Deep learning multi-model runoff prediction method and system embedded with runoff production mechanism

Abstract

The invention discloses a runoff production mechanism embedded deep learning multi-model runoff prediction method and system, and belongs to the field of hydraulic engineering application, and the method comprises the steps: obtaining data which comprises meteorological data and hydrological data; inputting the meteorological data into three neural networks respectively embedded with a GR4J runoff production mechanism, a common linear runoff production mechanism and an index bucket runoff production mechanism to obtain three initial runoff production processes; inputting the three initial runoff production processes into a long-short term memory neural network model to obtain three initial runoff production processes; a peak value constraint, a peak present time constraint and a monotonicity constraint are introduced into the long and short term memory neural network model; and inputting the three initial outflow processes into a Bayesian average model to obtain a predicted outflow process, and calculating to obtain a final outflow process in combination with an actually measured outflow process in hydrological data. According to the method, the uncertainty of model prediction can be reduced to the greatest extent, and the precision of runoff prediction is ensured.

Description

Technical Field

[0001] The present invention belongs to the application field of water conservancy engineering, and specifically relates to a deep learning multi-model runoff prediction method and system embedded with runoff generation mechanism. Background Art

[0002] Efficient and reliable hydrological models can accurately obtain runoff information and are an important technical support for water resources management and flood prevention and mitigation. Process-based hydrological models are equipped with explicit hydrological knowledge, such as conservation of mass and energy and runoff generation upon storage. Currently, the most widely used physical models are the Exponential Bucket Hydrologic Model (EXP-Hydro), the Simple Linear Model (SLM), and GR4J (Génie Rural à 4 paramètres Journalier), which can achieve good accuracy in predicting runoff processes. However, when physical models are applied to catchments where the hydrological conditions are inconsistent with the model assumptions, they may produce some unsatisfactory results.

[0003] With the rapid development of computer technology, machine learning (ML) models, particularly deep learning (DL) models, have become key tools for studying watershed responses to meteorological changes and for multi-basin runoff forecasting due to their superior data processing capabilities and computational efficiency. For example, random forests, support vector regression, artificial neural networks, and recurrent neural networks (RNNs) are commonly used ML methods for runoff forecasting. Long short-term memory (LSTM) neural networks are state-of-the-art methods for runoff forecasting, featuring specialized memory cells that selectively retain information for extended periods. However, many scholars in the hydrological community remain cautious about adopting ML methods. Because ML models are entirely data-driven, they inherently lack representation of internal states and physical processes. When applied to watersheds where hydrological processes are affected by external disturbances (such as climate change or human activities), ML models can produce unreliable predictions and significantly negatively impact prediction accuracy.

[0004] Given the physical interpretability provided by physical models and the powerful data mining capabilities of neural network models, the synergy between these two approaches has become an attractive research topic in recent years. Physically interpretable neural network models, also known as hybrid models, have attracted increasing attention. Hybrid models can be categorized as loosely coupled and tightly coupled. Loosely coupled approaches include: 1) using a neural network model as a postprocessor to correct errors in the process-based model; 2) replacing some hydrological processes in the physical model with a neural network model; 3) using the output data of the process-based model as synthetic data for training the neural network model; and 4) incorporating physical constraints as penalties into the loss function of the DL model. These computationally efficient approaches are relatively easy to implement, but they all require the inclusion of physical knowledge in the ML model in a closed-form representation. However, due to the highly dynamic nature of the Earth system, its intermediate processes are difficult to capture in closed-form representation. In contrast, tightly coupled models embed a hydrological model with numerous parameters into the framework by adding one or more physical layers, transforming the neural network into one that is more consistent with geophysical processes. However, this approach is more complex than loosely coupled approaches and places high demands on the programming skills of hydrologists.

