbidity initial field interpolation result of the water turbidity prediction model is obtained by the following method:

[0023] Get all grid nodes of water turbidity prediction model;

[0024] Obtain the measured values ​​of water turbidity at n monitoring points;

[0025] The measured water turbidity values ​​at n monitoring points are interpolated to each grid node of the water turbidity prediction model through the spatial interpolation method to obtain the turbidity interpolation result of each grid node of the water turbidity prediction model;

[0026] The turbidity interpolation result of each grid node of the water turbidity prediction model is used to generate the turbidity initial field interpolation result of the water turbidity prediction model.

[0027] In a preferred embodiment, in step S6, the turbidity initial field prediction result and the turbidity initial field interpolation result are integrated, including the following method steps:

[0028] S601, obtain the coordinates of n monitoring points: Monitored_Coordinates = {(x_i, y_i)};

[0029] Among them, x_i represents the horizontal coordinate of the i-th monitoring point, and y_i represents the vertical coordinate of the i-th monitoring point;

[0030] S602, obtaining the coordinates of each grid node of the water turbidity prediction model: Grid_Coordinates = {(x'_j, y'_j)};

[0031] Among them, x'_j represents the horizontal coordinate of the j-th grid node, and y'_j represents the vertical coordinate of the j-th grid node;

[0032] S603, calculate the shortest distance between each grid node and n monitoring points:

[0033] Distances_j=min_{i}(sqrt{(x'_j-x_i)^2+(y'_j-y_i)^2});

[0034] Wherein, Distances_j represents the shortest distance between the jth grid node and the n monitoring points;

[0035] S604, select the maximum distance among the closest distances between all grid nodes and n monitoring points:

[0036] Max_Distance=Max(Distances_j, j=1,2,3,…,m);

[0037] Among them, Max_Distance represents the maximum distance among the closest distances between all grid nodes and n monitoring points, and m represents the total number of grid nodes in the water turbidity prediction model;

[0038] S605, calculate the fusion weight of each grid node:

[0039] Weights_j=1-Distances_j / Max_Distance;

[0040] Among them, Weights_j represents the fusion weight of the jth grid node;

[0041] S606: Using the fusion weight of each grid node, the turbidity initial field prediction result and the turbidity initial field interpolation result are fused:

[0042] FusionTurbidity_j=Weights_j*PredictedTurbidity_j+(1-Weights_j)*InterpolationTurbidity_j;

[0043] Among them, FusionTurbidity_j represents the fusion turbidity of the j-th grid node, Weights_j represents the fusion weight of the j-th grid node, PredictedTurbidity_j represents the turbidity prediction result of the j-th grid node, and InterpolationTurbidity_j represents the turbidity interpolation result of the j-th grid node;

[0044] The fused turbidity of each grid node is used to obtain the fused turbidity initial field of the water turbidity prediction model.

[0045] In a preferred embodiment, in step S7, the fused turbidity initial field is physically constrained by the following method:

[0046] Setting a physical constraint range for the fused turbidity of each grid node in the fused turbidity initial field;

[0047] When the fused turbidity of a grid node exceeds the set physical constraint range, the fused turbidity of the grid node is truncated.

[0048] In a preferred embodiment, in step S7, the fused turbidity of each grid node in the fused turbidity initial field after physical constraints is smoothed.

[0049] In a preferred embodiment, in step S8, the overall uncertainty of the fused turbidity initial field is quantitatively analyzed by the following method:

[0050] S801, calculate the uncertainty value of the turbidity forecast of each grid node in the turbidity initial field forecast result: PredictedTurbidity_j['model_uncertainty'] = pred_var;

[0051] Among them, PredictedTurbidity_j['model_uncertainty'] represents the uncertainty value of the turbidity forecast of the jth grid node; pred_var represents the prediction variance of the machine learning model;

[0052] Wherein, j = 1, 2, 3, ..., m, m represents the total number of grid nodes of the water turbidity prediction model;

[0053] S802, calculating the uncertainty value of the interpolated turbidity of each grid node in the interpolation result of the turbidity initial field: InterpolationTurbidity_j['uncertainty']=Distances_j / Max_Distance;

[0054] Among them, InterpolationTurbidity_j['uncertainty'] represents the uncertainty value of the interpolated turbidity of the j-th grid node; Distances_j represents the closest distance between the j-th grid node and the n monitoring points; Max_Distance represents the maximum distance among the closest distances between all grid nodes and the n monitoring points;

[0055] Wherein, j = 1, 2, 3, ..., m, m represents the total number of grid nodes of the water turbidity prediction model;

[0056] S803. Calculate the average of the uncertainty value of the turbidity forecast of each grid node and the uncertainty value of the interpolated turbidity as the uncertainty value of the fused turbidity of each grid node in the fused turbidity initial field:

[0057] FusionTurbidity_j['Total_uncertainty']=

[0058] (PredictedTurbidity_j['model_uncertainty']+InterpolationTurbidity_j['uncertainty']) / 2;

[0059] Among them, FusionTurbidity_j['Total_uncertainty'] represents the uncertainty value of the fused turbidity of the jth grid node;

[0060] By setting the overall uncertainty threshold and combining the uncertainty value of the fused turbidity of each grid node in the turbidity initial field, the overall uncertainty of the fused turbidity initial field is quantitatively analyzed.

