Snow melting runoff phenomenon prediction method and system, storage medium and equipment

Abstract

The embodiment of the invention discloses a snow melting runoff phenomenon prediction method. The method comprises the following steps: acquiring environment weather data of a target area; inputting the environmental weather data into a ConvLSTM model to obtain a simulated snow water equivalent; inputting the environment weather data into a GRU model to obtain a simulated temperature value; inputting the environmental weather data, the simulated snow water equivalent and the simulated temperature value into a Transform model to obtain a first simulated runoff value; and determining a snow melting runoff phenomenon prediction result according to the first simulation runoff value. According to the method, multiple advanced deep learning models are comprehensively applied, and accurate prediction of the snow melting runoff phenomenon can be realized through multi-step data processing and fusion.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of hydrological forecasting, and in particular to a snowmelt runoff phenomenon prediction method, system, storage medium and equipment. BACKGROUND

[0002] Snowmelt runoff, as an important component of water resources, its accurate prediction is of great significance in many fields. In the water power industry, accurate snowmelt runoff prediction is the key basis for the optimal scheduling of hydropower stations, which helps to reasonably arrange power generation plans, improve power generation efficiency, ensure stable power supply, and at the same time can avoid equipment damage or energy waste caused by water inflow estimation errors. In the field of agriculture, snowmelt runoff provides water for irrigation, and accurate prediction can help farmers plan irrigation schemes in advance to ensure that crops get enough water in the critical growth stage, thereby improving crop yield and ensuring food security. In addition, for the ecological system, snowmelt runoff affects the ecological flow of rivers and is related to the stability of river ecological systems and the maintenance of biodiversity.

[0003] Accurate snowmelt runoff prediction is of great significance to water resources management, flood control and disaster reduction, and water power scheduling. However, traditional snowmelt runoff prediction methods often rely on empirical formulas and physical models, which are difficult to accurately capture the complexity and nonlinearity of the snowmelt process, and the prediction accuracy is limited. SUMMARY

[0004] Therefore, it is necessary to propose a snowmelt runoff phenomenon prediction method to solve the above problems.

[0005] A snowmelt runoff phenomenon prediction method, the method comprising the following steps:

[0006] Collecting environmental weather data of a target area;

[0007] Inputting the environmental weather data into a ConvLSTM model to obtain simulated snow water equivalent;

[0008] Inputting the environmental weather data into a GRU model to obtain simulated temperature values;

[0009] Inputting the environmental weather data, simulated snow water equivalent and simulated temperature values into a Transformer model to obtain first simulated runoff values;

[0010] Determining a snowmelt runoff phenomenon prediction result according to the first simulated runoff values.

[0011] In the above scheme, after determining the snowmelt runoff phenomenon prediction result according to the first simulated runoff values, the method further comprises:

[0012] Inputting the environmental weather data into a ConvLSTM model to obtain second simulated runoff values;

[0013] inputting the environmental weather data into the GRU model to obtain a third simulated runoff value;

[0014] performing weighted summation on the first simulated runoff value, the second simulated runoff value and the third simulated runoff value to determine a snowmelt runoff phenomenon prediction result.

[0015] In the scheme, the ConvLSTM model structure comprises an encoding network and a prediction network, and the prediction network comprises a deconvolution layer and a ConvLSTM layer.

[0016] inputting the environmental weather data into the encoding network to obtain corresponding spatiotemporal features;

[0017] performing up-sampling on the spatiotemporal features by using the deconvolution layer to obtain features of an original spatial resolution;

[0018] inputting the features of the original spatial resolution into the ConvLSTM layer to perform time series prediction and obtain simulated snow water equivalent and a second simulated runoff value.

[0019] In the scheme, the inputting the features of the original spatial resolution into the ConvLSTM layer to perform time series prediction and obtain simulated snow water equivalent and a second simulated runoff value specifically comprises:

[0020] determining relevant parameters of the prediction network according to the following formula:

[0021] f t =σ g (W xf *X t +W hf *H t-1 +W cf *C t-1 +b f )

[0022] i t =σ g (W xi *X t +W hi *H t-1 +W ci *C t-1 +b i )

[0023]

[0024] o t =σ g (W xo *X t +W ho *H t-1+W co *C t +b o )

[0025]

[0026] where σ g is a sigmoid activation function, tanh is a hyperbolic tangent activation function, X t is an input feature of a current time step, H t-1 is a hidden state of a previous time step, C t-1 is a memory cell of a previous step, f t , i t , C t , o t are a forget gate, an input gate, a memory cell and an output gate respectively, and W and b are model parameters.

