CN109376937B - Self-adaptive scheduling end-of-term water level prediction method based on ensemble empirical mode decomposition - Google Patents

Abstract

The invention discloses a scheduling end-of-term water level prediction method based on adaptive ensemble empirical mode decomposition, and relates to the technical field of reservoir operation scheduling and water conservancy informatization. The method fully considers the unsteadiness of the scheduling end-of-term water level sequence, converts the scheduling end-of-term water level into a plurality of groups of steady-state sequences by using a method of ensemble empirical mode decomposition, realizes the stabilization of the hydrological sequences, and provides the most basic data conditions for the conventional forecasting method. In addition, when the self-adaptive forecasting model provided by the invention is used for forecasting, the self-adaptive forecasting model is a rolling forecasting operation, so that the model is corrected in real time, the adaptability of the model is ensured, and the model foundation is ensured for accurate forecasting. Therefore, when the method provided by the invention is adopted, the established prediction model is used for predicting the water level at the end of the scheduling period, and the method has higher accuracy.

Description

Prediction method of end-of-period water level for adaptive scheduling based on ensemble empirical mode decomposition

technical field

The invention relates to the technical field of reservoir operation scheduling and water conservancy informatization, in particular to a method for predicting the water level at the end of the scheduling period based on adaptive ensemble empirical mode decomposition.

Background technique

The water level at the end of the operation period is an important part of the reservoir operation, and the prediction of the water level at the end of the operation period is of great significance to the residual benefit and risk assessment of the reservoir operation. The water level at the end of the reservoir operation period has trend, periodicity and randomness, showing an unsteady sequence. At the same time, the water level is affected by many factors such as climate variables, the amount of water inflow, the conditions of the underlying surface, and the dispatching method of the dispatching decision maker. Deterministic and uncertain multi-layer information, so to accurately predict the water level, it is necessary to conduct in-depth excavation of various influencing factors.

At present, there are few methods for predicting the water level at the end of the dispatch period, mainly through forecasting runoff and dispatching rules to estimate the water level at the end of the dispatch period, or through different intelligent optimization algorithms such as artificial neural network, support vector machine, linear regression, etc. for autoregressive model prediction. water level. However, these intelligent optimization algorithms are a black-box model, and it is difficult to identify the unsteady state of the water level. In addition, because these intelligent optimization algorithms ignore the hydrological physical mechanism in the water level process, the autocorrelation analysis of the water level will lead to over-prediction. Fitting, there is a distortion phenomenon.

SUMMARY OF THE INVENTION

The purpose of the present invention is to provide a method for predicting the water level at the end of the dispatch period based on the adaptive ensemble empirical mode decomposition, so as to solve the aforementioned problems existing in the prior art.

In order to achieve the above object, the technical scheme adopted in the present invention is as follows:

An adaptive scheduling end-of-period water level prediction method based on ensemble empirical mode decomposition, comprising the following steps:

S1, data preprocessing: extract a sequence of water levels at the end of the scheduling period with a length of L continuous multi-scheduling periods as the independent variable L _S , use the initial water level L ₀ at the early stage of scheduling, the water level L t at time _t and the inflow water volume W during the remaining period at time t Form the set of dependent variables X=[L ₀ , L _t , W];

S2, the steady-state decomposition of the independent variable: the independent variable L _S sequence is decomposed by the method of ensemble empirical mode decomposition to generate multiple sets of steady-state eigenmode functions IMF(n) and a set of residual sequences Res, which together form the self- Variable combination Y=[IMF(n), Res];

S3, determine the prediction factor: use the correlation between the dependent variable set X=[L ₀ , L _t , W] and the independent variable combination Y to determine the prediction decision factor, and implement it according to the following steps:

