A reservoir water level prediction method and prediction system based on lead-lag relationship

Abstract

This invention provides a reservoir water level prediction method and system based on a lead-lag relationship, relating to the field of hydrological prediction. The method comprises: obtaining raw hydrological time series data (including water level, flow, etc.), and using maximum mutual information time-lag correlation analysis to obtain leading hydrological elements and their time periods, as well as lagging hydrological elements and their time periods; temporally aligning the leading hydrological elements according to the leading time periods to construct a leading dataset; temporally aligning the lagging hydrological elements according to the lagging time periods, and if any lagging hydrological elements are missing, using a hydrological element prediction model based on a single-model, single-output, multi-step rolling strategy to obtain predicted values ​​and fill them in, thereby constructing a lagging dataset; and predicting the reservoir water level using a data-driven reservoir water level prediction model based on the leading and lagging datasets. This invention considers both leading and lagging hydrological elements to achieve high-precision and timely prediction of reservoir water levels.

Description

Technical Field

[0001] The present invention relates to the field of hydrological prediction, and in particular to a reservoir water level prediction method and prediction system based on a lead-lag relationship. Background Art

[0002] Reservoir water level fluctuations directly impact key areas such as water resources management, flood control, and power generation efficiency. Therefore, reservoir water level forecasting is crucial for ensuring the safety of water conservancy projects, ensuring the rational use of water resources, and formulating flood prevention and disaster reduction measures. Reservoir water levels are influenced by a combination of meteorological and hydrological factors, including flow, rainfall, and evaporation. These factors exhibit significant time lags and exhibit complex nonlinear interactions, resulting in extremely complex patterns of water level fluctuations. In flood control and drought relief, as well as water resources management practices, errors in flood season water level forecasting can lead to erroneous flood prevention decisions, threatening people's lives and property. During water resource allocation, forecasting errors can lead to water waste or supply shortages, impacting the sustainable development of regional economies and societies.

[0003] Traditional single-factor water level prediction models rely solely on historical water level data. When the forecast horizon is long, their results often suffer from insufficient timeliness and large errors. Existing research faces bottlenecks in handling the complex relationships between multiple factors, making it difficult to accurately reveal the inherent connections between each factor and water level, thus failing to meet the practical needs of high-precision water level prediction. Therefore, a solution is urgently needed to address these issues. Summary of the Invention

[0004] The purpose of the present invention is to provide a reservoir water level prediction method and prediction system based on lead-lag relationship, which can improve the low accuracy and low timeliness of the prediction results of the existing water level prediction methods.

[0005] In a first aspect, the present invention provides a reservoir water level prediction method based on a lead-lag relationship, comprising:

[0006] Obtaining raw hydrological time series data , and the maximum mutual information time-lag correlation analysis is used to obtain the advanced hydrological elements , number of leading periods , delayed hydrological elements and the number of lag periods ; The original hydrological time series data Including reservoir water level time series data ;

[0007] According to the number of leading periods For the advanced hydrological elements Perform time alignment and build an advanced dataset ;

[0008] According to the number of hysteresis periods For the lagging hydrological elements Perform time alignment. If there are any lagged hydrological elements that have not been obtained, the hydrological element prediction model is used to obtain the predicted values ​​based on the single model single output multi-step rolling strategy to fill in and construct a lagged dataset. ;

[0009] Based on the advanced dataset and lagged datasets ,The reservoir water level prediction model based on data-driven method is used to predict the reservoir water level.

[0010] The present invention provides a reservoir water level prediction method based on lead-lag relationship. The method obtains original hydrological time series data and uses maximum mutual information time-lag correlation analysis to obtain leading hydrological elements and their time period numbers, and lagging hydrological elements and their time period numbers; then, the leading hydrological elements and the lagging hydrological elements are time-aligned; finally, based on the aligned hydrological elements, a reservoir water level prediction model based on a data-driven method is used to achieve high-precision and high-efficiency prediction of the reservoir water level.

