Snowmelt runoff determination method, device, computer equipment and storage medium

Abstract

The present invention relates to the field of data processing technology, and discloses a method, device, computer equipment, and storage medium for determining snowmelt runoff. The method comprises: obtaining target data of a study area; dividing the study area into a plurality of computational grids according to preset grid units; inputting the target data into a pre-constructed hydrological model to obtain the output of the hydrological model; the output of the hydrological model includes the runoff generation result corresponding to each computational grid; determining the output of the hydrological model as the input of a pre-constructed target prediction model to generate the output of the target prediction model; the output of the target prediction model represents the predicted result of snowmelt runoff in the study area; the target prediction model includes an embedded multi-layer perception probe, which is used to analyze the nonlinear characteristics in the input of the target prediction model. The target prediction model of the present invention embeds a multi-layer perception probe, which improves the accuracy of snowmelt runoff determination.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a method, device, computer equipment and storage medium for determining snowmelt runoff. Background Art

[0002] Snowpack plays a key role in regulating seasonal hydrological variations, and its melt provides a vital water source for industrial and agricultural production in downstream areas. However, rising spring temperatures trigger a rapid increase in snowmelt runoff, posing a serious threat to the safety and property of downstream residents. Therefore, accurate monitoring of snowmelt runoff is crucial for early warning of snowmelt floods.

[0003] Traditional hydrological models mainly focus on outlet section flow in terms of functionality, and are unable to describe snowmelt runoff quickly and accurately.

[0004] Therefore, there is an urgent need to construct a method that can accurately quantify snowmelt runoff. Summary of the Invention

[0005] Therefore, the technical problem to be solved by the present invention is to overcome the problem that the related art is difficult to quickly and accurately determine snowmelt runoff.

[0006] In order to solve the above technical problems, the present invention provides a method for determining snowmelt runoff, which includes:

[0007] Obtain target data for the study area;

[0008] Dividing the study area into a plurality of computational grids according to preset grid units;

[0009] Inputting the target data into a pre-built hydrological model to obtain the output of the hydrological model; the output of the hydrological model includes the runoff result corresponding to each calculation grid;

[0010] Determining the output of the hydrological model as the input of a pre-built target prediction model to generate the output of the target prediction model; the output of the target prediction model represents the prediction result of snowmelt runoff in the study area;

[0011] The target prediction model includes embedded multi-layer perception probes, which are used to analyze nonlinear features in the input of the target prediction model.

[0012] In an optional embodiment, the target prediction model includes:

[0013] The input layer is used to standardize the input of the target prediction model to obtain the output of the input layer;

[0014] The output of the input layer is the input of the grid structure;

[0015] The grid structure includes: a plurality of grid units; the connection relationship between the plurality of grid units is constructed based on the time dimension and the space dimension;

[0016] The grid structure is used to analyze the output of the input layer to determine the output of the grid structure;

[0017] The output of the grid structure includes: the instantaneous features corresponding to each computational grid on a standard step size, the hidden states between adjacent computational grids, and the memory states corresponding to the long-term features; the output of the grid structure is the input of the multi-layer perception probe;

[0018] The multi-layer perception probe is specifically used to perform nonlinear characteristic analysis on the output of the grid structure to determine the output of the multi-layer perception probe;

[0019] The output of the multi-layer perception probe includes: flow, snowmelt time and snowmelt runoff of each computational grid at a standard step size;

[0020] The output of the multi-layer perception probe is the input of the output layer;

[0021] The output layer is used to perform feature mapping on the output of the multi-layer perception probe to determine the output of the output layer;

[0022] The output of the output layer includes: flow sequence, snowmelt time series and snowmelt runoff sequence on the standard step size scale corresponding to the outlet section.

[0023] In an optional embodiment, the grid unit includes: a forget gate, an input gate, a candidate memory state, a memory state, an output gate and a hidden state, and a gating mechanism determined according to the forget gate, the input gate, the candidate memory state, the memory state, the output gate and the hidden state.

[0024] In an optional embodiment, when the multi-layer perception probe is a perception probe with two hidden layers, the multi-layer perception probe is configured as follows:

[0025] h1=σ(W1·c t +b1);

[0026] h2=σ(W2·h1+b2);

[0027] Where h1 represents the output vector of the first hidden layer, h2 represents the output vector of the second hidden layer, σ represents the activation function, W1 represents the weight matrix of the first hidden layer, b1 represents the bias vector of the first hidden layer, W2 represents the weight matrix of the second hidden layer, b2 represents the bias vector of the second hidden layer, and c t Represents the output of the previous adjacent grid cell.

[0028] In an optional embodiment, the target prediction model further includes: a loss function;

[0029] The loss function is used to determine the loss value based on the predicted result and the actual result of the target prediction model; the loss value represents the size of the gap between the predicted result and the actual result of the target prediction model; the loss value is used to adjust the parameter value of the parameter of the target prediction model.

[0030] In an optional embodiment, the runoff results are in the form of a time series, including surface runoff, snowmelt, baseflow, net evapotranspiration, potential evapotranspiration and soil moisture.

[0031] In an optional embodiment, when the target data includes meteorological and hydrological data and underlying surface characteristic data, inputting the target data into a pre-constructed hydrological model to obtain the output of the hydrological model includes: inputting the meteorological and hydrological data and underlying surface characteristic data into a pre-constructed hydrological model to obtain the output of the hydrological model.

[0032] In a second aspect, the present invention provides a snowmelt runoff determination device, comprising:

[0033] The first processing module is used to obtain target data of the study area;

[0034] A second processing module is used to divide the study area into a plurality of calculation grids according to preset grid units;

[0035] The third processing module is used to input the target data into a pre-built hydrological model to obtain the output of the hydrological model; the output of the hydrological model includes the runoff result corresponding to each calculation grid;

[0036] a fourth processing module, configured to determine the output of the hydrological model as an input of a pre-built target prediction model to generate an output of the target prediction model; the output of the target prediction model represents a prediction result of snowmelt runoff in the study area;

[0037] The target prediction model includes embedded multi-layer perception probes, which are used to analyze nonlinear features in the input of the target prediction model.

[0038] In a third aspect, the present invention provides a computer device comprising: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the snowmelt runoff determination method of the first aspect or any corresponding embodiment thereof by executing the computer instructions.

[0039] In a fourth aspect, the present invention provides a computer-readable storage medium, wherein a single computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the snowmelt runoff determination method of the above-mentioned first aspect or any corresponding embodiment thereof.

