Prediction method and system for the impact of flood storage and detention area operation on flood level in the river section

Abstract

The present invention provides a method and system for predicting the impact of flood storage and detention area operation on the flood flow level of a river section. Based on historical typical flood sequences or artificially simulated flood sequences of flood storage and detention area operation and corresponding data on the impact of flood storage and detention area operation on the flood flow level of the river section, multiple flood storage and detention area operation samples are generated. Key factors that affect the effectiveness of flood storage and detention area operation are identified and optimized using an optimal information set selection wrapper. A flood storage and detention area operation impact prediction model is constructed based on training and verification using an artificial neural network. Using the flood storage and detention area operation impact prediction model, the impact of flood storage and detention area operation on the flood flow level of the river section is predicted under the water and engineering conditions of each flood storage and detention area operation. The present invention is simple, easy to use, and balances computational speed and accuracy, achieving high-precision and rapid prediction of the impact of flood storage and detention area operation on the flood flow level, thus having important engineering application value.

Description

Technical Field

[0001] The invention belongs to the technical field of water conservancy project calculation management, and relates to a method and system for predicting the impact of flood storage and detention area operation on flood discharge levels. Background Art

[0002] The construction of flood control projects in a river basin is an important means of combating the threat of flood disasters. Flood storage and detention areas are temporary flood storage sites and an important component of the river basin flood control engineering system. They are the "trump card" for ensuring flood control safety in the river basin. When a major flood occurs in a river basin, and despite the reasonable flood control by reservoirs and the sufficient flow of water through the river channel, the flood volume still exceeds the discharge capacity of the river channel, the flood storage and detention areas must be activated to store the excess flood volume, thereby avoiding levee breaches and devastating disasters. In other years, restricted production activities and daily life can be carried out within the flood storage and detention areas. Developing a scientific flood storage and detention area utilization plan is of great significance for maximizing the overall flood control safety of the river basin and reducing the impact of flood diversion on the socioeconomic situation. Rapidly and accurately predicting the impact of flood storage and detention area operation on river flood levels is the basis for developing a real-time flood storage and detention area utilization plan.

[0003] Currently, the operational impact of flood storage and detention areas is typically calculated using hydrodynamic models. These models generalize the river network and detention areas and solve the Saint-Venant equations to determine the water level and flow processes at each section. While these models offer high computational accuracy and a clear physical foundation, they utilize a large number of computational grids and complex algorithms to describe the spatiotemporal evolution of the hydrodynamic system. These models place high demands on data length, modeling paradigms, and computational power, resulting in complex implementation and slow computation, making them unsuitable for real-time computing. Compared to process-based hydrodynamic models, data-driven models such as support vector machines and artificial neural networks utilize statistical methods to extract correlations between variables, offering high computational efficiency. However, the accuracy of data-driven models depends heavily on the adequacy and representativeness of the data. The input factors and amount of data in data-driven models are crucial for balancing accuracy and ease of use. However, since detention areas are only operational during periods of extreme floods, operational data is very limited, making it difficult for data-driven models to obtain sufficient input data. This reduces the accuracy of data-driven models in calculating the operational impact of detention areas. Summary of the Invention

[0004] In order to solve the problems described in the background technology that the hydrodynamic model in the calculation of the impact of the use of flood storage and detention areas has high requirements on data length, modeling paradigm, and computing power conditions, the calculation speed is slow, does not meet the requirements of real-time calculation, and the data-driven model has difficulty in obtaining sufficient input data volume, the present invention proposes a method and system for predicting the impact of the use of flood storage and detention areas on the flood level of the river section.

[0005] The method of the present invention comprises:

[0006] Generate various flood storage and detention area utilization samples based on historical typical flood sequences or artificially simulated flood sequences and the corresponding data on the impact of flood storage and detention area utilization on the flood level of the river section;

[0007] Based on a variety of flood storage and detention area application samples, the optimal information set is used to select a wrapper, identify and optimize the key factors that affect the effectiveness of flood storage and detention area application.

[0008] Using various flood storage and detention area operation samples and key factors affecting their effectiveness as input factors, and the magnitude of water level reduction at important control stations during each period of the forecast period as the target factor, a flood storage and detention area operation impact prediction model was constructed through training and verification based on an artificial neural network.

[0009] Using the flood storage and detention area utilization impact prediction model, the impact of flood storage and detention area utilization on the flood discharge level of the river section is predicted under the water and working conditions of each flood storage and detention area.

[0010] Furthermore, the flood storage and detention area utilization sample includes the flood diversion utilization status of the flood storage and detention area and the flood storage and detention area utilization impact;

[0011] The flood diversion operation state of the flood storage and detention area is determined by the water inflow conditions and the engineering operation strategy; the method for obtaining the water inflow conditions is: according to the activation conditions of the flood storage and detention area, the water level of the important control station is selected from the historical typical flood sequence or the artificially simulated flood sequence to reach a certain magnitude of flood, so as to characterize different water inflow conditions; the method for obtaining the engineering operation strategy is: from the historical typical flood sequence or the artificially simulated flood sequence, variables including flood diversion timing, flood diversion flow and flood diversion duration are obtained, and the Latin hypercube sampling method is used to first consider the variation range of each variable separately, and each variable is globally sampled between the upper and lower limits to preliminarily generate the flood storage and detention area engineering operation strategy, and then consider the correlation between the variables, and according to the flood storage capacity limit of the flood storage and detention area and the randomly generated flood diversion flow, the flood diversion duration in the preliminarily generated flood storage and detention area flood diversion operation state is corrected to obtain the final flood storage and detention area engineering operation strategy;

[0012] The method for obtaining the impact of flood storage and detention area operation is as follows: the water level of the key control station is used to represent the flood discharge level of the river section; under the flood diversion operation state of each flood storage and detention area, a one-dimensional hydrodynamic model that has been calibrated and verified is run to calculate the water level process of the key control station under the flood diversion operation state of the flood storage and detention area; under various hydrological conditions, the same hydrodynamic model is run to calculate the water level process of the key control station under the condition that the flood storage and detention area does not divert floods; the impact of the flood diversion operation of the flood storage and detention area on the flood discharge level of the river section is compared and analyzed, and the impact of the flood diversion operation of the flood storage and detention area is obtained.

