Runoff forecasting method and system based on multi-model combination

Abstract

The invention discloses a runoff forecasting method and system based on multi-model combination. The method comprises the following steps: acquiring a forecasting data set of a to-be-studied area; preprocessing the forecast data set of the to-be-studied area to obtain a preprocessed data set; respectively inputting the preprocessed data set into different trained single runoff forecasting models for forecasting to obtain a plurality of single forecasting results; performing error correction on each single forecast result by using a Copula function to obtain each corrected single forecast result; inputting each corrected single forecast result into the trained BP neural network model for combination to obtain an initial runoff forecast result; performing error correction on the initial runoff forecasting result by using a Copula function to obtain a final runoff forecasting result; according to the method, the runoff forecasting precision can be improved, runoff forecasting errors in a long forecasting period are reduced, and the effective forecasting period of runoff forecasting is prolonged.

Description

Technical Field

[0001] The present invention relates to the technical field of runoff forecasting, and in particular to a runoff forecasting method and system based on multi-model combination. Background Art

[0002] Runoff forecasting technology provides indispensable forward-looking information and plays a vital role in water resource management, flood control and disaster prevention. In terms of water resource management, runoff forecasting technology can accurately predict future water flow conditions and provide a solid foundation for the scientific dispatch of reservoirs. Based on the results of runoff forecasting, managers can reasonably arrange the storage and release of reservoirs to achieve optimal allocation of water resources, avoid unnecessary waste, and ensure that there are sufficient water resources to meet the needs of human life and production during droughts or water shortages. In terms of flood control, by real-time monitoring and predicting changes in river flow, relevant departments can respond quickly and take necessary flood control measures, such as strengthening embankments and activating drainage systems, thereby effectively reducing the risks and losses caused by flood disasters and ensuring the safety of people's lives and property. In terms of disaster prevention, runoff forecasting technology provides strong support, especially in areas where geological disasters occur frequently, where rainfall runoff is often the key factor in triggering landslides, mudslides and other disasters; with the help of runoff forecasting, relevant departments can warn of these potential disaster risks in advance, and gain valuable prevention and response time for the masses and relevant departments, thereby further reducing the occurrence and impact of disasters. Therefore, runoff forecasting technology, with its accuracy and foresight, provides strong technical support for water resources management, flood control and disaster prevention, and is an important force in promoting the sustainable use of water resources and ensuring the safety of social and economic development.

[0003] At present, in runoff forecasting technology, the commonly used models are mainly divided into two categories: physical-driven models and data-driven models. The physical-driven model is based on hydrological principles and uses mathematical and physical equations to describe the occurrence and development of runoff in detail. However, the parameter calibration is complex, the calculation process is complex, the basin characteristic information is required to be detailed, and its application scope is not easy to expand to other fields. In contrast, the data-driven model does not deeply explore the internal mechanism of the hydrological process, but directly uses the input data to predict runoff. However, this type of model has high requirements for data quality and weak interpretation of the prediction results. Since runoff is affected by a variety of temporal and spatial factors such as hydrology, meteorology, topography, and human activities, its runoff generation and convergence process is quite complex, and a single runoff forecast model is often difficult to meet the requirements of forecast accuracy and forecast period at the same time. Summary of the invention

[0004] The purpose of the present invention is to provide a runoff forecasting method and system based on multi-model combination, which combines the forecast results of multiple single runoff forecasting models through a BP neural network model and uses the Coupla function to correct the forecast results, thereby improving the runoff forecast accuracy, reducing the runoff forecast error under long forecast periods, and extending the effective forecast period of runoff forecasts.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] In a first aspect, the present invention provides a runoff forecasting method based on a combination of multiple models, comprising:

[0007] Obtaining a forecast data set for the area to be studied, the forecast data set including hydrological, meteorological and topographic data in the current and past time periods, and meteorological data in the forecast period;

[0008] Preprocessing the forecast data set of the area to be studied to obtain a preprocessed data set;

[0009] The preprocessed data sets are input into different trained single runoff forecast models for prediction, and multiple single forecast results are obtained;

[0010] Using the Copula function, the error of each single forecast result is corrected to obtain each corrected single forecast result;

[0011] Input each corrected single forecast result into the trained BP neural network model for combination to obtain the initial runoff forecast result;

[0012] The Copula function is used to correct the error of the initial runoff forecast result to obtain the final runoff forecast result.

