A long-term runoff multi-period integrated prediction method based on factor decomposition optimization

Abstract

The present invention provides a long-term runoff multi-period integrated prediction method based on factor decomposition optimization, comprising selecting the month for predicting runoff and the corresponding forecast period step, spreading climate factors according to the prediction object period, selecting the maximum relevant monthly factor of each factor, setting the dimensionality reduction compensation contribution rate, reducing the dimensionality of the maximum relevant monthly factor matrix, constructing multiple machine learning regression models, setting priority evaluation indicators, determining the optimal contribution rate of each model through machine optimization, setting model numbers and a multi-objective evaluation priority mechanism, and realizing runoff asynchronous period integrated prediction; utilizing factor spreading correlation and dimensionality reduction to filter out redundant silver influences, and utilizing a multi-indicator priority evaluation mechanism to perform integrated prediction on multiple models, which can effectively overcome the instability of single model prediction and effectively improve the model prediction accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of runoff prediction, and in particular to a long-term runoff multi-period integrated prediction method based on factor decomposition optimization. Background Art

[0002] Current long-term runoff prediction methods primarily rely on modeling the relationships between multiple climate factors and runoff. However, the numerous factors influencing runoff in nature, and the interactions between these factors can overlap, increase model complexity and uncertainty, leading to cross-influence errors and model instability. For example, errors in individual factors can propagate and amplify each other, making the overall error difficult to control. Furthermore, the complex dependencies between factors can lead to cumulative errors if not accurately modeled. Due to the large number and sensitivity of parameters, the models exhibit high instability to small parameter changes, increasing computational resource requirements and further exacerbating model instability. Furthermore, since different long-term runoff models exhibit significant differences in runoff prediction performance over different time periods and forecast time steps, and no single model currently outperforms others at all scales, the use of a single model carries inherent predictive instability and uncertainty, resulting in larger errors in the final forecast data compared to models that integrate multiple methods, presenting certain limitations in its application. Summary of the Invention

[0003] The purpose of the present invention is to address the deficiencies of the above-mentioned prior art and provide a long-term runoff multi-period integrated prediction method based on factor decomposition optimization.

[0004] To achieve the above object, the present invention adopts the following technical solutions:

[0005] The present invention provides a long-term runoff multi-period integrated prediction method based on factor decomposition optimization, comprising the following steps:

[0006] S1. Determination of runoff prediction object: for the target site, select the month for runoff prediction and the corresponding forecast period step;

[0007] S2. Spreading of climate factors and determination of highly correlated months: Determine the factor period based on the forecast object, spread the climate factors according to the interannual cycle, select the maximum correlated monthly factor for each factor, and form a corresponding matrix;

[0008] S3. Optimal Monthly Factorization and Machine Optimization: Utilize principal component analysis (PCA) to reduce the dimensionality of the maximum correlation monthly factor matrix, set the dimensionality reduction compensation contribution rate, construct multiple machine learning regression models, set priority evaluation indicators, and determine the optimal dimensionality reduction compensation contribution rate for each machine learning model through machine optimization.

[0009] S4. Integrated prediction of runoff during asynchronous periods: set model numbers and a multi-objective evaluation priority mechanism, calculate the optimal model numbers for different months and forecast periods, and realize integrated prediction of runoff during asynchronous periods.

[0010] Furthermore, the S1 is specifically:

[0011] For each target site, set the corresponding predicted traffic month as , set the forecast period to , then the flow of the forecast object is expressed as ,in , indicating that predictions are made for any month in a year and 12 months. is a positive integer less than 12. For each forecast object flow , let any prediction object sample be , is the year corresponding to the flow sample.

[0012] Furthermore, the S2 is specifically:

[0013] S21. Set the sample data of the factors used to predict runoff to ,in, is the type of factor, is the month corresponding to the previous forecast factor sample, is the year corresponding to the factor sample;

[0014] S22. Based on the forecast object month, considering that climate factors have typical annual scale fluctuations, for any forecast object sample Determine the month corresponding to the previous forecast factor sample The value is , the specific determination method is:

[0015] ;

[0016] in, For the Months corresponding to factor samples The value of , and Represent the years corresponding to the factor samples The year corresponding to the traffic sample No. samples;

[0017] According to the determined factor set, for any prediction object sample There are sample sets of previous predictors , the expression is:

[0018] ;

[0019] S23. Based on the sample set of previous prediction factors , for each factor The monthly factors are respectively related to any prediction object sample Perform correlation and find the month with the largest correlation coefficient for each factor, and finally obtain the maximum monthly correlation factor sample set matrix:

[0020] ;

[0021] in, Represents any prediction object sample The corresponding maximum monthly correlation factor sample set matrix, For the type of factor The corresponding month has the largest correlation coefficient.