[0027] determining a hidden state of a ConvLSTM layer according to a feature of the original spatial resolution

[0028]

[0029] determining a simulated snow water equivalent and a second simulated runoff value through the hidden state

[0030]

[0031] where E ConvLSTM is the second simulated runoff value, SWE ConvLSTM is the simulated snow water equivalent,

[0032] W ConvLSTM，o is a convolution kernel weight of the ConvLSTM model, and b ConvLSTM，o is a bias term of the ConvLSTM model, is the hidden state.

[0033] In the above scheme, the GRU model includes an update gate and a reset gate.

[0034] determining a final hidden state of the GRU model according to functions corresponding to the update gate and the reset gate;

[0035] determining a simulated temperature value and a third simulated runoff value according to the final hidden state.

[0036] In the above scheme, the determining of the simulated temperature value and the third simulated runoff value according to the final hidden state specifically includes:

[0037] determining the final hidden state h t of the GRU model according to the following formula:​

[0038]

[0039] wherein is an element-wise multiplication, h t-1 is the hidden state of the previous time step;

[0040] According to the final hidden state h t determining a simulated temperature value and a third simulated runoff value:

[0041] (E GRU , T GRU ) = W GRU，o *h t +b GRU，o

[0042] wherein E GRU is the third simulated runoff value, T GRU is the simulated temperature value, W GRU，o is the convolution kernel weight of the GRU model, and b GRU，o is the bias term of the GRU model.

[0043] In the above scheme, the environmental weather data, simulated snow water equivalent and simulated temperature value are input into the Transformer model to obtain a first simulated runoff value, specifically including:

[0044] The environmental weather data, simulated snow water equivalent and simulated temperature value are input into the Transformer model to obtain a Transformer model hidden state

[0045] According to the Transformer model hidden state determining a first simulated runoff:

[0046]

[0047] wherein W Transformer，o is the weight of the Transformer model, b Transformer，o is the bias term of the Transformer model, and E Transformer is the first simulated runoff.

[0048] The present application proposes a snowmelt runoff phenomenon prediction system, which comprises a data acquisition unit, a ConvLSTM model unit, a GRU model unit, a Transformer model unit and a judgment unit.

[0049] The data acquisition unit is used for acquiring environmental weather data of a target region.

[0050] The ConvLSTM model unit is configured to input the environmental weather data into a ConvLSTM model to obtain simulated snow water equivalent.

[0051] The GRU model unit is configured to input the environmental weather data into a GRU model to obtain simulated temperature values.

[0052] The Transformer model unit is configured to input the environmental weather data, the simulated snow water equivalent, and the simulated temperature values into a Transformer model to obtain first simulated runoff values.

[0053] The judging unit is configured to determine a snowmelt runoff phenomenon prediction result according to the first simulated runoff values.

[0054] The present application provides a readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the following steps:

[0055] Collect environmental weather data of a target area.

[0056] Input the environmental weather data into a ConvLSTM model to obtain simulated snow water equivalent.

[0057] Input the environmental weather data into a GRU model to obtain simulated temperature values.

[0058] Input the environmental weather data, the simulated snow water equivalent, and the simulated temperature values into a Transformer model to obtain first simulated runoff values.

[0059] Determine a snowmelt runoff phenomenon prediction result according to the first simulated runoff values.

[0060] The present application provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer program causes the processor to perform the following steps:

[0061] Collect environmental weather data of a target area.

[0062] Input the environmental weather data into a ConvLSTM model to obtain simulated snow water equivalent.

[0063] Input the environmental weather data into a GRU model to obtain simulated temperature values.

[0064] Input the environmental weather data, the simulated snow water equivalent, and the simulated temperature values into a Transformer model to obtain first simulated runoff values.

