A waterway water regime forecasting method and system based on nested multi-neural network models

Abstract

The present invention discloses a method and system for forecasting water conditions in a waterway based on nesting multiple neural network models. The method includes training a meteorological forecast model using historical meteorological data, training a meteorological-hydrological mapping model using meteorological elements and historical hydrological data for a day predicted by the meteorological-hydrological mapping model, and training a hydrological-hydrodynamic mapping model using surface runoff flow and historical hydrodynamic data predicted by the meteorological-hydrological mapping model; and nesting multiple trained neural network models to predict the waterway water level, water depth, or flow velocity distribution for multiple days in the future based on meteorological elements for multiple consecutive days before the current day in the target watershed. The present invention aims to overcome the technology for predicting water depth and bridge navigation clearance for complex waterways, solve the problem of multi-day forecasting of shoal water depth and bridge navigation clearance, and achieve simple, efficient, accurate, physically interpretable, and robust predictions of waterway water level, water depth, or flow velocity distribution for multiple stations in the target watershed for multiple days in the future.

Description

Technical Field

[0001] The present invention relates to the technical field of waterway navigation water condition prediction, and in particular to a waterway water condition prediction method and system based on nested multi-neural network models. Background Art

[0002] Water transport plays a vital role in the transportation industry. With growing demand for shipping, building safe, convenient, efficient, green, economical, inclusive, and resilient waterways is crucial. Water levels, as a key indicator for maintaining waterway dimensions, directly determine channel size and serve as a crucial factor in guiding ship loading and ensuring safe navigation. Shoal depth, bridge clearance, and lock threshold depth are key factors in developing safe, convenient, efficient, and economical waterways. Extreme weather conditions that cause excessively high or low water levels can hinder safe water transport operations and easily cause ships to capsize or run aground, leading to safety risks and property losses. Accurately forecasting water conditions and protecting the "golden line" of waterway depth in the shipping network are crucial to ensuring safe and smooth waterways. However, current traditional waterway water level prediction methods offer good accuracy for one to two days, but struggle to meet the timeliness requirements of the waterway network's seven-day transit cycle. Summary of the Invention

[0003] The technical problem to be solved by the present invention is as follows: In response to the above-mentioned problems of the prior art, a method and system for forecasting water conditions in a waterway based on nested multi-neural network models are provided. The present invention aims to overcome the technology for predicting the water depth of complex waterways and the navigation clearance of bridges, solve the multi-day forecasting problem of the water depth of shallows and the navigation clearance of bridges, and realize simple, efficient, accurate, physically interpretable and robust prediction of part or all of the waterway water level, water depth or flow velocity distribution of multiple stations in the target basin for multiple days in the future.

[0004] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0005] A waterway water regime forecasting method based on nested multi-neural network models includes the following steps:

[0006] 1) Collect historical meteorological data of the target basin and use it to build a training dataset to train the weather forecast model. This will establish a mapping relationship between the meteorological elements of the past multiple consecutive days and the corresponding meteorological elements of the next day for the target basin.

[0007] 2) Collect historical hydrological data for the target basin. Use the daily meteorological elements predicted by the meteorological forecast model and the corresponding surface runoff flow in the historical hydrological data to construct a training dataset to train the meteorological-hydrological mapping model, so that it can establish a mapping relationship between daily meteorological elements and surface runoff flow for the target basin.

[0008] 3) Collect historical hydrodynamic data for the target basin, and use the surface runoff predicted by the meteorological-hydrological mapping model and part or all of the corresponding channel water level, water depth, or flow velocity distribution in the historical hydrodynamic data to construct a training dataset to train the hydrological-hydrodynamic mapping model, so that it can establish a mapping relationship between the surface runoff and part or all of the channel water level, water depth, or flow velocity distribution for a day for the target basin;

[0009] 4) Obtain meteorological elements for multiple consecutive days before the current day in the target basin, use the trained meteorological forecast model to predict the meteorological elements for the first day in the future using the meteorological elements for multiple consecutive days before the current day, and then use the predicted meteorological elements for the next day as known quantities to use the trained meteorological forecast model to roll-forward predict the meteorological elements for the next two days until the meteorological elements for multiple days in the future are predicted; for the predicted meteorological elements for multiple days in the future, use the trained meteorological-hydrological mapping model to predict the corresponding surface runoff flow for each day, and use the trained hydrological-hydrodynamic mapping model to obtain part or all of the corresponding channel water level, water depth or flow velocity distribution for the predicted surface runoff flow, thereby obtaining part or all of the channel water level, water depth or flow velocity distribution for multiple days in the future.