[0022] Furthermore, the details of S3 are:

[0023] S31, using principal component analysis method PCA to analyze the maximum monthly correlation factor sample set matrix Perform dimensionality reduction and set the factor dimensionality reduction compensation contribution rate to be ,but:

[0024] ;

[0025] in, is the sample set matrix of the maximum monthly correlation factor The projected principal component factor matrix after PCA dimensionality reduction is performed, and the number of principal components is determined by the factor dimensionality reduction compensation contribution rate control; is the projected principal component factor matrix The bth principal component factor of ; is the variance contribution rate of the bth principal component factor;

[0026] S32. Build a variety of machine learning regression model frameworks ,in, Types of machine learning regression models;

[0027] S33. Set the corresponding priority evaluation index for traffic prediction as follows: ;

[0028] S34, determine the optimal compensation rate for each machine learning through machine optimization, and project the principal component factor matrix Divide into training set and validation set, and project the principal component factor matrix of the training set Input multiple regression models, train the parameters of multiple machine learning regression models, and obtain the trained machine learning regression models , and then the projected principal component factor matrix of the validation set Input the trained machine learning regression model , get the prediction results on the validation set;

[0029] S35. Based on the prediction results of the validation set and the actual data, use the priority evaluation indicators Count the prediction effects of each model and obtain the priority evaluation index results of each model on the validation set ;

[0030] For each regression model, select the priority evaluation indicator results The corresponding factor dimension reduction compensation contribution rate at the optimal time The value is the final factor dimensionality reduction compensation contribution rate value, for the machine learning regression model The optimal dimensionality reduction compensation contribution rate of the factor is expressed as .

[0031] Furthermore, the S4 is specifically:

[0032] S41, based on the optimal factor dimensionality reduction compensation contribution rate Determine the final factor dataset , repeat step S34 to obtain the optimal parameter prediction model set ;

[0033] S42. Set multiple evaluation indicators for traffic prediction and set a priority evaluation mechanism for the multiple evaluation indicators; wherein the evaluation indicator matrix is ​​set as , Indicates the evaluation indicators;

[0034] For the evaluation index matrix, set the corresponding priority evaluation voting matrix , for the priority index, set the number of votes to , for the final indicator of preference, set the number of votes to 1;

[0035] S43, calculating the statistical results of multiple evaluation indicators for each model, performing priority scoring voting on the evaluation indicator set of each model, counting the cumulative votes, and determining the optimal model type;

[0036] Using the evaluation index matrix Calculate the optimal parameter prediction model set Multi-index evaluation results of various models ;

[0037] ;

[0038] For the first evaluation index , multi-index evaluation results The first column of the matrix The model corresponding to the best value in gets the number of votes , if there are multiple models of the first evaluation If they are the same, multiple models will get votes at the same time , the remaining models received votes of ;

[0039] The remaining evaluation indicators are voted on separately until the Evaluation indicators , count the total number of votes for each model for:

[0040] ;

[0041] in, Optimal parameter prediction model The final number of votes; Optimal parameter prediction model exist The number of votes on each evaluation indicator;

[0042] The model with the highest total number of votes is the final prediction model applicable to the current prediction object;

[0043] S44, repeat the operations from S1 to S43 for multiple prediction objects , determine the predicted traffic month and forecast period The model with the highest cumulative votes is the corresponding applicable model. , together build An optimal model is generated to form an integrated prediction scheme for runoff asynchronous periods;

[0044] S45, the actual factor data is substituted into the integrated prediction scheme, for different predicted flow months and forecast period The flow is adapted to the corresponding prediction model to obtain the final runoff prediction result.