S301, construct a set of candidate predictors: combine the three influencing factor sequences L ₀ , L _t and W in the dependent variable set X=[L ₀ , L _t , W] in separate, pairwise, and common ways, respectively , constitute seven groups of candidate predictor sets, respectively F ₁ =[L ₀ ], F ₂ =[L _t ], F ₃ =[W], F ₄ =[L ₀ ,L _t ], F ₅ =[L ₀ ,W], F ₆ =[L _t ,W], F ₇ =[L ₀ ,L _t ,W];

S302, performing correlation analysis on each group of data in the seven groups of candidate predictor sets and the independent variable combination Y=[IMF(n), Res] respectively, to obtain the correlation factors;

S303, according to the correlation factor obtained in S302, select the candidate predictor with the largest correlation with each group of sequences in Y=[IMF(n), Res] as the final predictor F of each group of independent variables, and complete the identification of the predictors ;

S4, construct an adaptive scheduling end-of-period water level forecast model according to the final forecast factor of each group of independent variables;

S5 , predicting the water level at the end of the dispatching period according to the forecasting model, and obtaining a prediction result of the water level at the end of the self-adaptive dispatching period.

Preferably, in S303, the candidate predictor with the greatest correlation with each group of sequences in Y=[IMF(n), Res] is selected as the final predictor F of each group of independent variables through T test.

Preferably, S4 includes the following steps:

S401, establishing a training sample set: the forecast determines the model training period length M and the verification period length (L-M) according to the size of the sample data. Since the model is a rolling forecast, the step size of the forecast period is 1 time step;

S402, establish a three-layer BP artificial neural Y=[IMF(n), Res] prediction model: the independent variable L _S of the water level sequence at the end of the dispatch period is decomposed based on the IMF(n), Res steady state sequence based on the empirical modal method, Establish n+1 forecasting models with their respective forecasting factors F, respectively forecast the independent variable L _S of the water level sequence at the end of the dispatch period according to these forecasting models, and obtain the forecast values of each group, which are IMF _f (n), Res _f ;

S403, according to the predicted value of each layer, calculate the reservoir water level L _(M+1)f at the end of the dispatching period predicted at the time M+1 according to the following formula:

Among them, IMF _f (j) is the predicted value of the j-th layer eigenmode function component, and Res _f is the predicted residual;

S404, add the predicted L _(M+1)f to the training sample, skip to S401, and perform the next round of water level forecasting until the forecasting at time L is completed, and the reservoir water level L _Lf at the end of the scheduling period predicted at time L is obtained ;

S405, compare the predicted L _(M+1)f ...L _Lf with the reservoir water level L _S at the end of the dispatch period from (M+1) to L time, and evaluate the forecast effect;

S406, if the evaluation effect meets the set threshold, the establishment of the adaptive model is completed, and the model is used to predict the water level of the reservoir at the end of the scheduling period in the future period; otherwise, the model parameters of the BP neural network are re-adjusted, and the model is re-modeled until the model is established. .

Preferably, in S405, the forecast effect is evaluated, and the evaluation indicators include Nash efficiency coefficient, relative error and pass rate;

The Nash efficiency coefficient Nash is calculated according to the following formula:

为实际调度期末水位的均值，Nash越接近1，预报越精准；Among them, L _f is the predicted water level at the end of the dispatch period,

is the mean value of the water level at the end of the actual dispatch period, the closer Nash is to 1, the more accurate the forecast;

The relative error MARE is calculated according to the following formula:

的序列长度，MARE越接近0，说明实测与预报值越接近，预报效果越精准，常认为MARE＜20％时，预报效果较好；Among them, N is

The closer the MARE is to 0, the closer the measured and predicted values are, and the more accurate the prediction effect is. It is often believed that when MARE < 20%, the prediction effect is better;

The pass rate QR is calculated according to the following formula:

Among them, n is the number of qualified forecasts; m is the total number of forecasts. When QR>80%, the forecast effect is considered to be better.