[0040] In a fifth aspect, the present invention provides a computer program product comprising computer instructions for causing a computer to execute the snowmelt runoff determination method of the first aspect or any corresponding embodiment thereof.

[0041] The technical solution provided by the present invention has the following technical effects:

[0042] The technical solution of the present invention divides the study area into multiple grid cells. This gridding approach facilitates discretization of complex geographic areas. This approach allows for more detailed consideration of the spatial heterogeneity of the study area, adapting to complex terrain, and improving model precision, accuracy, and reliability of predictions.

[0043] The target data is input into a pre-built hydrological model to obtain the model's output. This output is then used as the input for the pre-built target prediction model, generating the target prediction model's output. The target prediction model further processes and predicts the hydrological model's output, receiving the model's runoff output as input. The target prediction model's output represents the predicted snowmelt runoff within the study area. The target prediction model incorporates a multi-layer perception probe, which analyzes nonlinear characteristics within the target prediction model's input. Many relationships in hydrological processes are nonlinear, and the multi-layer perception probe can deeply explore these nonlinear relationships, enabling a better understanding and explanation of the model's behavior.

[0044] For example, you can discover which combinations of input data have the greatest impact on snowmelt runoff and how the interaction between these variables affects the final prediction results, thereby improving the accuracy of model predictions.

[0045] The combined use of multiple models, combining hydrological models and target prediction models, can leverage their respective strengths. Hydrological models simulate hydrological processes based on physical or empirical relationships, while target prediction models can leverage multi-layer perception probes to further extract features and make predictions based on the output of hydrological models. Combining the strengths of both can lead to more accurate snowmelt runoff predictions.

[0046] The technical solution of the present invention combines the advantages of grid division and multi-layer perception probes to provide more accurate and interpretable results for the prediction of snowmelt runoff. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in related technologies, the following briefly introduces the drawings required for use in the specific embodiments or related technical descriptions. Obviously, the drawings described below are 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.

[0048] Figure 1 is a flow chart of a method for determining snowmelt runoff according to an embodiment of the present invention;

[0049] Figure 2 It is a schematic diagram of the structure of the network framework of an embodiment of the present invention;

[0050] Figure 3 This is a statistical diagram of flow and snowfall at a hydrological station according to an embodiment of the present invention;

[0051] Figure 4 Schematic diagram of runoff simulation effect according to an embodiment of the present invention;

[0052] Figure 5 Schematic diagram of the difference between the snowmelt time output by the model and the image result according to an embodiment of the present invention;

[0053] Figure 6 is a schematic diagram of the contribution of snowmelt runoff to the total runoff output by the model of an embodiment of the present invention;

[0054] Figure 7 is a schematic diagram of a specific process of a method for determining snowmelt runoff according to an embodiment of the present invention;

[0055] Figure 8 is a schematic structural diagram of a snowmelt runoff determination device according to an embodiment of the present invention;

[0056] Figure 9 Schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0057] To make the purpose, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are 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 those skilled in the art without making creative efforts shall fall within the scope of protection of the present invention.

[0058] Snowpack regulates seasonal hydrological variations, and snowmelt provides water for industrial and agricultural production in downstream cities and residents. On the other hand, rising spring temperatures lead to a rapid increase in snowmelt runoff, threatening the safety of downstream residents. Therefore, accurately capturing snowmelt runoff is crucial for early warning of snowmelt floods. Currently, traditional hydrological models focus primarily on outlet flow, failing to quickly and accurately describe snowmelt runoff and its contribution. This leads to uncertainty in high-precision snowmelt runoff simulation, particularly in high-altitude watersheds with widespread seasonal snowpack. Meanwhile, data-driven deep learning approaches are emerging, offering fast computational speed and accuracy. However, a recognized obstacle to this approach is the lack of sufficient interpretability of hydrological processes through "black-box" models, which limits their further development. Currently, the most commonly used approach is coupling physical process models with deep learning approaches. Typically, such coupling approaches focus on bias correction in runoff simulation results, neglecting understanding of hydrological processes. This is particularly true when simulating the complex hydrological cycle in high-altitude and cold watersheds, where the role of snowpack in hydrological processes is often overlooked. Ignoring snowmelt runoff will affect the accuracy of runoff simulation during the spring flood season. Therefore, there is an urgent need for a modeling framework that can combine physical process models with deep learning and accurately quantify the temporal and magnitude characteristics of snowmelt runoff.

[0059] In view of this, embodiments of the present invention provide a method, apparatus, computer device, and storage medium for determining snowmelt runoff to solve problems in related technologies.

[0060] According to an embodiment of the present invention, an embodiment of a method for determining snowmelt runoff is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer device such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0061] Figure 1 4 is a flow chart of a method for determining snowmelt runoff according to an embodiment of the present invention.

[0062] like Figure 1 As shown, an embodiment of the present invention provides a method for determining snowmelt runoff, the method comprising:

[0063] S101: Acquire target data of the study area.

[0064] In this embodiment, the target data includes meteorological and hydrological data and underlying surface characteristic data, which can serve as inputs to the hydrological model. The target data can also include remote sensing imagery products of the snow water equivalent of the study area. The study area is the watershed where snowmelt runoff is to be determined. The remote sensing imagery products of the snow water equivalent of the study area can be acquired using remote sensing technology and can provide image data that reflects information about the water content of snowpack.

[0065] Meteorological and hydrological data include, but are not limited to, basin-wide rainfall, temperature, wind speed, and historical runoff data. Underlying surface characteristic data include, but are not limited to, digital elevation models (DEMs), land use, and soil types. Furthermore, to identify snow cover distribution and variability, microwave imagery-based snow water equivalent products were collected within the study area to analyze snow cover distribution and snow storage, providing a reference for model simulation results.

[0066] In this embodiment, S101 acquires target data for the study area. This specifically includes obtaining meteorological and hydrological data within the watershed, underlying surface characteristics, and remote sensing imagery of snow water equivalent in the study area. These initial data are preprocessed based on the scope of the study area to obtain the target data for the study area. The target data is the preprocessed data and can be daily data, i.e., data collected at daily intervals.

[0067] S102: Divide the study area into multiple computational grids according to preset grid units.

[0068] In this embodiment, the size of the preset grid unit can be set and modified according to actual conditions. As an example, the preset grid unit can be a 5 km×5 km grid unit.