[0011] Optionally, the maximum mutual information time-lag correlation analysis is used to obtain the advanced hydrological elements , number of leading periods , delayed hydrological elements and the number of lag periods When the original hydrological time series data is selected, the and the reservoir water level time series data Constructing a two-dimensional dataset , and use the grid division algorithm to calculate the corresponding maximum mutual information for different time lags, obtain the maximum information coefficient based on the maximum mutual information under different time lags, compare the maximum information coefficients corresponding to all time lags, and obtain the maximum value of the maximum information coefficient and its corresponding time lag; if the time lag is less than 0, then the original hydrological time series data For the advanced hydrological elements , the number of leading time periods is the current time lag; if the time lag is greater than 0, the current original hydrological time series data The hysteresis hydrological element , the number of lag periods is the current time lag.

[0012] Optionally, the advanced hydrological elements When performing time alignment, it includes: all the advanced hydrological elements Move the corresponding leading time periods backward in time sequence , get the advanced hydrological elements after time alignment :

[0013] ,

[0014] Each time-aligned advanced hydrological element Time series data of the reservoir water level Building an advanced dataset :

[0015] { D i ,t C } = [ Z 1 Z 2 ⋮ H i , 1 C ' H i , 2 C ' ⋮ Z T H i , T C ' ] ,

[0016] Merge all advanced datasets , obtain an advance dataset containing all time-aligned advance hydrological elements :

[0017] { D i ,t C ' } = [ Z 1 H 1 , 1 C ' H 2 , 1 C ' ⋯ H i , 1 C ' Z 2 H 1 , 2 C ' H 2 , 2 C ' ⋯ H i , 2 C ' ⋮ ⋮ ⋮ ⋮ Z T H 1 , T C ' H 2 , T C ' ⋯ H i , T C ' ] .

[0018] Optionally, the delayed hydrological elements When performing time alignment, it includes: all the lagged hydrological elements Move forward in chronological order the number of lag periods corresponding to each of them If there are any unobtained lagged hydrological elements, the hydrological element prediction model is used to obtain the predicted value according to the single model single output multi-step rolling strategy to fill the lagged hydrological elements, and each filled lagged hydrological element is Time series data of the reservoir water level Constructing a lagged dataset :

[0019] { D j ,t F } = [ Z 1 Z 2 ⋮ H j, 1 F ' H j, 2 F ' ⋮ Z T H j, 3 F ' ] ,

[0020] Merge all lagged datasets , obtain a lagged dataset containing all time-aligned lagged hydrological elements :

[0021] { D j ,t F ' } = [ Z 1 H 1 , 1 F ' H 2 , 1 F ' ⋯ H j , 1 F ' Z 2 H 1 , 2 F ' H 2 , 2 F ' ⋯ H j , 2 F ' ⋮ ⋮ ⋮ ⋮ Z T H 1 , T F ' H 2 , T F ' ⋯ H j , T F ' ] .

[0022] Optionally, when using a hydrological element prediction model to obtain a predicted value based on a single-model single-output multi-step rolling strategy, the following steps are included:

[0023] Step 1: The latest lagged hydrological element after time alignment The historical data of time steps are used as the input sequence, input into the hydrological element prediction model, and the predicted value of the first future time step is output; wherein, is the preset number of hysteresis periods;

[0024] Step 2: Remove the historical data of the earliest time step in the input sequence and add the predicted value of the first future time step to the end of the input sequence to form the updated input sequence;

[0025] Step 3: Input the updated input sequence into the hydrological element prediction model to obtain the predicted value of the next future time step;

[0026] Step 4. Repeat steps 2 to 3, removing the data of the earliest time step in the current input sequence in turn, and adding the latest predicted value to the end of the input sequence until the predicted values ​​of the required number of future time steps are obtained.

[0027] Optionally, the training method of the reservoir water level prediction model includes: defining an objective function, using the NSGA-II multi-objective optimization algorithm to optimize the parameters to be optimized and hyperparameters in the data-driven method to generate a Pareto optimal solution set; based on the TOPSIS multi-attribute decision-making method, calculating the closeness of each solution in the Pareto optimal solution set to the ideal solution and the negative ideal solution, sorting according to the closeness and selecting the optimal solution; and constructing a reservoir water level prediction model based on the optimal solution.