[0065] Determine a snowmelt runoff phenomenon prediction result according to the first simulated runoff values.

[0066] The snowmelt runoff phenomenon prediction method has the following beneficial effects: the snowmelt runoff phenomenon prediction method collects target area environment weather data, uses a ConvLSTM model to mine data space-time characteristics to accurately simulate snow water equivalent, uses a GRU model to grasp time sequence rules to obtain simulated temperature values, and uses a Transformer model to fuse environment weather data, simulated snow water equivalent and simulated temperature values by means of a powerful attention mechanism, comprehensively considers complex relationships among multiple factors, and thus obtains a first simulated runoff value which is high in accuracy and strong in adaptability, and the snowmelt runoff phenomenon prediction result determined according to the first simulated runoff value can effectively reflect an actual situation and provide a reliable basis for relevant decision-making. BRIEF DESCRIPTION OF DRAWINGS

[0067] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description only constitute some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0068] Among them:

[0069] Figure 1 It is a flowchart of the snowmelt runoff phenomenon prediction method in an embodiment;

[0070] Figure 2 It is a schematic diagram of the ConvLSTM model structure in an embodiment;

[0071] Figure 3 It is a spatial distribution comparison chart of the multi-year average snow water equivalent simulated by the ConvLSTM model in the test set and the remote sensing measured value;

[0072] Figure 4 It is a comparison chart of the simulated temperature and the measured temperature changing with time of the GRU model in the test set;

[0073] Fig. 5(A) is a exceedance probability distribution chart of the site simulation and measured flow based on linear conversion;

[0074] Fig. 5(B) is a exceedance probability distribution chart of the site simulation and measured flow based on logarithmic conversion. DETAILED DESCRIPTION

[0075] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments only constitute some embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0076] In the following description, numerous specific details are set forth in order to provide a more thorough understanding of the invention; however, it will be apparent to those skilled in the art that the invention may be practiced without one or more of these details; in other instances, certain technical features well-known in the art have not been described in order to avoid confusion with the invention. It should be understood that the invention can be practiced in different forms and should not be construed as limited to the embodiments set forth herein; rather, these embodiments are provided to make the disclosure thorough and complete and to fully convey the scope of the invention to those skilled in the art.

[0077] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the invention. When used herein, the singular forms “a,” “an,” and “the” are also intended to include the plural forms, unless the context clearly indicates otherwise. The terms “comprising” and / or “including,” when used in this specification, identify the presence of said features, integers, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups. When used herein, the term “and / or” includes any and all combinations of the associated listed items.

[0078] To fully understand the present invention, a detailed structure will be presented in the following description in order to illustrate the technical solution proposed by the present invention; optional embodiments of the present invention are described in detail below, however, in addition to these detailed descriptions, the present invention may have other embodiments.

[0079] like Figure 1 As shown, in one embodiment, a method for predicting snowmelt runoff is provided. This method includes steps S101 to S105, which are detailed below:

[0080] S101. Collect environmental weather data for the target area;

[0081] Comprehensive collection of environmental weather data for the target area, including meteorological data (temperature T, precipitation P, etc.), remote sensing data (snow depth DS and snow water equivalent SWE), underlying surface data, and hydrological data, provides a rich information foundation for subsequent model predictions, ensuring that the model can comprehensively consider various factors affecting snowmelt runoff. Precise data collection methods enable the acquisition of accurate meteorological conditions in the target area, reducing prediction bias caused by data errors.

[0082] S102. Input environmental weather data into the ConvLSTM model to obtain the simulated snow water equivalent;

[0083] The ConvLSTM model combines the advantages of Convolutional Neural Network (CNN) and Long Short-Term Memory Network (LSTM), and can effectively extract the spatio-temporal features of environmental weather data. For the snowmelt process, the changes of meteorological elements such as temperature and precipitation in time and space will affect the snow water equivalent, and the ConvLSTM model can capture these complex spatio-temporal variation rules, so as to more accurately simulate the dynamic changes of snow water equivalent. The ConvLSTM model has strong non-linear mapping ability, which can establish the complex relationship between environmental weather data and snow water equivalent, and improve the accuracy of simulating snow water equivalent.

[0084] In S103, environmental weather data is input into the GRU model to obtain a simulated temperature value.

