Lake water level prediction method, system, equipment and program based on hybrid model

Through the combination of optimal variational modal decomposition and integrated learning XGBoost model, the data demand and modeling complexity problems in traditional lake water level prediction are solved, efficient and accurate prediction of lake water level is achieved, parameter optimization process is simplified, and the Predicted automation.

CN119005419APending Publication Date: 2024-11-22NANJING INST OF GEOGRAPHY & LIMNOLOGY
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411071384.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-06
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

Traditional physical process models have fine data requirements and modeling complexity in lake water level prediction, and rely on subjective experience for determining the number of decompositions based on signal decomposition methods. The parameters of artificial intelligence model are optimized in complex, making it difficult to achieve efficient and accurate lake water level prediction.

Method used

Optimal variational modal decomposition decomposes the historical water level time series into multiple modal components, and combines the integrated learning XGBoost model and Optuna framework for hyperparameter optimization, automatic parameter adjustment and model training, and performs through modal components and influencing factors. Water level forecast.

Benefits of technology

It realizes objective, fast and accurate prediction of lake water levels, simplifies the parameter optimization process, and improves the degree of automation and accuracy of prediction.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119005419A_ABST
    Figure CN119005419A_ABST
Patent Text Reader

Abstract

The invention discloses a lake water level prediction method, system, equipment and program based on a hybrid model, and the method comprises the steps: obtaining actually measured historical water level time sequences of different monitoring points of a lake, and decomposing the historical water level time sequence obtained by each monitoring point into a plurality of modal components through optimal variational modal decomposition; respectively constructing an integrated learning XGBoost model for each monitoring point for model training by taking the plurality of modal components and the additional parameters as input and taking a time sequence corresponding to each to-be-predicted modal component as output, and optimizing hyper-parameters of the XGBoost model by adopting an Optuna framework; and summing the XGBoost model prediction result corresponding to each mode to obtain the water level prediction result of the corresponding monitoring point. By means of the method, the lake water level can be objectively, rapidly and accurately predicted.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of lake hydrology and data processing technology, and specifically relates to a lake water level prediction method, system, equipment and program based on a hybrid model. Background Art

[0002] Lakes are important carriers of the earth's material and energy cycles and have extremely important ecological value. Water level is one of the important hydrological indicators of lakes. It can reflect the health of lake wetlands, characterize lake ecological services and other functions, and is a direct indicator for assessing the risk of lake floods and droughts. Unlike the fluctuations in ocean or estuary water levels that are disturbed by astronomical tides, changes in lake water levels are mainly disturbed by meteorological and hydrological factors. When traditional physical process models predict lake water levels, they usually require discrete solutions to the control equations. The required data are sophisticated and multi-source, and the modeling technology is demanding. There may be deficiencies in real-time and accurate prediction of lake water levels.

[0003] Against the backdrop of climate change, the risk of floods and droughts in lakes around the world has been increasing, making real-time and efficient lake water level prediction particularly important. The lake water level series itself can be regarded as a complex signal fluctuation containing multiple frequencies, which is suitable for signal decomposition methods. Its prediction mathematically belongs to regression, which is one of the classic application directions of artificial intelligence algorithms. With the development of the concept of big data, the signal decomposition method is used to split the time series into multiple sub-signals with more regular fluctuations, improve the predictability of the time series, and then couple it with the artificial intelligence model, which has become one of the powerful means to solve the regression problem. In view of the above advantages, the coupled signal decomposition and artificial intelligence methods are of great application value in lake water level prediction. However, when the lake water level signal is decomposed based on the signal decomposition method, the determination of the number of decompositions is still subjective and empirical, and the optimization of the model parameters of the artificial intelligence method also depends on the modeler to build a framework for calibration, which is relatively complicated. Therefore, it is necessary to find a method to more reasonably and objectively determine the number of lake water level signal decompositions based on signal decomposition, simplify the parameter optimization process of the artificial intelligence method, and achieve efficient prediction of lake water level. Summary of the invention

[0004] The first purpose of the present invention is to overcome the deficiencies in the above-mentioned prior art and provide a lake water level prediction method based on a hybrid model.