[0069] S103: Input the target data into a pre-built hydrological model to obtain the output of the hydrological model.

[0070] In this embodiment, the output of the hydrological model includes runoff generation results for each computational grid. These runoff generation results are presented in a time series format and include surface runoff, snowmelt, baseflow, net evapotranspiration, potential evapotranspiration, and soil moisture. When the target data includes meteorological and hydrological data and underlying surface characteristic data, the target data is input into a pre-built hydrological model to obtain the hydrological model output. Specifically, the hydrological and hydrological data and underlying surface characteristic data are input into the pre-built hydrological model to obtain the hydrological model output.

[0071] In this embodiment, as an example, the hydrological model can be a variable infiltration capacity (VIC) model. When the target data is daily, the output of the hydrological model includes the daily runoff results for each computational grid. The output of the hydrological model is stored in a file format, specifically a grid file, which serves as data input for the next step, the multi-layer perception probe and hybrid neural network model framework.

[0072] S104: Determine the output of the hydrological model as the input of a pre-built target prediction model to generate the output of the target prediction model.

[0073] In this embodiment, the output of the target prediction model represents the prediction results of snowmelt runoff in the study area, specifically referring to the flow sequence, snowmelt time series, and snowmelt runoff sequence at the standard step size corresponding to the outlet section of the study area.

[0074] In this embodiment, the target prediction model includes embedded multi-layer perception probes, which are used to analyze nonlinear features in the input of the target prediction model.

[0075] In this embodiment, as an example, the target prediction model is a long short-term memory network (LSTM) model, which is constructed based on a multi-layer perception probe and a hybrid neural network model framework.

[0076] In this example, accurately simulating snowmelt runoff is crucial for flood mitigation efforts in basins with widespread seasonal snow cover. Currently, traditional hydrological models focus primarily on outlet flow and are unable to quickly and accurately describe snowmelt runoff and its contribution. While data-driven deep learning simulation technology offers high accuracy, it lacks sufficient interpretability for hydrological processes, limiting its further development.

[0077] In this regard, in order to combine the modeling advantages of physical process models and deep learning, reflect the understanding of hydrological processes under the premise of high-precision simulation, and accurately quantify the time and magnitude characteristics of snowmelt runoff, the present invention designs and constructs a multi-layer perception probe and hybrid neural network model framework, such as Figure 2 shown.

[0078] Model framework design: The model is constructed using machine learning algorithms to quickly calculate grid runoff data and quantify the magnitude and timing of snowmelt runoff at the outlet section.

[0079] In an optional embodiment, the target prediction model includes:

[0080] The input layer is used to standardize the input of the target prediction model to obtain the output of the input layer.

[0081] The output of the input layer is the input of the grid structure.

[0082] In this embodiment, as the data input port of the model framework, the input data required by the present invention is the runoff data output by the VIC model, and the input data is latitude-processed to meet the model input requirements.

[0083] For the output of the VIC model (the input of the target prediction model), its mathematical expression can be X∈R H×W×D , where H is the grid height, representing the vertical distribution of the study area. W is the grid width, representing the horizontal distribution of the study area. D is the characteristic dimension, including surface runoff, snowmelt, base flow, net evapotranspiration, potential evapotranspiration and soil moisture. Its dimensions are:

[0084]

[0085] in, Represents the runoff result corresponding to the grid point (computational grid) i row j column (i, j) at time step t.

[0086] The input of the target prediction model is standardized. A normalization scheme can be used to process the input data as the feature vector of the next input, so that the input data is between [0, 1], ensuring that different features are at the same level and avoiding unnecessary influence of certain features on model training.

[0087] The grid structure includes multiple grid cells. The connections between the multiple grid cells are constructed based on the time dimension and the space dimension. The grid structure is a two-dimensional grid structure. The grid cells are arranged in rows and columns. Each grid cell can receive the output of its adjacent grid cells in the time dimension and the space dimension as input.

[0088] The spatial distribution between multiple grid cells can be determined based on the computational grid distribution obtained by dividing the study area.

[0089] The grid structure is used to analyze the output of the input layer to determine the output of the grid structure.

[0090] The output of the grid structure includes: the instantaneous features corresponding to each computational grid at the standard step size, the hidden states between adjacent computational grids, and the memory states corresponding to the long-term features. The output of the grid structure serves as the input to the multi-layer perception probe.

[0091] In this embodiment, each grid unit includes: a forget gate, an input gate, a candidate memory state, a memory state, an output gate and a hidden state, and a gating mechanism determined according to the forget gate, the input gate, the candidate memory state, the memory state, the output gate and the hidden state.

[0092] LSTM is a commonly used method for time series forecasting. Its nonlinear and time-cyclical characteristics make it suitable for flow forecasting in the hydrological field. Traditional LSTM excels at processing one-dimensional sequence data but cannot demonstrate interdependencies in multidimensional data. Gridded LSTM (a grid-based structure) expands LSTM units (grid cells) across multiple dimensions to better capture complex dependencies between multidimensional inputs. The connections and interactions between LSTM units (grid cells) are consistent with the computational grid. Each computational grid can be considered to correspond to one LSTM unit. For a grid cell, its spatially adjacent grid cells correspond to the computational grids adjacent to the computational grid corresponding to that grid cell after meshing.

[0093] The grid structure design of the present invention includes three features. First: each LSTM unit is associated not only with the previous and next states of the time step, but also with its spatially adjacent grid units. Second: in the two-dimensional grid, the LSTM units are arranged in a row and column (H, W) structure, and each unit will receive the output from its temporal and spatial adjacent units as input information. Third: for each time step, LSTM will update the state of each grid unit while taking into account the temporal and previous dependencies and spatial adjacent dependencies. Due to its multi-dimensional characteristics, it can be recursively updated through the grid structure when receiving inputs from multiple directions. Therefore, this parallel structure enables the target prediction model to process high-dimensional data more quickly. The grid structure constructed by the embodiment of the present invention can achieve recursive updates.

[0094] Temporal and spatial correlation features: Traditional LSTM units mainly focus on the connection between the previous and next states in the time step, and perform information transmission and state updates along the time axis when processing sequence data. However, the gridded LSTM structure of the present invention is expanded on this basis. In addition to considering its own evolution in the time series, each LSTM unit also establishes connections with spatially adjacent pixels (that is, adjacent grid units). This means that when processing data, it will comprehensively analyze the changes in the current grid unit in the time process and the impact of the adjacent grid units in the surrounding space on it. When studying snowmelt runoff, if the study area is divided into grids, the snowmelt conditions of a certain calculation grid may be affected by factors such as the influx of meltwater from snow in adjacent calculation grids and heat transfer. This connection can more comprehensively reflect the actual hydrological and physical processes.