[0010] Optionally, the meteorological elements in step 1) include regional two-dimensional distribution of daily average temperature, humidity, rainfall and radiation values.

[0011] Optionally, when collecting historical hydrological data of the target basin in step 2), the surface runoff flow of the target basin under the two-dimensional distribution of different meteorological elements is calculated using a hydrological numerical model, and the calculation function expression of the surface runoff flow is:

[0012] ,

[0013] in, is the surface runoff flow, is the flow rate, is the sink flow, is the evaporation amount.

[0014] Optionally, the meteorological-hydrological mapping model is composed of two sub-models: a natural surface runoff sub-model and a hub scheduling interference sub-model. The natural surface runoff sub-model is used to map the corresponding initial surface runoff flow according to the input meteorological elements of the day, and the hub scheduling interference sub-model is used to predict the surface runoff flow correction value based on the initial surface runoff flow predicted by the natural surface runoff sub-model. The surface runoff flow output by the meteorological-hydrological mapping model is the sum of the initial surface runoff flow and the surface runoff flow correction value.

[0015] Optionally, the functional expression of the loss function used in the training of the meteorological-hydrological mapping model is:

[0016] ,

[0017] ,

[0018] in, is the loss of the meteorological-hydrological mapping model, is the hub scheduling interference sub-model, is the initial surface runoff flow predicted by the natural surface runoff sub-model, is the error of surface runoff flow predicted by the natural surface runoff sub-model, is the actual measured value of surface runoff flow.

[0019] Optionally, the network model adopted by the hub scheduling interference sub-model is a variational autoencoder network model; the weather forecast model, the natural surface runoff sub-model, and the hydrological-hydrodynamic mapping model are one of a convolutional neural network, a Fourier operator neural network, and a graph neural network.

[0020] Optionally, when collecting historical hydrodynamic data of the target basin in step 3), it includes conducting multiple sets of numerical simulations under different surface runoff flow boundaries of the target basin based on a preset channel hydrodynamic model, and calculating part or all of the channel water level, water depth or flow velocity distribution at multiple stations in the target basin.

[0021] In addition, the present invention also provides a waterway water condition forecasting system based on nested multiple neural network models, including a microprocessor and a memory connected to each other, and the microprocessor is programmed or configured to execute the waterway water condition forecasting method based on nested multiple neural network models.

[0022] In addition, the present invention also provides a computer-readable storage medium, which stores a computer program or instruction, and the computer program or instruction is programmed or configured to execute the waterway water condition forecasting method based on nested multiple neural network models through a processor.

[0023] In addition, the present invention also provides a computer program product, including a computer program or instructions, which are programmed or configured to execute the waterway water condition forecasting method based on nested multiple neural network models through a processor.

[0024] Compared with the prior art, the present invention can mainly achieve the following beneficial effects:

[0025] 1. The present invention adopts the nesting of meteorological forecast model, meteorological-hydrological mapping model and hydrological-hydrodynamic mapping model to realize waterway water situation forecast. Users only need to input meteorological data of the past and future days to train the above-mentioned nested multiple neural network models. The trained nested multiple neural network models can use the meteorological elements of the past days to perform the meteorological-hydrological-hydrodynamic mapping process for the future days. Once the model training is completed, the waterway depth and flow rate can be quickly predicted based on the meteorological elements, thereby avoiding frequent complex numerical simulation calculations in the meteorological-hydrological-hydrodynamic process, and effectively overcoming the complex waterway depth and bridge navigation clearance prediction technology and the multi-day forecast problem of shoal water depth and bridge navigation clearance. The solution is simple, efficient, intelligent and feasible.

[0026] 2. The waterway water regime forecasting method of the present invention is composed of a nested meteorological forecasting model, a meteorological-hydrological mapping model, and a hydrological-hydrodynamic mapping model. The entire forecasting method has strong physical interpretability and robustness. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 Schematic diagram of the basic process of the method of the embodiment of the present invention.

[0028] Figure 2 Schematic diagram of the principle of nesting multiple neural network models in an embodiment of the present invention.

[0029] Figure 3 Schematic diagram of the network structure of the meteorological-hydrological mapping model in an embodiment of the present invention.

