Underdam water level artificial neural network prediction method considering downstream multi-branch jacking effect

Abstract

The embodiment of the invention provides an under-dam water level artificial neural network prediction method considering a downstream multi-branch jacking effect, and relates to the technical field of water level prediction calculation, and the method comprises the steps: building a data set based on the historical operation data of a target hydro-junction and the historical hydrological data of downstream branches; decomposing data in the data set by using an empirical mode decomposition method based on time-varying filtering to obtain a supervised data set; based on the supervised data set, extracting features by using an external attention mechanism to obtain a feature data set; constructing and training based on the feature data set to obtain a GRU-KAN hybrid artificial neural network model; and on the basis of the GRU-KAN hybrid artificial neural network model and the real-time data, predicting the water level under the dam under the conditions of different discharge flows of the hydro-junction and different convergence flows of the downstream branch. According to the method, the problem of predicting the water level under the dam under the jacking action of multiple downstream branches is solved.

Description

Technical Field

[0001] The present application relates to the technical field of water level prediction calculation, and specifically, to an artificial neural network prediction method for the water level downstream of a dam considering the backwater effect of multiple downstream tributaries. Background Art

[0002] The change of the water level downstream of a water conservancy project will directly affect the navigation safety of ships. The water level downstream of the project and the discharge flow have obvious regularity under natural conditions. However, when there are large tributaries flowing into the downstream of the dam site, there is a close hydraulic connection between the main stream and the tributaries, and the water level downstream of the dam site is often affected by the backwater effect of the tributary rising water. Especially in the southwestern region of China, the river channels are mostly long and narrow, and the water level is extremely sensitive to the change of flow. A change in flow of several hundred cubic meters per second may cause a greater impact. During the real-time scheduling process of the project, the sudden change of the flow of the tributaries flowing into the downstream is likely to cause the amplitude change of the water level downstream of the dam to exceed the control requirements of the hourly and daily amplitude changes, thus affecting the navigation safety. Therefore, accurately predicting the water level downstream of the project considering the backwater effect of multiple downstream tributaries is an urgent problem to be solved currently.

[0003] Therefore, it is necessary to propose an artificial neural network prediction method for the water level downstream of a dam considering the backwater effect of multiple downstream tributaries to effectively ensure the safe, efficient and stable operation of the navigation structures. Summary of the Invention

[0004] An embodiment of the present application provides an artificial neural network prediction method for the water level downstream of a dam considering the backwater effect of multiple downstream tributaries. This method uses an empirical mode decomposition method based on time-varying filtering, an external attention mechanism, and a GRU-KAN hybrid network to construct a coupling model to solve the problem of predicting the water level downstream of the dam under the backwater effect of multiple downstream tributaries.

[0005] Other features and advantages of the present application will become apparent through the following detailed description, or will be partially learned through the practice of the present application.

[0006] According to the first aspect of the embodiments of the present application, an artificial neural network prediction method for the water level downstream of a dam considering the backwater effect of multiple downstream tributaries is provided, including:

[0007] Establishing a data set based on the historical operation data of the target water conservancy project and the historical hydrological data of the downstream tributaries;

[0008] Decomposing the data in the data set using an empirical mode decomposition method based on time-varying filtering to obtain a supervised data set;

[0009] Based on the supervised data set, using an external attention mechanism to extract features to obtain a feature data set;

[0010] Construct and train a GRU-KAN hybrid artificial neural network model based on the feature dataset;

[0011] Predict the downstream water level under different discharge rates of the water conservancy project and different confluence flows of the downstream tributaries based on the GRU-KAN hybrid artificial neural network model and real-time data.

[0012] In some embodiments of the present application, based on the foregoing solution, establish a dataset based on the historical operation data of the target water conservancy project and the historical hydrological data of the downstream tributaries, including:

[0013] Select the data that has a significant impact on the downstream water level from the historical operation data of the target water conservancy project and the historical hydrological data of the downstream tributaries as input features, and construct an input dataset;

[0014] Select the downstream water level as the output feature to obtain an output dataset;

[0015] Clean and integrate the data in the input dataset and the output dataset, delete outliers and interpolate data, and merge and construct the dataset.

[0016] In some embodiments of the present application, based on the foregoing solution, the data that has a significant impact on the downstream water level includes: the discharge rate of the project, the confluence flow of the downstream tributary, and the historical downstream water level.