[0095] Two-dimensional grid arrangement and information reception method: LSTM units are arranged in a two-dimensional grid according to a row and column (H, W) structure, forming a chessboard-like layout. Each grid unit receives input information from its temporally adjacent grid units (i.e., grid units at the same position at the previous and next moments) and spatially adjacent grid units (such as grid units at adjacent positions such as above, below, left, and right). This enables each grid unit to obtain richer information, no longer limited to its own historical information, but also including the real-time status information of surrounding grid units. For example, in the process of simulating hydrology, a grid unit can adjust its own state update based on information such as the soil moisture content and surface runoff conditions of surrounding grid units, thereby more accurately simulating processes such as the diffusion and convergence of water flow in space.

[0096] Multidimensional Update Mechanism and Efficient Processing: At each time step, the LSTM updates the state of each grid cell, taking into account both temporal dependencies and spatial dependencies. Because it can receive input from multiple directions, it performs recursive updates through its grid structure. For example, when updating the state of a grid cell, it first performs preliminary calculations based on its own previous and next time step states and the input from neighboring grid cells. This updated state then serves as one of the inputs for the next update of the neighboring grid cells, and this cycle repeats, forming a recursive update process. This multidimensional nature and parallel structure of recursive updates significantly enhance the model's ability to handle high-dimensional data. When faced with complex data containing a large number of grid cells and long time series, it can complete calculations and analysis more quickly than traditional models that only consider the temporal dimension or simply process spatial information, improving the model's operational efficiency and adaptability to complex data.

[0097] When the grid structure performs vector calculations on surface runoff, snowmelt, baseflow, net evapotranspiration, potential evapotranspiration, and soil moisture characteristics, the pre-calculated information of the two-dimensional latent vector is input to the next state according to the grid cell arrangement rule. At this time, the calculation of the memory vector is expanded to:

[0098]

[0099] Where m is the memory vector, ⌒ is the element-wise product, represents the control gate of grid cell (i, j) at time t, m i，j is the storage unit of grid unit (i, j), Represents the memory state of grid cell (i, j).

[0100] In more detail, for the gating mechanism of gridded LSTM, at each grid cell (i, j):

[0101]

[0102] in, and Respectively represent the forget gate, input gate, candidate memory state, memory state, output gate and hidden state. d is the dimension parameter, which represents the feature dimension of the six input data in this invention. σ and tanh are activation functions, W and b are mapping weights and bias parameters respectively. Specifically, Represents the mapping weight corresponding to the forget gate, Represents the bias parameter corresponding to the forget gate, W i d Represents the mapping weight corresponding to the input gate, Represents the bias parameter corresponding to the input gate, represents the mapping weight corresponding to the candidate memory state, represents the bias parameter corresponding to the candidate memory state, Represents the mapping weight corresponding to the output gate, Represents the bias parameter corresponding to the output gate, x i，j Indicates the current input, represents the hidden state at height i and width j-1, Represents the output of the forget gate in the vertical direction at height i-1 and width j, Represents the hidden state in the horizontal direction at height i and width j-1, Represents the hidden state in the vertical direction at height i-1 and width j, Represents the memory state in the horizontal direction at height i and width j-1, represents the memory state in the vertical direction at height i-1 and width j, and h and v represent the horizontal and vertical directions in the grid structure.

[0103] Forget Gate Used to control the retention or discard of memory information in the input data, according to The hidden state of the current input feature and the adjacent grid cells Perform weighted mapping and add bias parameters According to the σ activation function, the output of the forget gate is limited to [0, 1], which represents the forgetting ratio.

[0104] Input Gate Controls the degree of update of the memory state by the current input. Its calculation includes the activation function σ, the weighted mapping W, and the bias parameter b to determine the information that needs to be added to the memory state.

[0105] Candidate Memory Generate new potential information and prepare to update it to the memory state. The calculation includes the weighted mapping W and the bias parameter b. The tanh activation function is used to limit the value to [-1, 1] to retain the possibility of negative information and to represent the newly added memory.

[0106] Memory Status Represents the update of the memory state, using the output of the forget gate and the input gate to update the current memory state. Its calculation includes the memory update calculation in the horizontal (h) and vertical (v) directions in the network structure.

[0107] Calculate the output gate, determine the memory information of the current hidden state, and control the contribution of the hidden state in the memory state according to the activation function σ, weighted mapping W, and bias parameter b.

[0108] The current hidden state is updated according to the result of the output gate, the mapping weight is U, and the tanh function is used to activate the main information of the memory state as a time-step calculation.

[0109] Input to the gridded LSTM structure: feature vectors of surface runoff, snowmelt, baseflow, net evapotranspiration, potential evapotranspiration, and soil moisture processed by the normalization scheme.

[0110] Output of the gridded LSTM structure: hidden state containing grid-by-grid instantaneous features and neighborhood spatial correlations at the daily step scale and the memory state of the accumulated long-term features from the initial time step to the current time step

[0111] The multi-layer perception probe is specifically used to perform nonlinear characteristic analysis on the output of the grid structure to determine the output of the multi-layer perception probe.

[0112] The output of the multi-layer perception probe includes: the flow (specifically, runoff), snowmelt time, and snowmelt runoff at the standard step size for each computational grid. In this embodiment, when the target data is daily, the corresponding standard step size is the daily step size.

[0113] The output of the multi-layer perception probe is the input of the output layer.

[0114] In this example, to address the limited ability of the gridded LSTM structure's output to interpret and analyze the model's intermediate information, probes are introduced as a tool for gaining deeper understanding and analysis of the model's internal representations. Specifically, a probe is an additional classifier or regressor that typically processes the LSTM model's output cell states using a linear probe approach to assess whether the cell states in the model's intermediate layers contain information relevant to a specific task.

[0115] In this invention, two fully connected layers are added to the grid structure, and nonlinear probes are used to calculate the feature projection of the hidden layer state. The intermediate signal results are output to analyze the complex nonlinear features in the grid structure state vector. In order to apply nonlinear transformations to interpret complex patterns and relationships, when the multi-layer perception probes are the perception probes of the two hidden layers, the multi-layer perception probes are set to:

[0116] h1=σ(W1·c t +b1);

[0117] h2=σ(W2·h1+b2).