[0030] Figure 4 This is a comparison chart of the water depth of Yangsha Lake in the Xiangjiang River channel predicted in an embodiment of the present invention and the measured results.

[0031] Figure 5 This is a water level distribution map of the Xiangjiang River Basin predicted in an embodiment of the present invention. DETAILED DESCRIPTION

[0032] In order to enable those skilled in the art to better understand the technical solution of the present invention, the technical solution of the present invention will be further described in detail below with reference to the accompanying drawings in the embodiments of the present invention.

[0033] like Figure 1 As shown, the waterway water condition forecasting method based on the nesting of multiple neural network models in this embodiment includes the following steps:

[0034] 1) Collect historical meteorological data of the target basin and use it to build a training dataset to train the weather forecast model. This will establish a mapping relationship between the meteorological elements of the past multiple consecutive days and the corresponding meteorological elements of the next day for the target basin.

[0035] 2) Collect historical hydrological data for the target basin. Use the daily meteorological elements predicted by the meteorological forecast model and the corresponding surface runoff flow in the historical hydrological data to construct a training dataset to train the meteorological-hydrological mapping model, so that it can establish a mapping relationship between daily meteorological elements and surface runoff flow for the target basin.

[0036] 3) Collect historical hydrodynamic data for the target basin, and use the surface runoff predicted by the meteorological-hydrological mapping model and the corresponding channel water level, water depth, and flow velocity distribution in the historical hydrodynamic data to construct a training dataset to train the hydrological-hydrodynamic mapping model. This will establish a mapping relationship between the surface runoff, channel water level, water depth, and flow velocity distribution for a single day for the target basin. It should be noted that, depending on actual needs, only part of the channel water level, water depth, and flow velocity distribution data can be used to train the hydrological-hydrodynamic mapping model, allowing it to predict part of the channel water level, water depth, and flow velocity distribution.

[0037] 4) Use multiple neural network models to nest to predict the channel water level, water depth and flow velocity distribution for multiple days in the future based on the meteorological elements of the target basin for multiple consecutive days before the current day: obtain the meteorological elements of the target basin for multiple consecutive days before the current day, use the meteorological elements of the target basin for multiple consecutive days before the current day to predict the meteorological elements of the first day in the future, and then use the predicted meteorological elements of the next day as known quantities to use the trained meteorological forecast model to roll-forward predict the meteorological elements of the next two days until the meteorological elements of multiple days in the future are predicted; for the predicted meteorological elements of multiple days in the future, use the trained meteorological-hydrological mapping model to predict the corresponding surface runoff flow for each day, and use the trained hydrological-hydrodynamic mapping model to obtain the corresponding channel water level, water depth and flow velocity distribution of the predicted surface runoff flow, thereby obtaining the channel water level, water depth and flow velocity distribution for multiple days in the future.

[0038] It should be noted that, when training the weather forecast model in step 1), the number of consecutive days in the past can be configured according to actual needs, for example, as an optional implementation method, such as Figure 2 As shown, in this embodiment, the weather forecast model is trained to establish a mapping relationship between the weather elements of the past four consecutive days (including the current day) and the corresponding weather elements of the next day for the target basin. The training data used comes from the weather forecast database, which is generated by collecting historical reanalysis weather data. The features used in training are the weather elements of the past four days in the target basin, and the labels used in training are the weather elements of the next day in the target basin.

[0039] It should be noted that the meteorological elements in step 1) can adopt desired indicators based on actual needs. For example, as an optional implementation, the meteorological elements in step 1) of this embodiment include the regional two-dimensional distribution of the daily average temperature, humidity, rainfall, and radiation values. Therefore, for the meteorological forecast model, the input features are the regional two-dimensional distribution of the daily average temperature, humidity, rainfall, and radiation values ​​of the target watershed over the past four days (including the current day), and the output labels are the regional two-dimensional distribution of the daily average temperature, humidity, rainfall, and radiation values ​​of the target watershed for the next day. The input features and output labels are used to construct a database training operator neural network model or other neural network model, thereby constructing a meteorological forecast model for the target watershed.