[0013] Optionally, the single runoff forecast model is a physical mechanism model or a data-driven model; wherein the physical mechanism model includes the watershed hydrological model SWAT and the variable infiltration capacity model, and the data-driven model includes artificial neural networks, support vector machines and random forests.

[0014] Optionally, the preprocessing process includes:

[0015] Scale-convert the hydrological, meteorological and topographic data according to the forecast time scale to obtain the hydrological, meteorological and topographic data that match the forecast time scale;

[0016] The hydrological, meteorological and topographic data that match the forecast time scale are cleaned and normalized in turn to obtain the preprocessed results.

[0017] Optionally, the error correction process includes:

[0018] Obtaining the measured runoff values ​​corresponding to different forecast periods in the historical forecast results; wherein the historical forecast results are historical single forecast results or historical initial runoff forecast results; the historical single forecast results are obtained by predicting the historical hydrological, meteorological and topographic data in a single runoff forecast model; the historical initial runoff forecast results are obtained by inputting each historical single forecast result into a BP neural network model;

[0019] According to the measured runoff values ​​corresponding to different forecast periods, the forecast results are analyzed for errors, and the error series is obtained, that is, the set of forecast errors under different forecast periods.

[0020] The forecast errors under different forecast periods are used as the independent variables of the Copula function, and multiple forecasts are performed to obtain the error data matrix;

[0021] The kernel density estimation method is used to calculate the overall probability distribution density of forecast errors under different forecast periods, and it is used as the marginal distribution function of forecast errors under different forecast periods.

[0022] Based on the Copula function and the marginal distribution function of the forecast error under different forecast periods, the joint distribution function of the forecast error under different forecast periods is obtained.

[0023] Generate a random sequence based on the joint distribution function of forecast errors under different forecast periods to represent the cumulative distribution probability of the error distribution under different forecast periods;

[0024] Based on the random sequence, the inverse cumulative distribution function of the marginal distribution of forecast errors under different forecast periods is calculated to obtain the simulated forecast error sequence;

[0025] The current forecast result before error correction is added to the simulated forecast error sequence to obtain the current forecast result after error correction.

[0026] Optionally, the error data matrix is ​​represented as follows:

[0027]

[0028] Where E represents the error data matrix; e n，m It represents the forecast error generated by the nth forecast in the mth forecast period; the formula of forecast error is as follows:

[0029]

[0030] Among them, e i，j is the forecast error generated by the i-th forecast in the j-th forecast period; i, j are sequence numbers; F i，j and Q i，j They represent the predicted runoff value and the measured runoff value in the i-th forecast in the j-th forecast period respectively.

[0031] Optionally, the overall probability distribution density function of the forecast error under a certain forecast period is expressed as follows:

[0032]

[0033]

[0034] in, represents the overall probability distribution density function of the forecast error in the t-th forecast period; x t represents the forecast error in the tth forecast period; e w，t It represents the forecast error caused by the w-th forecast in the t-th forecast period; K' is the kernel function; h' is the window width.

[0035] Optionally, the joint distribution function of forecast errors at different forecast horizons is expressed as follows:

[0036]

[0037] in, Forecast error at different forecast periods The joint distribution function of is the forecast error x in the mth forecast period m The marginal distribution function of It is a Copula function.

[0038] Optionally, the BP neural network model includes an input layer, a plurality of hidden layers and an output layer;

[0039] The input layer includes l nodes, which are l corrected single forecast results and corresponding measured runoff values; wherein the l corrected single forecast results are obtained by predicting l different trained single runoff forecast models, l≥2;

[0040] The number of hidden layers can be adjusted according to the training results of the BP neural network model;

[0041] The output layer is the initial runoff forecast result.

[0042] Optionally, the training process of the BP neural network model includes:

[0043] Using each historical single forecast result data set after error correction and the corresponding measured runoff value as a training set; the historical single forecast result data set is predicted by a single runoff forecast model;

[0044] Train the BP neural network model based on the data in the training set to obtain the optimal model parameters;

[0045] According to the optimal model parameters, confirm the trained BP neural network model.