[0061] In a preferred embodiment, in step S9, the fused turbidity initial field is corrected by the following method:

[0062] S901, when the uncertainty value of the fused turbidity of a certain grid node is greater than the set overall uncertainty threshold, the average value of the fused turbidity of the 10 grid nodes that are closest to the grid node and less than the uncertainty threshold is calculated;

[0063] S902, using the calculated average value as the corrected fused turbidity of the grid node;

[0064] S903, repeating step S901 to step S902, correcting the fused turbidity of each grid node in the fused turbidity initial field, thereby obtaining a corrected fused turbidity initial field, and using it as a turbidity initial field setting scheme for the water body turbidity prediction model.

[0065] Another aspect of the present invention is to provide a turbidity initial field setting system for a water body turbidity prediction model, wherein the turbidity initial field setting system is used to execute a turbidity initial field setting method for a water body turbidity prediction model provided by the present invention.

[0066] Compared with the prior art, the present invention has the following beneficial effects:

[0067] The present invention proposes a method and system for setting the turbidity initial field of a water turbidity prediction model, which sets the turbidity initial field of the water turbidity prediction model based on the integration of machine learning and mathematical statistics, taking into account both accuracy and efficiency, thereby improving the accuracy of water turbidity prediction and reducing prediction errors.

[0068] The present invention proposes a method and system for setting the turbidity initial field of a water turbidity prediction model, which comprehensively utilizes the prediction results of machine learning, the simulation results of traditional water turbidity prediction models (such as hydrological and water quality models), and the results of mathematical and statistical analysis of measured data to achieve the rapid setting of the turbidity initial field of the water turbidity prediction model. It fully integrates the advantages of various model algorithms, takes into account both accuracy and efficiency, effectively improves the prediction accuracy of water turbidity, and provides a way to solve the problem of difficult turbidity prediction.

[0069] The present invention proposes a method and system for setting the turbidity initial field of a water turbidity prediction model, which screens out the factors that most significantly affect turbidity from multiple monitoring indicators through feature extraction and feature selection, thereby improving the reliability of the water turbidity prediction model.

[0070] The present invention proposes a method and system for setting the turbidity initial field of a water turbidity prediction model, which utilizes advanced machine learning algorithms to capture the nonlinear relationship between turbidity and other parameters such as hydrology, meteorology, and water quality, thereby improving prediction accuracy. Geostatistical methods (such as Kriging interpolation) are used to consider the spatial correlation of water turbidity, thereby making up for the deficiency of sparse monitoring points.

[0071] The present invention proposes a method and system for setting the turbidity initial field of a water turbidity prediction model, which fuses the machine learning prediction results and the spatial interpolation results by designing a reasonable weight function, thereby retaining the prediction ability of machine learning and taking into account the spatial distribution characteristics.

[0072] The present invention proposes a method and system for setting the turbidity initial field of a water turbidity prediction model, which introduces physical constraints to ensure that the prediction result conforms to the actual situation, and reduces spatial noise through smoothing processing, thereby improving the rationality of the water turbidity initial field.

[0073] The present invention proposes a method and system for setting the turbidity initial field of a water turbidity prediction model, which quantitatively analyzes the uncertainty of the water turbidity initial field setting by analyzing the prediction variance and the interpolation variance, and provides a basis for further optimization.

[0074] The present invention proposes a method and system for setting the turbidity initial field of a water turbidity prediction model, which combines machine learning and spatial statistics methods, makes full use of multi-source data information, and realizes high-precision setting of the turbidity initial field of the water turbidity prediction model.

[0075] The present invention proposes a method and system for setting the turbidity initial field of a water turbidity prediction model, which effectively improves the setting accuracy of the turbidity initial field of the water turbidity prediction model by integrating machine learning prediction and mathematical statistical analysis and introducing physical constraints and uncertainty analysis, and provides a new technical solution for improving the accuracy of water turbidity prediction.

[0076] The method and system for setting the turbidity initial field of a water turbidity prediction model proposed in the present invention have broad application prospects. It can be used in a water quality early warning system to provide more accurate initial conditions for real-time water quality early warnings, thereby improving the timeliness and accuracy of early warnings. It can be used in water resource management to assist water resource management departments in water quality assessment and decision-making, and optimize water resource scheduling. It can be used in ecological environmental protection to provide a scientific basis for the ecological environmental protection of lakes, rivers and other water bodies, and to assist in ecological restoration. It can be used in water supply treatment to provide water plants with more accurate raw water quality predictions, optimize treatment processes, and improve water supply quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] In order to more clearly illustrate the specific implementation of the present invention or the technical solution in the prior art, the following briefly introduces the drawings required for use in the specific implementation or the prior art description. Obviously, the drawings described below are some implementations of the present invention, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0078] Figure 1 The present invention discloses a flow chart of a method for setting a turbidity initial field of a water turbidity prediction model.