[0028] Optionally, the objective function includes: mean absolute error, root mean square error, mean square error, interquartile range correlation index, Nash-Sutcliffe efficiency coefficient, determination coefficient and maximum error.

[0029] In a second aspect, the present invention further provides a reservoir water level prediction system based on a lead-lag relationship, comprising:

[0030] Data acquisition module, to obtain raw hydrological time series data , and the maximum mutual information time-lag correlation analysis is used to obtain the advanced hydrological elements , number of leading periods , delayed hydrological elements and the number of lag periods ; The original hydrological time series data Including reservoir water level time series data ;

[0031] The advance alignment module, according to the number of the advance period For the advanced hydrological elements Perform time alignment and build an advanced dataset ;

[0032] The hysteresis alignment module is configured to align the hysteresis intervals according to the number of hysteresis intervals. For the lagging hydrological elements Perform time alignment. If there are any lagged hydrological elements that have not been obtained, the hydrological element prediction model is used to obtain the predicted values ​​based on the single model single output multi-step rolling strategy to fill in and construct a lagged dataset. ;

[0033] Water level prediction module, based on the advance data set and lagged datasets ,The reservoir water level prediction model based on data-driven method is used to predict the reservoir water level. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 A flow chart of a reservoir water level prediction method based on a lead-lag relationship provided by an embodiment of the present invention.

[0035] Figure 2 This is a structural diagram of a reservoir water level prediction system based on lead-lag relationship provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0036] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. Unless otherwise defined, the technical terms or scientific terms used herein should be the common meanings understood by people with ordinary skills in the field to which the present invention belongs.

[0037] See also Figure 1 The present invention provides a reservoir water level prediction method based on lead-lag relationship, comprising the following steps:

[0038] S1. Obtaining original hydrological time series data , and the maximum mutual information time-lag correlation analysis is used to obtain the advanced hydrological elements , number of leading periods , delayed hydrological elements and the number of lag periods ; The original hydrological time series data Including reservoir water level time series data ;

[0039] S2, according to the number of leading time periods For the advanced hydrological elements Perform time alignment and build an advanced dataset ;

[0040] S3, according to the number of hysteresis periods For the lagging hydrological elements Perform time alignment. If there are any lagged hydrological elements that have not been obtained, the hydrological element prediction model is used to obtain the predicted values ​​based on the single model single output multi-step rolling strategy to fill in and construct a lagged dataset. ;

[0041] S4. Based on the advanced dataset and lagged datasets ,The reservoir water level prediction model based on data-driven method is used to predict the reservoir water level.

[0042] In fact, the reservoir water level prediction method provided by the present invention obtains the original hydrological time series data and adopts the maximum mutual information number time lag correlation analysis to obtain the leading hydrological elements, the leading time period number, the lagging hydrological elements and the lagging time period number, and then performs time alignment on the hydrological elements according to the time period number. When aligning the lagging hydrological elements, there may be unacquired parts, and the hydrological element prediction model can be used to obtain the predicted value according to the single model single output multi-step rolling strategy for filling. Finally, a data set is constructed according to the aligned hydrological elements, and the reservoir water level prediction model based on the data-driven method is used to predict the reservoir water level. Therefore, it is possible to obtain a prediction result with high timeliness and low error, thereby achieving the purpose of high-precision prediction of the reservoir water level.

[0043] In some embodiments, in step S1, the original hydrological time series data and reservoir water level time series data are collected from historical hydrological data related to the Xijiang hub in the middle reaches of the Ganjiang River from December 2013 to August 2020. The historical hydrological data covers time series data such as outflow flow, Xijiang station flow, Dongbei station flow, Ji'an station flow, Xijiang station rainfall, Dongbei station rainfall, Ji'an station rainfall and Xijiang evaporation; among them, the original hydrological time series data include water level, flow, etc.

[0044] In some embodiments, when obtaining the original hydrological time series data in step S1, it is necessary to perform missing value correction and normalization processing on the original hydrological time series data.

[0045] Specifically, the linear interpolation method is used to correct missing values ​​of the original hydrological time series data. The formula is as follows:

[0046] ,

[0047] in, are the missing points to be interpolated, is the estimated value of the missing points to be interpolated, 、 are the two known data points before and after the missing point.