[0118] Where h1 represents the output vector of the first hidden layer, h2 represents the output vector of the second hidden layer, σ represents the activation function, W1 represents the weight matrix of the first hidden layer, b1 represents the bias vector of the first hidden layer, W2 represents the weight matrix of the second hidden layer, b2 represents the bias vector of the second hidden layer, and c t Represents the output of the previous adjacent grid cell, that is, the cell state output by the previous grid cell.

[0119] The calculation of h1 uses the cell state of the previous adjacent grid cell as input, and the outputs of the adjacent grid cells in the time dimension and the adjacent grid cells in the spatial dimension as the input of the current grid cell. It uses σ (ReLU activation function) for calculation, truncating negative values ​​and retaining positive values ​​to enhance feature sparsity.

[0120] h2 uses the output of the first hidden layer h1 as input and further refines the feature representation. The regression task targets the runoff volume, snowmelt time, and snowmelt runoff at the daily time scale for all computational grids within the study area. Snowmelt features are extracted based on the surge in runoff caused by rapid snowmelt during low rainfall events. Therefore, the decomposition of the h1 signal at each grid cell scale quickly identifies the amount of snowmelt and its timing. This step outputs the runoff volume, snowmelt time, and snowmelt runoff results for each computational grid within the study area at the daily time scale.

[0121] Input to the multi-layer perception probe: hidden states of the gridded LSTM and memory status

[0122] Outputs of the multi-layer perception probe: daily step runoff, snowmelt time, and snowmelt runoff for all computational grids.

[0123] The output layer is used to perform feature mapping on the output of the multi-layer perception probe to determine the output of the output layer.

[0124] The output layer includes the flow sequence, snowmelt time series, and snowmelt runoff series at the standard step size for the outlet section. The output layer uses a preset formula to determine the flow sequence, snowmelt time series, and snowmelt runoff series at the standard step size for each computational grid based on the flow (specifically, runoff), snowmelt time, and snowmelt runoff at the standard step size for each computational grid. The preset formula includes a weight for each computational grid, which is determined based on its importance and contribution to snowmelt runoff.

[0125] As an example, the flow series, snowmelt time series, and snowmelt runoff series at the outlet section on a daily time scale are time series descriptions of the hydrological conditions at the outlet of the study area:

[0126] Flow series: This refers to a sequence of flow rate values ​​at a watershed outlet at daily intervals. It reflects the temporal changes in the overall water output of a watershed. By monitoring and recording daily flow, we can understand the dynamic trends of water resources within the basin. For example, during the snowmelt season, flow may gradually increase due to melting snow, while during the non-snowmelt season, flow may be relatively stable or exhibit different fluctuation patterns due to other factors such as precipitation.

[0127] Snowmelt time series: This daily series defines the time when snow melt begins and occurs at different stages within the study area. This is crucial for studying snowmelt patterns and predicting the temporal course of snowmelt runoff. For example, it can be used to determine the approximate start date of snowmelt each year and the time range of peak snowmelt, aiding in water resource planning and flood warning preparation.

[0128] Snowmelt runoff series: This is a daily numerical sequence of runoff volume generated by melting snow and ultimately collected at the outlet of a watershed. It reflects the temporal changes in snowmelt's direct contribution to watershed runoff. Combined with snowmelt time series, it comprehensively reflects the generation, collection, and output of water during the snowmelt process, providing key data support for accurately assessing the contribution and role of snowmelt runoff in the total watershed runoff.

[0129] The output layer is responsible for converting the output of the multi-layer perception probe in the previous step into a one-dimensional flow sequence with a sequence dimension of y∈R T , where T is the daily time step and R is the flow value.

[0130] The input of the output layer is the features of h2 processed by the multi-layer perception probe. All grid information is integrated, and a weighted pooling scheme is used to perform feature mapping on the network features to obtain the daily step flow, snowmelt time and snowmelt runoff sequence of the outlet section.

[0131] As an example, data conversion basics:

[0132] The multi-layer perception probe outputs information about each computing grid, which contains rich features. However, in order to obtain outlet section data that meets the needs of actual applications and analysis, specific processing is required to convert it into a one-dimensional flow sequence.

[0133] Here we define the sequence dimension as y∈R T , where T represents the daily time step, meaning that this series organizes data according to daily time intervals. R is the flow value, indicating the flow data corresponding to each time step.

[0134] Weighted pooling and feature mapping: The output layer uses the h2 features processed by the multi-layer perception probe as input. The h2 features carry the results of in-depth analysis of the computational grid data, but these results are scattered across the computational grids.

[0135] A weighted pooling scheme is used to integrate all grid information. Weighted pooling comprehensively processes the characteristics of different computational grids based on a specific weight distribution strategy. For example, computational grids that are in key locations within the watershed (such as near the main river channel or in low-lying terrain) or that contribute significantly to runoff are given higher weights, while less important computational grids are given lower weights. In this way, the pooling operation can more reasonably reflect the degree of influence of each computational grid on the outlet cross-section results.

[0136] Feature mapping of network features involves converting the weighted pooled features into the final outlet section's daily step-by-step flow, snowmelt time, and snowmelt runoff series according to predetermined rules. This process may employ conversion formulas and algorithms based on physical principles or experience to map abstract features into hydrological data sequences with practical physical meaning. This yields results that intuitively reflect the hydrological conditions at the basin's outlet, providing critical data support for snowmelt runoff research and water resources management.

[0137] In an optional implementation, the target prediction model further includes: a loss function.

[0138] The loss function is used to determine the loss value based on the target prediction model's predictions and the actual results. The loss value represents the gap between the target prediction model's predictions and the actual results. The loss value is used to adjust the parameters of the target prediction model. The loss value can be expressed as the Nash efficiency coefficient.

[0139] In this embodiment, historical data can be obtained, a data set can be constructed based on the historical data, and a target prediction model can be trained. The accuracy can be evaluated according to a 70% training set and a 30% validation set. The input is a 365-day long sequence to predict the next 365-day runoff and snowmelt runoff.