[0013] Furthermore, the method for identifying and optimizing the key factors that affect the effectiveness of flood storage and detention areas includes:

[0014] Based on various flood storage and detention area application samples, a candidate factor set was constructed using the flow at each upstream control station, the water level at each downstream control station, the rise and fall rate, and the flood diversion flow, flood diversion days, and impact period of the flood storage and detention area in the period before the flood storage and detention area was put into use as possible influencing factors.

[0015] According to the data of the impact of the alternative factor set and the use of flood storage and detention areas on the flood level of the river section, the optimal information set is used to select the wrapper, and the key factors that affect the effect of lowering the flood level are selected from the alternative factor set.

[0016] Furthermore, the optimization objectives in the optimal information set selection wrapper include: the prediction accuracy of the factor set for the target variable f 1( S i ) Highest, factor set dimension f 2( S i ) Minimum, correlation between factor set and target variable f 3( S i ) Maximum, factor set redundancy f 4( S i ) is the lowest; the factor set has the lowest prediction accuracy for the target variable f 1( S i ) Maximum sum factor set dimension f 2( S i ) is the main optimization goal, and the correlation between the factor set and the target variable f 3( S i ) Maximum sum factor set redundancy f 4( S i ) is the lowest secondary optimization goal; in the process of selecting the optimal information set wrapper, the decision variables are set to 0 and 1 variables, representing the factors in each candidate factor set X Whether it is selected, the factors whose decision variables are all 1 in each non-dominated solution are the selected factors, forming the set , a non-dominated sorting genetic algorithm is used to optimize multiple non-dominated alternative factor solutions at one time to form a non-dominated solution set. Then, based on non-quantitative factors including whether the observation data is easy to obtain and whether the observation is reliable, an appropriate solution is selected from the non-dominated solution set as the optimal key factor affecting the flood level lowering effect.

[0017] Furthermore, the factor set has a prediction accuracy for the target variable f 1( S i ) Use the target variable to predict the value The symmetric uncertainty SU between the measured value y indicates that the factor set has a high prediction accuracy for the target variable. f 1( S i )The highest expression is as follows:

[0018] (1),

[0019] (2),

[0020] (3),

[0021] Where, S i is the factor set, x i For each factor in the factor set, X is the set of candidate factors, ; H (‧) is the information entropy of the variable; M is the number of hidden layer nodes, Represents the randomly generated mth hidden layer node sigmoid activation function parameter, Represents the output weight to be calibrated;

[0022] The factor set dimension f 2( S i ) i.e. the number of factors in the factor set | S i |, the dimension of the factor set f 2( S i )The lowest expression is as follows:

[0023] (4);

[0024] The correlation between the factor set and the target variable f 3( S i ) Use factor centering for each factor x i ( x i ∈ S i ) and the target variable y The sum of the symmetric uncertainty of the factor set and the correlation between the target variable f 3( S i )The largest expression is as follows:

[0025] (5),

[0026] Where, S i is the factor set, x i For each factor in the factor set, X is the set of candidate factors, , y is the target variable;

[0027] The factor set redundancy f 4( S i ) is the sum of the symmetric uncertainties of each factor in the factor set, and the redundancy of the factor set is f 4( S i )The lowest expression is as follows:

[0028] (6).

[0029] Furthermore, the process of constructing the flood storage and detention area utilization impact prediction model includes:

[0030] Using various flood storage and detention area application samples and key factors affecting the flood level reduction effect as input factors, and the water level reduction amplitude of important control stations in each period during the forecast period as the target factor, the number of hidden layers and the form of transfer function are determined according to the general approximation theorem. The number of hidden neural networks is determined through repeated experiments. A two-layer feedforward network with a sigmoid function as the hidden layer neuron activation function and a linear function as the output layer neuron activation function is used to approximate the multidimensional mapping problem and determine the artificial neural network structure. The expression is as follows:

[0031] (7),

[0032] Where, For the moment t + τ Predicted changes in control station water levels caused by flood discharge in the flood storage and detention area, , T For the foreseeable period, N is the number of hidden layer neurons with sigmoid function as activation function, For the moment t The input vector of , consists of the key influencing factors identified, is the artificial neural network parameter, the number is N×(2+D)+1, where D is the dimension of the input space, that is, the number of key influencing factors;

[0033] The Levenberg-Marquardt training algorithm is used to optimize the parameters of the artificial neural network. In each training session, a training set, a smaller validation set, and a test set are set. The training set is used to calculate the gradient and update the network; the validation set is used to determine whether the prediction accuracy stops improving and when to stop training; the test set has no effect on the training and independently measures the performance of the trained network. When the error metric of the validation set is stable or the prediction error is acceptable, the training iteration is terminated, and the artificial neural network structure and corresponding parameter values ​​are determined, which is used as a prediction model for the impact of flood storage and detention area utilization.