[0045] The beneficial effects of the present invention are as follows: considering the high temporal variability of runoff and the complex time lag of the interaction between multiple spheres of the Earth system, different influencing factors are considered for runoff at different times and different forecast periods, and redundant factors are filtered out through dimensionality reduction. At the same time, considering that the prediction effects of different models on runoff at different time periods and different forecast period steps are significantly different, a model integration method with multi-index priority evaluation is proposed, integrating the prediction model with the best effect in each month and time period step. This can effectively overcome the instability of single model prediction in asynchronous time periods, thereby maximizing the advantages of multiple models and further improving the model prediction accuracy compared to traditional single model prediction results. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 This is a flowchart of a long-term runoff multi-period integrated prediction method based on factor decomposition optimization. DETAILED DESCRIPTION

[0047] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, 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 and are not intended to limit the present invention.

[0048] See also Figure 1 , a long-term runoff multi-period integrated prediction method based on factor decomposition optimization, including the following steps:

[0049] S1. Determination of runoff prediction object: for the target site, select the month for runoff prediction and the corresponding forecast period step;

[0050] S2. Spreading of climate factors and determination of highly correlated months: Determine the factor period based on the forecast object, spread the climate factors according to the interannual cycle, select the maximum correlated monthly factor for each factor, and form a corresponding matrix;

[0051] S3. Optimal Monthly Factorization and Machine Optimization: Utilize principal component analysis (PCA) to reduce the dimensionality of the maximum correlation monthly factor matrix, set the dimensionality reduction compensation contribution rate, construct multiple machine learning regression models, set priority evaluation indicators, and determine the optimal dimensionality reduction compensation contribution rate for each machine learning model through machine optimization.

[0052] S4. Integrated prediction of runoff during asynchronous periods: set model numbers and a multi-objective evaluation priority mechanism, calculate the optimal model numbers for different months and forecast periods, and realize integrated prediction of runoff during asynchronous periods.

[0053] For each target site, set the corresponding predicted traffic month as , set the forecast period to , then the flow of the forecast object is expressed as ,in , indicating that predictions are made for any month in a year and 12 months. is a positive integer less than 12. For each forecast object flow , let any prediction object sample be , is the year corresponding to the flow sample.

[0054] In this example, the monthly runoff data from 1962 to 2023 at the Shigu Station in the Yangtze and Jinsha River basins are used as the prediction dataset. The prediction months are selected as January to December, the forecast period is 1 to 3 months, the factor sample length is 62 years from 1962 to 2023, and the flow rate is processed as a percentage of the flow deviation from the mean for prediction.

[0055] The S2 is specifically:

[0056] S21. Set the sample data of the factors used to predict runoff to ,in, is the type of factor, is the month corresponding to the previous forecast factor sample, is the year corresponding to the factor sample;

[0057] In this embodiment, the factors used are 130 monthly climate factor data from the National Climate Center from 1961 to 2023, including 88 atmospheric circulation indices, 26 sea temperature indices, and 16 other types of climate indices, which can more comprehensively characterize the global climate system signal. After screening, there are 75 factors in total. At the same time, the previous actual flow is added, and there are a total of 76 factors. The factors are processed into factor anomalies for prediction.

[0058] S22. Based on the forecast object month, considering that climate factors have typical annual scale fluctuations, for any forecast object sample Determine the month corresponding to the previous forecast factor sample The value is , the specific determination method is:

[0059] ;

[0060] in, For the Months corresponding to factor samples The value of , and Represent the years corresponding to the factor samples The year corresponding to the traffic sample No. samples;

[0061] According to the determined factor set, for any prediction object sample There are sample sets of previous predictors , the expression is:

[0062] ;

[0063] In this embodiment, 76 factors are flattened to obtain a total of 76*12=912 early factors.

[0064] S23. Based on the sample set of previous prediction factors , for each factor The monthly factors are respectively related to any prediction object sample Perform correlation and find the month with the largest correlation coefficient for each factor, and finally obtain the maximum monthly correlation factor sample set matrix:

[0065] ;

[0066] in, Represents any prediction object sample The corresponding maximum monthly correlation factor sample set matrix, For the type of factor The corresponding month has the largest correlation coefficient.

[0067] In this embodiment, for each type of factor, the month with the greatest impact on the predicted monthly traffic is extracted as an independent factor from the 76*12 factors, thereby obtaining 76 highly correlated traffic influencing factors.