The beneficial effects of the present invention are as follows: the method for predicting the water level at the end of the dispatch period based on the self-adaptive ensemble empirical mode decomposition provided by the embodiment of the present invention fully considers the non-steady state of the water level sequence at the end of the dispatch period, by using the method of ensemble empirical modal decomposition The water level at the end of the dispatch period is converted into multiple sets of steady-state sequences, which realizes the steady state of the hydrological sequence and provides the most basic data conditions for conventional forecasting methods. In addition, when the adaptive forecasting model provided by the present invention is used for forecasting, it is a rolling forecasting operation, which enables the model to realize real-time correction, ensures the adaptability of the model, and ensures the model foundation for accurate forecasting. Therefore, by using the method provided by the present invention, the established prediction model has high accuracy when used for predicting the water level at the end of the dispatching period.

Description of drawings

1 is a schematic flowchart of a method for predicting water level at the end of a dispatch period based on adaptive set empirical modal decomposition provided by the present invention;

Fig. 2 is the schematic flow chart of the collective empirical mode decomposition method adopted by the present invention;

3 is a schematic diagram of a sequence after the water level sequence is decomposed based on the ensemble empirical mode decomposition method;

Fig. 4 is a schematic diagram of the prediction effect of the water level at the end of the adaptive scheduling period.

Detailed ways

In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only used to explain the present invention, but not to limit the present invention.

In order to solve the problem that the uncertainty of incoming water cannot be quantified in order to solve the problem that the water level is an unsteady sequence that limits conventional forecasting methods. The invention provides a method for predicting the water level at the end of the dispatching period with self-adaptive ensemble empirical mode decomposition, which effectively stabilizes the water level at the end of the dispatching period, excavates the influencing factors that respond to the water level, and explicitly and quantifies the uncertainty of the incoming water. The impact on the water level at the end of the reservoir dispatching period, and the self-adaptive forecasting model, keeps the forecasting model continuously updated, adapts to the water level change process at the end of the dispatching period, and maintains the stability of the model.

As shown in FIG. 1 , an embodiment of the present invention provides an adaptive scheduling end-of-period water level prediction method based on ensemble empirical mode decomposition, including the following steps:

S1, data preprocessing: extract a sequence of water levels at the end of the scheduling period with a length of L continuous multi-scheduling periods as the independent variable L _S , use the initial water level L ₀ at the early stage of scheduling, the water level L t at time _t and the inflow water volume W during the remaining period at time t Form the set of dependent variables X=[L ₀ , L _t , W];

S2, the steady-state decomposition of the independent variable: the independent variable L _S sequence is decomposed by the method of ensemble empirical mode decomposition to generate multiple sets of steady-state eigenmode functions IMF(n) and a set of residual sequences Res, which together form the self- Variable combination Y=[IMF(n), Res];

S3, determine the prediction factor: use the correlation between the dependent variable set X=[L ₀ , L _t , W] and the independent variable combination Y to determine the prediction decision factor, and implement it according to the following steps:

S301, construct a set of candidate predictors: combine the three influencing factor sequences L ₀ , L _t and W in the dependent variable set X=[L ₀ , L _t , W] in separate, pairwise, and common ways, respectively , constitute seven groups of candidate predictor sets, respectively F ₁ =[L ₀ ], F ₂ =[L _t ], F ₃ =[W], F ₄ =[L ₀ ,L _t ], F ₅ =[L ₀ ,W], F ₆ =[L _t ,W], F ₇ =[L ₀ ,L _t ,W];

S302, performing correlation analysis on each group of data in the seven groups of candidate predictor sets and the independent variable combination Y=[IMF(n), Res] respectively, to obtain the correlation factors;

S303, according to the correlation factor obtained in S302, select the candidate predictor with the largest correlation with each group of sequences in Y=[IMF(n), Res] as the final predictor F of each group of independent variables, and complete the identification of the predictors ;

S4, construct an adaptive scheduling end-of-period water level forecast model according to the final predictor of each set of independent variables, and evaluate the constructed model. If the forecast result meets the set threshold, the constructed model will be used as the final forecast model, otherwise, re-run Adjust the model parameters of the BP neural network, and re-model until the model is established;

S5: Predict the water level at the end of the dispatch period according to the final forecast model, and obtain a forecast result of the water level at the end of the dispatch period.