[0140] In this embodiment, as an example, the basin above a hydrological station is taken as the research object. The basin area of ​​the study area is nearly 50,000 km2, with the lowest altitude exceeding 4,000 meters and the average altitude exceeding 5,000 meters. More than 50% of the basins have seasonal snow accumulation lasting more than 30 days. Snowmelt caused by rising temperatures during the flood season significantly affects runoff. Statistics show that snowmelt runoff accounts for more than 20%. Figure 3 The flow process line and snowfall statistics for the study area are shown. Based on the existing historical runoff data, the measured flow data of the hydrological stations from January 1, 1982 to December 31, 2008 were collected as a reference for the verification of the target prediction model.

[0141] The Nash Sutcliffe efficiency (NSE) can be calculated based on the daily step-length discharge series of the outlet section output by the target prediction model training and the measured runoff series at a hydrological station. The loss function is defined according to the following scheme to optimize the model parameters:

[0142]

[0143] Where Loss orig is the loss function of the model, Represents the prediction result of the target prediction model, Q outlet Indicates the true result. Loss total Is a combined loss function that includes probes. α represents a parameter that controls the strength of regularization. A larger value means stronger regularization, which is used to reduce model complexity and prevent overfitting. The ρ parameter represents the ratio of Lasso regression to Ridge regression, ranging from 0 to 1, where 1 represents L1 regularization and 0 represents L2 regularization. i || represents the weight W i A grid search method is used to test the combination of α and ρ from the parameter pool to determine the optimal value.

[0144] According to the loss function definition, torch-2.0.1+cuda117 version was used, implemented in a programming language, and run on the operating system. To prevent overfitting, a dropout layer was added between the connection layers of the model, with a value of 0.2. The early stopping technique was also set, with a patience value of 10. The Adam algorithm was used to optimize the model parameters, with 50 epochs selected. The best performance values ​​of the loss function parameters were α=1.0 and β=0.25. In this example, the output data of the VIC model was used as the input of the grid LSTM. The training set was 1982-1999, and the validation set was 2000-2008. The runoff simulation effect is as follows Figure 4 shown.

[0145] As an example, based on the snowmelt time and snowmelt runoff simulation scheme, the snowmelt time and snowmelt runoff from the output of the multi-layer perception probe can be compared with the snow water equivalent product based on microwave imagery.

[0146] The amount of snowmelt in each computational grid needs to be calculated based on the daily snow water equivalent (SWE) in the study area obtained from the image:

[0147] SWE d-1 -SWE d =SWE loss .

[0148] Among them, SWE d-1 Indicates that the SWE is greater than the given d SWE one day earlier, SWE loss Represents the daily snowmelt rate. Generate grid-by-grid image snowmelt characteristic data and compare it with the probe output. All comparisons are performed at the grid scale. Figure 5 and Figure 6 The temporal error between the model output and the image, as well as the corresponding proportion of snowmelt runoff, are displayed. This invention can quickly and efficiently quantify the temporal and magnitude characteristics of snowmelt runoff in the basin above a hydrological station.

[0149] Based on the multi-layer perception probe and hybrid neural network framework constructed above, the results calculated from the imagery products and the actual results at the watershed outlet were used as the evaluation criteria. The snowmelt runoff quantification standard was based on matching the output of the multi-layer perception probe with the snowmelt characteristics calculated from the imagery. All comparisons were performed at the grid scale.

[0150] When evaluating the target prediction model, gridded snowmelt information calculated from remote sensing imagery of the study area's snow water equivalent and actual measured runoff data at the basin's outlet were used as a reference for measuring model performance. Specific quantitative criteria were determined by comparing the output of the model's multi-layer perception probes with snowmelt characteristics calculated from remote sensing imagery products, with this comparison performed at the scale of each computational grid.

[0151] The implementation process is as follows:

[0152] Data Preparation: First, we obtained microwave-based snowwater equivalent products for the study area. These images were used to calculate snowmelt for each grid. The calculation method was: (where represents one day earlier than the given value and represents the daily snowmelt rate), thereby generating grid-by-grid image snowmelt characteristic data. Simultaneously, a constructed model framework (including the VIC model and the target prediction model) was used to obtain flow, snowmelt time, and snowmelt runoff results for each grid at a daily step. The output signals of the multi-layer perception probes served as a key source of comparative data.

[0153] Comparison and Matching: At the grid level, snowmelt characteristics calculated from imagery (such as snowmelt amount and snowmelt time) are compared one by one with the corresponding snowmelt characteristics output by the multi-layer perception probe. For example, the actual snowmelt start time for each grid is compared with the model's predicted start time, and the actual snowmelt amount is compared with the model's calculated snowmelt amount. This comparison provides an intuitive view of the model's prediction accuracy for snowmelt runoff at each computational grid. If the model output is close to the imagery result in both time and magnitude, the model's prediction performance for that grid is good. Otherwise, further analysis and model improvement are required.

[0154] Evaluation and optimization: Based on the comparison results, calculate relevant evaluation indicators, such as the Nash efficiency coefficient (NSE), etc., to quantify the overall performance of the model. If the prediction results of the model on a large number of grids match the image data highly, and the NSE value is close to 1, it indicates that the model performance is good. If the degree of match is low and the NSE value deviates greatly from 1, the parameters or structure of the model need to be adjusted. For example, in the present invention, by adjusting the parameters of the gridded LSTM (such as the dropout layer setting, the parameters of the Adam algorithm, etc.) and the relevant parameters of the multi-layer perception probe (such as the weight matrix and the bias vector, etc.), and retraining the model, the model's quantification accuracy of snowmelt runoff is continuously improved, so that it can more accurately reflect the actual snowmelt runoff situation.

[0155] This invention aims to address the shortcomings of existing physical process models and deep learning solutions in considering snowmelt runoff insufficiently. It proposes a framework based on multi-layer perception probes and hybrid neural network coupling to quantify the time and magnitude characteristics of snowmelt runoff, so as to quickly extract snowmelt runoff in complex high-altitude watersheds. The overall process of the technical solution of this invention is as follows Figure 7 shown.

[0156] This paper discloses a technical solution for quantifying snowmelt runoff based on a multi-layer perception probe and hybrid neural network framework to rapidly extract snowmelt runoff in complex, high-altitude watersheds. The present invention constructs a framework combining a physical hydrological model with a neural network and adds a multi-layer perception probe as a tool for interpreting the model's output signals. Supported by remote sensing data, the present invention accurately quantifies the onset of snowmelt and the corresponding runoff contribution compared to traditional hydrological models. The present invention demonstrates higher precision in extracting snowmelt runoff characteristics in alpine watersheds, enhances understanding of nonlinear hydrological processes, and provides reliable snowmelt predictions for watersheds with extensive snow cover.