[0068] The details of S3 are:

[0069] S31, using principal component analysis method PCA to analyze the maximum monthly correlation factor sample set matrix Perform dimensionality reduction and set the factor dimensionality reduction compensation contribution rate to be ,but:

[0070] ;

[0071] in, is the sample set matrix of the maximum monthly correlation factor The projected principal component factor matrix after PCA dimensionality reduction is performed, and the number of principal components is determined by the factor dimensionality reduction compensation contribution rate control; is the projected principal component factor matrix The bth principal component factor of ; is the variance contribution rate of the bth principal component factor;

[0072] S32. Build a variety of machine learning regression model frameworks ,in, Types of machine learning regression models;

[0073] The models used in this embodiment include a multiple linear regression model, a random forest model, and an improved Beluga optimization algorithm-bidirectional long short-term memory neural network model MWOA-BiLSTM.

[0074] S33. Set the corresponding priority evaluation index for traffic prediction as follows: ;

[0075] The priority evaluation indicator used in this embodiment is the deviation sign consistency score PS, which represents the consistency rate between the predicted flow deviation and the actual flow deviation sign. If the prediction and the actual situation are completely consistent, the score is 100, and if they are completely inconsistent, the score is 0.

[0076] S34, determine the optimal compensation rate for each machine learning through machine optimization, and project the principal component factor matrix Divide into training set and validation set, and project the principal component factor matrix of the training set Input multiple regression models, train the parameters of multiple machine learning regression models, and obtain the trained machine learning regression models , and then the projected principal component factor matrix of the validation set Input the trained machine learning regression model , get the prediction results on the validation set;

[0077] In this embodiment, the training set and the validation set are divided into two groups according to a ratio of 0.7 and 0.3 in number of samples in the training set and the validation set, respectively.

[0078] S35. Based on the prediction results of the validation set and the actual data, use the priority evaluation indicators Count the prediction effects of each model and obtain the priority evaluation index results of each model on the validation set ;

[0079] For each regression model, select the priority evaluation indicator results The corresponding factor dimension reduction compensation contribution rate at the optimal time The value is the final factor dimensionality reduction compensation contribution rate value, for the machine learning regression model The optimal dimensionality reduction compensation contribution rate of the factor is expressed as .

[0080] The S4 is specifically:

[0081] S41, based on the optimal factor dimensionality reduction compensation contribution rate Determine the final factor dataset , repeat step S34 to obtain the optimal parameter prediction model set ;

[0082] S42. Set multiple evaluation indicators for traffic prediction and set a priority evaluation mechanism for the multiple evaluation indicators; wherein the evaluation indicator matrix is ​​set as , Indicates the evaluation indicators;

[0083] For the evaluation index matrix, set the corresponding priority evaluation voting matrix , for the priority index, set the number of votes to , for the final indicator of preference, set the number of votes to 1;

[0084] The evaluation indicators used in this embodiment are the deviation sign consistency score PS, the root mean square error score RMSE and the mean absolute error score MAE. The vote number for the mean absolute error score PS is set to 3, the root mean square error score RMSE is set to 2, and the mean absolute error score MAE is set to 1.

[0085] S43, calculating the statistical results of multiple evaluation indicators for each model, performing priority scoring voting on the evaluation indicator set of each model, counting the cumulative votes, and determining the optimal model type;

[0086] Using the evaluation index matrix Calculate the optimal parameter prediction model set Multi-index evaluation results of various models ;

[0087] ;

[0088] For the first evaluation index , multi-index evaluation results The first column of the matrix The model corresponding to the best value in gets the number of votes , if there are multiple models of the first evaluation If they are the same, multiple models will get votes at the same time , the remaining models received votes of ;

[0089] The remaining evaluation indicators are voted on separately until the Evaluation indicators , count the total number of votes for each model for:

[0090] ;

[0091] in, Optimal parameter prediction model The final number of votes; Optimal parameter prediction model exist The number of votes on each evaluation indicator;

[0092] The model with the highest total number of votes is the final prediction model applicable to the current prediction object;

[0093] S44, repeat the operations from S1 to S43 for multiple prediction objects , determine the predicted traffic month and forecast period The model with the highest cumulative votes is the corresponding applicable model. , together build An optimal model is generated to form an integrated prediction scheme for runoff asynchronous periods;

[0094] S45, the actual factor data is substituted into the integrated prediction scheme, for different predicted flow months and forecast period The flow is adapted to the corresponding prediction model to obtain the final runoff prediction result.