In the above method, the unsteady nature of the water level sequence at the end of the dispatch period is fully considered, and the method of ensemble empirical mode decomposition is used to convert the water level at the end of the dispatch period into multiple sets of steady-state sequences, realizing the steady state of the hydrological sequence, which is a conventional forecasting method. Provides the most basic data conditions.

In the present invention, in S303, through the T test, the candidate predictor with the greatest correlation with each group of sequences in Y=[IMF(n), Res] is selected as the final predictor F of each group of independent variables.

In a preferred embodiment of the present invention, S4 may include the following steps:

S401, establishing a training sample set: the forecast determines the model training period length M and the verification period length (L-M) according to the size of the sample data. Since the model is a rolling forecast, the step size of the forecast period is 1 time step;

S402, establish a three-layer BP artificial neural Y=[IMF(n), Res] prediction model: the independent variable L _S of the water level sequence at the end of the dispatch period is decomposed based on the IMF(n), Res steady state sequence based on the empirical modal method, Establish n+1 forecasting models with their respective forecasting factors F, respectively forecast the independent variable L _S of the water level sequence at the end of the dispatch period according to these forecasting models, and obtain the forecast values of each group, which are IMF _f (n), Res _f ;

S403, according to the predicted value of each layer, calculate the reservoir water level L _(M+1)f at the end of the dispatching period predicted at the time M+1 according to the following formula:

Among them, IMF _f (j) is the predicted value of the j-th layer eigenmode function component, and Res _f is the predicted residual;

S404, add the predicted L _(M+1)f to the training sample, skip to S401, and perform the next round of water level forecasting until the forecasting at time L is completed, and the reservoir water level L _Lf at the end of the scheduling period predicted at time L is obtained ;

S405, compare the predicted L _(M+1)f ...L _Lf with the reservoir water level L _S at the end of the dispatch period from (M+1) to L time, and evaluate the forecast effect;

S406, if the evaluation effect meets the set threshold, the establishment of the adaptive model is completed, and the model is used to predict the water level of the reservoir at the end of the scheduling period in the future period; otherwise, the model parameters of the BP neural network are re-adjusted, and the model is re-modeled until the model is established. .

It can be seen that, in the present invention, an adaptive forecasting model is established by means of a rolling forecasting operation, which can make the model realize real-time correction, ensure the adaptability of the model, and ensure the basis for the model to be used for accurate forecasting of the reservoir water level at the end of the scheduling period in the future period. .

Wherein, in S405, the forecast effect is evaluated, and the evaluation indicators include Nash efficiency coefficient, relative error and pass rate;

The Nash efficiency coefficient Nash is calculated according to the following formula:

为实际调度期末水位的均值，Nash越接近1，预报越精准；Among them, L _f is the predicted water level at the end of the dispatch period,

is the mean value of the water level at the end of the actual dispatch period, the closer Nash is to 1, the more accurate the forecast;

The relative error MARE is calculated according to the following formula:

的序列长度，MARE越接近0，说明实测与预报值越接近，预报效果越精准，常认为MARE＜20％时，预报效果较好；Among them, N is

The closer the MARE is to 0, the closer the measured and predicted values are, and the more accurate the prediction effect is. It is often believed that when MARE < 20%, the prediction effect is better;

The pass rate QR is calculated according to the following formula:

Among them, n is the number of qualified forecasts; m is the total number of forecasts. When QR>80%, the forecast effect is considered to be better.

specific embodiment

In the present invention, the water level prediction process at the end of the dispatch period of the Longyangxia Reservoir in the upper reaches of the Yellow River is selected as an example to verify the effect of the content of the present invention, and the following steps are specifically used to implement:

Step 1, data preprocessing.