[0005] To achieve the above object, the present invention adopts the following technical solution:

[0006] A lake water level prediction method based on a hybrid model, comprising:

[0007] Obtain the measured historical water level time series at different monitoring points of the lake, and decompose the historical water level time series obtained at each monitoring point into multiple modal components through optimal variational mode decomposition;

[0008] Taking the multiple modal components and additional parameters as input and the time series corresponding to each modal component to be predicted as output, an integrated learning XGBoost model is constructed for each monitoring point for model training, and the XGBoost model hyperparameters are optimized using the Optuna framework; wherein the additional parameters are variables that affect or reflect changes in lake water levels;

[0009] The prediction results of the XGBoost model corresponding to each mode are summed to obtain the water level prediction results of the corresponding monitoring points.

[0010] As a preferred implementation, the additional parameters include historical water level fluctuation information of lake monitoring points and historical flow information affecting lake water level changes. The historical flow information affecting lake water level changes refers to the historical flow of rivers that have water exchange with lakes.

[0011] As a preferred implementation, the additional parameters are the historical water levels of each mode at the lake monitoring point L days before the prediction start time, and the historical flow of the river that has water flow exchange with the lake L days before the prediction start time.

[0012] As a preferred implementation, L=5-10.

[0013] As a preferred implementation, the acquired historical water level time series is decomposed into multiple modal components by optimal variational modal decomposition, and the number of decomposed modal components is determined based on the following method:

[0014] Preset different numbers of modal components 2, 3, n, ..., N and perform modal decomposition based on the preset numbers of modal components;

[0015] If the difference between the center frequency of the last modal component decomposed based on the current number of modal components n and the previous number of modal components n-1 is less than the preset value, n-1 is taken as the final number of modal components; 3≤n≤N.

[0016] As a preferred implementation, during model training, the data is divided into a training set and a test set according to a preset ratio, the training set data is used for model training, and after the model training is completed, the test set data is used for accuracy evaluation to determine the optimal combination of additional parameters.

[0017] A second object of the present invention is to provide a lake water level prediction system based on a hybrid model, comprising:

[0018] The data acquisition module obtains the measured historical water level time series of different monitoring points of the lake and the historical flow information of the rivers that have water exchange with the lake;

[0019] The modal decomposition module decomposes the historical water level time series obtained at each monitoring point into multiple modal components through optimal variational modal decomposition;

[0020] The water level prediction module takes multiple modal components and historical flow information of rivers that exchange water with lakes as input, and outputs the time series corresponding to each modal component to be predicted. An integrated learning XGBoost model is constructed for each monitoring point for model training, and the Optuna framework is used to optimize the XGBoost model hyperparameters. The prediction results of the XGBoost model corresponding to each mode are summed to obtain the water level prediction results of the corresponding monitoring point.

[0021] A third object of the present invention is to provide a computer device, comprising a memory, a processor and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the above method.

[0022] A fourth object of the present invention is to provide a computer-readable storage medium having a computer program / instruction stored thereon, which implements the steps of the above method when the computer program / instruction is executed by a processor.

[0023] A fifth object of the present invention is to provide a computer program product, comprising a computer program / instruction, which implements the steps of the above method when executed by a processor.

[0024] The present invention has the following beneficial effects:

[0025] (1) The present invention uses optimal variational modal decomposition to decompose the acquired historical water level time series into multiple modal components, objectively determines the number of decompositions of the lake water level signal, and uses the Optuna framework to automatically adjust the parameters of the machine learning model;

[0026] (2) The present invention combines the objectively decomposed modal components and the additional parameters that affect or reflect the lake water level fluctuations, and performs learning prediction based on the integrated learning XGBoost model. The algorithm is simple, efficient and has a high degree of automation, which can achieve objective, rapid and accurate prediction of lake water levels. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 is a flow chart of an embodiment of the present invention.

