Method for allocating water quantity in arid region based on river damage simulation

Abstract

The invention belongs to the technical field of arid region water volume allocation, and discloses an arid region water volume allocation method based on river loss simulation, which comprises the following steps: calculating the actually measured river loss amount of a drainage basin according to the actually measured runoff data of the drainage basin in an arid region; constructing a river loss prediction model of the drainage basin according to the hydrometeorological characteristics of the drainage basin and the corresponding actually measured river loss amount; performing river loss prediction on the drainage basin according to the river loss prediction model to obtain river loss prediction information; and setting constraint conditions according to the river loss prediction information, constructing a water resource optimization allocation model of the arid region based on the constraint conditions, and generating a water quantity allocation scheme of the arid region according to the water resource optimization allocation model. According to the method, high-precision prediction of the river loss is realized through the river loss prediction model of machine learning, and the river loss prediction model is embedded into the water resource optimal allocation model, so that the arid region water allocation method based on the river loss process simulation and suitable for the high-river-loss basin is constructed, and the water resource allocation of the arid region is more reasonable.

Description

Technical Field

[0001] The invention discloses a water allocation method for arid areas based on river loss simulation, belonging to the technical field of water allocation in arid areas. Background Art

[0002] River loss, or water loss from a river channel, refers to the amount of water lost due to various natural factors as runoff flows along the river channel. This lost water no longer directly contributes to the subsequent runoff process, significantly impacting the management and allocation of water resources. River loss is particularly significant in arid and semi-arid regions, where the underlying surface of the river channel is highly permeable and evaporative. Therefore, accurate and rational calculation of river loss is crucial for water resource allocation.

[0003] The main factors affecting river loss include the geometric characteristics of the river channel (such as its length, width, and longitudinal gradient), the geological conditions of the riverbed and banks (such as soil type and permeability), climatic conditions (such as precipitation and evaporation), and hydrological conditions (such as upstream water inflow and flow velocity). Traditional methods for calculating river loss include the river loss rate method, the unit river length method, and the nonlinear equation fitting method. While widely used in practice, these methods each have their own shortcomings.

[0004] River loss rate method: Calculates river loss by setting a fixed river loss rate. This method is simple and easy to implement, but ignores the differences in river losses in different river sections and under different hydrological conditions, resulting in problems with water reserves in the calculation results. Unit river length river loss method: Expresses river loss as the loss per unit river length. Although it takes into account the influence of river length, it fails to fully reflect the impact of key factors such as upstream water volume and water flow velocity on river loss. Therefore, the calculation results are not accurate enough. Fitting nonlinear equation method: Describes the relationship between river loss and various influencing factors by fitting nonlinear equations. This method is more flexible in theory, but due to the complexity of actual hydrological conditions, it is often difficult to find a high-precision fitting equation, resulting in low accuracy in the calculation results. Summary of the Invention

[0005] The present invention aims to provide a method for water allocation in arid regions based on river loss simulation to address the technical problem of low calculation accuracy caused by existing river loss calculation methods. To achieve the above-mentioned object, the present invention proposes a method for water allocation in arid regions based on river loss simulation. The specific scheme is as follows:

[0006] A water allocation method for arid areas based on river loss simulation, comprising:

[0007] S1. Calculating the measured river loss in the watershed according to the measured runoff data of the watershed in the arid area;

[0008] S2. constructing a river loss prediction model for the river basin based on the hydrological and meteorological characteristics of the river basin and the corresponding measured river loss amount;

[0009] S3. Predicting river damage in the basin according to the river damage prediction model to obtain river damage prediction information;

[0010] S4. Establishing constraint conditions according to the river loss prediction information, constructing a water resource optimization allocation model for the arid area based on the constraint conditions, and generating a water allocation plan for the arid area according to the water resource optimization allocation model.

[0011] Preferably, the S1 specifically includes:

[0012] Obtain the upstream section inflow, downstream section outflow and interval water diversion of the watershed in the arid area;

[0013] The measured river losses in the basin are calculated based on the water balance principle.