[0048] Specifically, the minimum-maximum normalization method is used to map the range of all original hydrological time series data to the [0,1] standardized interval. The formula is as follows:

[0049] ,

[0050] in, is the normalized value, is the original data, is the minimum value of the original data, is the maximum value of the original data.

[0051] In some embodiments, when the maximum mutual information number time lag correlation analysis is used in step S1 to obtain the leading hydrological elements, the number of leading time periods, the lagging hydrological elements, and the number of lagging time periods, a certain original hydrological time series data and the reservoir water level time series data are sequentially selected to construct a two-dimensional data set, and the corresponding maximum mutual information is calculated using a grid partitioning algorithm for different time lags. The maximum information coefficient is obtained based on the maximum mutual information under different time lags, and the maximum information coefficient corresponding to all time lags is compared to obtain the maximum value of the maximum information coefficient and its corresponding time lag; if the time lag is less than 0, the current original hydrological time series data is the leading hydrological element, and the number of leading time periods is the current time lag; if the time lag is greater than 0, the current original hydrological time series data is the lagging hydrological element, and the number of lagging time periods is the current time lag. The specific process is as follows:

[0052] S11, set the range of the number of leading time periods to [ - n ,-1 ] , the range of the number of lag periods is [ 1, m ] ; Select a certain original hydrological time series data , and combined with reservoir water level time series data Construct a sample set based on a given two-dimensional random variable , , .in, and is the maximum number of leading periods and the maximum number of lagging periods set in advance, is the time series length of the data;

[0053] S12. Time series data and Analyze and determine its maximum and minimum values ​​respectively. 、 In the range of values, according to the arithmetic progression, indivual, different values, so the sample space can be divided into List Gridding rows to generate totals A two-dimensional grid with grid cells. and The two-dimensional random variable sample set , count the number of sample data points falling into each grid, and calculate the number of sample data points falling into the first grid by these numbers. Column and The frequencies in the row grid are used to deduce the sample set of two-dimensional random variables exist 、 The marginal probability distribution on and , and the joint probability distribution ;

[0054] S13. According to the edge probability distribution and joint probability distribution obtained above, the mutual information index of the current partition is derived :

[0055] ,

[0056] Normalize the mutual information index to obtain the maximum mutual information :

[0057] ,

[0058] Two-dimensional random variable sample set The maximum information coefficient Defined as the normalized maximum mutual information under this partition:

[0059] ,

[0060] S14: Within the range of the set number of leading and lagging time periods, change the number of lag days in sequence ( exist [ - n ,-1 ] and [ 1, m ] For each time lag , the original hydrological time series data According to the time lag Perform time offset, reconstruct the sample set, repeat S12 to S13, and obtain the reservoir water level time series data With the original hydrological time series data of value;

[0061] S15. Compare all calculated value, find the maximum value among them . The corresponding time lag , which is the time lag between the hydrological element and the reservoir water level. , then the hydrological element is ahead of the reservoir water level, recorded as the number of leading periods ;like , then the hydrological element lags behind the reservoir water level, recorded as the lag period number The above operation is performed for each hydrological element to identify the leading hydrological elements with a time lag of less than 0 with the reservoir water level. and the number of leading periods , and lagged hydrological elements with a time lag greater than 0 and the number of lag periods .

[0062] In some embodiments, when time-aligning the leading hydrological elements in step S2, all leading hydrological elements are shifted backward in chronological order by the number of leading periods corresponding to each of them, and each time-aligned leading hydrological element is combined with the reservoir water level time series data to construct an advanced dataset. All leading datasets are then merged to obtain an advanced dataset containing all time-aligned leading hydrological elements. The specific process is as follows:

[0063] S21, record the reservoir water level time series data as , the advanced hydrological elements are And the corresponding number of leading periods is , the advanced hydrological elements after time alignment are recorded as ,in is the time series length of the data; for advanced hydrological elements , because it is ahead of the reservoir water level period, so the advanced hydrological elements need to be moved backward in time sequence The locations are as follows:

[0064] ,

[0065] S22, combined with time-aligned advanced hydrological elements and reservoir water level time series data Building an advanced dataset , you can Represented as a The matrix is ​​as follows:

[0066] { D i ,t C } = [ Z 1 Z 2 ⋮ H i , 1 C ' H i , 2 C ' ⋮ Z T H i , T C ' ] ,

[0067] S23, the advanced data set at this time Each row of data in represents the reservoir water level and the aligned advanced hydrological element data value at the same time point. Repeat S21 to S22 for the remaining advanced hydrological elements according to their corresponding number of advanced periods, and merge all the advanced data sets to obtain the advanced data set containing all the time-aligned advanced hydrological elements. ,as follows:

[0068] { D i ,t C ' } = [ Z 1 H 1 , 1 C ' H 2 , 1 C ' ⋯ H i , 1 C ' Z 2 H 1 , 2 C ' H 2 , 2 C ' ⋯ H i , 2 C ' ⋮ ⋮ ⋮ ⋮ Z T H 1 , T C ' H 2 , T C ' ⋯ H i , T C ' ] .

[0069] In some embodiments, when time-aligning the lagged hydrological elements in step S3, all lagged hydrological elements are shifted forward in chronological order by the number of lag periods corresponding to each of them. If there are any lagged hydrological elements that have not been obtained, the hydrological element prediction model is used to obtain predicted values ​​according to a single-model single-output multi-step rolling strategy to fill the lagged hydrological elements. Each filled lagged hydrological element is then combined with the reservoir water level time series data to construct a lagged dataset. All lagged datasets are merged to obtain a lagged dataset containing all time-aligned lagged hydrological elements. The specific process is as follows:

[0070] S31, record the reservoir water level time series data as , the lagging hydrological elements are And the corresponding lag period number is , the lagged hydrological elements after time alignment are recorded as ,in is the time series length of the data; for lagged hydrological elements , because it lags behind the reservoir water level time periods, so the delayed hydrological elements need to be moved forward in time sequence The locations are as follows:

[0071] ,

[0072] Among them, if If the lagged hydrological elements of the time period are not obtained, a single model single output multi-step rolling strategy is used to obtain the predicted value filling The single-model single-output multi-step rolling strategy is as follows:

[0073] S311. Lagged hydrological elements after time alignment The End time steps of historical data as the input sequence , input the input sequence into the trained hydrological element prediction model to obtain the predicted value of the first time step in the future , if the long short-term memory network LSTM is used to build a data-driven prediction model, it can be expressed as ;

[0074] S312, the predicted value Add to input sequence , replacing the input sequence The earliest historical data of the time step in the update sequence is formed ; If the original input sequence is , after adding the predicted value ;

[0075] S313, inputting the updated input sequence into the hydrological element prediction model to obtain the predicted value for the next future time step;

[0076] S314, repeat steps S312 to S313, add the latest predicted value to the input sequence to replace the data of its earliest time step, until the predicted values ​​of the required number of future time steps are obtained. Suppose the input sequence for the nth prediction is , input it into the hydrological element prediction model to obtain the predicted value of the nth future time step ,Right now: ;

[0077] S32, the filled lagged hydrological elements are combined with the reservoir water level time series data according to the same time sequence and structure requirements. Constructing a lagged dataset ,as follows:

[0078] { D j ,t F } = [ Z 1 Z 2 ⋮ H j, 1 F ' H j, 2 F ' ⋮ Z T H j, 3 F ' ] ,

[0079] S33, at this time Each row of data in represents the reservoir water level and the aligned lagged hydrological element data value at the same time point. Repeat S31 to S32 for the remaining lagged hydrological elements according to their corresponding lag period number and predicted value, and merge all lagged data sets to obtain a lagged data set containing all time-aligned lagged hydrological elements. ,as follows:

[0080] { D j ,t F ' } = [ Z 1 H 1 , 1 F ' H 2 , 1 F ' ⋯ H j , 1 F ' Z 2 H 1 , 2 F ' H 2 , 2 F ' ⋯ H j , 2 F ' ⋮ ⋮ ⋮ ⋮ Z T H 1 , T F ' H 2 , T F ' ⋯ H j , T F ' ] .