[0046] In the second aspect, a runoff forecasting system based on a combination of multiple models includes:

[0047] An acquisition module is used to acquire a forecast data set for the area to be studied, wherein the forecast data set includes hydrological, meteorological and topographic data in the current and past time periods, and meteorological data in the forecast period;

[0048] A preprocessing module is used to preprocess the forecast data set of the area to be studied to obtain a preprocessed data set;

[0049] The single prediction module is used to input the preprocessed data set into different trained single runoff prediction models for prediction, and obtain multiple single prediction results;

[0050] The first correction module is used to use the Copula function to perform error correction on each single forecast result to obtain each corrected single forecast result;

[0051] The combination module is used to input each corrected single forecast result into the trained BP neural network model for combination to obtain the initial runoff forecast result;

[0052] The second correction module is used to use the Copula function to perform error correction on the initial runoff forecast result to obtain the final runoff forecast result.

[0053] Compared with the prior art, the present invention has the following beneficial effects:

[0054] The present invention provides a runoff forecasting method and system based on multi-model combination. The method firstly predicts multiple single runoff forecasting models to obtain multiple single forecasting results, and uses Copula function to correct the single forecasting results; then uses BP neural network model to combine the corrected single forecasting results to obtain initial runoff forecasting results; finally uses Copula function to correct the initial runoff forecasting results to obtain final runoff forecasting results; the method not only reduces the uncertainty that may exist in single runoff forecasting model by integrating the advantages of different models, but also improves the accuracy and reliability of runoff forecasting; it also adopts the serial application of BP neural network model and error correction to solve the problem of low forecasting accuracy under long forecast period, and effectively prolongs the forecast period of runoff forecasting. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 It is a schematic diagram of a flow chart of a runoff forecasting method based on a combination of multiple models in one embodiment of the present invention;

[0056] Figure 2Shown is a schematic diagram of the error correction process in one embodiment of the present invention;

[0057] Figure 3 The figure shows a schematic diagram of forecast results before correction with a forecast period of 1 day in one embodiment of the present invention;

[0058] Figure 4 The figure shows a schematic diagram of a forecast result after correction with a forecast period of 1 day in one embodiment of the present invention;

[0059] Figure 5 The figure shows a schematic diagram of forecast results before correction with a forecast period of 5 days in one embodiment of the present invention;

[0060] Figure 6 The figure shows a schematic diagram of a forecast result after correction with a forecast period of 5 days in one embodiment of the present invention;

[0061] Figure 7 The figure shows a schematic diagram of forecast results before correction with a forecast period of 10 days in one embodiment of the present invention;

[0062] Figure 8 The figure shows a schematic diagram of a forecast result after correction with a forecast period of 10 days in an embodiment of the present invention. DETAILED DESCRIPTION

[0063] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present invention.

[0064] Example 1

[0065] like Figure 1 As shown, the embodiment of the present invention introduces a runoff forecasting method based on multi-model combination, comprising the following steps:

[0066] S1: Obtain a forecast data set for the area to be studied, wherein the forecast data set includes hydrological, meteorological and topographic data in the current and past time periods, and meteorological data in the forecast period;

[0067] S2: preprocessing the forecast data set of the area to be studied to obtain a preprocessed data set;

[0068] S3: input the preprocessed data sets into different trained single runoff forecast models for prediction, and obtain multiple single forecast results;

[0069] S4: Using the Copula function, error correction is performed on each single forecast result to obtain each corrected single forecast result;

[0070] S5: inputting each corrected single forecast result into the trained BP neural network model for combination to obtain the initial runoff forecast result;

[0071] S6: Use the Copula function to correct the error of the initial runoff forecast result to obtain the final runoff forecast result.

[0072] This embodiment provides a runoff forecasting method based on multi-model combination, which can combine the forecast results of multiple single runoff forecasting models through the BP neural network model, and use the Coupla function to correct the forecast results, thereby improving the runoff forecast accuracy, reducing the runoff forecast error under long forecast periods, and extending the effective forecast period of runoff forecasts.