[0034] Based on the above method, the present invention proposes a system for predicting the impact of flood storage and detention area utilization on the flood flow level of the river section, including a flood storage and detention area utilization sample generation module, a key factor identification and optimization module, a flood storage and detention area utilization impact prediction model construction module, and a flood storage and detention area utilization impact prediction module.

[0035] The flood storage and detention area utilization sample generation module is used to generate a variety of flood storage and detention area utilization samples based on historical typical flood sequences or artificially simulated flood sequences of flood storage and detention area utilization and the corresponding data on the impact of flood storage and detention area utilization on the flood level of the river channel in this river section.

[0036] The key factor identification and optimization module is used to select a wrapper based on multiple flood storage and detention area application samples through an optimal information set to identify and optimize key factors that affect the application effect of the flood storage and detention area.

[0037] The flood storage and detention area utilization impact prediction model construction module is used to use a variety of flood storage and detention area utilization samples and key factors affecting the flood storage and detention area utilization effect as input factors, and the water level reduction amplitude of important control stations in each period during the forecast period as the target factor, and to train and verify based on artificial neural networks to construct a flood storage and detention area utilization impact prediction model.

[0038] The module for predicting the impact of the use of flood storage and detention areas on flood flow levels is used to use the flood storage and detention area use impact prediction model to predict the impact of the use of flood storage and detention areas on the flood flow levels of the river section under the water conditions and working conditions of each flood storage and detention area.

[0039] The present invention also proposes a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the method and system for predicting the impact of the use of flood storage areas on the flood level of the river section as described above.

[0040] Compared with the existing technology, the present invention has the following advantages: (1) It combines the advantages of hydrodynamic models and data-driven models to achieve high-precision and rapid prediction of flood level impact; (2) It obtains a large amount of sample data on the use of flood storage and detention areas through historical typical flood sequences or artificially simulated flood sequences, overcomes the limitation of insufficient measured data, and the constructed model is simple and easy to use; (3) By selecting the wrapper through the optimal information set, it excavates and clarifies the key factors that determine the effect of the use of flood storage and detention areas, and improves the prediction accuracy of the subsequently constructed model; (4) It provides an effective technical means for the prediction of the effect of the use of flood storage and detention areas in the post-engineering period, and has high engineering application and promotion value.

[0041] In summary, the present invention solves the problems existing in the hydrodynamic model and data-driven model in the calculation of the impact of the use of flood storage and detention areas. The present invention not only overcomes the limitation of insufficient measured data by artificially simulating a large amount of sample data, but also explores key influencing factors and constructs an efficient prediction model. The model is simple and easy to use, and takes into account both calculation speed and accuracy, thereby achieving high-precision and rapid prediction of the impact of flood discharge levels, and has important engineering application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 Flow chart of the method of the present invention.

[0043] Figure 2 Select a matrix for the key influencing factors of flood diversion in a flood storage and detention area.

[0044] Figure 3 The correlation between the predicted value and target value of the flood diversion effect of a flood storage area on a hydrological station in different sample sets.

[0045] Figure 4 This is the cumulative distribution curve of the prediction error of a flood storage area for a hydrological station.

[0046] Figure 5 This is the first example of calculating the water level drop at a hydrological station after flood diversion in a flood storage area using different models under typical flood conditions.

[0047] Figure 6 This is the second example of calculating the change process of water level drop at a hydrological station after flood diversion in a flood storage area using different models under typical flood conditions.

[0048] Figure 7 This is the third example of calculating the change process of water level drop at a hydrological station after flood diversion in a flood storage area using different models under typical flood conditions.

[0049] Figure 8 This is a system architecture diagram of the present invention. DETAILED DESCRIPTION

[0050] In order to make the technical problems, technical solutions and beneficial effects to be solved by this application more clearly understood, this application is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0051] Example 1

[0052] The method for predicting the impact of flood storage and detention areas on the flood level of the river section is as follows: Figure 1 As shown, the details are as follows.

[0053] First, based on the historical typical flood sequences or artificially simulated flood sequences of flood storage and detention area utilization and the corresponding data on the impact of flood storage and detention area utilization on the flood level of the river section, a variety of flood storage and detention area utilization samples are generated.

[0054] Specifically, the flood storage and detention area utilization samples include the flood diversion utilization status of the flood storage and detention area and the impact of the flood storage and detention area utilization.

[0055] The flood diversion status of the flood storage and detention area is determined by the inflow conditions and the engineering operation strategy. The inflow conditions are obtained by selecting floods at key control stations that reach a certain magnitude from historical typical flood sequences or artificially simulated flood sequences, based on the activation conditions of the flood storage and detention area. The engineering operation strategy is obtained by obtaining variables including diversion timing, diversion flow, and diversion duration from historical typical flood sequences or artificially simulated flood sequences. Using the Latin hypercube sampling method, the range of each variable is first considered individually, and each variable is globally sampled between the upper and lower limits to initially generate the flood storage and detention area engineering operation strategy. Then, considering the correlation between variables, the diversion duration in the initially generated flood storage and detention area diversion operation status is modified based on the flood storage capacity limit and the randomly generated diversion flow, resulting in the final flood storage and detention area engineering operation strategy.

[0056] The method for obtaining the impact of flood storage and detention area operation is as follows: the water level of the key control station is used to represent the flood discharge level of the river section; under the flood diversion operation status of each flood storage and detention area, a one-dimensional hydrodynamic model that has been calibrated and verified is run to calculate the water level process of the key control station under the flood diversion operation status of the flood storage and detention area; under various hydrological conditions, the same hydrodynamic model is run to calculate the water level process of the key control station under the condition that the flood storage and detention area does not divert floods; the impact of the flood diversion operation of the flood storage and detention area on the flood discharge level of the river section is compared and analyzed, and the impact of the flood diversion operation of the flood storage and detention area is obtained.