[0014] Preferably, S2 specifically includes:

[0015] S21, constructing a feature sliding window to extract the feature vector of the temporal hydrological and meteorological characteristics of the basin;

[0016] S22. Using the characteristic vector as input and the corresponding measured river loss amount as a label, a river loss prediction model for the basin is constructed.

[0017] Preferably, after S21 and before S22, the following steps are further included:

[0018] Performing multidimensional clustering on the feature vectors using a stratified sampling strategy of multidimensional feature clustering;

[0019] Extracting a preset proportion of samples from each layer of the multidimensional clustering as a test set, and the remaining samples as a training set;

[0020] Correspondingly, in S22, a river loss prediction model for the basin is constructed using the feature vector in the training set as input and the corresponding measured river loss amount as a label.

[0021] Preferably, the S22 further includes:

[0022] S221. Based on the historical training result set, dynamically update the sampling distribution of the hyperparameters of the river loss prediction model to be constructed, and generate a candidate set of hyperparameters;

[0023] S222: Generate a dynamic parameter adjustment instruction set for the river loss prediction model to be constructed based on the candidate set of hyperparameters;

[0024] S223. Optimize the hyperparameters of the river loss prediction model according to the dynamic parameter adjustment instruction set.

[0025] Preferably, the S221 specifically includes:

[0026] Automatically adjusting hyperparameters of the river loss prediction model to be constructed through a Bayesian optimization framework;

[0027] Updating the sampling distribution of the hyperparameters based on historical parameter test results;

[0028] Based on the updated sampling distribution, a candidate set of hyperparameters for the river loss prediction model to be constructed is generated.

[0029] Preferably, S222 specifically includes:

[0030] Eliminate the parameter groups that do not meet the criteria from the hyperparameter candidate set according to a preset loss convergence threshold, and record the set of parameter groups that meet the criteria as the first candidate parameter set;

[0031] Based on the parameter performance verification matrix, screening a parameter group whose accuracy meets a preset standard from the first candidate parameter group;

[0032] Based on the screened parameter group, a dynamic parameter adjustment instruction set of the river loss prediction model to be constructed is generated.

[0033] Preferably, setting constraints according to the river loss prediction information and constructing a water resource optimization allocation model for the arid area based on the constraints specifically includes:

[0034] S41, constructing a water resource allocation model according to the water demand of multiple users in the arid area;

[0035] S42, constructing a reservoir operation model according to the reservoir operation rules in the arid area;

[0036] S43. Couple the water resource allocation model and the reservoir scheduling model, establish water loss constraints based on the river loss prediction information, and construct a water resource optimization and allocation model.

[0037] Preferably, the hydrological and meteorological characteristics include:

[0038] Historical water inflow of the river section, current water inflow of the river section, current water diversion volume, current groundwater depth of the river section and current evaporation and emission.

[0039] Preferably, coupling the water resource allocation model and the reservoir scheduling model specifically includes:

[0040] The reservoir discharge and node water supply in the arid area are used as decision variables, and the model coupling is performed with the minimum water shortage of multiple users in the arid area as the optimization goal.

[0041] Beneficial effects: The present invention achieves high-precision prediction of river losses through a river loss prediction model based on machine learning, and embeds it into a water resource optimization allocation model to construct a water allocation method for arid areas based on river loss process simulation, which is suitable for high river loss basins, making water resource allocation in arid areas more reasonable. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 Calculation results of river loss in the Aral-Xinqiman River section in the embodiment of the present invention;

[0043] Figure 2 Calculation results of river loss in the Xinqiman-Yingbaza River section in the embodiment of the present invention;

[0044] Figure 3 This is the calculation result of the Yingbaza-Wusiman River section in the embodiment of the present invention;

[0045] Figure 4 The calculation result of river loss in the Usman-Qiala River section in the embodiment of the present invention;

[0046] Figure 5 This is a structural diagram of a deep neural network in an embodiment of the present invention;

[0047] Figure 6 A comparison chart of the measured and predicted river losses in the Aral-Xinqiman River section in an embodiment of the present invention;