[0073] In this embodiment, step S1 is a forecast data set of the area to be studied, and the forecast data set includes hydrological, meteorological and topographic data in the current and past time periods, and meteorological data in the forecast period; the hydrological, meteorological and topographic data include but are not limited to runoff, rainfall, evaporation, soil moisture, terrain characteristics, vegetation coverage data, etc.;

[0074] Specifically, the hydrological, meteorological and topographic data obtained need to cover the upstream catchment area of ​​the forecast section;

[0075] Among them, the time scale requirements are as follows:

[0076] If the forecast time scale is hourly and daily, the historical runoff, rainfall, evaporation and soil moisture in the acquired data should be at least hourly, and the rest of the data can be daily;

[0077] If the forecast time scale is ten days, the historical runoff, rainfall, evaporation and soil moisture in the acquired data should be at least on a daily scale, and the remaining data can be on a ten-day scale;

[0078] If the forecast time scale is monthly, all data obtained should be at least on a monthly scale.

[0079] In this embodiment, step S2 preprocesses the forecast data set of the area to be studied to obtain a preprocessed data set, wherein the preprocessing step specifically includes:

[0080] S21: Obtain the set forecast time scale;

[0081] S22: Scale-convert the hydrological, meteorological and topographic data according to the forecast time scale to obtain the hydrological, meteorological and topographic data matching the forecast time scale;

[0082] Specifically, the forecast time scale meets the following requirements:

[0083] If the forecast time scale is hourly and daily, the historical runoff, rainfall, evaporation and soil moisture in the acquired data should be at least hourly, and the rest of the data can be daily;

[0084] If the forecast time scale is ten days, the historical runoff, rainfall, evaporation and soil moisture in the acquired data should be at least on a daily scale, and the remaining data can be on a ten-day scale;

[0085] If the forecast time scale is monthly, all data obtained should be at least on a monthly scale.

[0086] Specifically, when the time scale of the acquired hydrological, meteorological and topographic data does not match the forecast time scale, the runoff and soil moisture data need to be accumulated and averaged, and the rainfall and evaporation data need to be accumulated to obtain the data of the corresponding time scale. Taking the forecast time scale as hours as an example, the formula is as follows:

[0087]

[0088]

[0089] in, Represents hourly runoff or soil moisture data, Represents daily runoff or soil moisture data, is the number of hourly runoff or soil moisture data. If it is hourly runoff data, then is 24. is the hourly rainfall or evaporation data, is the daily rainfall or evaporation data, n* is the number of hourly rainfall or evaporation data, if it is hourly rainfall or evaporation data, then n* is 24.

[0090] S23: Clean and normalize the hydrological, meteorological and topographic data that match the forecast time scale in turn to obtain preprocessed results.

[0091] Specifically, data cleaning processing includes checking whether the data is complete, that is, checking whether there are missing values, outliers or duplicate values; processing outliers, that is, they can be processed by deletion, replacement (such as using the mean, median or interpolation method) or retention but marking them as outliers; deleting duplicate values ​​to avoid affecting subsequent analysis; filling missing values, that is, selecting appropriate filling methods according to the nature of the data and the distribution of missing values, such as mean filling, median filling, interpolation methods (such as linear interpolation, polynomial interpolation), etc.

[0092] Specifically, data normalization processing includes: selecting appropriate normalization methods according to the nature and distribution characteristics of hydrological, meteorological and topographic data that match the forecast time scale, such as minimum-maximum normalization (scaling the data to between 0 and 1), Z-score normalization (converting the data to a distribution with a mean of 0 and a standard deviation of 1), etc.

[0093] In this embodiment, in step S3, the preprocessed data sets are respectively input into different trained single runoff forecasting models for prediction to obtain multiple single forecasting results, wherein the single runoff forecasting model is a physical mechanism model or a data-driven model; wherein the physical mechanism model includes the watershed hydrological model SWAT and the variable infiltration capacity model, and the data-driven model includes artificial neural network, support vector machine and random forest.

[0094] Specifically, if the present invention constructs two single runoff forecast models, the two models may be based on the same or different principles, for example: two different physics-driven models, two different data-driven models, one physics-driven model and one data-driven model.