[0057] In this embodiment, for a certain flood storage and detention area, a large number of samples of the impact of the operation of a certain flood storage and detention area on the water level of a certain hydrological station are randomly generated under different hydrological conditions and operation strategies. The operation state parameters of a certain flood storage and detention area and the sampling range of each parameter are shown in Table 1. Considering the activation conditions of a certain flood storage and detention area, from the historical measured typical large floods of different magnitudes and flood compositions, the main flood season flood process corresponding to the water level of another hydrological station exceeding the activation water level is selected as the inflow boundary set to reflect the changes in hydrological conditions. The typical large floods selected include floods in 1931, 1935, 1954, 1968, 1996, and 1998. The gate control and diversion strategy of a certain flood storage and detention area is described by the activation time, diversion flow and diversion duration. The activation time is discretized over the entire flood process. For example, during the 1935 flood, the activation time of a flood storage and detention area was any time point between 6:00 AM on July 8, 1935, and 6:00 AM on July 14, 1935. The diversion flow rate is discretized from 0 to the maximum diversion flow rate of the flood storage and detention area. The diversion duration is discretized from 0 to 10 days. First, a Latin hypercube sampling method is used to perform global sampling between the upper and lower limits of each variable, generating 600 diversion operation states. The diversion duration of each sample is then adjusted based on the storage capacity limit of the flood storage and detention area. Under each flood storage and detention area operation state, a calibrated one-dimensional hydrodynamic model is used to simulate the water level process at a hydrological station after diversion of the flood storage and detention area. Under the same hydrological conditions, the same hydrodynamic model is used to simulate the water level process at a hydrological station without diversion of the flood storage and detention area. The water level drop at the hydrological station before and after diversion is compared to analyze the impact of diversion on the water level process at the hydrological station. Based on the current state of effective hydrological forecasting in the lower Yangtze River basin, this model's forecast period is set to seven days. Based on the real-time scheduling step size, the forecast time step is set to six hours, resulting in a total of 29 sampling points within the forecast period. Combining the operational state variables and the water level drawdown at a hydrological station, an operational sample of a specific flood storage and detention area was constructed, resulting in a total of 17,400 simulation samples.

[0058] Table 1 Operational parameters and variation range of a flood storage area

[0059]

[0060] Then, based on a variety of flood storage and detention area application samples, the wrapper is selected through the optimal information set to identify and optimize the key factors affecting the application effect of flood storage and detention areas.

[0061] Specifically, based on the hydrological factors that influence the effectiveness of flood storage and detention area operations, including upstream inflow (depicting flood magnitude), water level (depicting water inertia), downstream support, and the rate of water level fluctuation (depicting flood stages), as well as the engineering operation strategies of flood storage and detention areas, and combining the ease of use of key factors with the complex spatial correlation between stations, a set of candidate factors was constructed, using the flow at each upstream control station in the river section before the flood storage and detention area was activated, the water level and fluctuation rate at each downstream control station, and the flood diversion flow, number of days, and impact period of the flood storage and detention area as possible influencing factors. Based on the candidate factor set and data on the impact of flood storage and detention area operation on the river channel flood level in this river section, a wrapper was selected using the optimal information set, and key factors that influence the effect of flood level reduction were selected from the candidate factor set.

[0062] The optimization objectives in the optimal information set selection wrapper include: the prediction accuracy of the factor set for the target variable f 1( S i ) Highest, factor set dimension f 2( S i ) Minimum, correlation between factor set and target variable f 3( S i ) Maximum, factor set redundancy f 4( S i ) is the lowest. The factor set has the lowest prediction accuracy for the target variable. f 1( S i ) Maximum sum factor set dimension f 2( S i ) is the main optimization goal, which aims to ensure the high prediction accuracy of the factor set for the target variable while streamlining the number of factors screened. Correlation between the factor set and the target variable f 3( S i ) Maximum sum factor set redundancy f 4( S i ) is the secondary optimization goal, which is to expand the set of non-dominated solutions and provide more options for subsequent screening factors based on expert experience.

[0063] In the optimization process of the optimal information set selection wrapper, the decision variables are set to 0 and 1 variables, representing the factors in each candidate factor set X. Whether it is selected, the factors whose decision variables are all 1 in each non-dominated solution are the selected factors, forming the set , a non-dominated sorting genetic algorithm is used to optimize multiple non-dominated alternative factor solutions at one time to form a non-dominated solution set. Then, based on non-quantitative factors including whether the observation data is easy to obtain and whether the observation is reliable, an appropriate solution is selected from the non-dominated solution set as the optimal key factor affecting the flood level lowering effect.

[0064] More specifically, the factor set has a significant impact on the prediction accuracy of the target variable. f 1( S i ) Use the target variable to predict the value The symmetric uncertainty SU between the measured value y indicates that the greater the prediction accuracy, the stronger the predictive power of the factor set for the target variable. SU is an important concept in information theory and is calculated from the information entropy H(‧) of the variable. SU equals 0, indicating that the two variables are completely unrelated, and SU equals 1, indicating that the two variables are completely correlated. f 1( S i )The highest expression is as follows:

[0065] (1),

[0066] (2),

[0067] (3),

[0068] Where, S i is the factor set, x i is each factor in the factor set. Formula (3) is based on the factor set S i Each selected factor As input, the target variable prediction is predicted using the extreme learning machine model , M is the number of hidden layer nodes, X is the set of candidate variables, Represents the randomly generated mth hidden layer node sigmoid activation function parameter, Represents the output weight to be calibrated.