Selecting the Longyangxia Reservoir in the upper reaches of the Yellow River from January 1, 2010 to December 31, 2016, the water level sequence at the end of the dispatch period with a dispatch period of months is the independent variable L _S , the water level at the beginning of the dispatch period in the same dispatch period is L ₀ , and the real-time water level sequence The inflow volume W of the remaining period at time L _t and t constitutes a series of dependent variable sets X=[L ₀ , L _t , W];

Step 2, the steady-state decomposition of the independent variable.

Determine the basic parameters based on the adaptive ensemble empirical mode decomposition method: noise variance (Nstd=0.2), noise group number (NE=100), number of iterations ( _MaxIter =500). The state decomposition method is decomposed (see Figure 2) to generate n-layer steady-state eigenmode functions IMF(1), IMF(2), ..., IMF(n) and 1-layer residual sequence Res, each group of sequences All are discrete functions, which are linear or nonlinear sequences of different frequencies, which together form the independent variable combination Y=[IMF(1),IMF(2),...,IMF(n),Res], the specific relational formula as follows:

Among them, n is the number of components of the eigenmode function, and IMF(j) is the eigenmode sequence of the jth layer.

In the embodiment of the present invention, the daily L _S of the Longyangxia Reservoir in the upper reaches of the Yellow River is decomposed from January 1, 2010 to December 31, 2016, wherein, after the decomposition, a steady-state sequence of 11 layers from high frequency to low frequency is generated, The result of the eigenmode sequence IMF(1) of the first layer is shown in Fig. 3. It can be seen from Fig. 3 that the decomposed sequence is a high-frequency steady-state sequence as the input of the prediction model.

Step 3, determine the predictor. The prediction decision factor is determined by the correlation between the dependent variable set X=[L ₀ , L _t , W] and the independent variable Y, including the following sub-steps:

(3-1) Construct a set of candidate predictors. The three influencing factor sequences L ₀ , L _t and W in the dependent variable set X=[L ₀ , L _t , W] are combined individually, in pairs, and together to form a total of 7 groups of alternative forecasts Set of factors, respectively F ₁ =[L ₀ ], F ₂ =[L _t ], F ₃ =[W], F ₄ =[L ₀ ,L _t ], F ₅ =[L ₀ ,W],F ₆ =[L _t ,W], F ₇ =[L ₀ ,L _t ,W].

(3-2) Analyze the 7 groups of candidate predictor sets and each group of data in the independent variable combination Y=[IMF(n), Res] respectively, carry out correlation analysis, and calculate the correlation coefficient according to the following formula;

Among them, Cov(X, Y) is the covariance of X and Y, Var[X] is the variance of X, and Var[Y] is the variance of Y.

(3-3) Set a significant level of α=0.025, carry out T test on the correlation coefficient between the sequence and meteorological factors, and calculate as follows, the factors that pass the hypothesis test are considered to be significantly related. Select the candidate predictor with the greatest correlation with each group of series in Y=[IMF(n), Res] as the final predictor F of each group of independent variables, and complete the identification of the predictor;

Among them, n is the number of data samples, and r is the correlation coefficient of the Pearson correlation coefficient.

Step 4: Build an adaptive scheduling end-of-period water level forecast model, and implement it according to the following steps:

(4-1) Establish a training sample set. The prediction will determine the model training period length M and the validation period length (L-M) according to the size of the sample data. Since the model is a rolling forecast, the step of the forecast period is one time step. In the embodiment of the present invention, the training period may be from January 1, 2010 to December 31, 2014, and the verification period may be 2015. January 1st to December 31st, 2016.

(4-2) Establish a three-layer BP artificial neural Y=[IMF(n), Res] prediction model. Contains input layer, hidden layer and output layer. Set the BP artificial neural network parameters.