[0095] In this embodiment, the multivariate regression model is defined as 1, the random forest model is defined as 2, and the improved Beluga optimization algorithm-bidirectional long short-term memory neural network model MWOA-BiLSTM is defined as 3. The final integrated prediction scheme and the evaluation results of the integrated prediction for the runoff from January to December at Shigu Station in the forecast period from January to March are shown in Table 1. The evaluation results of the integrated model prediction for each flow month and forecast period are shown in Table 2. Overall, good results were achieved after integration.

[0096] Table 1 Final integrated prediction scheme and evaluation results of integrated prediction

[0097]

[0098] Table 2 Evaluation results of the integrated model prediction for each flow month and forecast period

[0099]

[0100] In summary, the present invention adopts a long-term runoff asynchronous period integrated prediction method based on multi-factor decomposition optimization, taking into account the high temporal variability of runoff and the complex time lag of the interaction between multiple spheres of the earth system, and considering different influencing factors for runoff at different times and different forecast periods, and filtering the influence of redundant factors by dimensionality reduction. At the same time, considering that the prediction effects of different models on runoff at different time periods and different forecast period steps are significantly different, a model integration method with multi-index priority evaluation is proposed, and the prediction model with the best effect is integrated in each month and time period step. It can effectively overcome the instability of single model prediction in asynchronous time periods, thereby maximizing the advantages of multiple models, and can further improve the model prediction accuracy compared with the traditional single model prediction results, which is conducive to the application and promotion of long-term runoff prediction services.

[0101] The above-described embodiments merely illustrate the implementation methods of the present invention. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art would be able to make numerous modifications and improvements without departing from the spirit of the present invention, all of which fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the appended claims.

Claims

1. A long-term runoff multi-period integrated prediction method based on factor decomposition optimization, characterized by: The following steps are involved: S1. Determination of runoff prediction object: for the target site, select the month for runoff prediction and the corresponding forecast period step; S2. Climate factor spreading and determination of highly correlated months: Determine the factor period based on the forecast object, spread the climate factors of the previous 12 months according to the interannual cycle, select the monthly factor with the maximum correlation between each factor and the forecast object, and form a corresponding matrix; S3. Optimal Monthly Factor Decomposition and Machine Optimization: Utilize principal component analysis (PCA) to reduce the dimensionality of the maximum correlation month factor matrix, set the factor dimensionality reduction compensation contribution rate, and control the number of principal components obtained by dimensionality reduction through the factor dimensionality reduction compensation contribution rate. Construct multiple machine learning regression models, input the reduced principal component factor matrix into the model for training, set priority evaluation indicators, and statistically analyze the prediction effects of each model based on the priority evaluation indicators. Through machine optimization, select the factor dimensionality reduction compensation contribution rate that optimizes the priority evaluation indicator as the optimal dimensionality reduction compensation contribution rate for the corresponding model. S4. Integrated prediction of runoff in asynchronous periods: Based on the optimal dimensionality reduction compensation contribution rate of each model, the optimal parameter prediction model set is trained, the model number and multi-objective evaluation priority mechanism are set, and the optimal model number is counted for different months and forecast periods to achieve integrated prediction of runoff in asynchronous periods.

2. The method for long-term runoff multi-period integrated prediction based on factor decomposition optimization according to claim 1 is characterized in that: The S1 is specifically: For each target site, set the corresponding predicted traffic month as , set the forecast period to , then the flow of the forecast object can be expressed as ,in , indicating that predictions are made for any month in a year and 12 months. is a positive integer less than 12. For each forecast object flow , let any prediction object sample be , is the year corresponding to the flow sample.