[0069] Factor set dimension f 2( S i ) i.e. the number of factors in the factor set | S i |, the fewer factors in a factor set, the easier it is to use. Factor set dimension f 2( S i )The lowest expression is as follows:

[0070] (4).

[0071] Correlation between factor set and target variable f 3( S i ) Use factor centering for each factor x i ( x i ∈ S i ) and the target variable y The greater the correlation between the factor set and the target variable, the more it can reflect the change of the target variable. f 3( S i )The largest expression is as follows:

[0072] (5),

[0073] Where, S i is the factor set, x i For each factor in the factor set, X is the set of candidate factors, , y is the target variable.

[0074] Factor set redundancy f 4( S i ) is the sum of the symmetrical uncertainties of each factor in the factor set. The smaller the redundancy of the factor set, the smaller the similarity of each variable in the factor set, and the greater the diversity of the selected factors. f 4( S i )The lowest expression is as follows:

[0075] (6).

[0076] In this embodiment, the upstream inflow that describes the flood magnitude, the water level that describes the inertia of the water body, the water level fluctuation rate that describes the flood stage, and the downstream water level that describes the supporting effect are comprehensively considered. The complex spatial correlation between stations is considered to construct a candidate factor set, as shown in Table 2. k The observations at time intervals, kis the flood propagation time from this station to a certain hydrological station. Other variables are the observed values ​​at the time of flood diversion. The water level drop within the forecast period of a certain hydrological station is taken as the target variable. The optimal information set selection wrapper algorithm is used to analyze the key influencing factors of the flood diversion effect of a certain flood storage area in the alternative factor set. The non-dominated sorting genetic algorithm is used as the multi-objective evolutionary algorithm for optimization and solution. The target evaluation number is set to 5000 in this algorithm. In the extreme learning machine algorithm, the number of hidden layer neurons with sigmoid function as the activation function is set to 50, and 10-fold cross validation is performed. Taking into account the randomness of the algorithm, the prediction is repeated 5 times under each input factor, and the average value of each prediction result is used to estimate the prediction accuracy of each model.

[0077] Table 2 Alternative factor sets for optimal information set selection wrapper algorithm analysis

[0078]

[0079] Weighing the optimal number of factor sets and the prediction accuracy of the factor sets, we select the factor sets with prediction accuracy within 60% of the highest accuracy. Each candidate subset consists of Figure 2 The "Selection Matrix" is shown. Each row represents a factor set, with the identified key influencing factors marked with gray rectangles. The factor sets are sorted in ascending order of dimensionality from top to bottom. Within each dimensionality, the factor sets are sorted in ascending order of predictive accuracy. The marker color varies with the dimensionality of the subset, with lighter gray indicating a smaller subset dimension. In this example, the factor set dimensions range from 2 to 5. The predictive accuracy of each factor set is marked on the right, with the highest predictive accuracy indicated in red.

[0080] The vertical bars connecting multiple rows of gray markers reflect the correlation of the predictor with the target variable. The flood storage and detention area diversion strategy variables, namely diversion flow and diversion duration, are included in each subset, forming an uninterrupted vertical bar. The time variable is included in each subset with an accuracy greater than 0.6. This indicates that these variables cannot be replaced by other input factors, otherwise the prediction accuracy will be greatly reduced. These variables are strongly correlated factors affecting the flood storage and detention area's diversion effect. Factor set 7 has the highest prediction accuracy and consists of the water level of a certain downstream hydrological station, the water level of a certain hydrological station, the diversion flow, the diversion duration, and the time sequence number. Compared with factor set 7, the accuracy of factor set 6 is reduced by less than 2%. Therefore, the water level of a certain downstream hydrological station is replaced by the water level of a certain hydrological station, which is a more critical control station. The reliability of the observation value acquisition of factor set 6 is improved, and factor set 6 is the selected factor set.

[0081] Then, a flood storage and detention area utilization impact prediction model was constructed based on artificial neural network training and verification, using various flood storage and detention area utilization samples and key factors affecting the flood level reduction effect as input factors, and the water level reduction amplitude of important control stations in each period during the forecast period as target factors.

[0082] Specifically, various flood storage and detention area utilization samples and key factors affecting the flood level reduction effect are used as input factors, and the water level reduction amplitude of important control stations in each period during the forecast period is used as the target factor. In order to minimize the amount of calculation, the minimum structure with acceptable error is adopted. The number of hidden layers and the form of transfer function are determined according to the general approximation theorem. The number of hidden neural networks is determined through repeated experiments. A two-layer feedforward network with a sigmoid function as the hidden layer neuron activation function and a linear function as the output layer neuron activation function is used to approximate the multidimensional mapping problem and determine the artificial neural network structure. The expression is as follows:

[0083] (7),

[0084] Where, For the moment t + τ Predicted changes in control station water levels caused by flood discharge in the flood storage area, , T For the foreseeable period, N is the number of hidden layer neurons with sigmoid function as activation function, For the moment t The input vector of , consists of the key influencing factors identified, is the artificial neural network parameter, and its number is N×(2+D)+1, where D is the dimension of the input space, that is, the number of key influencing factors.

[0085] The Levenberg-Marquardt training algorithm is used to optimize the parameters of the artificial neural network. In each training, a training set, a smaller validation set, and a test set are set. The training set is used to calculate the gradient and update the network; the validation set is used to determine whether the prediction accuracy stops improving and when to stop training; the test set has no effect on the training and independently measures the performance of the trained network. To avoid overfitting and save training time, the training iteration is terminated when the error metric of the validation set is stable or the prediction error is acceptable. The artificial neural network structure and corresponding parameter values ​​are determined and used as a prediction model for the impact of flood storage and detention area utilization.