[0040] When training the meteorological-hydrological mapping model in step 2), the training data used is from a meteorological-hydrological mapping information database generated using a physics-based numerical hydrological model. The features used in training are the meteorological elements of the target basin, and the labels used in training are the surface runoff flow of the target basin. In this embodiment, when collecting historical hydrological data for the target basin in step 2), the hydrological numerical model is used to calculate the surface runoff flow of the target basin under the two-dimensional distribution of different meteorological elements. The calculation function expression for the surface runoff flow is:

[0041] ,

[0042] in, is the surface runoff flow, is the flow rate, is the sink flow, It should be noted that the numerical hydrological models for runoff, sink, and evaporation are all well-known models, and a distributed or semi-distributed hydrological model can be used as needed, so their implementation details are not detailed here.

[0043] like Figure 3 As shown, the meteorological-hydrological mapping model in this embodiment consists of two sub-models: a natural surface runoff sub-model (Model 2) and a hub scheduling interference sub-model (Model 3). The natural surface runoff sub-model is used to map the corresponding initial surface runoff flow based on the input meteorological elements of the day. The hub scheduling interference sub-model is used to predict the surface runoff flow correction value based on the initial surface runoff flow predicted by the natural surface runoff sub-model. The surface runoff flow output by the meteorological-hydrological mapping model is the sum of the initial surface runoff flow and the surface runoff flow correction value. Among them, the role of Model 3 is to superimpose the probabilistic error caused by hub scheduling uncertainty on the surface runoff results output by Model 2, thereby enhancing the robustness of surface runoff prediction.

[0044] Therefore, in this embodiment, when training the meteorological-hydrological mapping model, a high-precision hydrological numerical model is used to calculate the surface runoff flow of the target basin under the two-dimensional distribution of different meteorological elements, a meteorological-hydrological mapping information database is constructed, and a natural surface runoff sub-model (Model 2) is constructed based on an operator neural network model or other models, and training is performed based on the database; the difference between the predicted value and the measured value of Model 2 after training is saved, and the predicted value of Model 2 and the difference are paired as the training database of the hub scheduling interference sub-model (Model 3), and a variational autoencoder network model or other models is used to construct the hub scheduling interference sub-model (Model 3), and the database is used to train the hub scheduling interference sub-model.

[0045] In this embodiment, the function expression of the loss function used in the training of the meteorological-hydrological mapping model is:

[0046] ,

[0047] ,

[0048] in, is the loss of the meteorological-hydrological mapping model, is the hub scheduling interference sub-model, is the initial surface runoff flow predicted by the natural surface runoff sub-model, is the error of surface runoff flow predicted by the natural surface runoff sub-model, is the actual measured value of surface runoff flow.

[0049] In this embodiment, the network model used by the hub scheduling interference sub-model is a variational autoencoder network model. In addition, other neural network models can also be used as needed; the weather forecast model, the natural surface runoff sub-model, and the hydrological-hydrodynamic mapping model are one of the convolutional neural network, the Fourier operator neural network, and the graph neural network. In addition, other neural network models can also be used as needed.

[0050] In this embodiment, the collection of historical hydrodynamic data for the target watershed in step 3) involves conducting multiple numerical simulations based on a preset waterway hydrodynamic model for different surface runoff flow boundaries within the target watershed, and calculating the distribution of waterway water levels, water depths, and flow velocities at multiple stations within the target watershed. Based on the refined waterway hydrodynamic model, multiple numerical simulations are conducted for different surface runoff flow boundaries within the target waterway, calculating water regime information such as water levels, water depths, and flow velocities at multiple stations within the target waterway. A surface runoff flow-hydraulic information mapping database is constructed, and this database is used to train an operator neural network to construct a hydrological-hydrodynamic mapping model (referred to as Model 4).

[0051] After the above model training is completed, the spatial two-dimensional distribution of the average daily temperature, humidity, rainfall and radiation values ​​of the target basin in the past four days (including the current day) is input into the meteorological forecast model, and the spatial two-dimensional distribution of the average daily temperature, humidity, rainfall and radiation values ​​of the next day is output. Similarly, the forecast is rolled out day by day until the meteorological information for the next 7 days is obtained; the predicted meteorological information is input into the natural surface runoff sub-model (Model 2), and then the predicted value of Model 2 is input into the hub scheduling interference sub-model (Model 3), and the predicted value of Model 2 is added to the output value of Model 3 to obtain the surface runoff flow prediction value, and the predicted surface flow value is used as the input of the hydrological-hydrodynamic mapping model (Model 4), and finally the waterway water conditions information is output, including the waterway water level, water depth and flow velocity distribution of multiple stations in the target basin.