[0079] Figure 2 This is a distribution diagram of the turbidity initial field forecast results of the J reservoir machine learning model in one embodiment of the present invention.

[0080] Figure 3 It is a distribution diagram of the interpolation results of the initial turbidity field obtained by spatial interpolation of the J reservoir in one embodiment of the present invention.

[0081] Figure 4 In one embodiment of the present invention, the J reservoir fuses the turbidity initial field prediction results obtained by the machine learning model and the turbidity initial field interpolation results obtained by spatial interpolation to generate a fused turbidity initial field distribution map.

[0082] Figure 5 It is a general uncertainty distribution diagram of the fused turbidity initial field of the J reservoir in one embodiment of the present invention.

[0083] Figure 6 It is an initial turbidity field distribution map (corrected fused initial turbidity field distribution map) of the J reservoir water turbidity prediction model in one embodiment of the present invention.

[0084] Figure 7 The water turbidity initial field obtained by the water turbidity initial field setting method of the present invention, the water turbidity initial field obtained by spatial interpolation and the water turbidity initial field obtained by the existing water turbidity initial field setting method are respectively input into the J reservoir water turbidity prediction model and the water turbidity change of the J reservoir in the next 24 hours is predicted, and finally the turbidity prediction result change trend of the J reservoir water transfer area using different initial field setting methods is obtained. DETAILED DESCRIPTION

[0085] In order to make the above and other features and advantages of the present invention more clear, the present invention is further described below in conjunction with the accompanying drawings. It should be understood that the specific embodiments given herein are for the purpose of explaining to those skilled in the art and are only exemplary and not restrictive.

[0086] Combination Figure 1 According to an embodiment of the present invention, a method for setting the turbidity initial field of a water turbidity prediction model is provided. In this embodiment, J Reservoir is taken as an example. A water turbidity prediction model and other water quality hydrodynamic models (mechanism models) have been established in J Reservoir. In this embodiment, the calculation area of ​​the water turbidity prediction model of J Reservoir is divided into 6538 triangular grids, with a total of 11483 grid nodes. There are 5 monitoring points in the study area of ​​the water turbidity prediction model, and the hydrological, water quality and meteorological monitoring data of J Reservoir are monitored through the 5 monitoring points.

[0087] In the following embodiment, the water turbidity prediction model of the J reservoir is taken as the initial calculation time at 0:00 on May 1, 2023, and the turbidity initial field of the water turbidity prediction model at this time is set as an example to illustrate the method for setting the turbidity initial field of a water turbidity prediction model provided by the present invention.

[0088] Combination Figure 1According to an embodiment of the present invention, a method for setting the turbidity initial field of a water turbidity prediction model is provided, comprising the following method steps:

[0089] Step S1: Collect multi-source data such as hydrology, water quality, and meteorology, and pre-process the data.

[0090] The collected data should include the hydrological, water quality and meteorological monitoring data of all monitoring points in the water turbidity prediction model study area. The monitoring data include but are not limited to turbidity, pH, dissolved oxygen, water temperature, conductivity, flow velocity, flow direction, water depth, wind speed, wind direction, rainfall, etc.

[0091] For example, in this embodiment, there are 5 hydrological, water quality and meteorological monitoring points in the study area of ​​the water turbidity prediction model, and the hourly monitoring data (hydrological, water quality and meteorological monitoring data) of the 5 monitoring points from January 1 to March 31, 2023 are obtained. The monitoring data types include: turbidity, chlorophyll a, water temperature, pH, dissolved oxygen, water level, water depth, flow velocity, flow direction, wind speed, wind direction, and rainfall. 12 data type features.

[0092] Preprocessing of the collected data includes: data screening and cleaning, removing noise and outliers, and improving data quality.

[0093] For example, in this embodiment, data screening and cleaning are performed on the 12 data type features of turbidity, chlorophyll a, water temperature, pH, dissolved oxygen, water level, water depth, flow velocity, flow direction, wind speed, wind direction, and rainfall, and outliers are replaced to improve data quality.

[0094] Step S2: extract features from the preprocessed data and select a feature subset from all the features.

[0095] The preprocessed data is converted into a new and more representative feature representation through feature extraction. Feature extraction methods include mathematical transformation, dimensionality reduction, feature engineering technology, etc.

[0096] Feature selection methods are used to screen out the most valuable feature subsets with high correlation with the target feature turbidity from all features, reduce the dimension of the feature space, and improve the generalization ability and interpretability of the machine learning model.