[0086] In this example, 17,400 samples from a flood storage area were randomly divided into training (70%), validation (15%), and test (15%) subsets. When the number of hidden neurons exceeds 10, the marginal benefit of increasing the number of hidden neurons decreases, so the number of hidden neurons is set to 10.

[0087] The correlation coefficients between the predicted values ​​and target values ​​of the training, validation and test sets are as follows: Figure 3 As shown in , the correlation coefficients of each subset are all higher than 0.99, indicating that the prediction performance is good and there is no overfitting. The cumulative distribution curve of the absolute value of the prediction error is shown in Figure 4 The results show that the probability of the prediction error being less than 3 cm is 90%.

[0088] The process of the impact of flood diversion in a certain flood storage and detention area on the water level of a certain hydrological station under typical flood conditions was further compared with the MIKE model to verify the prediction accuracy of the flood storage and detention area application impact prediction model constructed by the present invention. The basin-wide floods in 1954 and 1998 and the regional flood in 1935 were selected for testing. When the water level forecast at a certain hydrological station reached 34.4m, the flood storage and detention area was activated. In order to test the accuracy of the prediction results of the flood diversion impact prediction model constructed under arbitrary flood diversion conditions, during the flood process in which the water level at another hydrological station exceeded 34.4m, the flood diversion time was randomly selected, and the flood diversion flow and flood diversion duration were randomly generated to construct a flood diversion scenario. Different models in each scenario calculate the change process of the water level drop at a certain hydrological station after the flood diversion in a certain flood storage and detention area, as shown in the figure below. Figure 5 、 Figure 6 and Figure 7 As shown, Figure 5 On July 27, 1954, a flood storage area diverted floodwaters at a flow rate of 5320 m³ / s for 3.15 days. Figure 6 For example: On July 25, 1998, a flood storage area diverted floodwaters at a flow rate of 5480 m³ / s for 3.95 days; Figure 7 Figure 1: On July 13, 1935, a flood storage and detention area diverted floodwater at a flow rate of 3160 m³ / s for 3.95 days. On the primary axis, the black solid line represents the water level drop at a hydrological station simulated by the MIKE model in the sample, while the red solid line represents the water level drop at the hydrological station predicted by the constructed model. On the secondary axis, the gray bar chart represents the forecast error for each time period. As shown in the figure, under three typical scenarios, the water level drop trends at a hydrological station simulated by the flood storage and detention area impact prediction model and the MIKE model are consistent. Over the 7-day forecast period, the forecast error for each time period under the 1954 flood diversion scenario was no more than 3.5 cm, the forecast error for each time period under the 1998 flood diversion scenario was no more than 3 cm, and the forecast error for each time period under the 1935 flood diversion scenario was no more than 1 cm. Overall, the constructed flood storage and detention area impact prediction model accurately predicts the water level drop at a hydrological station after the flood storage and detention area diverts floodwater.

[0089] A 7-day simulation was conducted using the traditional one-dimensional hydrodynamic MIKE model to obtain the impact of the operation of a flood storage and detention area on the water level of a hydrological station. The calculation time was about 1.5 minutes. The calculation time of the constructed flood storage and detention area operation impact prediction model was within 1 second, which greatly reduced the calculation cost while maintaining accuracy.

[0090] Finally, the flood storage and detention area operation impact prediction model is used to predict the impact of flood discharge level on the river section under the water and working conditions of each flood storage and detention area.

[0091] Example 2

[0092] The flood storage area uses the flood level prediction system for the river section, the structure diagram is as follows Figure 8 As shown in FIG, it consists of a flood storage and detention area utilization sample generation module, a key factor identification and optimization module, a flood storage and detention area utilization impact prediction model construction module, and a flood storage and detention area utilization impact prediction module.

[0093] The flood storage and detention area utilization sample generation module is used to generate a variety of flood storage and detention area utilization samples based on the historical typical flood sequence or artificially simulated flood sequence of flood storage and detention area utilization and the corresponding data on the impact of flood storage and detention area utilization on the flood level of the river section.

[0094] The key factor identification and optimization module is used to select the packager based on a variety of flood storage and detention area application samples through the optimal information set, and identify and optimize the key factors that affect the application effect of flood storage and detention areas.

[0095] The module for constructing a flood storage and detention area utilization impact prediction model is used to use a variety of flood storage and detention area utilization samples and key factors that affect the flood level reduction effect as input factors, and the water level reduction amplitude of important control stations in each period during the forecast period as target factors. It is trained and verified based on artificial neural networks to construct a flood storage and detention area utilization impact prediction model.

[0096] The module for predicting the impact of the use of flood storage and detention areas on flood flow levels is used to use the flood storage and detention area use impact prediction model to predict the impact of the use of flood storage and detention areas on the flood flow levels of the river section under the water conditions and working conditions of each flood storage and detention area.

[0097] The specific implementation of each module in this system is consistent with that described in Example 1 and will not be repeated here.

[0098] Example 3

[0099] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the method for predicting the impact of the use of flood storage and detention areas on the flood level of a river channel in a river section as described in Example 1, and the system for predicting the impact of the use of flood storage and detention areas on the flood level of a river channel in a river section as described in Example 2.

[0100] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of the present application may be implemented in various computer languages, such as object-oriented programming languages ​​Java, C++, Python, and interpreted scripting languages ​​like JavaScript.