[0095] Specifically, the training process of a single runoff forecast model includes:

[0096] Obtain historical data sets for the area to be studied, i.e. hydrological, meteorological and topographic data from the past several years;

[0097] Preprocessing the historical data set of the area to be studied to obtain a preprocessed historical data set; wherein the preprocessing process refers to steps S21-S23;

[0098] The preprocessed historical data set is divided into a training set and a test set, where the last year of the historical data set is used as the test set, and the data of the remaining periods are used as the training set.

[0099] When the constructed single runoff forecast model is a physically driven model, the model structure is first defined, including the simulation of hydrological processes such as rainfall-runoff process, evapotranspiration process, soil moisture dynamics, etc., and the state variables and parameters of the model are determined; then the training set is used to adjust the model parameters until the runoff forecast value output by the model is well matched to the actual runoff value; finally, the model forecast effect is verified through the test set.

[0100] When the constructed single runoff forecast model is a data-driven model, the input factors of the model are first determined. The input factors include at least runoff and rainfall data, and the output factors are runoff forecast values. Different input factors can be adjusted to form a variety of forecast schemes. Then, the model parameters are determined through the training set, and the effects of different forecast schemes are verified through the test set.

[0101] Specifically, when testing the model, the mean relative error (MARE) is used as the evaluation index of the forecast scheme effect, where the value of MARE is , which reflects the degree of deviation between the predicted value and the measured value. The closer its value is to 0, the better the model prediction effect is. The MARE calculation formula is:

[0102]

[0103] Among them, F i，j and Q i，j They represent the predicted runoff value and the measured runoff value of the i-th forecast in the j-th forecast period respectively; N represents the total number of samples in the test set, that is, the number of measured runoff values.

[0104] In this embodiment, step S4 uses the Copula function to perform error correction on each single forecast result to obtain each corrected single forecast result; wherein the error correction process includes: based on the simulated forecast error sequence, performing error correction on the uncorrected forecast result; such as Figure 2 As shown, the steps for obtaining the simulation forecast error sequence are as follows:

[0105] S01: Obtain the measured runoff values ​​corresponding to different forecast periods in the historical forecast results;

[0106] Wherein, when the historical forecast result in step S4 is a historical single forecast result, it is obtained by predicting in a single runoff forecast model through historical hydrological, meteorological and topographic data;

[0107] When the historical forecast result in step S6 is a historical initial runoff forecast result, it is obtained by inputting each historical single forecast result into the BP neural network model.

[0108] S02: According to the measured runoff values ​​corresponding to different forecast periods, the forecast results are analyzed for errors to obtain the error sequence, i.e., the set of forecast errors under different forecast periods;

[0109] S03: Taking the forecast errors under different forecast periods as the independent variables of the Copula function, multiple forecasts are performed to obtain the error data matrix;

[0110] Specifically, the error data matrix is ​​expressed as follows:

[0111]

[0112] Where E represents the error data matrix; e n，m It represents the forecast error caused by the nth forecast in the mth forecast period; n is the total number of forecasts, and m is the total number of forecast periods;

[0113] The formula for the forecast error is as follows:

[0114]

[0115] Among them, e i，j is the forecast error generated by the i-th forecast in the j-th forecast period; i, j are sequence numbers; F i，j and Q i，j They represent the predicted runoff value and the measured runoff value in the i-th forecast in the j-th forecast period respectively.

[0116] S04: Calculate the overall probability distribution density of forecast error under different forecast periods using the kernel density estimation method, and use it as the marginal distribution function of forecast error under different forecast periods;

[0117] Specifically, the overall probability distribution density function of the forecast error in a certain forecast period is expressed as follows:

[0118]

[0119]

[0120] in, represents the overall probability distribution density function of the forecast error in the t-th forecast period; x t represents the forecast error in the tth forecast period; e w，t It represents the forecast error caused by the w-th forecast in the t-th forecast period; K' is the kernel function; h' is the window width; t is the sequence number.

[0121] S05: Based on the Copula function and the marginal distribution function of the forecast error under different forecast periods, the joint distribution function of the forecast error under different forecast periods is obtained;

[0122] Specifically, the joint distribution function of forecast errors under different forecast periods is expressed as follows:

[0123]

[0124] in, Forecast error at different forecast periods The joint distribution function of is the forecast error x in the mth forecast period m The marginal distribution function of It is a Copula function.