[0097] For example, in this embodiment, feature extraction is performed from the 12 data type features monitored, including turbidity, chlorophyll a, water temperature, pH, dissolved oxygen, water level, water depth, flow velocity, flow direction, wind speed, wind direction, and rainfall. Twelve features are extracted, plus the 12 data type features originally monitored, for a total of 24 features.

[0098] The feature importance analysis was performed on the 24 features, and the top 12 features in terms of feature importance were selected from the 24 features as the feature subset.

[0099] Step S3: Use the feature subset screened out in step S2 as input variables and water turbidity as the target variable to train the machine learning model.

[0100] The machine learning model of the present invention can select a linear model, a decision tree, a support vector machine, a neural network model, and an integrated learning model. It should be understood that the machine learning model is not limited to this, and those skilled in the art can select a more appropriate machine learning model based on the specific actual situation.

[0101] In this embodiment, the random forest model, XGBoost model, Adaboost+XGBoost stacking model, TPOT model, and TPOT+random forest stacking model were cross-validated to evaluate the robustness of the machine learning model, and it was found that the TPOT model was optimal. The machine learning model of the present invention is preferably the TPOT model.

[0102] According to an embodiment of the present invention, the feature subset screened out in step S2 is used as an input variable and the water turbidity is used as a target variable to train a machine learning model (preferably a TPOT model).

[0103] In a specific embodiment, the feature subset screened in step S2 is divided into a training set and a test set. The training set R 2 is 0.89, the test set R 2 The average absolute error of the test set is 4.7NTU (NTU, Nitrogen Turbidity Units), and the average relative error is 21.36%, and the optimal parameter combination of the machine learning model (TPOT model) is obtained.

[0104] Step S4: using the machine learning model trained in step S3, obtain the turbidity initial field prediction result of the water turbidity prediction model.

[0105] Specifically, according to an embodiment of the present invention, the turbidity initial field prediction result of the water turbidity prediction model is obtained by the following method:

[0106] Get all grid nodes of the water turbidity prediction model.

[0107] For example, in this embodiment, the water turbidity prediction model and other water quality hydrodynamic models (mechanism models) established by J Reservoir have a total of 11483 grid nodes, and 11483 grid node information is obtained. The grid node information of the water turbidity prediction model includes the coordinates of each grid node.

[0108] The mechanism model refers to a model established through mathematical modeling methods based on the understanding of the internal mechanisms and principles of the physical system. The mechanism model referred to in the present invention includes both water turbidity prediction models and other water quality hydrodynamic models.

[0109] For each grid node, the data required for the feature subset screened out in step S2 is extracted from the simulation results of other water quality hydrodynamic models (mechanism models) and meteorological forecast data as the input data of the machine learning model of each grid node (the trained TPOT model with the optimal parameter combination), and the turbidity prediction result of each grid node of the water turbidity prediction model is output.

[0110] The turbidity initial field prediction result of the water body turbidity prediction model is generated by using the turbidity prediction result of each grid node of the water body turbidity prediction model.

[0111] In this embodiment, the turbidity prediction result of each grid node of the water turbidity prediction model is used to generate the initial turbidity field of the water turbidity prediction model of the J reservoir at 0:00 on May 1, 2023, as shown in FIG. Figure 2 shown.

[0112] Since there are only 5 monitoring points in the study area of ​​the water turbidity prediction model, the number of monitoring points is limited, and it is impossible to implement hydrological, water quality and meteorological monitoring of each grid node of the water turbidity prediction model. For this reason, the present invention extracts data of the data type contained in the feature subset from the simulation results of other water quality and hydrodynamic models (mechanism models) according to the feature subset screened out in step S2, including chlorophyll a, water temperature, pH, dissolved oxygen, water level, flow velocity, flow direction, water depth and other data, and uses these data as input data of the machine learning model to output the turbidity prediction result of each grid node of the water turbidity prediction model.

[0113] In some embodiments, the mechanism model also uses a WRF wind field model or a meteorological forecast model to obtain meteorological data type characteristics such as cumulative rainfall and cumulative effective wind speed, so as to enrich the input data of the machine learning model and improve the forecast accuracy and interpretability of the machine learning model.

[0114] Step S5: perform spatial interpolation using the turbidity data measured at the monitoring points to obtain an interpolation result of the turbidity initial field of the water turbidity prediction model.

[0115] Specifically, according to an embodiment of the present invention, the turbidity initial field interpolation result of the water turbidity prediction model is obtained by the following method:

[0116] Get all grid nodes of the water turbidity prediction model.

[0117] For example, in this embodiment, the water turbidity prediction model established in the J reservoir has a total of 11483 grid nodes, and the 11483 grid node information is obtained. The grid node information of the water turbidity prediction model includes the coordinates of each grid node.

[0118] Obtain the measured values ​​of water turbidity at n monitoring points.