[0101] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0102] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0103] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0104] Although the preferred embodiments of the present application have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present application.

[0105] Obviously, those skilled in the art may make various changes and modifications to this application without departing from the spirit and scope of this application. Thus, if these modifications and variations of this application fall within the scope of the claims of this application and their equivalents, this application is intended to include these modifications and variations.

Claims

1. The method for predicting the impact of flood storage and detention areas on the flood level of the river section is characterized by: include: Generate various flood storage and detention area utilization samples based on historical typical flood sequences or artificially simulated flood sequences and the corresponding data on the impact of flood storage and detention area utilization on the flood level of the river section; Based on a variety of flood storage and detention area application samples, the optimal information set is used to select a wrapper, identify and optimize the key factors that affect the effectiveness of flood storage and detention area application. Using various flood storage and detention area operation samples and key factors affecting their effectiveness as input factors, and the magnitude of water level reduction at important control stations during each period of the forecast period as the target factor, a flood storage and detention area operation impact prediction model was constructed through training and verification based on an artificial neural network. Using the flood storage and detention area operation impact prediction model, under the water and working conditions of each flood storage and detention area, the impact of the operation of the flood storage and detention area on the flood flow level of the river section is predicted; The method for identifying and optimizing key factors that affect the effectiveness of flood storage and detention areas includes: constructing a set of candidate factors based on multiple flood storage and detention area utilization samples, taking the flow of each upstream control station in the river section before the flood storage and detention area is activated, the water level of each downstream control station, the rise and fall rate, and the flood diversion flow, flood diversion days, and impact period of the flood storage and detention area as possible influencing factors; selecting a wrapper using an optimal information set based on the set of candidate factors and data on the impact of the flood storage and detention area operation on the flood level of the river channel in the river section, and optimizing the key factors that affect the effectiveness of the flood storage and detention area from the set of candidate factors; The optimization objectives in the optimal information set selection wrapper include: the prediction accuracy of the factor set for the target variable f 1( S i ) Highest, factor set dimension f 2( S i ) Minimum, correlation between factor set and target variable f 3( S i ) Maximum, factor set redundancy f 4( S i ) is the lowest; the factor set has the lowest prediction accuracy for the target variable f 1( S i ) Maximum sum factor set dimension f 2( S i ) is the main optimization goal, and the correlation between the factor set and the target variable f 3( S i ) Maximum sum factor set redundancy f 4( S i ) is the lowest secondary optimization goal; in the process of selecting the optimal information set wrapper, the decision variables are set to 0 and 1 variables, representing the factors in each candidate factor set X Whether it is selected, the factors whose decision variables are all 1 in each non-dominated solution are the selected factors, forming the set , a non-dominated sorting genetic algorithm is used to optimize multiple non-dominated alternative factor solutions at one time to form a non-dominated solution set. Then, based on non-quantitative factors including whether the observation data is easy to obtain and whether the observation is reliable, an appropriate solution is selected from the non-dominated solution set as the optimal key factor affecting the flood level lowering effect.

2. The method for predicting the impact of flood storage and detention area operation on the flood level of a river section according to claim 1 is characterized by: The flood storage and detention area utilization sample includes the flood storage and detention area flooding utilization status and the flood storage and detention area utilization impact; The flood diversion operation state of the flood storage and detention area is determined by the water inflow conditions and the engineering operation strategy; the method for obtaining the water inflow conditions is: according to the activation conditions of the flood storage and detention area, the water level of the important control station is selected from the historical typical flood sequence or the artificially simulated flood sequence to reach a certain magnitude of flood, so as to characterize different water inflow conditions; the method for obtaining the engineering operation strategy is: from the historical typical flood sequence or the artificially simulated flood sequence, variables including flood diversion timing, flood diversion flow and flood diversion duration are obtained, and the Latin hypercube sampling method is used to first consider the variation range of each variable separately, and each variable is globally sampled between the upper and lower limits to preliminarily generate the flood storage and detention area engineering operation strategy, and then consider the correlation between the variables, and according to the flood storage capacity limit of the flood storage and detention area and the randomly generated flood diversion flow, the flood diversion duration in the preliminarily generated flood storage and detention area flood diversion operation state is corrected to obtain the final flood storage and detention area engineering operation strategy; The method for obtaining the impact of flood storage and detention area operation is as follows: the water level of the key control station is used to represent the flood discharge level of the river section; under the flood diversion operation state of each flood storage and detention area, a one-dimensional hydrodynamic model that has been calibrated and verified is run to calculate the water level process of the key control station under the flood diversion operation state of the flood storage and detention area; under various hydrological conditions, the same hydrodynamic model is run to calculate the water level process of the key control station under the condition that the flood storage and detention area does not divert floods; the impact of the flood diversion operation of the flood storage and detention area on the flood discharge level of the river section is compared and analyzed, and the impact of the flood diversion operation of the flood storage and detention area is obtained.