[0052] In order to verify the waterway water regime forecasting method based on the nesting of multiple neural network models in this embodiment, Yangsha Lake in the Xiangjiang River waterway is used as the target basin in this embodiment. The waterway water regime forecasting method based on the nesting of multiple neural network models in this embodiment is used to conduct experiments on its historical data. The comparison between the water depth of Yangsha Lake and the measured results is shown in the figure below. Figure 4 The Xiangjiang River Basin was used as the target basin, and the waterway water regime forecasting method based on the nested multi-neural network model of this embodiment was used to conduct experiments based on its historical data. The final waterway water level distribution map of the Xiangjiang River Basin is shown in the figure below. Figure 5 See Figure 4 and Figure 5 It can be seen that the waterway water situation forecasting method based on the nesting of multiple neural network models in this embodiment can avoid frequent complex numerical simulation calculations in the meteorological-hydrological-hydrodynamic process, and realize rapid, efficient and accurate prediction of the waterway water level, water depth and flow velocity distribution of the target river basin in the next few days based on the meteorological factors of the past years.

[0053] In summary, the waterway water condition forecasting method based on the nesting of multiple neural network models in this embodiment constructs training samples, guides the training of the waterway water condition forecasting model based on physical information, and uses a rolling prediction model to predict the water depth of the waterway for the next 7 days. Users only need to input meteorological data for the past four days and the next seven days into the forecasting system, and the trained waterway water condition forecasting model can map the meteorological-hydrological-hydrodynamic processes of the past four days and the next seven days, thereby realizing rapid prediction of the waterway water condition. The above-mentioned waterway water condition forecasting system is composed of a nested meteorological forecast model, a meteorological-hydrological mapping model, and a hydrological-hydrodynamic mapping model, wherein the meteorological-hydrological mapping model is composed of two sub-models: natural surface runoff and artificial hub scheduling, and is combined with variational autoencoding to introduce the uncertainty of hub regulation. Therefore, the entire forecasting system has strong physical interpretability and robustness, and can provide stable water condition information for the next seven days for ship navigation in complex waterways. The method of this embodiment overcomes the difficult problems of water depth prediction technology and bridge navigation clearance height prediction technology for complex high-level waterways, and multi-day forecast of shallow water depth and bridge navigation clearance height. It can effectively promote the transformation of water transport safety to pre-emptive prevention, improve social service, dispatching command and emergency response capabilities, and provide solutions for building a safe, convenient, efficient, green, economical, inclusive and resilient sustainable transportation system.

[0054] In addition, this embodiment also provides a waterway water condition forecasting system based on nested multiple neural network models, including a microprocessor and a memory connected to each other, and the microprocessor is programmed or configured to execute the waterway water condition forecasting method based on nested multiple neural network models.

[0055] In addition, this embodiment also provides a computer-readable storage medium, which stores a computer program or instruction, and the computer program or instruction is programmed or configured to execute the waterway water condition forecasting method based on nested multiple neural network models through a processor.

[0056] In addition, this embodiment also provides a computer program product, including a computer program or instructions, which are programmed or configured to execute the waterway water condition forecasting method based on nested multiple neural network models through a processor.