[0048] Figure 7 A comparison chart of the measured and predicted river loss results for the Xinqiman-Yingbaza River section in an embodiment of the present invention;

[0049] Figure 8 A comparison chart of the measured and predicted river losses in the Yingbaza-Usman River section in an embodiment of the present invention;

[0050] Figure 9 A comparison chart of the measured and predicted river loss results for the Usman-Qiala River section in an embodiment of the present invention;

[0051] Figure 10 This is a schematic diagram of a water resource configuration node in an embodiment of the present invention;

[0052] Figure 11 Schematic diagram of the operation process of the aggregated reservoir in each typical year in an embodiment of the present invention, where (i) represents the operation process in a wet year, (j) represents the operation process in a normal year, and (k) represents the operation process in a dry year;

[0053] Figure 12 Schematic diagram of the operation process of the Paman Reservoir in each typical year in an embodiment of the present invention, wherein (i) represents the operation process in a wet year, (j) represents the operation process in a normal year, and (k) represents the operation process in a dry year;

[0054] Figure 13 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION

[0055] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and do not limit the scope of protection of the present invention.

[0056] A water allocation method for drought-affected areas based on river loss simulation, such as Figure 13 As shown, the method includes:

[0057] S1. Calculating the measured river loss in the watershed according to the measured runoff data of the watershed in the arid area;

[0058] Furthermore, the upstream section inflow, downstream section outflow and interval water diversion of the river section in the arid area are obtained; and the measured river loss of the river section in the arid area is calculated based on the water balance principle.

[0059] Specifically, the watershed in the arid area includes multiple river sections, and the runoff of each river section is directly or indirectly connected. In the study of the upper and middle reaches of the Tarim River, the watershed nodes such as Figure 10 shown.

[0060] Specifically, the measured runoff data can include the inflow of the upstream section of the river section, the outflow of the downstream section, and the water diversion between the upstream and downstream sections. Therefore, the measured river loss of the river basin can be calculated based on the water balance principle. Specifically, in the study of the upper and middle reaches of the Tarim River, the calculated river losses of the Aral-Xinqiman River section, the Xinqiman-Yingbaza River section, the Yingbaza-Usman River section and the Usman-Qala River section are as follows: Figure 1 、 Figure 2 、 Figure 3 and Figure 4 shown.

[0061] Specifically, the calculation formula for the measured river loss is as follows:

[0062]

[0063] Where:

[0064] is the river loss of the river section, that is, the measured river loss (10 8 m 3 );

[0065] and is the runoff of the upper and lower sections (10 8 m 3 );

[0066] is the water diversion volume of the river section (10 8 m 3 ).

[0067] S2. constructing a river loss prediction model for the river basin based on the hydrological and meteorological characteristics of the river basin and the corresponding measured river loss amount;

[0068] Furthermore, S2 may specifically be:

[0069] S21, constructing a feature sliding window to extract the feature vector of the temporal hydrological and meteorological characteristics of the basin;

[0070] S22. Using the characteristic vector as input and the corresponding measured river loss amount as a label, a river loss prediction model for the basin is constructed.

[0071] Specifically, obtaining input features involves obtaining time-series hydrometeorological characteristics. These characteristics can include historical water inflow, current water inflow, current water diversion, current groundwater depth, and current evapotranspiration. Other features include meteorological data (precipitation, temperature, humidity, and wind speed), coastal vegetation cover index, and river channel morphology coefficient.

[0072] Specifically, the historical inflow of a river segment represents the total amount of water flowing into the basin during a specific historical period, reflecting the impact of previous inflows on the current water storage status of the river segment. The current inflow of a river segment represents the total amount of water flowing into the target basin during the current period. This directly affects the water balance of the river segment and is a core input for river loss calculations. The current water diversion represents the amount of water artificially diverted from the target basin during the current period, including irrigation and industrial water withdrawals. This reflects the direct reduction in residual water volume in the river segment due to human intervention. The current groundwater depth of a river segment represents the average depth of the groundwater level around the river segment during the current period. This reflects the interaction between groundwater and river water (such as seepage recharge or discharge). Generally, a river segment with a shallower current groundwater depth is more susceptible to seepage into the groundwater. The current evapotranspiration represents the sum of surface evaporation and vegetation transpiration during the current period. This directly affects water loss in the river segment, and is particularly significant in arid regions.