3. The method for long-term runoff multi-period integrated prediction based on factor decomposition optimization according to claim 1 is characterized in that: The S2 is specifically: S21. Set the sample data of the factors used to predict runoff to ,in, is the type of factor, is the month corresponding to the previous forecast factor sample, is the year corresponding to the factor sample; S22. Based on the forecast object month, considering that climate factors have annual scale fluctuations, for any forecast object sample Determine the month corresponding to the previous forecast factor sample The value is , the specific determination method is: ； in, For the Months corresponding to factor samples The value of , and Represent the years corresponding to the factor samples The year corresponding to the traffic sample No. samples; According to the determined factor set, for any prediction object sample There are sample sets of previous predictors , the expression is: ； S23, based on the sample set of previous prediction factors , for each factor The monthly factors are respectively related to any prediction object sample Perform correlation and find the month with the largest correlation coefficient for each factor, and finally obtain the maximum monthly correlation factor sample set matrix: ； in, Represents any prediction object sample The corresponding maximum monthly correlation factor sample set matrix, For the type of factor The corresponding month has the largest correlation coefficient.

4. The method for long-term runoff multi-period integrated prediction based on factor decomposition optimization according to claim 1 is characterized in that: The details of S3 are: S31, using PCA method to analyze the maximum monthly correlation factor sample set matrix Perform dimensionality reduction and set the factor dimensionality reduction compensation contribution rate to be ,but: ； in, is the sample set matrix of the maximum monthly correlation factor The projected principal component factor matrix after PCA dimensionality reduction, the number of principal components is determined by the factor dimensionality reduction compensation contribution rate control; is the projected principal component factor matrix The bth principal component factor of ; is the variance contribution rate of the bth principal component factor; S32. Build a variety of machine learning regression model frameworks ,in, Types of machine learning regression models; S33. Set the corresponding priority evaluation index for traffic prediction as follows: ; S34, determine the optimal compensation rate for each machine learning through machine optimization, and project the principal component factor matrix Divide into training set and validation set, and project the principal component factor matrix of the training set Input multiple regression models, train the parameters of multiple machine learning regression models, and obtain the trained machine learning regression models , and then the projected principal component factor matrix of the validation set Input the trained machine learning regression model , get the prediction results on the validation set; S35. Based on the prediction results of the validation set and the actual data, use the priority evaluation indicators Count the prediction effects of each model and obtain the priority evaluation index results of each model on the validation set ; For each regression model, select the priority evaluation indicator results The corresponding factor dimension reduction compensation contribution rate at the optimal time The value is the final factor dimensionality reduction compensation contribution rate value, for machine learning regression model The optimal dimensionality reduction compensation contribution rate of the factor is expressed as .

5. The method for long-term runoff multi-period integrated prediction based on factor decomposition optimization according to claim 4 is characterized in that: The S4 is specifically: S41, based on the optimal factor dimensionality reduction compensation contribution rate Determine the final factor dataset , repeat step S34 to obtain the optimal parameter prediction model set ; S42. Set multiple evaluation indicators for traffic prediction and set a priority evaluation mechanism for the multiple evaluation indicators; wherein the evaluation indicator matrix is ​​set as , Indicates the evaluation indicators; For the evaluation index matrix, set the corresponding priority evaluation voting matrix ,For the indicator with the priority tendency, the number of votes set is higher than the number of votes set for the indicator with the last tendency; S43, calculating the statistical results of multiple evaluation indicators for each model, performing priority scoring voting on the evaluation indicator set of each model, counting the cumulative votes, and determining the optimal model type; Using the evaluation index matrix Calculate the optimal parameter prediction model set Multi-index evaluation results of various models ; ； For the first evaluation index , multi-index evaluation results The first column of the matrix The model corresponding to the best value in gets the number of votes , if there are multiple models of the first evaluation If they are the same, multiple models will get votes at the same time , the remaining models received votes of ; The remaining evaluation indicators are voted on separately until the Evaluation indicators , count the total number of votes for each model for: ； in, Optimal parameter prediction model The final number of votes; Optimal parameter prediction model exist The number of votes on each evaluation indicator; The model with the highest total number of votes is the final prediction model applicable to the current prediction object; S44, repeat the operations from S1 to S43 for multiple prediction objects , determine the predicted traffic month and forecast period The model with the highest cumulative votes is the corresponding applicable model. , together build An optimal model is generated to form an integrated prediction scheme for runoff asynchronous periods; S45, the actual factor data is substituted into the integrated prediction scheme, for different predicted flow months and forecast period The flow is adapted to the corresponding prediction model to obtain the final runoff prediction result.

Images