[0005] In addition, there are many sources of uncertainty in runoff prediction using neural network models, such as parameter uncertainty, model structure uncertainty, and input data uncertainty. These uncertainties will ultimately directly affect the model output results. Hydrological scholars have proposed a multi-model ensemble forecasting method to quantify and reduce the uncertainty of hydrological forecasts. By applying a set of different models for hydrological forecasting and then integrating the forecast results using post-processing methods, it is possible to capture the different characteristics of the hydrological series and take into account structural uncertainty, thereby effectively improving the forecast accuracy. The Bayesian Model Averaging (BMA) is a commonly used multi-model ensemble forecasting method that can be applied to runoff forecasting to reduce the uncertainty generated in the forecasting process.

[0006] Therefore, it is necessary to design a deep learning multi-model runoff prediction method and system that embeds the runoff generation mechanism to address the above problems. Summary of the Invention

[0007] The purpose of the present invention is to provide a deep learning multi-model runoff prediction method that embeds the runoff generation mechanism to address the various sources of uncertainty and accuracy issues in runoff prediction. Three different runoff generation mechanisms are embedded in the neural network, and the LSTM neural network is used to replace the confluence process. In addition, the Bayesian average model is combined with the measured outflow process in the hydrological data to calculate the weights of the three models and obtain the final outflow process, which ensures the integrity and interpretability of the overall process, makes the runoff predicted by the deep learning model more in line with hydrological principles, and improves the prediction accuracy, computational efficiency and user-friendliness.

[0008] According to one aspect of this specification, a deep learning multi-model runoff prediction method embedded with runoff generation mechanism is provided, comprising:

[0009] S1. Acquire data, wherein the data includes meteorological data and hydrological data;

[0010] S2. Input meteorological data into three neural networks embedded with the GR4J runoff generation mechanism, the ordinary linear runoff generation mechanism, and the exponential bucket runoff generation mechanism, respectively, to obtain three preliminary runoff generation processes;

[0011] S3. Inputting the three preliminary flow generation processes into a long short-term memory neural network model to obtain three preliminary flow outflow processes; the long short-term memory neural network model introduces a peak size constraint, a peak occurrence time constraint, and a monotonicity constraint;

[0012] S4. Input the three preliminary outflow processes into the Bayesian average model to obtain the predicted outflow process, and combine it with the measured outflow process in the hydrological data to calculate the final outflow process.

[0013] Furthermore, the process of embedding the flow generation mechanism into the neural network in S2 includes:

[0014] Using physical formulas Replace nonlinear activation functions in neural network models ,parameter Replace some weight coefficients in the neural network model , soil moisture content Replace state variables , and obtain a neural network embedded in the flow generation mechanism.

[0015] Furthermore, the neural network embedded in the flow generation mechanism is expressed as:

[0016]

[0017]

[0018] in, They are rainfall, sunshine duration, and temperature. is the soil moisture content, is the updated formula for soil moisture content, is the update formula of each flow generation mechanism, are the physical parameters in each runoff generation mechanism.

[0019] Furthermore, the S3 further includes:

[0020] By adding peak size constraints, peak time constraints, and monotonicity constraints to the original loss function of the long short-term memory neural network model, the new loss function expression is:

[0021]

[0022] in, is the original loss function, is the peak size constraint, is the peak time constraint, is the monotonicity constraint, It is a custom weight.

[0023] Furthermore, the S3 further includes:

[0024] The peak size constraint is calculated by calculating the relative error between the annual simulated peak and the measured peak, and adding the corresponding weight to the loss function;

[0025] The peak time constraint is calculated by calculating the relative error between the annual simulated peak time and the measured peak time, and adding it to the loss function according to the corresponding weight;

[0026] The monotonicity constraint is based on the monotonic relationship between runoff generation and runoff inflow, and the loss is calculated using a response function.

[0027] Furthermore, the S4 further includes:

[0028] The posterior distribution is obtained based on the predicted outflow process and the measured outflow process in the hydrological data, and the mean and variance of the posterior distribution are iteratively calculated using the expectation maximization algorithm to obtain the weights of the three neural networks;

[0029] Based on the weights of the three neural networks and the three preliminary outflow processes, the final outflow process is obtained.

[0030] Furthermore, the final outflow process is expressed as:

[0031]

[0032] in, Represents the final outflow process, represents the weights of the three neural networks, represents the initial outflow process of the three neural networks, .