[0028] Figure 2 It is a schematic diagram of measured data in an embodiment of the present invention.

[0029] Figure 3 Schematic diagram of the results of the OVMD model at Hukou Station according to an embodiment of the present invention.

[0030] Figure 4Schematic diagram of different input variable combinations in an embodiment of the present invention.

[0031] Figure 5 It is a comparison chart of the prediction results and measured values ​​corresponding to different input variable combinations in the embodiment of the present invention. DETAILED DESCRIPTION

[0032] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present invention.

[0033] Example 1

[0034] The method flow of the present invention is as follows Figure 1 As shown, the following steps are included:

[0035] S1 obtains the measured historical water level time series at different monitoring points of the lake, and decomposes the historical water level time series obtained at each monitoring point into multiple modal components (IMFs) through optimal variational mode decomposition (OVMD);

[0036] The algorithm expression of the optimal variational mode decomposition (OVMD) model is:

[0037]

[0038] Where k represents the serial number of the signal decomposed by the optimal variational mode decomposition (OVMD) model; t represents time; u k (t) is the kth modal component, which is an AM / FM signal with simpler regularity than the initial signal; K is the number of modal components defined by the user, i.e., the number of OVMD decomposed signals.

[0039] The number of decomposed signals in the optimal variational mode decomposition (OVMD) model is determined based on the following method:

[0040] Set different numbers of modal components (K) and observe the center frequency of the last modal component. If the difference between the center frequency of the last modal component under the current number of modal components (K) and the previous number of modal components (K-1) is less than the preset threshold, stop enumeration and take the corresponding K-1 as the final number of modal components.

[0041] S2 takes the multiple modal components and additional parameters as input, and the time series corresponding to each modal component to be predicted as output, and constructs an integrated learning XGBoost model for each monitoring point for model training; wherein the additional parameters are variables that affect or reflect the change of lake water level, and in this embodiment, different variable combinations are selected for model training;

[0042] S3 divides the input and output data into training sets and test sets according to a certain ratio, trains the ensemble learning XGBoost model based on the training set data, and uses the Optuna framework to optimize the XGBoost model hyperparameters;

[0043] S4 obtains the final integrated learning XGBoost model based on the hyperparameter value optimized by the Optuna framework, and sums the prediction results of each modal component with the integrated learning XGBoost model as the prediction value of the lake water level by the OVMD-XGBoost-Optuna hybrid model;

[0044] The algorithm expression of the integrated learning XGBoost model is:

[0045]

[0046] In the formula, u I (t) is the modal component value of the OVMD model predicted by the ensemble learning XGBoost model; I and i are the number of decision tree models constructed by the ensemble learning XGBoost model and the corresponding decision tree model serial number; X is the input layer parameter matrix; M i The topological structure learned by the i-th decision tree model of the ensemble learning XGBoost model; i Represents the model output of the i-th decision tree.

[0047] The hyperparameters of the integrated learning XGBoost model are optimized and determined by using the open source hyperparameter optimization Optuan framework in the Python environment.

[0048] S5 evaluates the accuracy of the prediction results based on the test set data and identifies the optimal combination of input variables.

[0049] Example 2

[0050] This embodiment takes the water level prediction of Poyang Lake as an example for detailed explanation. Poyang Lake receives water from its five rivers (Xiushui River, Ganjiang River, Fuhe River, Xinjiang River, Raohe River), and its water is injected into the Yangtze River from the lake mouth waterway. Therefore, it is also affected by the water from the Yangtze River, and there is water exchange between it and the five rivers and the Yangtze River.

[0051] Figure 2 This is a schematic diagram of the measured data of this embodiment, in which the flow of the Yangtze River is reflected by the measured flow of the Hankou station, and the flow of the five rivers is reflected by the measured flow of Waizhou, Dufengkeng, Lijiadu, Hushan, Meigang and Wanjiabu.