[0081] In some embodiments, the training method for the reservoir water level prediction model in step S4 includes: defining multiple objective functions for comprehensively evaluating the performance of the reservoir water level prediction model; optimizing the objective functions using the NSGA-II multi-objective optimization algorithm; and precisely defining the parameters and hyperparameters that need to be optimized in the data-driven approach. After NSGA-II optimization, a Pareto optimal solution set is generated. The TOPSIS multi-attribute decision-making method is used to comprehensively consider multiple attribute dimensions, calculate the closeness of each solution in the Pareto optimal solution set to the ideal solution and the negative ideal solution, rank the alternative solutions based on the closeness, and select the solution with the highest closeness as the optimal solution. Finally, a reservoir water level prediction model based on the data-driven approach is constructed based on the optimal solution. The objective functions may include mean absolute error (MAE), root mean square error (RMSE), mean square error (MSE), interquartile range correlation index (QR), Nash-Sutcliffe efficiency coefficient (NS), coefficient of determination (R²), and maximum error (MIE). Furthermore, hydrological element prediction models can also be trained by referring to this training method.

[0082] See also Figure 2 The present invention provides a reservoir water level prediction system based on lead-lag relationship, comprising the following steps:

[0083] Data acquisition module 100, obtains original hydrological time series data , and the maximum mutual information time-lag correlation analysis is used to obtain the advanced hydrological elements , number of leading periods , delayed hydrological elements and the number of lag periods ; The original hydrological time series data Including reservoir water level time series data ;

[0084] The advance alignment module 200, according to the number of the advance period For the advanced hydrological elements Perform time alignment and build an advanced dataset ;

[0085] The hysteresis alignment module 300 is configured to align the time intervals according to the hysteresis period. For the lagging hydrological elements Perform time alignment. If there are any lagged hydrological elements that have not been obtained, the hydrological element prediction model is used to obtain the predicted values ​​based on the single model single output multi-step rolling strategy to fill in and construct a lagged dataset. ;

[0086] The water level prediction module 400, based on the advance data set and lagged datasets ,The reservoir water level prediction model based on data-driven method is used to predict the reservoir water level.

[0087] While the embodiments of the present invention have been described in detail above, it will be apparent to those skilled in the art that various modifications and variations of these embodiments are possible. However, it should be understood that such modifications and variations are within the scope and spirit of the present invention as set forth in the claims. Furthermore, the invention described herein is susceptible to other embodiments and may be practiced or implemented in a variety of ways.

Claims

1. A reservoir water level prediction method based on lead-lag relationship, characterized in that: include: Obtaining raw hydrological time series data , and the maximum mutual information time-lag correlation analysis is used to obtain the advanced hydrological elements , number of leading periods , delayed hydrological elements and the number of lag periods When the original hydrological time series data is selected, the and the reservoir water level time series data Constructing a two-dimensional dataset , and use the grid division algorithm to calculate the corresponding maximum mutual information for different time lags, obtain the maximum information coefficient based on the maximum mutual information under different time lags, compare the maximum information coefficients corresponding to all time lags, and obtain the maximum value of the maximum information coefficient and its corresponding time lag; if the time lag is less than 0, then the original hydrological time series data For the advanced hydrological elements , the number of leading time periods is the current time lag; if the time lag is greater than 0, the current original hydrological time series data The hysteresis hydrological element , the number of lag periods is the current time lag; the original hydrological time series data Including reservoir water level time series data ; According to the number of leading periods For the advanced hydrological elements Perform time alignment and build an advanced dataset ; According to the number of hysteresis periods For the lagging hydrological elements Perform time alignment. If there are any lagged hydrological elements that have not been obtained, the hydrological element prediction model is used to obtain the predicted values ​​based on the single model single output multi-step rolling strategy to fill in and construct a lagged dataset. ; Based on the advanced dataset and lagged datasets ,The reservoir water level prediction model based on data-driven method is used to predict the reservoir water level.