[0033] According to one aspect of this specification, a deep learning multi-model runoff prediction system embedded with runoff generation mechanism is provided, comprising:

[0034] A data acquisition module, used to acquire data, including meteorological data and hydrological data;

[0035] The preliminary runoff generation process module is used to input meteorological data into three neural networks embedded with the GR4J runoff generation mechanism, the ordinary linear runoff generation mechanism, and the exponential bucket runoff generation mechanism, respectively, to obtain three preliminary runoff generation processes;

[0036] A preliminary outflow process module is used to input the three preliminary outflow processes into a long short-term memory neural network model to obtain three preliminary outflow processes; the long short-term memory neural network model introduces peak size constraints, peak appearance time constraints, and monotonicity constraints;

[0037] The outflow process module is used to input the three preliminary outflow processes into the Bayesian average model to obtain the predicted outflow process, and combine it with the measured outflow process in the hydrological data to calculate the final outflow process.

[0038] According to one aspect of the present specification, an electronic device is provided, comprising a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the deep learning multi-model runoff prediction method embedded with the runoff generation mechanism are implemented.

[0039] According to one aspect of the present specification, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of the deep learning multi-model runoff prediction method embedded in the runoff generation mechanism are implemented.

[0040] Compared with the prior art, the present invention has the following beneficial effects:

[0041] 1. The present invention proposes to integrate three different flow generation mechanisms into a neural network, which not only overcomes the shortcomings of the harsh preconditions of ordinary physical models, but also overcomes the shortcomings of the lack of interpretability of machine learning models.

[0042] 2. By embedding the runoff generation mechanism into the neural network and replacing the confluence process with the LSTM model, the present invention constructs three neural networks and three long-short-term memory neural network models with physical constraints. This can ensure the integrity of the overall process, high computational efficiency and user-friendliness, making the runoff predicted by multiple models more consistent with hydrological principles.

[0043] 3. The present invention uses the Bayesian average model to process the three preliminary outflow processes, combines the measured outflow processes in the hydrological data, calculates the final outflow process, and realizes the prediction of runoff. It can not only minimize the uncertainty of the model prediction to the greatest extent, but also ensure the accuracy of the prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0045] Figure 1 is a flow chart of a method according to an embodiment of the present invention;

[0046] Figure 2 This is a diagram of the architecture of a recurrent neural network embedded with a flow generation mechanism in an embodiment of the present invention;

[0047] Figure 3 Schematic diagram of the outflow process of the GR4J flow generation mechanism embedded in an embodiment of the present invention;

[0048] Figure 4 Schematic diagram of the outflow process of embedding a common linear flow generation mechanism in an embodiment of the present invention;

[0049] Figure 5 Schematic diagram of the outflow process of the embedded index bucket flow generation mechanism in an embodiment of the present invention;

[0050] Figure 6 Schematic diagram of the outflow process after integration using the BMA model in an embodiment of the present invention. DETAILED DESCRIPTION

[0051] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0052] like Figure 1As shown, an embodiment of the present invention provides a deep learning multi-model runoff prediction method embedded in the runoff generation mechanism, including: step 1, basic data preparation; step 2, constructing three neural network models based on the runoff generation mechanism; step 3, constructing an LSTM model with physical constraints as a confluence module; step 4, inputting the preliminary outflow process calculated by the three neural network models into the BMA model to calculate the final multi-model runoff prediction result.

[0053] Specifically, the embodiment of the present invention also provides the specific content of step 1. The basic data mainly includes a meteorological data set (daily rainfall, daily maximum / minimum temperature, sunshine duration, etc.) that drives the neural network and a hydrological data set (measured daily runoff) that supervises the training of the neural network. The data set is divided into a training set and a validation set in a ratio of 7:3.

[0054] Specifically, this embodiment of the present invention obtains daily-scale hydrometeorological and watershed attribute data from the open-source, large-scale CAMELS dataset of watershed properties and meteorology. This dataset includes 671 US watersheds with minimal human disturbance. Only 569 watersheds with complete, continuous hydrological observations from October 1, 1985, to December 31, 2008 were considered. The neural network model requires rainfall, temperature, and sunshine duration as input to predict target runoff. The training period was 1985-2001, and the validation period was 2002-2008, with a training-to-validation ratio of 7:3. This embodiment of the present invention uses only one of these watersheds (numbered 01013500) as an example to demonstrate the method.