[0052] Taking Hukou, Xingzi, Duchang and Tangyin stations as examples, the corresponding OVMD-XGBoost-Optuna hybrid models were trained and constructed respectively, and the water level prediction accuracy of the OVMD-XGBoost-Optuna hybrid model in different areas of Poyang Lake was tested.

[0053] like Figure 1 As shown, the method of this embodiment includes the following steps:

[0054] (1) Obtain the measured historical water level time series of each station (monitoring point) in Poyang Lake, and decompose the historical water level time series obtained at each station into multiple modal components through optimal variational mode decomposition.

[0055] When the OVMD method is used to perform modal decomposition of the water level at each station, the number of decomposed modes is determined based on the following method:

[0056] Set different numbers of modal components K, and observe the center frequency of the last modal component under different K values. If the center frequency of the last modal component under the current number of modal components K and the previous number of modal components K-1 is close, stop enumeration and take the corresponding previous number of modal components as the final value.

[0057] Taking the Hukou Station of Poyang Lake as an example, the center frequency values ​​of each mode when the OVMD model takes different mode numbers are shown in Table 1. When the mode number increases from 6 to 7, the difference in the center frequency of the last mode of the OVMD model under the two settings is 0.0002, which is very close and less than the preset threshold of 0.0005. Therefore, the final mode number of VMD at Hukou Station is 6, and the corresponding decomposition results are as follows Figure 3 The remaining sites also use the same method to determine the number of decomposed modes. In this embodiment, the number of decomposed modes at each site is finally 6.

[0058] Table 1 The central frequencies of each mode when the OVMD model takes different mode numbers (K) at the Hukou station of Poyang Lake

[0059]

[0060] (2) Taking the multiple modal components and additional parameters as input and the time series corresponding to each modal component to be predicted as output, an integrated learning XGBoost model is constructed for each monitoring point to perform model training.

[0061] Additional parameters are variables that affect or reflect changes in lake water levels. Different combinations of input variables may lead to different prediction accuracies. This embodiment provides different combinations of input variables, which are tested and optimized. The schematic diagram of different combinations of input variables is shown in FIG. Figure 4 shown.

[0062] Input variable combination 1 considers the external disturbance variables that affect the water level fluctuations. On the basis of the former, input variable combination 2 further considers the historical water level information of each mode of the station to be predicted. Taking Poyang Lake as an example, since it is affected by the flow of the Five Rivers and the Yangtze River at the same time, input variable combination 1 contains the flow information of the Five Rivers and the Yangtze River one week before the start of the prediction, which is used to predict the water level of each station in Poyang Lake in the next day. The historical information of the station water level has an important reference value for its subsequent changes under the influence of flow. Therefore, on the basis of input variable combination 1, input variable combination 2 further considers the fluctuation information of each modal component of OVMD one week before the start of the prediction.

[0063] (3) The input data is divided into a training set and a test set according to a preset ratio. In this embodiment, the data from 2003 to 2016 is divided into the training set data, and the data from 2017 to 2019 is divided into the test set data. The ratio of the training set to the test set data is about 8:2.

[0064] (4) The parameters of the ensemble learning XGBoost model are optimized based on the Optuna framework in the Python environment. Based on the hyperparameter values ​​of the ensemble learning XGBoost model optimized based on the Optuna framework, the XGBoost prediction model corresponding to each mode of each station in Poyang Lake can be trained to obtain the OVMD model. The sum of the prediction results of the XGBoost prediction model corresponding to each mode of the same station is the water level prediction result of the corresponding station.

[0065] (5) Based on the OVMD-XGBoost-Optuna hybrid model constructed above, the accuracy of the constructed prediction model was tested using the test set data.