[0057] Those skilled in the art should understand that the technical solution provided by the present invention may be in the form of a method, a system, or a computer program product. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-readable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The present invention 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 invention. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of processes and / or boxes in the flowchart and / or block diagram, may be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the functions described in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 These computer program instructions can also be stored in a computer-readable memory that can guide a computer or other programmable data processing device to work in a specific way, so that the instructions stored in the computer-readable memory produce a product including the instruction device, which implements the function specified in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0058] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiment. All technical solutions based on the concept of the present invention are within the scope of protection of the present invention. It should be noted that for those skilled in the art, various improvements and modifications that do not depart from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A waterway water regime forecasting method based on nested multi-neural network models, characterized in that: The steps include: 1) Collect historical meteorological data of the target basin and use it to build a training dataset to train the weather forecast model. This will establish a mapping relationship between the meteorological elements of the past multiple consecutive days and the corresponding meteorological elements of the next day for the target basin. 2) Collect historical hydrological data for the target basin. Use the daily meteorological elements predicted by the meteorological forecast model and the corresponding surface runoff flow in the historical hydrological data to construct a training dataset to train the meteorological-hydrological mapping model, so that it can establish a mapping relationship between daily meteorological elements and surface runoff flow for the target basin. 3) Collect historical hydrodynamic data for the target basin, and use the surface runoff predicted by the meteorological-hydrological mapping model and part or all of the corresponding channel water level, water depth, or flow velocity distribution in the historical hydrodynamic data to construct a training dataset to train the hydrological-hydrodynamic mapping model, so that it can establish a mapping relationship between the surface runoff and part or all of the channel water level, water depth, or flow velocity distribution for a day for the target basin; 4) Obtain meteorological elements for multiple consecutive days before the current day in the target basin, use the trained meteorological forecast model to predict the meteorological elements for the first day in the future using the meteorological elements for multiple consecutive days before the current day, and then use the predicted meteorological elements for the next day as known quantities to use the trained meteorological forecast model to roll-forward predict the meteorological elements for the next two days until the meteorological elements for multiple days in the future are predicted; for the predicted meteorological elements for multiple days in the future, use the trained meteorological-hydrological mapping model to predict the corresponding surface runoff flow for each day, and use the trained hydrological-hydrodynamic mapping model to obtain part or all of the corresponding channel water level, water depth or flow velocity distribution for the predicted surface runoff flow, thereby obtaining part or all of the channel water level, water depth or flow velocity distribution for multiple days in the future.

2. The waterway water regime forecasting method based on nested multi-neural network models according to claim 1, characterized in that: The meteorological elements in step 1) include the regional two-dimensional distribution of daily average temperature, humidity, rainfall and radiation values.

3. The waterway water regime forecasting method based on nested multi-neural network models according to claim 1, characterized in that: When collecting historical hydrological data of the target basin in step 2), the hydrological numerical model is used to calculate the surface runoff flow of the target basin under the two-dimensional distribution of different meteorological elements. The calculation function expression of the surface runoff flow is: ， in, is the surface runoff flow, is the flow rate, is the sink flow, is the evaporation amount.

4. The waterway water regime forecasting method based on nested multi-neural network models according to claim 1, characterized in that: The meteorological-hydrological mapping model consists of two sub-models: a natural surface runoff sub-model and a hub scheduling interference sub-model. The natural surface runoff sub-model is used to map the corresponding initial surface runoff flow according to the input meteorological elements of the day. The hub scheduling interference sub-model is used to predict the surface runoff flow correction value based on the initial surface runoff flow predicted by the natural surface runoff sub-model. The surface runoff flow output by the meteorological-hydrological mapping model is the sum of the initial surface runoff flow and the surface runoff flow correction value.

5. The waterway water regime forecasting method based on nested multiple neural network models according to claim 4, characterized in that: The functional expression of the loss function used in the training of the meteorological-hydrological mapping model is: ， ， in, is the loss of the meteorological-hydrological mapping model, is the hub scheduling interference sub-model, is the initial surface runoff flow predicted by the natural surface runoff sub-model, is the error of surface runoff flow predicted by the natural surface runoff sub-model, is the actual measured value of surface runoff flow.

6. The waterway water regime forecasting method based on nested multiple neural network models according to claim 5, characterized in that: The network model used by the hub scheduling interference sub-model is a variational autoencoder network model; the weather forecast model, natural surface runoff sub-model, and hydrological-hydrodynamic mapping model are one of a convolutional neural network, a Fourier operator neural network, and a graph neural network.

7. The waterway water regime forecasting method based on nested multiple neural network models according to claim 1, characterized in that: When collecting historical hydrodynamic data of the target basin in step 3), it includes conducting multiple sets of numerical simulations under different surface runoff flow boundaries of the target basin based on the preset channel hydrodynamic model, and calculating part or all of the channel water level, water depth or flow velocity distribution at multiple stations in the target basin.

8. A waterway water condition forecasting system based on nested multiple neural network models, comprising a microprocessor and a memory connected to each other, characterized in that: The microprocessor is programmed or configured to execute the waterway water condition forecasting method based on nesting of multiple neural network models as described in any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program or instruction stored therein, characterized in that: The computer program or instruction is programmed or configured to execute the waterway water condition forecasting method based on nesting of multiple neural network models as described in any one of claims 1 to 7 through a processor.

10. A computer program product comprising a computer program or instructions, characterized in that The computer program or instruction is programmed or configured to execute the waterway water condition forecasting method based on nesting of multiple neural network models as described in any one of claims 1 to 7 through a processor.

Images