[0125] S06: Generate a random sequence based on the joint distribution function of forecast errors under different forecast periods, representing the cumulative distribution probability of the error distribution under different forecast periods;

[0126] S07: Based on the random sequence, the inverse cumulative distribution function of the marginal distribution of the forecast error under different forecast periods is calculated to obtain the simulated forecast error sequence;

[0127] Specifically, random sequence ;

[0128] The simulated forecast error sequence X' is expressed as follows:

[0129]

[0130] in, It represents the inverse cumulative distribution function of the marginal distribution of the forecast error for the mth forecast period.

[0131] Specifically, based on the simulated forecast error sequence, the uncorrected forecast results are corrected, including:

[0132] Add the current forecast result before error correction and the simulated forecast error sequence to obtain the current forecast result after error correction;

[0133] The formula is as follows:

[0134]

[0135]

[0136]

[0137] Where F is the current forecast result before error correction; M is the current forecast result after error correction; n，m It represents the forecast value of the nth forecast after error correction in the mth forecast period.

[0138] Among them, the current forecast result after error correction in step S4 is the corrected single forecast result; and the current forecast result after error correction in step S6 is the final runoff forecast result.

[0139] In this embodiment, in step S5, each corrected single forecast result and the corresponding measured runoff value are input into the trained BP neural network model for combination to obtain an initial runoff forecast result. Specifically, the BP neural network model includes an input layer, multiple hidden layers and an output layer; the input layer includes l nodes, which are l corrected single forecast results and corresponding measured runoff values ​​respectively; wherein l corrected single forecast results are obtained by predicting l different trained single runoff forecast models, l≥2; wherein the number of hidden layers can be adjusted according to the training results of the BP neural network model; the output layer is the initial runoff forecast result;

[0140] Specifically, the zth node of the input layer It is expressed as follows:

[0141]

[0142]

[0143]

[0144] in, M represents the forecast value of the zth single runoff forecast model corrected under m different forecast periods, that is, the forecast result of the zth single runoff forecast model after error correction; z is the sequence number, z∈l; z n，m It represents the forecast value of the nth forecast of the zth single runoff forecast model after error correction in the mth forecast period; Representation and The corresponding runoff measurement results; Q z n，m It represents the measured runoff value of the nth forecast of the zth single runoff forecast model in the mth forecast period.

[0145] Specifically, the main parameters that need to be adjusted when training the BP neural network model include the learning step, hidden layer and momentum factor. During the training process, the learning step is set to 0.01-0.1, the number of hidden layers is set to 3-5, and the momentum factor is set to 0.9-0.99. The training set data of the BP neural network model is input into the BP neural network, and the main parameters are continuously adjusted until MARE is minimized.

[0146] In this embodiment, step S6 uses the Copula function to perform error correction on the initial runoff forecast result to obtain the final runoff forecast result, wherein the error correction process is the same as the error correction process in step S4 and will not be repeated here.

[0147] Example 2

[0148] This embodiment provides a runoff forecasting method based on multi-model combination, taking the daily runoff forecast of the Danba section of the Dadu River as an example, with a forecast period of 1-10 days. The historical runoff data of the Danba hydrological station from January 2010 to August 2022 and the historical rainfall data of 23 rainfall stations in the upstream area of ​​the Danba section are collected. The original data is at the hourly scale, and the daily scale data is obtained after accumulation processing.

[0149] The data from January 2010 to August 2021 are used as the training set, and the data from September 2021 to August 2022 are used as the test set. The historical rainfall data of 23 rain gauges and the historical runoff data of Danba Hydrological Station are used as input, and the predicted runoff of Danba Hydrological Station is used as output to construct two different single runoff forecasting models (data-driven models), namely, the LSTM model and the random forest model.

[0150] The data from January 2010 to August 2021 are used as the historical sample library, and the data from September 2021 to August 2022 are used as the test set to build a similarity runoff forecast model (physical driven model);

[0151] Based on the above three single runoff forecast models, rolling forecast is used to obtain the runoff forecast results with a forecast period of 1-10 days.