[0157] It should be noted that the contents not described in detail in the specification of the present invention belong to the common knowledge of those skilled in the art.

[0158] Figure 8 2 is a schematic structural diagram of a snowmelt runoff determination device according to an embodiment of the present invention.

[0159] The present invention provides a snowmelt runoff determination device, such as Figure 8 As shown, the snowmelt runoff determination device includes:

[0160] The first processing module 11 is used to obtain target data of the study area.

[0161] The second processing module 12 is configured to divide the study area into a plurality of computational grids according to preset grid units.

[0162] The third processing module 13 is used to input the target data into a pre-built hydrological model to obtain the output of the hydrological model. The output of the hydrological model includes the runoff results corresponding to each calculation grid.

[0163] The fourth processing module 14 is configured to use the output of the hydrological model as the input of a pre-built target prediction model to generate an output of the target prediction model. The output of the target prediction model represents the prediction result of snowmelt runoff in the study area.

[0164] The target prediction model includes embedded multi-layer perceptron probes, which are used to analyze nonlinear features in the input of the target prediction model.

[0165] In an optional embodiment, the target prediction model includes:

[0166] The input layer is used to standardize the input of the target prediction model to obtain the output of the input layer.

[0167] The output of the input layer is the input of the grid structure.

[0168] The grid structure includes: multiple grid units. The connection relationship between the multiple grid units is constructed based on the time dimension and the space dimension.

[0169] The grid structure is used to analyze the output of the input layer to determine the output of the grid structure.

[0170] The output of the grid structure includes: the instantaneous features corresponding to each computational grid at the standard step size, the hidden states between adjacent computational grids, and the memory states corresponding to the long-term features. The output of the grid structure serves as the input to the multi-layer perception probe.

[0171] The multi-layer perception probe is specifically used to perform nonlinear characteristic analysis on the output of the grid structure to determine the output of the multi-layer perception probe.

[0172] The outputs of the multi-layer perception probe include: flow, snowmelt time and snowmelt runoff at a standard step size for each computational grid.

[0173] The output of the multi-layer perception probe is the input of the output layer.

[0174] The output layer is used to perform feature mapping on the output of the multi-layer perception probe to determine the output of the output layer.

[0175] The output of the output layer includes: flow series, snowmelt time series and snowmelt runoff series at the standard step size corresponding to the outlet section.

[0176] In an optional embodiment, the grid unit includes: a forget gate, an input gate, a candidate memory state, a memory state, an output gate and a hidden state, and a gating mechanism determined according to the forget gate, the input gate, the candidate memory state, the memory state, the output gate and the hidden state.

[0177] In an optional embodiment, when the multi-layer perception probe is a perception probe with two hidden layers, the multi-layer perception probe is set to:

[0178] h1=σ(W1·c t +b1);

[0179] h2=σ(W2·h1+b2).

[0180] Where h1 represents the output vector of the first hidden layer, h2 represents the output vector of the second hidden layer, σ represents the activation function, W1 represents the weight matrix of the first hidden layer, b1 represents the bias vector of the first hidden layer, W2 represents the weight matrix of the second hidden layer, b2 represents the bias vector of the second hidden layer, and c t Represents the output of the previous adjacent grid cell.

[0181] In an optional implementation, the target prediction model further includes: a loss function.

[0182] The loss function is used to determine the loss value based on the target prediction model's predictions and the actual results. The loss value represents the gap between the target prediction model's predictions and the actual results. The loss value is used to adjust the parameter values ​​of the target prediction model.

[0183] In an alternative embodiment, the runoff results are in the form of a time series, including surface runoff, snowmelt, baseflow, net evapotranspiration, potential evapotranspiration, and soil moisture.

[0184] In an optional embodiment, the third processing module 13 is specifically configured to input the meteorological and hydrological data and the underlying surface characteristic data into a pre-built hydrological model to obtain the output of the hydrological model when the target data includes meteorological and hydrological data and underlying surface characteristic data.

[0185] The further functional description of each of the above modules and units is the same as that of the above corresponding embodiments and will not be repeated here.

[0186] The present invention also provides a computer device. Figure 9 , Figure 9 Schematic diagram of the hardware structure of the computer device according to the embodiment of the present invention. Figure 9 As shown, the computer device includes: one or more processors 10, a memory 20, and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. The various components are connected to each other using different buses and can be installed on a common motherboard or installed in other ways as needed. The processor can process instructions executed in the computer device, including instructions stored in or on the memory to display graphical information of a graphical user interface (GUI) on an external input / output device (such as a display device coupled to the interface). In an optional embodiment, if necessary, multiple processors and / or multiple buses can be used together with multiple memories and multiple memories. Similarly, multiple computer devices can be connected, and each device provides some necessary operations (for example, as a server array, a group of blade servers, or a multi-processor device). Figure 9 A processor 10 is taken as an example.

[0187] The processor 10 may be a central processing unit, a network processor, or a combination thereof. The processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit, a programmable logic device, or a combination thereof. The programmable logic device may be a complex programmable logic device, a field programmable gate array, a general purpose array logic, or any combination thereof.

[0188] The memory 20 stores instructions that can be executed by at least one processor 10, so as to enable at least one processor 10 to execute the method shown in the above embodiment.

[0189] The memory 20 may include a program storage area and a data storage area, wherein the program storage area may store an operating device, an application required for at least one function. The data storage area may store data created according to the use of the computer device, etc. In addition, the memory 20 may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In an optional embodiment, the memory 20 may optionally include a memory remotely arranged relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0190] The memory 20 may include volatile memory, such as random access memory. The memory may also include non-volatile memory, such as flash memory, a hard disk, or a solid-state drive. The memory 20 may also include a combination of the above types of memory.

[0191] The computer device further includes a communication interface 30 for the computer device to communicate with other devices or a communication network.

[0192] The embodiment of the present invention also provides a computer-readable storage medium. The method according to the embodiment of the present invention can be implemented in hardware, firmware, or implemented as a computer code that can be recorded in a storage medium, or downloaded through a network and originally stored in a remote storage medium or a non-temporary machine-readable storage medium and will be stored in a local storage medium, so that the method described herein can be stored in such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only storage memory, a random access memory, a flash memory, a hard disk or a solid-state drive, etc. Further, the storage medium can also include a combination of the above-mentioned types of memory. It can be understood that a computer, a processor, a microprocessor controller or programmable hardware includes a storage component that can store or receive software or computer code. When the software or computer code is accessed and executed by a computer, a processor or hardware, the method shown in the above embodiment is implemented.