[0055] Specifically, this embodiment of the present invention also provides the details of step 2, embedding the GR4J runoff generation mechanism, the common linear runoff generation mechanism, and the exponential bucket runoff generation mechanism into neurons of a recurrent neural network, thereby constructing neural network models that embed the GR4J runoff generation mechanism, the common linear runoff generation mechanism, and the exponential bucket runoff generation mechanism. Simultaneously, the meteorological dataset preprocessed in step 1 is input as the neural network's feature value, and the three neural network models are trained to obtain three preliminary runoff generation processes.

[0056] Specifically, the formulas for the three flow generation mechanisms in step 2 are as follows:

[0057] (1) GR4J runoff mechanism

[0058] Effective rainfall and residual evaporation capacity Calculation:

[0059]

[0060]

[0061] in, represents the rainfall at the current time step t, Represents the actual evaporation at the current time step t.

[0062] like >0, then effective rainfall part of The remaining part will enter the runoff reservoir and the remaining part will enter the runoff-generating reservoir; is the rainfall that replenishes the runoff reservoir, expressed by soil moisture content composition.

[0063]

[0064] in, represents a dimensionless parameter.

[0065] like =0, >0, is the evapotranspiration of the runoff reservoir.

[0066]

[0067] Soil moisture content Continuously updated, the calculation formula is:

[0068]

[0069] Therefore, the flow rate of the runoff reservoir Calculated by the following formula:

[0070]

[0071] Subtract The water storage capacity of the runoff reservoir is:

[0072]

[0073] Total production flow The update formula is:

[0074]

[0075] (2) Ordinary linear runoff mechanism

[0076] Actual evapotranspiration The calculation formula is:

[0077]

[0078]

[0079] in, represents potential evapotranspiration, represents the groundwater storage depth at the current time step t, The maximum allowable storage depth of groundwater.

[0080] Soil moisture content The updated expression is:

[0081]

[0082] Flow rate of each part The updated expression is:

[0083]

[0084]

[0085]

[0086] The final flow rate is:

[0087]

[0088] in, The maximum soil water content, Indicates the maximum allowable storage depth of groundwater, represents the maximum capacity of groundwater, represents the runoff caused by rainfall, z is a parameter that characterizes the groundwater loss, and k is a dimensionless parameter. It is a parameter that characterizes the soil moisture resistance coefficient and affects the resistance of soil moisture to evapotranspiration.

[0089] (3) Index Bucket Flow Mechanism

[0090] and Usually rainfall and daily temperature Calculation yields:

[0091]

[0092]

[0093] in, Indicates the amount of snowmelt, Indicates rainfall.

[0094] Snowmelt and temperature and snow capacity related, is a parameter that characterizes the heat-day factor, is the maximum temperature, is the minimum temperature:

[0095]

[0096] Evaporation Calculated as potential evapotranspiration part of the watershed, which depends on the current soil moisture content in the basin. Relative to maximum water storage capacity Size:

[0097]

[0098] Snow capacity and soil moisture content The update formula is:

[0099]

[0100]

[0101] The total flow is the base flow in the collection bucket When the water collection bucket is saturated, the excess storage flow The sum of , respectively expressed as:

[0102]

[0103]

[0104] (twenty four)

[0105] Specifically, the embedded flow generation mechanism process in step 2 is as follows:

[0106] First, the traditional recurrent neural network formula is:

[0107] (25)

[0108] (26)

[0109] in, For input, is the cell state, is the cell state at the previous moment, For output, are the nonlinear activation functions of cell state and output, respectively. are the weight matrices of cell state and output, respectively. is the bias between the cell state and output.

[0110] Then the flow generation mechanism is embedded in the variant recurrent neural network, and the formula is as follows:

[0111] (27)

[0112] (28)

[0113] in, They are rainfall, sunshine duration, and temperature. is the soil moisture content, is the soil moisture content at the previous moment, is the updated formula for soil moisture content, is the updated formula for flow production, are the physical parameters in each runoff generation mechanism.