[0073] Constructing a feature sliding window may include:

[0074] Specifically, a time window of a preset step size is constructed and slid according to the preset step size. In this embodiment, in order not to destroy the natural correlation between the input features, no further feature screening is performed.

[0075] Standardize and normalize the temporal hydrological and meteorological features to unify the dimensions of different features. It is also necessary to align all features to the same river section and the same time period to avoid spatial or temporal misalignment. In this embodiment, the water inflow of the river section last month (runoff of the previous month), the water inflow of the river section this month (runoff of the current month), the water diversion this month (water diversion), the groundwater depth this month (groundwater), and the evapotranspiration this month (evaporation) are used as inputs in the input layer, and the river loss obtained by the above calculation is used as a label. The river loss prediction model is trained based on the machine learning model, and its formula is as follows:

[0076]

[0077] Where:

[0078] For the water coming from the river section last month;

[0079] This is the water inflow of the river section this month;

[0080] The amount of water diverted this month;

[0081] The groundwater depth this month;

[0082] is the evaporation volume for this month;

[0083] This is the river loss for this month;

[0084] is the previous step size of the feature sliding window.

[0085] It should be noted that the above inputs and their corresponding labels are divided into training sets and test sets.

[0086] In this embodiment, the machine learning model specifically uses a deep neural network. Specifically, the input layer, hidden layer, and output layer of the deep neural network are set, the previous step size, learning rate, number of training periods, number of hidden layers, activation function, output features and other hyperparameters of the deep neural network are initialized, and the weights and biases of the deep neural network are optimized. The structural diagram of the deep neural network is shown in FIG. Figure 5 shown.

[0087] To avoid model deviation due to insufficient samples in sparse areas in the multidimensional feature space, this embodiment further includes the following steps after S21:

[0088] Performing multidimensional clustering on the feature vectors using a stratified sampling strategy of multidimensional feature clustering;

[0089] Extracting a preset proportion of samples from each layer of the multidimensional clustering as a test set, and the remaining samples as a training set;

[0090] Correspondingly, in S22, a river loss prediction model for the basin is constructed using the feature vector in the training set as input and the corresponding measured river loss amount as a label.

[0091] Specifically, the K-Means clustering algorithm is used to divide all samples into 10 "layers" with similar feature spaces. Within each layer, if the number of samples is greater than one, a 20% test sample is randomly selected (ensuring at least one test sample). The remaining samples are included in the training set, while layers with only one sample are directly assigned to the training set. This method can significantly improve the model's generalization ability and predictive stability.

[0092] In this embodiment, the specific implementation steps of the K-Means clustering process are as follows: first, the input data is standardized and preprocessed; then, random initialization is performed. Center points (i.e., 10 samples are randomly selected as initial cluster centers). Then, the iterative optimization phase begins, which specifically includes:

[0093] Sample allocation: Assign each sample to the nearest cluster center to form a temporary clustering result;

[0094] Center point update: recalculate the center point of each cluster based on the current clustering results.

[0095] The iteration termination conditions are any of the following: the cluster center no longer changes significantly (converges); the preset maximum number of iterations is reached.

[0096] If the termination condition is not met, the optimization is continued; if the condition is met, the data is divided into different levels according to the final clustering results.

[0097] After stratification, the training and test sets are proportionally divided within each stratum. Finally, the training sets of all strata are combined as the global training set, and all test sets are combined as the global test set. If the clustering results do not meet expectations (e.g., convergence), the center points must be randomly initialized again and the above process repeated until the termination criteria are met before stratification and data set division can be continued.