[0066] The comparison between the predicted values ​​and measured values ​​of the OVMD-XGBoost-Optuna hybrid model under different input variable combinations is shown in Figure 5 The root mean square error and maximum absolute error statistics are shown in Table 2. In general, the predicted values ​​of the OVMD-XGBoost-Optuna hybrid model corresponding to input variable combination 2 are better in line with the measured values. The root mean square error range of this combination at each station in Poyang Lake is 0.06m~0.08m, which is the preferred input variable combination.

[0067] Table 2 Comparison of statistical accuracy of predicting water levels at each station in Poyang Lake using different combinations of input variables using the VMD-XGBoost-Optuna hybrid model

[0068]

[0069] In summary, this embodiment decomposes the complex water level fluctuation of Poyang Lake into 6 simpler and more regular water level fluctuation signals (modal components) based on the OVMD model, sets different input variable combinations, and further learns and predicts each modal component after decomposition based on the integrated learning XGBoost model. The sum of the predicted values ​​is the water level prediction result of Poyang Lake. By optimizing the input variable combination, the OVMD-XGBoost-Optuna hybrid model constructed in this embodiment realizes the accurate prediction of the water level of Poyang Lake.

Claims

1. A lake water level prediction method based on a hybrid model, characterized in that: include: Obtain the measured historical water level time series at different monitoring points of the lake, and decompose the historical water level time series obtained at each monitoring point into multiple modal components through optimal variational mode decomposition; Taking the multiple modal components and additional parameters as input and the time series corresponding to each modal component to be predicted as output, an integrated learning XGBoost model is constructed for each monitoring point for model training, and the XGBoost model hyperparameters are optimized using the Optuna framework; wherein the additional parameters are variables that affect or reflect changes in lake water levels; The prediction results of the XGBoost model corresponding to each mode are summed to obtain the water level prediction results of the corresponding monitoring points.

2. The method according to claim 1, characterized in that The additional parameters include historical water level fluctuation information of lake monitoring points and historical flow information of rivers that have water exchange with lakes.

3. The method according to claim 1 or 2, characterized in that: The additional parameter is the time before the prediction start time. L The historical water levels of each mode at the daytime lake monitoring point and the time before the forecast start L Daily historical flow of rivers that exchange water with lakes.

4. The method according to claim 3, characterized in that Said L =5~10.

5. The method according to claim 1, characterized in that In the process of decomposing the acquired historical water level time series into multiple modal components by optimal variational modal decomposition, the number of decomposed modal components is determined based on the following method: Preset different numbers of modal components 2 , 3 , n , …, N and performing modal decomposition based on a preset number of modal components; If based on the current number of modal components n and the number of the previous modal components n -1 If the difference of the center frequency of the last modal component obtained by decomposition is less than the preset threshold, then n -1 is the number of final modal components; 3≤ n ≤ N .

6. The method according to claim 1, characterized in that During model training, the data is divided into training set and test set according to a preset ratio. The training set data is used to train the model, and after the model training is completed, the test set data is used to evaluate the accuracy and determine the optimal combination of additional parameters.

7. A lake water level prediction system based on a hybrid model, characterized in that: include: The data acquisition module obtains the measured historical water level time series of different monitoring points of the lake and the historical flow information of the river monitoring points that have water exchange with the lake; The modal decomposition module decomposes the historical water level time series obtained at each monitoring point into multiple modal components through optimal variational modal decomposition; The water level prediction module takes multiple modal components and historical flow information of rivers that exchange water with lakes as input, and outputs the time series corresponding to each modal component to be predicted. An integrated learning XGBoost model is constructed for each monitoring point for model training, and the Optuna framework is used to optimize the XGBoost model hyperparameters. The prediction results of the XGBoost model corresponding to each mode are summed to obtain the water level prediction results of the corresponding monitoring point.

8. A computer device comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program / instruction stored thereon, characterized in that: When the computer program / instructions are executed by a processor, the steps of the method described in any one of claims 1 to 6 are implemented.

10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the method described in any one of claims 1 to 6 are implemented.