The independent variable L _S of the water level sequence at the end of the dispatch period, based on the IMF(n), Res steady-state sequence decomposed by the empirical modal method, establish n+1 forecast models with their respective forecast factors F respectively. The independent variable L _S of the water level sequence at the end of the dispatch period is forecasted, and the forecast values of each group are obtained, which are IMF _f (n) and Res _f respectively.

(4-3) Calculate the reservoir water level L _(M+1)f at the end of the dispatching period predicted at the time of M+1 according to the forecast value of each layer. The calculation formula is as follows:

where IMF _f (j) is the predicted value of the j-th layer eigenmode function component, and Res _f is the predicted residual.

(4-4) Add the predicted L _(M+1)f to the training sample, skip to step (4-1), and carry out the next round of water level prediction until the prediction at time L is completed, and the predicted value at time L is obtained. Reservoir water level L _Lf at the end of the dispatch period.

(4-5) Compare the predicted L _(M+1)f ...L _Lf with the reservoir water level L _S at the end of the dispatch period from (M+1) to L time, and evaluate the forecast effect. The evaluation indicators are as follows: indivual:

Nash efficiency coefficient:

为实际调度期末水位的均值。Nash越接近1，预报越精准。Among them, L _f is the predicted water level at the end of the dispatch period,

is the mean value of the water level at the end of the actual dispatch period. The closer Nash is to 1, the more accurate the forecast.

Relative error:

的序列长度。MARE越接近0，说明实测与预报值越接近，预报效果越精准，常认为MARE＜20％时，效果较好。Among them, N is

sequence length. The closer MARE is to 0, the closer the measured and predicted values are, and the more accurate the prediction effect is. It is often believed that when MARE is less than 20%, the effect is better.

Pass rate:

Among them, n is the number of qualified forecasts; m is the total number of forecasts. When QR>80%, it can be considered that the prediction effect is better.

In the embodiment of the present invention, the prediction result of the model is shown in Figure 4, and the evaluation result is shown in the following table:

(4-6) As can be seen from Figure 4 and the above table, if the forecast effect of the water level at the end of the dispatch period is good, the adaptive model is established, and the model can be used to forecast the water level of the reservoir at the end of the dispatch period in the future.

By adopting the above technical solutions disclosed in the present invention, the following beneficial effects are obtained: The method for predicting the water level at the end of the dispatch period based on the adaptive set empirical mode decomposition provided by the embodiment of the present invention fully considers the non-steady state of the water level sequence at the end of the dispatch period. , using the method of ensemble empirical mode decomposition to convert the water level at the end of the dispatch period into multiple sets of steady-state sequences, realizing the steady state of the hydrological sequence and providing the most basic data conditions for conventional forecasting methods. In addition, when the adaptive forecasting model provided by the present invention is used for forecasting, it is a rolling forecasting operation, which enables the model to realize real-time correction, ensures the adaptability of the model, and ensures the model foundation for accurate forecasting. Therefore, by using the method provided by the present invention, the established prediction model has high accuracy when used for predicting the water level at the end of the dispatching period.

The above are only the preferred embodiments of the present invention. It should be pointed out that for those skilled in the art, without departing from the principles of the present invention, several improvements and modifications can be made. It should be regarded as the protection scope of the present invention.

Claims (2)