[0098] To efficiently search for hyperparameters of the model, S22 of this embodiment further includes:

[0099] S221. Based on the historical training result set, dynamically update the sampling distribution of the hyperparameters of the river loss prediction model to be constructed, and generate a candidate set of hyperparameters;

[0100] Specifically, the hyperparameters of the river loss prediction model to be constructed are automatically adjusted through a Bayesian optimization framework;

[0101] Updating the sampling distribution of the hyperparameters based on historical parameter test results;

[0102] Based on the updated sampling distribution, a candidate set of hyperparameters for the river loss prediction model to be constructed is generated.

[0103] S222: Generate a dynamic parameter adjustment instruction set for the river loss prediction model to be constructed based on the candidate set of hyperparameters;

[0104] Specifically, according to a preset loss convergence threshold, the parameter groups that do not meet the criteria in the hyperparameter candidate set are eliminated, and the set of parameter groups that meet the criteria is recorded as the first candidate parameter group;

[0105] Based on the parameter performance verification matrix, screening a parameter group whose accuracy meets a preset standard from the first candidate parameter group;

[0106] Based on the screened parameter group, a dynamic parameter adjustment instruction set of the river loss prediction model to be constructed is generated.

[0107] S223: Optimizing the hyperparameters of the river loss prediction model according to the dynamic parameter adjustment instruction set. This method improves the prediction performance of the model.

[0108] In this example, the Optuna hyperparameter optimization process is used. Specifically, the objective function and hyperparameter search space are defined, an optimization algorithm is selected, and an iterative optimization loop is initiated. In each iteration, Optuna generates a hyperparameter combination based on the current sampling strategy, evaluates model performance based on the objective function, and updates the parameter search direction based on the evaluation results. During the loop, the number of iterations is continuously monitored. If the number reaches a preset maximum (e.g., 100), the search is terminated and the globally optimal hyperparameter combination and its corresponding objective function value are output. If the upper limit is not reached, the parameter generation, performance evaluation, and model update process are repeated until the termination condition is met. The final output contains the optimal parameter combination and the corresponding optimal objective function value.

[0109] After training the river loss prediction model, the process also includes loading the trained model and performing a single-step rolling simulation on the test set. After testing the model, the process also includes calculating the simulation error and evaluating the simulation results. In this embodiment, the evaluation results are shown in Table 1. As can be seen from Table 1, the simulation accuracy of this embodiment is higher than that of the traditional polynomial model.

[0110] Table 1: Accuracy results of two models

[0111]

[0112] S3. Predicting river damage in the basin according to the river damage prediction model to obtain river damage prediction information;

[0113] Specifically, in the study of the upper and middle reaches of the Tarim River, the measured river losses in each river section were compared with the predicted river losses, such as Figure 6 、 Figure 7 、 Figure 8 and Figure 9 shown.

[0114] S4. Establishing constraint conditions according to the river loss prediction information, constructing a water resource optimization allocation model for the arid area based on the constraint conditions, and generating a water allocation plan for the arid area according to the water resource optimization allocation model.

[0115] S41, constructing a water resource allocation model according to the water demand of multiple users in the arid area;

[0116] Specifically, S41 may include:

[0117] S411. Collect data on water users in the basin, clarify their needs at different times, and determine agricultural and ecological water requirements.

[0118] S412. Based on the characteristics and distribution of water conservancy projects in the basin, the reservoirs, different water users, rivers, etc. in the basin are unified into a topological node diagram. Specifically, in the study of the upper and middle reaches of the Tarim River, the node diagram is as follows: Figure 10 shown.

[0119] S413. Analyze the water resources allocation in the river basin in different typical years. In this embodiment, the water resources allocation in the river basin is specifically shown in Table 2.

[0120] Table 2 Water resources allocation in each typical year

[0121]

[0122] S42, constructing a reservoir operation model according to the reservoir operation rules in the arid area;

[0123] Specifically, S42 may include:

[0124] S421. Collect the characteristic values ​​of plain reservoirs in the basin to clarify the role and operation rules of plain reservoirs.