3. The method for predicting the impact of flood storage and detention area operation on flood level in a river section according to claim 1 is characterized by: The factor set has a certain prediction accuracy for the target variable. f 1( S i ) Use the target variable to predict the value The symmetric uncertainty SU between the measured value y indicates that the factor set has a high prediction accuracy for the target variable. f 1( S i )The highest expression is as follows: （1）， （2）， （3）， Where, S i is the factor set, x i For each factor in the factor set, X is the set of candidate factors, ; H (‧) is the information entropy of the variable; M is the number of hidden layer nodes, Represents the randomly generated mth hidden layer node sigmoid activation function parameter, Represents the output weight to be calibrated; The factor set dimension f 2( S i ) i.e. the number of factors in the factor set | S i |, the dimension of the factor set f 2( S i )The lowest expression is as follows: （4）； The correlation between the factor set and the target variable f 3( S i ) Use factor centering for each factor x i ( x i ∈ S i ) and the target variable y The sum of the symmetric uncertainty of the factor set and the correlation between the target variable f 3( S i )The largest expression is as follows: （5）， Where, S i is the factor set, x i For each factor in the factor set, X is the set of candidate factors, , y is the target variable; The factor set redundancy f 4( S i ) is the sum of the symmetric uncertainties of each factor in the factor set, and the redundancy of the factor set is f 4( S i )The lowest expression is as follows: （6）。 4. The method for predicting the impact of flood storage and detention area operation on flood level in a river section according to claim 1 is characterized by: The construction process of the flood storage and detention area operation impact prediction model includes: Using various flood storage and detention area application samples and key factors affecting the flood level reduction effect as input factors, and the water level reduction amplitude of important control stations in each period during the forecast period as the target factor, the number of hidden layers and the form of transfer function are determined according to the general approximation theorem. The number of hidden neural networks is determined through repeated experiments. A two-layer feedforward network with a sigmoid function as the hidden layer neuron activation function and a linear function as the output layer neuron activation function is used to approximate the multidimensional mapping problem and determine the artificial neural network structure. The expression is as follows: （7）， Where, For the moment t + τ Predicted changes in control station water levels caused by flood discharge in the flood storage and detention area, , T For the foreseeable period, N is the number of hidden layer neurons with sigmoid function as activation function, For the moment t The input vector of , consists of the key influencing factors identified, is the artificial neural network parameter, the number is N×(2+D)+1, where D is the dimension of the input space, that is, the number of key influencing factors; The Levenberg-Marquardt training algorithm is used to optimize the parameters of the artificial neural network. In each training session, a training set, a smaller validation set, and a test set are set. The training set is used to calculate the gradient and update the network; the validation set is used to determine whether the prediction accuracy stops improving and when to stop training; the test set has no effect on the training and independently measures the performance of the trained network. When the error metric of the validation set is stable or the prediction error is acceptable, the training iteration is terminated, and the artificial neural network structure and corresponding parameter values ​​are determined, which is used as a prediction model for the impact of flood storage and detention area utilization.

5. A system for predicting the impact of flood storage and detention areas on flood levels in a river section, implementing the method of any one of claims 1 to 4, characterized in that: It includes a flood storage and detention area application sample generation module, a key factor identification and optimization module, a flood storage and detention area application impact prediction model construction module, and a flood storage and detention area application impact prediction module on flood level. The flood storage and detention area utilization sample generation module is used to generate a variety of flood storage and detention area utilization samples based on the historical typical flood sequence or artificially simulated flood sequence of the flood storage and detention area utilization and the corresponding data on the impact of the flood storage and detention area utilization on the flood level of the river channel in the river section; The key factor identification and optimization module is used to select a wrapper through the optimal information set based on multiple flood storage and detention area application samples, identify and optimize the key factors that affect the application effect of the flood storage and detention area; the identification and optimization method of the key factors that affect the application effect of the flood storage and detention area includes: based on multiple flood storage and detention area application samples, taking the flow of each upstream control station of the river section before the flood storage and detention area is put into use, the water level, the rise and fall rate of each downstream control station, and the flood diversion flow, flood diversion days, and impact period of the flood storage and detention area as possible influencing factors, to construct an alternative factor set; based on the alternative factor set and the data on the impact of the flood storage and detention area application on the flood level of the river channel of the river section, the optimal information set is used to select the wrapper, and the key factors that affect the application effect of the flood storage and detention area are optimized from the alternative factor set; the optimization objectives in the optimal information set selection wrapper include: the prediction accuracy of the factor set for the target variable f 1( S i ) Highest, factor set dimension f 2( S i ) Minimum, correlation between factor set and target variable f 3( S i ) Maximum, factor set redundancy f 4( S i ) is the lowest; the factor set has the lowest prediction accuracy for the target variable f 1( S i ) Maximum sum factor set dimension f 2( S i ) is the main optimization goal, and the correlation between the factor set and the target variable f 3( S i ) Maximum sum factor set redundancy f 4( S i ) is the lowest secondary optimization goal; in the process of selecting the optimal information set wrapper, the decision variables are set to 0 and 1 variables, representing the factors in each candidate factor set X Whether it is selected, the factors whose decision variables are all 1 in each non-dominated solution are the selected factors, forming the set , a non-dominated sorting genetic algorithm is used to optimize multiple non-dominated candidate factor solutions at one time to form a non-dominated solution set. Then, based on non-quantitative factors such as whether the observation data is easy to obtain and whether the observation is reliable, an appropriate solution is selected from the non-dominated solution set as the optimal key factor affecting the flood level reduction effect; The flood storage and detention area operation impact prediction model construction module is used to use a variety of flood storage and detention area operation samples and key factors affecting the flood storage and detention area operation effect as input factors, and the water level reduction amplitude of important control stations in each period during the forecast period as the target factor, and to train and verify based on an artificial neural network to construct a flood storage and detention area operation impact prediction model; The module for predicting the impact of the use of flood storage and detention areas on flood flow levels is used to use the flood storage and detention area use impact prediction model to predict the impact of the use of flood storage and detention areas on the flood flow levels of the river section under the water conditions and working conditions of each flood storage and detention area.

6. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method for predicting the impact of the flood storage area on the flood level of the river section as described in any one of claims 1 to 4 is implemented.

Images