[0119] For example, in this embodiment, there are 5 monitoring points in the study area of ​​the water turbidity prediction model established for the J reservoir, and the actual measured values ​​of the water turbidity at the 5 monitoring points are obtained.

[0120] The measured water turbidity values ​​at n monitoring points are interpolated to each grid node of the water turbidity prediction model through the spatial interpolation method to obtain the turbidity interpolation result of each grid node of the water turbidity prediction model.

[0121] For example, in this embodiment, the measured water turbidity values ​​at five monitoring points are interpolated to 11483 grid nodes of the water turbidity prediction model to obtain the turbidity interpolation result of each grid node of the water turbidity prediction model.

[0122] In some embodiments, the spatial interpolation method uses Kriging interpolation.

[0123] The turbidity interpolation result of each grid node of the water turbidity prediction model is used to generate the turbidity initial field interpolation result of the water turbidity prediction model.

[0124] In this embodiment, the turbidity interpolation result of each grid node of the water turbidity prediction model is used to generate the turbidity initial field interpolation result of the water turbidity prediction model of the J reservoir at 0 o'clock on May 1, 2023, as shown in FIG. Figure 3 shown.

[0125] Step S6: Fusing the turbidity initial field prediction result and the turbidity initial field interpolation result to obtain a fused turbidity initial field.

[0126] Specifically, according to an embodiment of the present invention, the fusion of the turbidity initial field prediction result and the turbidity initial field interpolation result includes the following method steps:

[0127] Step S601, obtain the coordinates of n monitoring points: Monitored_Coordinates = {(x_i, y_i)};

[0128] Among them, x_i represents the horizontal coordinate of the i-th monitoring point, and y_i represents the vertical coordinate of the i-th monitoring point.

[0129] Step S602, obtaining the coordinates of each grid node of the water turbidity prediction model: Grid_Coordinates = {(x'_j, y'_j)};

[0130] Among them, x'_j represents the horizontal coordinate of the j-th grid node, and y'_j represents the vertical coordinate of the j-th grid node.

[0131] Step S603: Calculate the shortest distance between each grid node and n monitoring points:

[0132] Distances_j=min_{i}(sqrt{(x'_j-x_i)^2+(y'_j-y_i)^2});

[0133] Wherein, Distances_j represents the shortest distance between the jth grid node and the n monitoring points. In this embodiment, n=5, j=1,2,3,…,m, m=11483.

[0134] Step S604: Select the maximum distance among the closest distances between all grid nodes and n monitoring points:

[0135] Max_Distance=Max(Distances_j, j=1,2,3,…,m);

[0136] Wherein, Max_Distance represents the maximum distance among the shortest distances between all grid nodes and n monitoring points, and m represents the total number of grid nodes of the water turbidity prediction model. In this embodiment, n=5, m=11483.

[0137] Step S605: Calculate the fusion weight of each grid node:

[0138] Weights_j=1-Distances_j / Max_Distance;

[0139] Among them, Weights_j represents the fusion weight of the j-th grid node.

[0140] The fusion weight of each grid node of the present invention is based on linear scaling based on distance, that is, the closer the distance, the greater the weight.

[0141] Step S606: using the fusion weight of each grid node, fuse the turbidity initial field prediction result and the turbidity initial field interpolation result:

[0142] FusionTurbidity_j=Weights_j*PredictedTurbidity_j+(1-Weights_j)*InterpolationTurbidity_j;

[0143] Among them, FusionTurbidity_j represents the fusion turbidity of the j-th grid node, Weights_j represents the fusion weight of the j-th grid node, PredictedTurbidity_j represents the turbidity prediction result of the j-th grid node, and InterpolationTurbidity_j represents the turbidity interpolation result of the j-th grid node;

[0144] The fused turbidity of each grid node is used to obtain the fused turbidity initial field of the water turbidity prediction model.

[0145] In this implementation, the fused turbidity of each grid node is used to obtain the fused turbidity initial field of the water turbidity forecast model of the J reservoir at 0 o'clock on May 1, 2023, as shown in Figure 4.

[0146] Step S7: Perform physical constraints and smoothing on the fused turbidity initial field.

[0147] Specifically, according to an embodiment of the present invention, the fused turbidity initial field is physically constrained by the following method:

[0148] Set the physical constraint range for the fused turbidity of each grid node in the fused turbidity initial field: [Turbidity_min, Turbidity_max].

[0149] Among them, Turbidity_min represents the minimum value of the integrated turbidity physical constraint range, and Turbidity_max represents the maximum value of the integrated turbidity physical constraint range.

[0150] In a preferred embodiment, Turbidity_min is set to 0 NTU, and Turbidity_max is set to a reasonable value according to the variation pattern of historical turbidity data in the study area of ​​the water turbidity prediction model.

[0151] When the fused turbidity of a grid node exceeds the set physical constraint range, the fused turbidity of the grid node is truncated.