[0125] S422. Clarify the operating constraints of plain reservoirs within the scheduling cycle. Constraints include water balance constraints, storage capacity constraints, reservoir discharge constraints, etc.

[0126] S423. Establish a plain reservoir operation model to analyze the reservoir operation process in different typical years. Specifically, in the study of the upper and middle reaches of the Tarim River, the operation process of different reservoirs (Juhe Reservoir and Paman Reservoir) in different typical years is as follows: Figure 11 、 Figure 12 shown.

[0127] S43. Couple the water resource allocation model and the reservoir scheduling model, establish water loss constraints based on the river loss prediction information, and construct a water resource optimization and allocation model.

[0128] Specifically, the reservoir discharge and node water supply in the arid area are used as decision variables, and the model coupling is performed with minimizing the water shortage of multiple users in the arid area as the optimization goal.

[0129] The objective function of the water resources optimization allocation model is as follows:

[0130]

[0131] Where:

[0132] Optimize the goal of minimizing water shortage for water users;

[0133] For the Period Agricultural water scarcity in each region;

[0134] For the Period The ecological water shortage in the region.

[0135] During the water resources optimization and allocation cycle, the reservoir operation constraints and water resources allocation constraints are as follows:

[0136] The reservoir water balance constraint is as follows:

[0137]

[0138] Where:

[0139] It is The year The storage capacity of the reservoir at the end of the time period;

[0140] It is years The inflow to the reservoir during the time period;

[0141] It is Year Downflow within a period of time.

[0142] The reservoir capacity constraint is as follows:

[0143]

[0144] Where:

[0145] For the Reservoir capacity during the period, billion m 3 ;

[0146] is the minimum storage capacity of the reservoir, 10,000 m 3 ;

[0147] is the maximum storage capacity of the reservoir, 10,000 m 3 .

[0148] The downstream flow constraint is as follows:

[0149]

[0150] Where:

[0151] yes Minimum downstream flow requirement within the time period;

[0152] It is years The outflow from the reservoir at the end of the period;

[0153] yes The maximum downstream flow demand during the period.

[0154] The node water balance constraint is as follows:

[0155]

[0156] Where:

[0157] For the period The amount of water in the main stream after the node has supplied water, 10,000 m 3 ;

[0158] For the period Node inflow flow, m 3 / s;

[0159] For the period Node water supply, 10,000 m 3 ;

[0160] For the period The river loss amount calculated by the node through the river loss prediction model is the river loss prediction information.

[0161] It should be noted that in actual applications, the river loss prediction models corresponding to different runoffs in arid areas are different. The river loss corresponding to each node in this conditional constraint is predicted based on the river loss prediction model corresponding to the node.

[0162] The node water supply capacity constraint is as follows:

[0163]

[0164] Where:

[0165] For the period Node water supply, 10,000 m 3 ;

[0166] For the period Flow capacity of the water diversion hub, m 2 / s.

[0167] Non-negative constraints, specifically water supply, all variables in the water resources optimization allocation model are non-negative.

[0168] The present invention achieves high-precision prediction of river losses through a river loss prediction model based on machine learning, and embeds it into a water resource optimization and allocation model to construct a water allocation method for arid areas based on river loss process simulation suitable for high river loss basins, making water resource allocation in arid areas more reasonable. It specifically optimizes the water supply plan of the water diversion hub so that the reservoir can efficiently release the stored water, assisting the water diversion hub in solving problems such as insufficient water supply in the basin, thereby achieving rational water use by various water-using departments in the basin and promoting high-quality and sustainable development of the basin's economy and ecology.

[0169] The above descriptions are merely several embodiments of the present invention and do not constitute any form of limitation to the present invention. Although the present invention is disclosed as above in terms of preferred embodiments, they are not intended to limit the present invention. Any technician familiar with the present profession who, without departing from the scope of the technical solution of the present invention, makes slight changes or modifications using the technical contents disclosed above are equivalent to equivalent implementation cases and fall within the scope of the technical solution.