[0017] In some embodiments of the present application, based on the foregoing solution, decompose the data in the dataset using the empirical mode decomposition method based on time-varying filtering to obtain a supervised dataset, including:

[0018] Perform maximum-minimum normalization processing on each input feature and output feature;

[0019] Use the empirical mode decomposition method based on time-varying filtering to decompose each input feature into multiple high-frequency components and low-frequency components to obtain a new input dataset;

[0020] Combine the new input dataset and the output dataset to form a new dataset, and convert it into a supervised dataset in input-output form, where the supervised dataset includes an input matrix composed of time steps and input features and an output matrix composed of time steps and output features.

[0021] In some embodiments of the present application, based on the foregoing solution, based on the supervised dataset, use an external attention mechanism to extract features to obtain a feature dataset, including:

[0022] Use the external attention mechanism to calculate the weights of each input feature value at each time step of the input matrix to obtain a weight matrix with the same dimension as the input matrix;

[0023] Perform the Hadamard product of the weight matrix and the input matrix to obtain a feature data set that serves as the input matrix of the GRU-KAN hybrid model, where the feature data set includes a training set and a test set.

[0024] In some embodiments of the present application, based on the foregoing solution, the GRU-KAN hybrid artificial neural network model constructed and trained based on the feature data set includes:

[0025] Use Kaiming initialization to determine the initial parameters of the model, use the training set data to obtain the model output through forward propagation, use the error backpropagation algorithm and the Adam optimization algorithm to adjust the GRU network parameters, use the LBFGS algorithm to adjust the KAN network parameters and the weights on each edge;

[0026] After multiple rounds of training, after the loss function value on the test set converges basically, determine the model structure and parameters.

[0027] In some embodiments of the present application, based on the foregoing solution, predicting the downstream water level under different discharge flows of the water conservancy project and different inflow flows of the downstream tributaries based on the GRU-KAN hybrid artificial neural network model includes:

[0028] Input the real-time data into the GRU-KAN hybrid artificial neural network model for prediction, and at set time intervals, add the new data generated after the previous model training to the training set for iterative training to obtain a new model. Compare the performance of the new and old models. If the new model has better performance, use the new model to replace the currently used old model for prediction.

[0029] The technical solution of the present application uses an empirical mode decomposition method based on time-varying filtering, an external attention mechanism, and a GRU-KAN hybrid network to construct a coupling model, which solves the problem of predicting the downstream water level under the backwater effect of multiple downstream tributaries.

[0030] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] The accompanying drawings herein are incorporated into the specification and constitute a part of the specification, showing embodiments consistent with the present application, and are used together with the specification to explain the principles of the present application. Obviously, the accompanying drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts. In the drawings:

[0032] Figure 1Shows a schematic flow chart of an artificial neural network prediction method for the water level downstream of a dam considering the backwater effect of multiple downstream tributaries according to an embodiment of the present application;

[0033] Figure 2 Shows a schematic diagram of the upstream and downstream relationship of Water Conservancy Project A according to an embodiment of the present application;

[0034] Figure 3 Shows a schematic diagram of the format of a data set according to an embodiment of the present application;

[0035] Figure 4 Shows a schematic diagram of the format of a data set after normalization processing according to an embodiment of the present application;

[0036] Figure 5 Shows a schematic diagram of a decomposition result according to an embodiment of the present application. Detailed implementation manners

[0037] Now, example embodiments will be described more fully with reference to the accompanying drawings. However, the example embodiments can be implemented in various forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that this application will be more complete and comprehensive, and will fully convey the concept of the example embodiments to those skilled in the art.

[0038] In addition, the described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. In the following description, numerous specific details are provided to give a thorough understanding of the embodiments of this application. However, those skilled in the art will realize that the technical solutions of this application can be practiced without one or more of the specific details, or other methods, components, devices, steps, etc. can be adopted. In other cases, well-known methods, devices, implementations, or operations are not shown or described in detail to avoid obscuring aspects of this application.

[0039] The block diagrams shown in the drawings are only functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software form, or in one or more hardware modules or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices.

[0040] The flowcharts shown in the drawings are only illustrative and do not necessarily include all the content and operations / steps, nor do they necessarily have to be executed in the described order. For example, some operations / steps can be decomposed, and some operations / steps can be combined or partially combined, so the actual execution order may change according to the actual situation.

[0041] To make the objectives, technical solutions and advantages of the present invention more clear, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0042] The following will describe in detail some embodiments of the present application with reference to the accompanying drawings. Without conflict, the embodiments described below and the features in the embodiments may be combined with each other.