[0152] The forecast results of LSTM, random forest and similarity runoff forecast models are compared with the actual runoff measured values. The forecast error in the 1-10 day forecast period is used as the independent variable of the Copula function. The distribution law of the forecast error is obtained through the Copula function to realize the correction of the runoff forecast results of LSTM, random forest and similarity runoff forecast models.

[0153] The error-corrected forecast results of LSTM, random forest and similarity runoff forecast models and the measured runoff values ​​are input into the BP neural network to obtain the initial forecast results.

[0154] Compare the initial forecast results with the measured runoff data, use the Copula function to correct the forecast results of the combined model, and obtain the final forecast results. Figure 3-Figure 8 As shown in the figure, the forecast results of the initial forecast results before and after the error correction for the forecast period of 1 day, 5 days and 10 days are shown respectively; by comparing with the measured values, it can be clearly seen that the corrected curve is more in line with the measured curve, thus having a higher accuracy;

[0155] Among them, the accuracy evaluation table of the initial forecast results before and after error correction is as follows:

[0156]

[0157] In the table, NS (Nash coefficient), RMSE (root mean square error), MAE (mean absolute error), and MARE (mean relative error) are evaluation indicators. From the above table, it can be seen that using the Copula function for error correction is very necessary to improve the forecast accuracy;

[0158] Example 3

[0159] This embodiment provides a runoff forecasting system based on a combination of multiple models, including:

[0160] An acquisition module is used to acquire a forecast data set for the area to be studied, wherein the forecast data set includes hydrological, meteorological and topographic data in the current and past time periods, and meteorological data in the forecast period;

[0161] A preprocessing module is used to preprocess the forecast data set of the area to be studied to obtain a preprocessed data set;

[0162] The single prediction module is used to input the preprocessed data set into different trained single runoff prediction models for prediction, and obtain multiple single prediction results;

[0163] The first correction module is used to use the Copula function to perform error correction on each single forecast result to obtain each corrected single forecast result;

[0164] The combination module is used to input each corrected single forecast result into the trained BP neural network model for combination to obtain the initial runoff forecast result;

[0165] The second correction module is used to use the Copula function to perform error correction on the initial runoff forecast result to obtain the final runoff forecast result.

[0166] Example 4

[0167] This embodiment provides a computer-readable storage medium storing a computer program, which, when executed, implements the runoff forecasting method based on multi-model combination described in embodiment 1 of claim 1.

[0168] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0169] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0170] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0171] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the enlightenment of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the purpose of the present invention and the claims, which all fall within the protection of the present invention.

Claims

1. A runoff forecasting method based on multi-model combination, characterized in that: include: Obtaining a forecast data set for the area to be studied, the forecast data set including hydrological, meteorological and topographic data in the current and past time periods, and meteorological data in the forecast period; Preprocessing the forecast data set of the area to be studied to obtain a preprocessed data set; The preprocessed data sets are input into different trained single runoff forecast models for prediction, and multiple single forecast results are obtained; Using the Copula function, the error of each single forecast result is corrected to obtain each corrected single forecast result; Input each corrected single forecast result into the trained BP neural network model for combination to obtain the initial runoff forecast result; The Copula function is used to correct the error of the initial runoff forecast result to obtain the final runoff forecast result.

2. The runoff forecasting method based on multi-model combination according to claim 1 is characterized in that: The single runoff forecast model is a physical mechanism model or a data-driven model; the physical mechanism model includes the basin hydrological model SWAT and the variable infiltration capacity model, and the data-driven model includes artificial neural network, support vector machine and random forest.

3. The runoff forecasting method based on multi-model combination according to claim 1 is characterized in that: The pre-processing process includes: Scale-convert the hydrological, meteorological and topographic data according to the forecast time scale to obtain the hydrological, meteorological and topographic data that match the forecast time scale; The hydrological, meteorological and topographic data that match the forecast time scale are cleaned and normalized in turn to obtain the preprocessed results.