2. The reservoir water level prediction method according to claim 1, characterized in that: For the advanced hydrological elements When performing time alignment, it includes: all the advanced hydrological elements Move the corresponding leading time periods backward in time sequence , get the advanced hydrological elements after time alignment : ， Each time-aligned advanced hydrological element Time series data of the reservoir water level Building an advanced dataset : ， Merge all advanced datasets , obtain an advance dataset containing all time-aligned advance hydrological elements : 。 3. The reservoir water level prediction method according to claim 1, characterized in that: For the lagging hydrological elements When performing time alignment, it includes: all the lagged hydrological elements Move forward in chronological order the number of lag periods corresponding to each of them If there are any unobtained lagged hydrological elements, the hydrological element prediction model is used to obtain the predicted value according to the single model single output multi-step rolling strategy to fill the lagged hydrological elements, and each filled lagged hydrological element is Time series data of the reservoir water level Constructing a lagged dataset : ， Merge all lagged datasets , obtain a lagged dataset containing all time-aligned lagged hydrological elements : 。 4. The reservoir water level prediction method according to claim 1, characterized in that: When using the hydrological element prediction model to obtain the predicted value based on a single model single output multi-step rolling strategy, it includes: Step 1: The latest lagged hydrological element after time alignment The historical data of time steps are used as the input sequence, input into the hydrological element prediction model, and the predicted value of the first future time step is output; wherein, is the preset number of hysteresis periods; Step 2: Remove the historical data of the earliest time step in the input sequence and add the predicted value of the first future time step to the end of the input sequence to form the updated input sequence; Step 3: Input the updated input sequence into the hydrological element prediction model to obtain the predicted value of the next future time step; Step 4. Repeat steps 2 to 3, removing the data of the earliest time step in the current input sequence in turn, and adding the latest predicted value to the end of the input sequence until the predicted values ​​of the required number of future time steps are obtained.

5. The reservoir water level prediction method according to claim 1, characterized in that: The training method of the reservoir water level prediction model includes: defining an objective function, optimizing the parameters to be optimized and hyperparameters in the data-driven method using the NSGA-II multi-objective optimization algorithm, and generating a Pareto optimal solution set; calculating the closeness of each solution in the Pareto optimal solution set to the ideal solution and the negative ideal solution based on the TOPSIS multi-attribute decision-making method, sorting and selecting the optimal solution according to the closeness; and constructing a reservoir water level prediction model based on the optimal solution.

6. The reservoir water level prediction method according to claim 5, characterized in that: The objective functions include mean absolute error, root mean square error, mean square error, interquartile range correlation index, Nash-Sutcliffe efficiency coefficient, determination coefficient and maximum error.

7. A reservoir water level prediction system based on lead-lag relationship, characterized in that: include: Data acquisition module, to obtain raw hydrological time series data , and the maximum mutual information time-lag correlation analysis is used to obtain the advanced hydrological elements , number of leading periods , delayed hydrological elements and the number of lag periods When the original hydrological time series data is selected, the and the reservoir water level time series data Constructing a two-dimensional dataset , and use the grid division algorithm to calculate the corresponding maximum mutual information for different time lags, obtain the maximum information coefficient based on the maximum mutual information under different time lags, compare the maximum information coefficients corresponding to all time lags, and obtain the maximum value of the maximum information coefficient and its corresponding time lag; if the time lag is less than 0, then the original hydrological time series data For the advanced hydrological elements , the number of leading time periods is the current time lag; if the time lag is greater than 0, the current original hydrological time series data The hysteresis hydrological element , the number of lag periods is the current time lag; the original hydrological time series data Including reservoir water level time series data ; The advance alignment module, according to the number of the advance period For the advanced hydrological elements Perform time alignment and build an advanced dataset ; The hysteresis alignment module is configured to align the hysteresis intervals according to the number of hysteresis intervals. For the lagging hydrological elements Perform time alignment. If there are any lagged hydrological elements that have not been obtained, the hydrological element prediction model is used to obtain the predicted values ​​based on the single model single output multi-step rolling strategy to fill in and construct a lagged dataset. ; Water level prediction module, based on the advance data set and lagged datasets ,The reservoir water level prediction model based on data-driven method is used to predict the reservoir water level.

Images