[0043] See Figure 1 , which shows a schematic flowchart of a method for predicting the water level downstream of a dam considering the backwater effect of multiple downstream tributaries according to an embodiment of the present application.

[0044] As Figure 1 shown, a method for predicting the water level downstream of a dam considering the backwater effect of multiple downstream tributaries is presented, which specifically includes steps S100 to S500.

[0045] Refer to Figure 1 , step S100, establish a data set based on the historical operation data of the target water conservancy project and the historical hydrological data of the downstream tributaries.

[0046] It should be noted that the data sampling intervals in the data set should be consistent and not change with time, and there should be no null values.

[0047] In some feasible embodiments, based on the foregoing solution, the establishment of the data set based on the historical operation data of the target water conservancy project and the historical hydrological data of the downstream tributaries includes:

[0048] Select the data that has a significant impact on the water level downstream of the dam from the historical operation data of the target water conservancy project and the historical hydrological data of the downstream tributaries as input features, and construct an input data set;

[0049] Select the water level downstream of the dam as the output feature to obtain an output data set;

[0050] Clean and integrate the data in the input data set and the output data set, delete outliers and interpolate data, and combine and construct the data set.

[0051] It should be noted that cleaning, integrating, deleting anomalies, and interpolating the input dataset and the output dataset can ensure that the historical operation data of the target water conservancy project and the historical hydrological data of the downstream tributaries have a consistent and unchanged sampling interval in the time dimension. The format of the organized dataset should be [timesteps; input features], where timesteps represents the time step and input features represents the input features.

[0052] Exemplarily, in this embodiment, the data that significantly affects the downstream water level includes: the discharge from the project, the inflow from the downstream tributaries, and the historical downstream water level, etc.

[0053] Continue to refer to Figure 1 , step S200, decompose the data in the dataset using the empirical mode decomposition method based on time-varying filtering to obtain a supervised dataset.

[0054] It can be understood that in this embodiment, each feature sequence in the input dataset can be decomposed into multiple high-frequency components (LHFs) and one low-frequency component (LLF).

[0055] In some feasible embodiments, based on the foregoing solution, the using the empirical mode decomposition method based on time-varying filtering to decompose the data in the dataset to obtain a supervised dataset includes:

[0056] Perform maximum-minimum normalization processing on each input feature and output feature;

[0057] Use the empirical mode decomposition method based on time-varying filtering to decompose each input feature into multiple high-frequency components and low-frequency components to obtain a new input dataset;

[0058] Combine the new input dataset and the output dataset to form a new dataset, and convert it into a supervised dataset in the input-output form, where the supervised dataset includes an input matrix composed of time steps and input features and an output matrix composed of time steps and output features.

[0059] It can be understood that in this embodiment, the obtained input matrix is a two-dimensional matrix [timesteps; input features], where timesteps represents the time step and input features represents the input features; the obtained output matrix is also a two-dimensional matrix [timesteps; output features], where output features represents the output features.

[0060] Continue to refer to Figure 1, step S300, based on the supervised data set, use an external attention mechanism to extract features and obtain a feature data set.

[0061] It can be understood that using the external attention mechanism for feature extraction processing can assign weights to each eigenvalue at each moment in each sample of the input matrix.

[0062] In some feasible embodiments, based on the foregoing solution, the step of based on the supervised data set, using an external attention mechanism to extract features and obtain a feature data set includes:

[0063] Use an external attention mechanism to calculate the weights of each input eigenvalue at each time step of the input matrix, and obtain a weight matrix with the same dimension as the input matrix;

[0064] Perform a Hadamard product of the weight matrix and the input matrix to obtain a feature data set that is the input matrix of the GRU-KAN hybrid model, where the feature data set includes a training set and a test set.

[0065] Continue to refer to Figure 1 , step S400, based on the feature data set, construct and train a GRU-KAN hybrid artificial neural network model.

[0066] In some feasible embodiments, based on the foregoing solution, the constructing the GRU-KAN hybrid artificial neural network model includes:

[0067] Use Kaiming initialization to determine the initial parameters of the model, use the training set data to obtain the model output through forward propagation, use the error backpropagation algorithm and the Adam optimization algorithm to adjust the GRU network parameters, and use the LBFGS algorithm to adjust the KAN network parameters and the weights on each edge;

[0068] After multiple rounds of training, when the loss function value on the test set converges basically, determine the model structure and parameters.