[0114] Specifically, during the training process, the weight coefficients and physical parameters of the recurrent neural network are updated through the neural network and adjusted by the back propagation of the loss function to achieve the update of soil moisture content (i.e., cell state variables), so that the update of neurons strictly abides by the water balance principle and enhances the interpretability of machine learning.

[0115] Specifically, if Figure 2 As shown, the embodiment of the present invention provides a recurrent neural network architecture before and after the embedded flow generation mechanism. Figure 2 (b) to (d) illustrate the mechanisms of GR4J runoff generation, ordinary linear runoff generation, and exponential bucket runoff generation, respectively. The input variables of each model, the physical parameters that replace some of the weights of the neural network, and the state variables of the model are introduced. By inputting the rainfall, sunshine duration, and temperature of the basin numbered 01013500 into the neural network model, the runoff generation process of the basin and the changing process of soil moisture content can be obtained.

[0116] Specifically, an embodiment of the present invention also provides the specific content of step 3, which respectively connects two LSTM models to the three neural network models embedded with the flow generation mechanism, and adds physical constraints such as peak size, peak time and monotonicity of flow generation and convergence to the loss function of the LSTM. The preliminary flow generation process is input into the LSTM model, and the preliminary outflow processes of the three models are calculated respectively.

[0117] Specifically, the peak size constraint is mainly achieved by calculating the relative error between the annual simulated peak and the measured peak, and adding it to the loss function according to the corresponding weight. The formula is as follows:

[0118] (29)

[0119] in, is the peak value of the simulated flow rate, is the peak value of the measured flow rate.

[0120] Specifically, the peak time constraint is mainly achieved by calculating the relative error between the annual simulated peak time and the measured peak time, and adding it to the loss function according to the corresponding weight. The formula is as follows:

[0121] (30)

[0122] in, is the peak time of simulated traffic, is the peak time of the measured flow rate.

[0123] Specifically, the runoff generation and runoff confluence monotonicity constraint means that when runoff generation increases, runoff confluence also increases accordingly. There is a monotonic relationship between runoff generation and runoff confluence, as shown in the following formula:

[0124] (31)

[0125] (32)

[0126] in, is the discharge at the current time step, It is a small increase in flow rate. is the response function, excluding monotonic terms and retaining non-monotonic terms.

[0127] All the above physical constraints are supplementary modifications to the loss function. The unmodified loss function is:

[0128] (33)

[0129] in, is the measured flow rate for a single time step, is the simulated flow rate for a single time step, is the average value of the simulated flow rate.

[0130] The loss function after adding physical constraints is:

[0131] (34)

[0132] in, are custom weights, and the best loss function weights are determined by trial and error: .

[0133] Specifically, after the LSTM model with physical constraints is constructed, it is connected to the neural network embedded in the runoff generation mechanism described in step 2. The preliminary runoff generation process is input into the LSTM model. Meteorological factors such as rainfall, temperature, and daily duration are used as feature values. The measured runoff is used as the supervised training data set to drive the LSTM training. The hyperparameters in the training process are determined by trial and error, including: learning rate = 0.001, packet loss rate = 0.1, batch size = 128, and number of iterations = 500. Finally, the outflow process is calculated, as shown in Figure 2. Figure 3-Figure 5 shown.

[0134] Specifically, an embodiment of the present invention also provides the specific content of step 4, inputting the preliminary outflow process calculated by the three neural network models into the Bayesian average model, calculating the posterior distribution of the preliminary outflow process and the measured outflow process, calculating the weight of each model, and combining the weights to calculate the final outflow process.

[0135] Specifically, the three preliminary outflow processes calculated in step 3 are selected as ensemble members and input into the BMA model. The probability density function (PDF) of the Bayesian average probability prediction model is shown below:

[0136] (35)

[0137] in, (k=1,2,3) is the prediction set of all models, represents the predictor variable, is the observation data with a data length of t, Is a given prediction model and observational data The posterior distribution of . Let the weight of the model be ,have , for The variance of , then formula (35) can be expressed as:

[0138] (36)

[0139] The mean and variance of the posterior distribution predicted by the BMA model are expressed as follows:

[0140] (37)

[0141] (38)

[0142] The variance of the BMA model is essentially a measure of BMA uncertainty. The BMA parameters are generally calculated iteratively using the expectation maximization algorithm. and .