[0152] For example, when the fused turbidity of a certain grid node is less than Turbidity_min, the fused turbidity of the grid node is set to Turbidity_min. When the fused turbidity of a certain grid node is greater than Turbidity_max, the fused turbidity of the grid node is set to Turbidity_max.

[0153] When the fused turbidity of a grid node is within the range of [Turbidity_min, Turbidity_max], the fused turbidity of the grid node will not be processed.

[0154] According to an embodiment of the present invention, the fused turbidity of each grid node in the fused turbidity initial field after physical constraints is smoothed to reduce local abnormal fluctuations and make the result more continuous and reliable.

[0155] Step S8: quantitatively analyze the overall uncertainty of the fused turbidity initial field and set a reasonable overall uncertainty threshold.

[0156] Specifically, according to an embodiment of the present invention, the overall uncertainty of the fused turbidity initial field is quantitatively analyzed by the following method:

[0157] Step S801, calculate the uncertainty value of the turbidity forecast of each grid node in the turbidity initial field forecast result: PredictedTurbidity_j['model_uncertainty']=pred_var;

[0158] Among them, PredictedTurbidity_j['model_uncertainty'] represents the uncertainty value of the turbidity forecast of the jth grid node; pred_var represents the prediction variance of the machine learning model;

[0159] Wherein, j = 1, 2, 3, ..., m, and m represents the total number of grid nodes of the water turbidity prediction model.

[0160] Step S802, calculating the uncertainty value of the interpolated turbidity of each grid node in the turbidity initial field interpolation result: InterpolationTurbidity_j['uncertainty']=Distances_j / Max_Distance;

[0161] Among them, InterpolationTurbidity_j['uncertainty'] represents the uncertainty value of the interpolated turbidity of the j-th grid node; Distances_j represents the closest distance between the j-th grid node and the n monitoring points, and Max_Distance represents the maximum distance among the closest distances between all grid nodes and the n monitoring points;

[0162] Wherein, j = 1, 2, 3, ..., m, and m represents the total number of grid nodes of the water turbidity prediction model.

[0163] Step S803: Calculate the average of the uncertainty value of the turbidity forecast of each grid node and the uncertainty value of the interpolated turbidity as the uncertainty value of the fused turbidity of each grid node in the fused turbidity initial field:

[0164] FusionTurbidity_j['Total_uncertainty']=

[0165] (PredictedTurbidity_j['model_uncertainty']+InterpolationTurbidity_j['uncertainty']) / 2;

[0166] Among them, FusionTurbidity_j['Total_uncertainty'] represents the uncertainty value of the fused turbidity of the jth grid node.

[0167] By setting the overall uncertainty threshold and combining the uncertainty value of the fused turbidity of each grid node in the turbidity initial field, the overall uncertainty of the fused turbidity initial field is quantitatively analyzed.

[0168] In this embodiment, the overall uncertainty threshold is set to 50, and the overall uncertainty distribution of the fused turbidity initial field of the water turbidity forecast model of the J reservoir at 0:00 on May 1, 2023 is quantitatively analyzed by integrating the uncertainty value of the fused turbidity of each grid node in the fused turbidity initial field, as shown in Figure 5 shown.

[0169] Step S9: using the overall uncertainty threshold, correcting the fused turbidity initial field; using the corrected fused turbidity initial field as the turbidity initial field of the water body turbidity prediction model.

[0170] Specifically, according to an embodiment of the present invention, the fused turbidity initial field is corrected by the following method:

[0171] Step S901: When the uncertainty value of the fused turbidity of a certain grid node is greater than the set overall uncertainty threshold, the average value of the fused turbidity of the 10 grid nodes that are closest to the grid node and less than the uncertainty threshold is calculated;

[0172] Step S902, using the calculated average value as the corrected fused turbidity of the grid node;

[0173] In some embodiments, when the uncertainty value of the fused turbidity of a certain grid node is higher than a set overall uncertainty threshold, a method such as local weighted regression may be used for correction.

[0174] Step S903, repeating steps S901 to S902, correcting the fused turbidity of each grid node in the fused turbidity initial field, thereby obtaining a corrected fused turbidity initial field, and using it as a turbidity initial field setting scheme for the water turbidity prediction model.

[0175] In this embodiment, the final obtained water turbidity initial field (corrected fused turbidity initial field) of the J reservoir water turbidity forecast model at 0:00 on May 1, 2023 is as follows: Figure 6 shown.

[0176] In this embodiment, the water turbidity initial field of the J reservoir water turbidity prediction model at 0:00 on May 1, 2023 obtained by the water turbidity initial field setting method of the present invention, the water turbidity initial field obtained by spatial interpolation, and the water turbidity initial field obtained by the existing water turbidity initial field setting method are respectively input into the J reservoir water turbidity prediction model and the water turbidity change of the J reservoir water in the next 24 hours is predicted, and finally the change trend of the turbidity prediction results of the J reservoir water transfer area using different initial field setting methods is obtained, as shown in FIG. Figure 7 shown.