[0069] Continue to refer to Figure 1 , step S500, based on the GRU-KAN hybrid artificial neural network model and real-time data, predict the downstream water level under different discharge flows of the water conservancy project and different confluence flows of the downstream tributaries.

[0070] In some feasible embodiments, based on the foregoing solution, the predicting the downstream water level under different discharge flows of the water conservancy project and different confluence flows of the downstream tributaries based on the GRU-KAN hybrid artificial neural network model and real-time data includes:

[0071] Input the real-time data into the GRU-KAN hybrid artificial neural network model for prediction, and at set time intervals, add the new data generated after the previous model training to the training set for iterative training to obtain a new model. Compare the performance of the new and old models. If the new model has better performance, replace the currently used old model with the new model for prediction.

[0072] The following provides a specific example of the method in the actual application process.

[0073] Select a certain A water control project in the upper reaches of a certain river as an example. Use the data of the discharge of the water control project, the inflow of the downstream tributaries, and the rainfall from January 2022 to August 2024 as the historical data set. The time step for water level prediction is 1 hour. There is tributary 1 joining about 1.9 km downstream and tributary 2 joining about 28 km downstream, resulting in a complex relationship between water level and flow downstream of the A water control project and great difficulty in regulation. Since the A water control project has navigation structures, to ensure the shipping safety downstream of the A water control project, it is necessary to accurately predict the water level downstream of the A water control project. The upstream and downstream relationship of the A water control project is as Figure 2 shown.

[0074] The specific steps are as follows:

[0075] S1: Select the discharge of the A water control project, the flow of tributary 1, the flow of tributary 2, and the historical downstream water level as the input features of the model, and select the downstream water level of the A water control project as the output feature;

[0076] S2: Clean and integrate the data, delete obvious outliers, and interpolate the data to ensure that the sampling intervals of the discharge of the A water control project, the flow of tributary 1, the flow of tributary 2, and the historical downstream water level data are consistent and unchanged in the time dimension, and organize to obtain the data set format for input to step S3 as Figure 3 shown;

[0077] S3: Perform min-max normalization processing on each input feature and output feature, and organize to obtain the data set format for input to step S4 as Figure 4 shown;

[0078] S4: Use TVFEMD to decompose the sequences of the discharge of the A water control project (QA), the flow of tributary 1 (Q1), the flow of tributary 2 (Q2), and the historical downstream water level (H) into 6 high-frequency components (LHFs) and 1 low-frequency component (LLF) respectively, to obtain a new input data set. Taking the discharge of the A water control project as an example, the decomposition result is as Figure 5 shown;

[0079] S5: Combine the decomposition results of each input feature and the downstream historical water level sequence to form a new data set. Use 80% of the data in the data set as the training set and 20% of the data as the test set. With a time step of 1 hour, input 7 days of input feature data and output the water level prediction results for the next 1 day, converting it into a supervised data set in the input-output form. Finally, the input matrix of the model is a two-dimensional matrix [168; 28], and the output matrix is a two-dimensional matrix [24; 1].

[0080] S6: Use the external attention mechanism to calculate the weights of each frequency component at each time step of the input matrix to obtain a weight matrix with the same dimension as the input matrix;

[0081] S7: Perform the Hadamard product of the weight matrix and the input matrix to obtain the input matrix of the GRU-KAN hybrid model;

[0082] S8: Use Kaiming initialization to determine the initial parameters of the GRU-KAN hybrid model. Use the input matrix data on the training set to obtain the model output through forward propagation. Use the error backpropagation algorithm and the Adam optimization algorithm to adjust the GRU network parameters. Use the LBFGS algorithm to adjust the KAN network parameters and the weights on each edge. Use the test set data to calculate the loss function value and evaluate the model prediction effect;

[0083] S9: After multiple rounds of training, when the loss function value on the test set converges basically, determine the model structure and parameters;

[0084] S10: Use the real-time input of the discharge of Hydraulic Project A, the flow of Tributary 1, the flow of Tributary 2, and the downstream historical water level sequence in the past 7 days to perform rolling prediction on the water level downstream of the dam. The prediction error is as Figure 3 shown;

[0085] S11: Every once in a while, add the recently collected data of the discharge of Hydraulic Project A, the flow of Tributary 1, the flow of Tributary 2, and the downstream water level sequence after the last model training to the data set, and continue to train on the basis of the existing model to obtain a new model. Compare the performance of the new model with that of the existing model. If the performance of the new model is better, then use the new model to replace the currently used model for prediction.