4. The runoff forecasting method based on multi-model combination according to claim 1 is characterized in that: The error correction process includes: Obtaining the measured runoff values ​​corresponding to different forecast periods in the historical forecast results; wherein the historical forecast results are historical single forecast results or historical initial runoff forecast results; the historical single forecast results are obtained by predicting the historical hydrological, meteorological and topographic data in a single runoff forecast model; the historical initial runoff forecast results are obtained by inputting each historical single forecast result into a BP neural network model; According to the measured runoff values ​​corresponding to different forecast periods, the forecast results are analyzed for errors, and the error series is obtained, that is, the set of forecast errors under different forecast periods. The forecast errors under different forecast periods are used as the independent variables of the Copula function, and multiple forecasts are performed to obtain the error data matrix; The kernel density estimation method is used to calculate the overall probability distribution density of forecast errors under different forecast periods, and it is used as the marginal distribution function of forecast errors under different forecast periods. Based on the Copula function and the marginal distribution function of the forecast error under different forecast periods, the joint distribution function of the forecast error under different forecast periods is obtained. Generate a random sequence based on the joint distribution function of forecast errors under different forecast periods to represent the cumulative distribution probability of the error distribution under different forecast periods; Based on the random sequence, the inverse cumulative distribution function of the marginal distribution of forecast errors under different forecast periods is calculated to obtain the simulated forecast error sequence; The current forecast result before error correction is added to the simulated forecast error sequence to obtain the current forecast result after error correction.

5. The runoff forecasting method based on multi-model combination according to claim 4 is characterized in that: The error data matrix is ​​expressed as follows: ； Where E represents the error data matrix; e n，m It represents the forecast error generated by the nth forecast in the mth forecast period; the formula of forecast error is as follows: ； Among them, e i，j is the forecast error generated by the i-th forecast in the j-th forecast period; i, j are sequence numbers; F i，j and Q i，j They represent the predicted runoff value and the measured runoff value in the i-th forecast in the j-th forecast period respectively.

6. The runoff forecasting method based on multi-model combination according to claim 5 is characterized in that: The overall probability distribution density function of the forecast error in a certain forecast period is expressed as follows: ； ； in, represents the overall probability distribution density function of the forecast error in the t-th forecast period; x t represents the forecast error in the tth forecast period; e w，t It represents the forecast error caused by the w-th forecast in the t-th forecast period; K' is the kernel function; h' is the window width.

7. The runoff forecasting method based on multi-model combination according to claim 6 is characterized in that: The joint distribution function of forecast error under different forecast periods is expressed as follows: ； in, Forecast error at different forecast periods The joint distribution function of is the forecast error x in the mth forecast period m The marginal distribution function of It is a Copula function.

8. The runoff forecasting method based on multi-model combination according to claim 1 is characterized in that: The BP neural network model includes an input layer, multiple hidden layers and an output layer; The input layer includes l nodes, which are l corrected single forecast results and corresponding measured runoff values; wherein the l corrected single forecast results are obtained by predicting l different trained single runoff forecast models, l≥2; The number of hidden layers can be adjusted according to the training results of the BP neural network model; The output layer is the initial runoff forecast result.

9. The runoff forecasting method based on multi-model combination according to claim 1, characterized in that: The training process of the BP neural network model includes: Using each historical single forecast result data set after error correction and the corresponding measured runoff value as a training set; the historical single forecast result data set is predicted by a single runoff forecast model; Train the BP neural network model based on the data in the training set to obtain the optimal model parameters; According to the optimal model parameters, confirm the trained BP neural network model.

10. A runoff forecasting system based on multi-model combination, characterized in that: include: An acquisition module is used to acquire a forecast data set for the area to be studied, wherein the forecast data set includes hydrological, meteorological and topographic data in the current and past time periods, and meteorological data in the forecast period; A preprocessing module is used to preprocess the forecast data set of the area to be studied to obtain a preprocessed data set; The single prediction module is used to input the preprocessed data set into different trained single runoff prediction models for prediction, and obtain multiple single prediction results; The first correction module is used to use the Copula function to perform error correction on each single forecast result to obtain each corrected single forecast result; The combination module is used to input each corrected single forecast result into the trained BP neural network model for combination to obtain the initial runoff forecast result; The second correction module is used to use the Copula function to perform error correction on the initial runoff forecast result to obtain the final runoff forecast result.

Images