[0177] Figure 7 In the figure, the black curve is the measured water turbidity change trend in the next 24 hours at 0:00 on May 1, 2023.

[0178] The blue curve is the water turbidity change trend in the next 24 hours at 0:00 on May 1, 2023, directly predicted by the existing water turbidity forecast model.

[0179] The green curve is the turbidity initial field set by the traditional interpolation method and input into the water turbidity forecast model, which predicts the turbidity change trend of the water body in the next 24 hours at 0:00 on May 1, 2023.

[0180] The red curve is the turbidity initial field (corrected fused turbidity initial field) input into the water turbidity prediction model of the present invention, and the predicted water turbidity change trend in the next 24 hours at 0:00 on May 1, 2023.

[0181] Depend on Figure 7 It can be seen that the turbidity initial field of the water turbidity prediction model set by the present invention has an accuracy of 86.58% for the future water turbidity prediction results, the existing water turbidity prediction model has an accuracy of 49.40% for the future water turbidity prediction results, and the turbidity initial field of the water turbidity prediction model set by the traditional interpolation method of the monitoring points alone has an accuracy of 61.32% for the future water turbidity prediction results. The turbidity initial field of the water turbidity prediction model set by the present invention can significantly improve the accuracy of the water turbidity prediction model for the water turbidity prediction results.

[0182] After testing, it is found that the turbidity initial field setting method of the water body turbidity prediction model provided by the present invention can effectively improve the prediction accuracy of the water body turbidity prediction model.

[0183] According to an embodiment of the present invention, a turbidity initial field setting system for a water body turbidity prediction model is provided, which is used to execute a turbidity initial field setting method for a water body turbidity prediction model provided by the present invention.

[0184] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may change, modify, replace and vary the above embodiments within the scope of the present invention.

Claims

1. A method for setting the turbidity initial field of a water turbidity prediction model, characterized in that: The turbidity initial field setting method comprises the following method steps: S1. Collect hydrological, water quality and meteorological data from multiple sources and pre-process the data; S2, extracting features from the preprocessed data and selecting feature subsets from all features; S3, using the feature subset screened in step S2 as input variables and water turbidity as the target variable to train the machine learning model; S4, using the machine learning model trained in step S3 to obtain the turbidity initial field prediction result of the water turbidity prediction model; S5, using the measured turbidity data of the monitoring points to perform spatial interpolation to obtain the turbidity initial field interpolation result of the water turbidity prediction model; S6, fusing the turbidity initial field prediction result and the turbidity initial field interpolation result to obtain a fused turbidity initial field; S7, physically constraining and smoothing the fused turbidity initial field; S8, quantitatively analyzing the overall uncertainty of the fused turbidity initial field and setting the overall uncertainty threshold; S9, using the overall uncertainty threshold, correct the fused turbidity initial field; The corrected fused turbidity initial field is used as the turbidity initial field of the water turbidity prediction model.

2. The method for setting the turbidity initial field according to claim 1, characterized in that: In step S1, the collected data is preprocessed including: data screening and cleaning, and removing noise and outliers.

3. The method for setting the turbidity initial field according to claim 1, characterized in that: In step S4, the turbidity initial field prediction result of the water turbidity prediction model is obtained by the following method: Get all grid nodes of water turbidity prediction model; For each grid node, extract the data required for the feature subset screened in step S2 from the mechanism model simulation results and the meteorological forecast data as the input data of the machine learning model of each grid node; Input data of the machine learning model of each grid node into the machine learning model, and output the turbidity prediction result of each grid node of the water turbidity prediction model; The turbidity initial field prediction result of the water turbidity prediction model is obtained by using the turbidity prediction result of each grid node of the water turbidity prediction model.

4. The method for setting the turbidity initial field according to claim 1, characterized in that: In step S5, the turbidity initial field interpolation result of the water turbidity prediction model is obtained by the following method: Get all grid nodes of water turbidity prediction model; Obtain the measured values ​​of water turbidity at n monitoring points; The measured water turbidity values ​​at n monitoring points are interpolated to each grid node of the water turbidity prediction model through the spatial interpolation method to obtain the turbidity interpolation result of each grid node of the water turbidity prediction model; The turbidity interpolation result of each grid node of the water turbidity prediction model is used to obtain the turbidity initial field interpolation result of the water turbidity prediction model.