[0143] After calculating the weights of each model, the final outflow process is calculated based on the weights:

[0144] (39)

[0145] in, is the weight of each model, is the initial outflow process of each model ( ). Calculated , substituting into the above formula we get:

[0146] (40)

[0147] The outflow process of the three models is integrated using the BMA model, and the outflow process after integration is finally calculated. , can minimize the uncertainty of deep learning methods and improve prediction accuracy, such as Figure 6 shown.

[0148] The implementation basis of each embodiment of the present invention is achieved through programmed processing by a device with processor functionality. Therefore, in engineering practice, the technical solutions and functions of each embodiment of the present invention are encapsulated into various modules. Based on this reality, on the basis of the above-mentioned embodiments, an embodiment of the present invention provides a deep learning multi-model runoff prediction system embedded in a runoff generation mechanism. This system is used to implement a deep learning multi-model runoff prediction method embedded in a runoff generation mechanism described in the above-mentioned method embodiment.

[0149] The system includes: a data acquisition module for acquiring data, wherein the data includes meteorological data and hydrological data; a preliminary runoff process module for inputting meteorological data into three neural networks respectively embedded with the GR4J runoff mechanism, the ordinary linear runoff mechanism and the exponential bucket runoff mechanism to obtain three preliminary runoff processes; a preliminary outflow process module for inputting the three preliminary runoff processes into a long short-term memory neural network model to obtain three preliminary outflow processes; the long short-term memory neural network model introduces peak size constraints, peak time constraints and monotonicity constraints; and an outflow process module for inputting the three preliminary outflow processes into a Bayesian average model to obtain a predicted outflow process, and combining the measured outflow process in the hydrological data to calculate the final outflow process.

[0150] The embodiment of the present invention provides a deep learning multi-model runoff prediction system with embedded runoff mechanism. In the existing technology, runoff prediction has multiple sources of uncertainty and accuracy problems. Several modules are used to embed three different runoff mechanisms into the neural network, and the LSTM neural network is used to replace the confluence process. In addition, the Bayesian average model is combined with the measured outflow process in the hydrological data to calculate the weights of the three models and obtain the final outflow process, which ensures the integrity and interpretability of the overall process, makes the runoff predicted by the deep learning model more in line with hydrological principles, and improves the prediction accuracy, computational efficiency and user-friendliness.

[0151] Based on the same inventive concept as the above-mentioned embodiment, an embodiment of the present invention also provides an electronic device, including a memory and a processor, the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement a deep learning multi-model runoff prediction method embedded in the runoff generation mechanism proposed in the above-mentioned embodiment.

[0152] An embodiment of the present invention also provides a computer-readable storage medium storing a computer program. When executed by a processor, this program overcomes various sources of uncertainty and accuracy issues in runoff prediction, ensuring the integrity and interpretability of the overall process. This allows runoff predictions from deep learning models to better align with hydrological principles, improving prediction accuracy, computational efficiency, and user-friendliness.

[0153] The storage medium can be any non-volatile storage device such as a hard disk, a solid-state drive, a flash drive, an optical disk, etc., which is used to store computer program code and necessary data files. The stored computer program includes: a data acquisition module, a preliminary flow generation process module, a preliminary flow outflow process module, and a flow outflow process module.

[0154] Finally, it should be noted that the above specific embodiments are merely representative examples of the present invention. Obviously, the present invention is not limited to the above specific embodiments and is susceptible to numerous variations. Any simple modifications, equivalent variations, and modifications to the above specific embodiments based on the technical essence of the present invention shall be deemed to fall within the scope of protection of the present invention.