[0086] Therefore, the present invention adopts the above-mentioned artificial neural network prediction method for the water level downstream of the dam considering the backwater effect of multiple downstream tributaries, and uses the empirical mode decomposition method based on time-varying filtering, the external attention mechanism, and the GRU-KAN hybrid network to construct a coupling model, which solves the problem of predicting the water level downstream of the dam under the backwater effect of multiple downstream tributaries.

[0087] Other embodiments of the present application will be readily contemplated by those skilled in the art after considering the specification and practicing the embodiments disclosed herein. The present application is intended to cover any variations, uses, or adaptations of the present application, which follow the general principles of the present application and include common general knowledge or conventional technical means in the technical field not disclosed in the present application. It should be understood that the present application is not limited to the exact structures described above and shown in the drawings, and various modifications and changes can be made without departing from its scope. The scope of the present application is only limited by the appended claims.

Claims

1. An artificial neural network prediction method for the downstream water level considering the backwater effect of multiple tributaries, characterized in that, Including: Establish a data set based on the historical operation data of the target water conservancy project and the historical hydrological data of the downstream tributaries; Decompose the data in the data set using the empirical mode decomposition method based on time-varying filtering to obtain a supervised data set; Based on the supervised data set, use an external attention mechanism to extract features to obtain a feature data set; Construct and train a GRU-KAN hybrid artificial neural network model based on the feature data set; Based on the GRU-KAN hybrid artificial neural network model and real-time data, predict the water level downstream of the dam under different discharge rates of the water conservancy project and different inflow rates of the downstream tributaries.

2. The method according to claim 1, wherein The establishment of the data set based on the historical operation data of the target water conservancy project and the historical hydrological data of the downstream tributaries includes: Select the data that has a significant impact on the water level downstream of the dam from the historical operation data of the target water conservancy project and the historical hydrological data of the downstream tributaries as input features, and construct an input data set; Select the water level downstream of the dam as the output feature to obtain an output data set; Clean and integrate the data in the input data set and the output data set, delete outliers and interpolate data, and merge and construct the data set.

3. The method according to claim 2, wherein The data that has a significant impact on the water level downstream of the dam includes: the discharge rate of the project, the inflow rate of the downstream tributary, and the historical water level downstream.

4. The method according to claim 2, wherein The decomposition of the data in the data set using the empirical mode decomposition method based on time-varying filtering to obtain a supervised data set includes: Perform maximum-minimum normalization processing on each input feature and output feature; Use the empirical mode decomposition method based on time-varying filtering to decompose each input feature into multiple high-frequency components and low-frequency components to obtain a new input data set; Combine the new input data set and the output data set into a new data set and convert it into a supervised data set in input-output form, where the supervised data set includes an input matrix composed of time steps and input features and an output matrix composed of time steps and output features.

5. The method according to claim 4, wherein Based on the supervised data set, use an external attention mechanism to extract features To obtain a feature data set, including: Use an external attention mechanism to calculate the weights of each input feature value at each time step of the input matrix to obtain a weight matrix with the same dimension as the input matrix; Perform a Hadamard product on the weight matrix and the input matrix to obtain a feature data set that serves as the input matrix of the GRU-KAN hybrid model, where the feature data set includes a training set and a test set.

6. The method according to claim 5, wherein The construction and training of the GRU-KAN hybrid artificial neural network model based on the feature data set includes: Use Kaiming initialization to determine the initial parameters of the model, use the training set data to obtain the model output through forward propagation, use the error backpropagation algorithm and the Adam optimization algorithm to adjust the GRU network parameters, and use the LBFGS algorithm to adjust the KAN network parameters and the weights on each edge; After multiple rounds of training, when the loss function value on the test set converges basically, determine the model structure and parameters.

7. The method according to claim 6, characterized in that The prediction of the water level downstream of the dam under different discharge rates of the water conservancy project and different inflow rates of the downstream tributaries based on the GRU-KAN hybrid artificial neural network model and real-time data includes: Input the real-time data into the GRU-KAN hybrid artificial neural network model for prediction, and at regular intervals, add the new data generated after the previous model training to the training set for iterative training to obtain a new model. Compare the performance of the new and old models. If the performance of the new model is better, replace the currently used old model with the new model for prediction.

Images