[0193] A portion of the present invention may be applied as a computer program product, such as a computer program instruction, which, when executed by a computer, can call or provide the method and / or technical solution according to the present invention through the operation of the computer. Those skilled in the art should understand that the form in which the computer program instruction exists in a computer-readable medium includes, but is not limited to, a source file, an executable file, an installation package file, etc. Accordingly, the way in which the computer program instruction is executed by the computer includes, but is not limited to: the computer directly executes the instruction, or the computer compiles the instruction and then executes the corresponding compiled program, or the computer reads and executes the instruction, or the computer reads and installs the instruction and then executes the corresponding installed program. Here, the computer-readable medium may be any available computer-readable storage medium or communication medium that can be accessed by the computer.

[0194] Although the embodiments of the present invention have been described with reference to the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present invention. Such modifications and variations are all within the scope defined by the appended claims.

Claims

1. A method for determining snowmelt runoff, characterized in that: include: Obtain target data for the study area; Dividing the study area into a plurality of computational grids according to preset grid units; Inputting the target data into a pre-built hydrological model to obtain the output of the hydrological model; the output of the hydrological model includes the runoff result corresponding to each calculation grid; Determining the output of the hydrological model as the input of a pre-built target prediction model to generate the output of the target prediction model; the output of the target prediction model represents the prediction result of snowmelt runoff in the study area; The target prediction model includes an embedded multi-layer perception probe, wherein the multi-layer perception probe is used to analyze nonlinear features in the input of the target prediction model; Wherein, the target prediction model includes: Input layer, output layer, the output of the input layer is the input of the grid structure; The grid structure includes: a plurality of grid units; the connection relationship between the plurality of grid units is constructed based on the time dimension and the space dimension; The grid structure is used to analyze the output of the input layer to determine the output of the grid structure; The output of the grid structure includes: the instantaneous features corresponding to each computational grid on a standard step size, the hidden states between adjacent computational grids, and the memory states corresponding to the long-term features; the output of the grid structure is the input of the multi-layer perception probe; The multi-layer perception probe is specifically used to perform nonlinear characteristic analysis on the output of the grid structure to determine the output of the multi-layer perception probe; The output of the multi-layer perception probe includes: flow, snowmelt time and snowmelt runoff of each computational grid at a standard step size; The output of the multi-layer perception probe is the input of the output layer.

2. The method according to claim 1, characterized in that The input layer is used to perform standardization processing on the input of the target prediction model to obtain the output of the input layer; The output layer is used to perform feature mapping on the output of the multi-layer perception probe to determine the output of the output layer; The output of the output layer includes: flow sequence, snowmelt time series and snowmelt runoff sequence on the standard step size scale corresponding to the outlet section.

3. The method according to claim 1, characterized in that The grid unit includes: a forget gate, an input gate, a candidate memory state, a memory state, an output gate and a hidden state, and a gating mechanism determined according to the forget gate, the input gate, the candidate memory state, the memory state, the output gate and the hidden state.

4. The method according to claim 1, wherein When the multi-layer perception probe is a perception probe with two hidden layers, the multi-layer perception probe is set to: h1=σ(W1·c t +b1); h2=σ(W2·h1+b2); Where h1 represents the output vector of the first hidden layer, h2 represents the output vector of the second hidden layer, σ represents the activation function, W1 represents the weight matrix of the first hidden layer, b1 represents the bias vector of the first hidden layer, W2 represents the weight matrix of the second hidden layer, b2 represents the bias vector of the second hidden layer, and c t Represents the output of the previous adjacent grid cell.

5. The method according to claim 1, wherein The target prediction model further includes: a loss function; The loss function is used to determine the loss value based on the predicted result and the actual result of the target prediction model; the loss value represents the size of the gap between the predicted result and the actual result of the target prediction model; the loss value is used to adjust the parameter value of the parameter of the target prediction model.

6. The method according to claim 1, characterized in that The runoff results are in the form of time series, including surface runoff, snowmelt, baseflow, net evapotranspiration, potential evapotranspiration and soil moisture.

7. The method according to claim 1, characterized in that When the target data includes meteorological and hydrological data and underlying surface characteristic data, inputting the target data into a pre-built hydrological model to obtain the output of the hydrological model includes: inputting the meteorological and hydrological data and underlying surface characteristic data into a pre-built hydrological model to obtain the output of the hydrological model.

8. A snowmelt runoff determination device, characterized in that: include: The first processing module is used to obtain target data of the study area; A second processing module is used to divide the study area into a plurality of calculation grids according to preset grid units; The third processing module is used to input the target data into a pre-built hydrological model to obtain the output of the hydrological model; the output of the hydrological model includes the runoff result corresponding to each calculation grid; a fourth processing module, configured to determine the output of the hydrological model as an input of a pre-built target prediction model to generate an output of the target prediction model; the output of the target prediction model represents a prediction result of snowmelt runoff in the study area; The target prediction model includes an embedded multi-layer perception probe, wherein the multi-layer perception probe is used to analyze nonlinear features in the input of the target prediction model; Wherein, the target prediction model includes: Input layer, output layer, the output of the input layer is the input of the grid structure; The grid structure includes: a plurality of grid units; the connection relationship between the plurality of grid units is constructed based on the time dimension and the space dimension; The grid structure is used to analyze the output of the input layer to determine the output of the grid structure; The output of the grid structure includes: the instantaneous features corresponding to each computational grid on a standard step size, the hidden states between adjacent computational grids, and the memory states corresponding to the long-term features; the output of the grid structure is the input of the multi-layer perception probe; The multi-layer perception probe is specifically used to perform nonlinear characteristic analysis on the output of the grid structure to determine the output of the multi-layer perception probe; The output of the multi-layer perception probe includes: flow, snowmelt time and snowmelt runoff of each computational grid at a standard step size; The output of the multi-layer perception probe is the input of the output layer.

9. A computer device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the snowmelt runoff determination method according to any one of claims 1 to 7 by executing the computer instructions.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the snowmelt runoff determination method according to any one of claims 1 to 7.

Images