Claims

1. A deep learning multi-model runoff prediction method embedded in runoff generation mechanism, characterized by: include: S1. Acquire data, wherein the data includes meteorological data and hydrological data; S2. Input meteorological data into three neural networks embedded with the GR4J runoff generation mechanism, the ordinary linear runoff generation mechanism, and the exponential bucket runoff generation mechanism, respectively, to obtain three preliminary runoff generation processes; S3. Inputting the three preliminary flow generation processes into a long short-term memory neural network model to obtain three preliminary flow outflow processes; the long short-term memory neural network model introduces a peak size constraint, a peak occurrence time constraint, and a monotonicity constraint; S4. Input the three preliminary outflow processes into the Bayesian average model to obtain the predicted outflow process, and combine it with the measured outflow process in the hydrological data to calculate the final outflow process.

2. A deep learning multi-model runoff prediction method embedded in runoff generation mechanism according to claim 1, characterized in that: The process of embedding the flow generation mechanism into the neural network in S2 includes: Using physical formulas Replace nonlinear activation functions in neural network models ,parameter Replace some weight coefficients in the neural network model , soil moisture content Replace state variables , and obtain a neural network embedded in the flow generation mechanism.

3. The deep learning multi-model runoff prediction method embedded in the runoff generation mechanism according to claim 2 is characterized in that: The neural network embedded in the flow generation mechanism is expressed as: , , in, They are rainfall, sunshine duration, and temperature. is the soil moisture content, is the updated formula for soil moisture content, is the update formula of each flow generation mechanism, are the physical parameters in each runoff generation mechanism.

4. The deep learning multi-model runoff prediction method embedded in runoff generation mechanism according to claim 1 is characterized in that: Said S3 further includes: By adding peak size constraints, peak time constraints, and monotonicity constraints to the original loss function of the long short-term memory neural network model, the new loss function expression is: , in, is the original loss function, is the peak size constraint, is the peak time constraint, is the monotonicity constraint, It is a custom weight.

5. The deep learning multi-model runoff prediction method embedded in runoff generation mechanism according to claim 1 is characterized in that: Said S3 further includes: The peak size constraint is calculated by calculating the relative error between the annual simulated peak and the measured peak, and adding the corresponding weight to the loss function; The peak time constraint is calculated by calculating the relative error between the annual simulated peak time and the measured peak time, and adding it to the loss function according to the corresponding weight; The monotonicity constraint is based on the monotonic relationship between runoff generation and runoff inflow, and the loss is calculated using a response function.

6. The deep learning multi-model runoff prediction method embedded in runoff generation mechanism according to claim 1 is characterized in that: Said S4 further includes: The posterior distribution is obtained based on the predicted outflow process and the measured outflow process in the hydrological data, and the mean and variance of the posterior distribution are iteratively calculated using the expectation maximization algorithm to obtain the weights of the three neural networks; Based on the weights of the three neural networks and the three preliminary outflow processes, the final outflow process is obtained.

7. The deep learning multi-model runoff prediction method embedded in runoff generation mechanism according to claim 6 is characterized in that: The final outflow process is expressed as: , in, Represents the final outflow process, represents the weights of the three neural networks, represents the initial outflow process of the three neural networks, .

8. A deep learning multi-model runoff prediction system embedded in runoff generation mechanism, characterized by: include: A data acquisition module, used to acquire data, including meteorological data and hydrological data; The preliminary runoff generation process module is used to input meteorological data into three neural networks embedded with the GR4J runoff generation mechanism, the ordinary linear runoff generation mechanism, and the exponential bucket runoff generation mechanism, respectively, to obtain three preliminary runoff generation processes; A preliminary outflow process module is used to input the three preliminary outflow processes into a long short-term memory neural network model to obtain three preliminary outflow processes; the long short-term memory neural network model introduces peak size constraints, peak appearance time constraints, and monotonicity constraints; The outflow process module is used to input the three preliminary outflow processes into the Bayesian average model to obtain the predicted outflow process, and combine it with the measured outflow process in the hydrological data to calculate the final outflow process.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the deep learning multi-model runoff prediction method embedded with runoff generation mechanism described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the deep learning multi-model runoff prediction method embedded with runoff generation mechanism described in any one of claims 1 to 7 are implemented.

Images