Claims

1. A water allocation method for arid areas based on river loss simulation, characterized in that: include: S1. Calculating the measured river loss in the watershed according to the measured runoff data of the watershed in the arid area; S2. constructing a river loss prediction model for the river basin based on the hydrological and meteorological characteristics of the river basin and the corresponding measured river loss amount; S3. Predicting river damage in the basin according to the river damage prediction model to obtain river damage prediction information; S4. Establishing constraint conditions according to the river loss prediction information, constructing a water resource optimization allocation model for the arid area based on the constraint conditions, and generating a water allocation plan for the arid area according to the water resource optimization allocation model.

2. The method for water allocation in arid areas according to claim 1, characterized in that: Said S1 specifically includes: Obtain the upstream section inflow, downstream section outflow and interval water diversion of river sections in arid areas; The measured river loss of the river section in the basin is calculated based on the water balance principle.

3. The method for water allocation in arid areas according to claim 1, characterized in that: S2 specifically includes: S21, constructing a feature sliding window to extract the feature vector of the temporal hydrological and meteorological characteristics of the basin; S22. Using the characteristic vector as input and the corresponding measured river loss amount as a label, a river loss prediction model for the basin is constructed.

4. The method for water allocation in arid areas according to claim 3, characterized in that: After S21 and before S22, the following further includes: Performing multidimensional clustering on the feature vectors using a stratified sampling strategy of multidimensional feature clustering; Extracting a preset proportion of samples from each layer of the multidimensional clustering as a test set, and the remaining samples as a training set; Correspondingly, in S22, a river loss prediction model for the basin is constructed using the feature vector in the training set as input and the corresponding measured river loss amount as a label.

5. The method for water allocation in arid areas according to claim 3, characterized in that: The S22 further includes: S221. Based on the historical training result set, dynamically update the sampling distribution of the hyperparameters of the river loss prediction model to be constructed, and generate a candidate set of hyperparameters; S222: Generate a dynamic parameter adjustment instruction set for the river loss prediction model to be constructed based on the candidate set of hyperparameters; S223. Optimize the hyperparameters of the river loss prediction model according to the dynamic parameter adjustment instruction set.

6. The method for water allocation in arid areas according to claim 5, characterized in that: The S221 specifically includes: Automatically adjusting hyperparameters of the river loss prediction model to be constructed through a Bayesian optimization framework; Updating the sampling distribution of the hyperparameters based on historical parameter test results; Based on the updated sampling distribution, a candidate set of hyperparameters for the river loss prediction model to be constructed is generated.

7. The method for water allocation in arid areas according to claim 5, characterized in that: S222 specifically includes: Eliminate the parameter groups that do not meet the criteria from the hyperparameter candidate set according to a preset loss convergence threshold, and record the set of parameter groups that meet the criteria as the first candidate parameter set; Based on the parameter performance verification matrix, screening a parameter group whose accuracy meets a preset standard from the first candidate parameter group; Based on the screened parameter group, a dynamic parameter adjustment instruction set of the river loss prediction model to be constructed is generated.

8. The method for water allocation in arid areas according to claim 1, characterized in that: Establishing constraint conditions according to the river loss prediction information, and constructing a water resource optimization allocation model for the arid area based on the constraint conditions, specifically including: S41, constructing a water resource allocation model according to the water demand of multiple users in the arid area; S42, constructing a reservoir operation model according to the reservoir operation rules in the arid area; S43. Couple the water resource allocation model and the reservoir scheduling model, establish water loss constraints based on the river loss prediction information, and construct a water resource optimization and allocation model.

9. The method for water allocation in arid areas according to claim 1, characterized in that: The hydrometeorological characteristics include: Historical water inflow of the river section, current water inflow of the river section, current water diversion volume, current groundwater depth of the river section and current evaporation and emission.

10. The method for water allocation in arid areas according to claim 8, characterized in that: Coupling the water resource allocation model and the reservoir operation model specifically includes: The reservoir discharge and node water supply in the arid area are used as decision variables, and the model coupling is performed with the minimum water shortage of multiple users in the arid area as the optimization goal.

Images