5. The method for setting the turbidity initial field according to claim 1, characterized in that: In step S6, the turbidity initial field prediction result and the turbidity initial field interpolation result are integrated, including the following method steps: S601, obtain the coordinates of n monitoring points: Monitored_Coordinates = {(x_i, y_i)}; Among them, x_i represents the horizontal coordinate of the i-th monitoring point, and y_i represents the vertical coordinate of the i-th monitoring point; S602, obtaining the coordinates of each grid node of the water turbidity prediction model: Grid_Coordinates = {(x'_j, y'_j)}; Among them, x'_j represents the horizontal coordinate of the j-th grid node, and y'_j represents the vertical coordinate of the j-th grid node; S603, calculate the shortest distance between each grid node and n monitoring points: Distances_j=min_{i}(sqrt{(x'_j-x_i)^2+(y'_j-y_i)^2}); Wherein, Distances_j represents the shortest distance between the jth grid node and the n monitoring points; S604, obtaining the maximum distance among the closest distances between all grid nodes and n monitoring points: Max_Distance=Max(Distances_j, j=1,2,3,…,m); Among them, Max_Distance represents the maximum distance among the closest distances between all grid nodes and n monitoring points, and m represents the total number of grid nodes in the water turbidity prediction model; S605, calculate the fusion weight of each grid node: Weights_j=1-Distances_j / Max_Distance; Among them, Weights_j represents the fusion weight of the jth grid node; S606: Using the fusion weight of each grid node, the turbidity initial field prediction result and the turbidity initial field interpolation result are fused: FusionTurbidity_j=Weights_j*PredictedTurbidity_j+(1-Weights_j)*InterpolationTurbidity_j; Among them, FusionTurbidity_j represents the fusion turbidity of the j-th grid node, Weights_j represents the fusion weight of the j-th grid node, PredictedTurbidity_j represents the turbidity prediction result of the j-th grid node, and InterpolationTurbidity_j represents the turbidity interpolation result of the j-th grid node; The fused turbidity of each grid node is used to obtain the fused turbidity initial field of the water turbidity prediction model.

6. The method for setting the turbidity initial field according to claim 1, characterized in that: In step S7, the fused turbidity initial field is physically constrained by the following method: Setting a physical constraint range for the fused turbidity of each grid node in the fused turbidity initial field; When the fused turbidity of a grid node exceeds the set physical constraint range, the fused turbidity of the grid node is truncated.

7. The method for setting the turbidity initial field according to claim 1, characterized in that: In step S7, the fused turbidity of each grid node in the fused turbidity initial field after physical constraints is smoothed.

8. The method for setting the turbidity initial field according to claim 1, characterized in that: In step S8, the overall uncertainty of the fused turbidity initial field is quantitatively analyzed by the following method: S801, calculate the uncertainty value of the turbidity forecast of each grid node in the turbidity initial field forecast result: PredictedTurbidity_j['model_uncertainty'] = pred_var; Among them, PredictedTurbidity_j['model_uncertainty'] represents the uncertainty value of the turbidity forecast of the jth grid node; pred_var represents the prediction variance of the machine learning model; Wherein, j = 1, 2, 3, ..., m, m represents the total number of grid nodes of the water turbidity prediction model; S802, calculating the uncertainty value of the interpolated turbidity of each grid node in the interpolation result of the turbidity initial field: InterpolationTurbidity_j['uncertainty']=Distances_j / Max_Distance; Among them, InterpolationTurbidity_j['uncertainty'] represents the uncertainty value of the interpolated turbidity of the j-th grid node; Distances_j represents the closest distance between the j-th grid node and the n monitoring points; Max_Distance represents the maximum distance among the closest distances between all grid nodes and the n monitoring points; Wherein, j = 1, 2, 3, ..., m, m represents the total number of grid nodes of the water turbidity prediction model; S803. Calculate the average of the uncertainty value of the turbidity forecast of each grid node and the uncertainty value of the interpolated turbidity as the uncertainty value of the fused turbidity of each grid node in the fused turbidity initial field: FusionTurbidity_j['Total_uncertainty']= (PredictedTurbidity_j['model_uncertainty']+InterpolationTurbidity_j['uncertainty']) / 2; Among them, FusionTurbidity_j['Total_uncertainty'] represents the uncertainty value of the fused turbidity of the jth grid node; By setting the overall uncertainty threshold and combining the uncertainty value of the fused turbidity of each grid node in the turbidity initial field, the overall uncertainty of the fused turbidity initial field is quantitatively analyzed.

9. The method for setting the turbidity initial field according to claim 1, characterized in that: In step S9, the fused turbidity initial field is corrected by the following method: S901, when the uncertainty value of the fused turbidity of a certain grid node is greater than the set overall uncertainty threshold, the average value of the fused turbidity of the 10 grid nodes that are closest to the grid node and less than the uncertainty threshold is calculated; S902, using the calculated average value as the corrected fused turbidity of the grid node; S903, repeating step S901 to step S902, correcting the fused turbidity of each grid node in the fused turbidity initial field, thereby obtaining a corrected fused turbidity initial field, and using it as a turbidity initial field setting scheme for the water body turbidity prediction model.

10. A turbidity initial field setting system for a water turbidity prediction model, characterized in that: The turbidity initial field setting system is used to execute the turbidity initial field setting method described in any one of claims 1 to 9.

Images