1. a self-adaptive scheduling end-of-term water level prediction method based on set empirical modal decomposition, is characterized in that, comprises the steps: S1, data preprocessing: extract a sequence of water levels at the end of the scheduling period with a length of L continuous multi-scheduling periods as the independent variable L _S , use the initial water level L ₀ at the early stage of scheduling, the water level L t at time _t and the inflow water volume W during the remaining period at time t Form the set of dependent variables X=[L ₀ , L _t , W]; S2, the steady-state decomposition of the independent variable: the independent variable L _S sequence is decomposed by the method of ensemble empirical mode decomposition to generate multiple sets of steady-state eigenmode functions IMF(n) and a set of residual sequences Res, which together form the self- Variable combination Y=[IMF(n), Res]; S3, determine the prediction factor: use the correlation between the dependent variable set X=[L ₀ , L _t , W] and the independent variable combination Y to determine the prediction decision factor, and implement it according to the following steps: S301, construct a set of candidate predictors: combine the three influencing factor sequences L ₀ , L _t and W in the dependent variable set X=[L ₀ , L _t , W] in separate, pairwise, and common ways, respectively , constitute seven groups of candidate predictor sets, respectively F ₁ =[L ₀ ], F ₂ =[L _t ], F ₃ =[W], F ₄ =[L ₀ ,L _t ], F ₅ =[L ₀ ,W], F ₆ =[L _t ,W], F ₇ =[L ₀ ,L _t ,W]; S302, performing correlation analysis on each group of data in the seven groups of candidate predictor sets and the independent variable combination Y=[IMF(n), Res] respectively, to obtain the correlation factors; S303, according to the correlation factor obtained in S302, select the candidate predictor with the largest correlation with each group of sequences in Y=[IMF(n), Res] as the final predictor F of each group of independent variables, and complete the identification of the predictors ; S4, construct an adaptive scheduling end-of-period water level forecast model according to the final forecast factor of each group of independent variables; S5, predicting the water level at the end of the dispatching period according to the forecasting model, to obtain an adaptive dispatching period end water level forecasting result; S4 includes the following steps: S401, establishing a training sample set: the forecast determines the model training period length M and the verification period length (L-M) according to the size of the sample data. Since the model is a rolling forecast, the step size of the forecast period is 1 time step; S402, establish a three-layer BP artificial neural Y=[IMF(n), Res] prediction model: the independent variable L _S of the water level sequence at the end of the dispatch period is decomposed based on the IMF(n), Res steady state sequence decomposed by the empirical modal method, Establish n+1 forecasting models with their respective forecasting factors F, respectively forecast the independent variable L _S of the water level sequence at the end of the dispatch period according to these forecasting models, and obtain the forecast values of each group, which are IMF _f (n), Res _f ; S403, according to the predicted value of each layer, calculate the reservoir water level L _(M+1)f at the end of the dispatching period predicted at the time M+1 according to the following formula:

Among them, IMF _f (j) is the predicted value of the j-th layer eigenmode function component, and Res _f is the predicted residual; S404, add the predicted L _(M+1)f to the training sample, skip to S401, and perform the next round of water level forecasting until the forecasting at time L is completed, and the reservoir water level L _Lf at the end of the scheduling period predicted at time L is obtained ; S405, compare the predicted L _(M+1)f ...L _Lf with the reservoir water level L _S at the end of the dispatch period from (M+1) to L time, and evaluate the forecast effect; S406, if the evaluation effect meets the set threshold, the establishment of the adaptive model is completed, and the model is used to predict the water level of the reservoir at the end of the scheduling period in the future period; otherwise, the model parameters of the BP neural network are re-adjusted, and the model is re-modeled until the model is established. ; In S303, through the T test, the candidate predictor with the greatest correlation with each group of series in Y=[IMF(n), Res] is selected as the final predictor F of each group of independent variables. 2. The self-adaptive scheduling end-of-period water level prediction method based on ensemble empirical mode decomposition according to claim 1, characterized in that, in S405, the prediction effect is evaluated, and the evaluation indicators include Nash efficiency coefficient, relative error and Pass rate; The Nash efficiency coefficient Nash is calculated according to the following formula:

Among them, L _f is the predicted water level at the end of the dispatch period,

is the mean value of the water level at the end of the actual dispatch period, the closer Nash is to 1, the more accurate the forecast; The relative error MARE is calculated according to the following formula:

Among them, N is

The closer the MARE is to 0, the closer the measured and predicted values are, and the more accurate the prediction effect is. When MARE<20%, the prediction effect is better; The pass rate QR is calculated according to the following formula:

Among them, n is the number of qualified forecasts; m is the total number of forecasts. When QR>80%, the forecast effect is considered to be better.

Images