[0085] The GRU model is a simplified variant of the LSTM model, which has the characteristics of simple structure and high computational efficiency. When processing environmental weather data to obtain a simulated temperature value, the GRU model can quickly complete the calculation under the premise of ensuring a certain prediction accuracy, improving the efficiency of the entire prediction process. The GRU model can capture the long-term dependence relationship in the environmental weather data. The change of temperature is often affected by the previous meteorological conditions, and the GRU model can learn the patterns in the historical data to better predict the future temperature trend, thereby providing more accurate simulated temperature values.

[0086] In S104, environmental weather data, simulated snow water equivalent, and simulated temperature value are input into the Transformer model to obtain a first simulated runoff value.

[0087] The Transformer model is based on the self-attention mechanism and can capture global information in the input data. In the snowmelt runoff prediction, there is a complex interaction relationship between environmental weather data, simulated snow water equivalent, and simulated temperature value, and the Transformer model can consider the global influence of these factors to more accurately predict the runoff value.

[0088] In S105, a snowmelt runoff phenomenon prediction result is determined according to the first simulated runoff value.

[0089] The first simulated runoff value provides important reference information about the snowmelt runoff phenomenon. Through the analysis of the simulated runoff value, the intensity and duration characteristics of the snowmelt runoff can be predicted, which provides a decision basis for water resource management, such as reasonably arranging the reservoir storage and discharge plan, and optimizing the allocation and utilization of water resources.

[0090] In some embodiments, after determining the snowmelt runoff phenomenon prediction result according to the first simulated runoff value, the method further includes:

[0091] In S106, environmental weather data is input into the ConvLSTM model to obtain a second simulated runoff value.

[0092] Specifically, the ConvLSTM model can extract the spatio-temporal features in the environmental weather data, capture the potential influence of meteorological changes in different regions and at different times on runoff, and thus generate the second simulated runoff value, providing a runoff prediction result considering the time and space dimensions for subsequent fusion prediction.

[0093] Inputting the environmental weather data into the GRU model to obtain a third simulated runoff value;

[0094] In snowmelt runoff prediction, meteorological data often has time series characteristics, and the GRU model can efficiently process these sequence data to extract features related to runoff and generate the third simulated runoff value.

[0095] The first simulated runoff value, the second simulated runoff value, and the third simulated runoff value are weighted and summed to determine the snowmelt runoff phenomenon prediction result.

[0096] By weighting and summing the simulated runoff values generated by the three different models, the advantages of each model can be fully utilized. Each model may have better performance in different aspects, such as the ConvLSTM model in spatio-temporal feature processing, the GRU model in sequence data processing, and the Transformer model (mentioned earlier for obtaining the first simulated runoff value) in global information capture. Weighted summation can integrate these advantages, reduce the bias of a single model, and improve the accuracy of the prediction result.

[0097] Preferably, the weights in the weighted summation can be adjusted according to actual conditions. In actual applications, the influencing factors of snowmelt runoff may vary in different regions and seasons. By adjusting the weights, different application scenarios can be better adapted to, and the prediction result can better meet the actual needs. For example, in some areas, temperature changes have a more significant impact on snowmelt runoff, and the weight of the model based on temperature-related models (such as the GRU model, which may be more affected by temperature) can be appropriately increased.

[0098] Preferably, since the ConvLSTM model needs to process multi-dimensional time series data containing rich spatial information, preprocessing is required before runoff prediction, and the specific process is as follows:

[0099] (1) Missing data filling. A general interpolation method is used to flexibly handle missing data in different dimensions for meteorological data, underlying surface data, and remote sensing data. Taking precipitation data as an example, for a three-dimensional data set P(x, y, t), where x and y are spatial coordinates and t is the time dimension. If the time dimension data is missing, linear interpolation can be used with adjacent time points.

[0100]

[0101] Here, t1 and t2 are two time points adjacent to the missing data point t, and P(x,y,t1) and P(x,y,t2) are known precipitation data. If a value is missing at a certain spatial location, bilinear interpolation is used:

[0102]

[0103] (2) Data normalization. Z-score normalization was used to normalize the watershed grid-by-grid data. For each grid (i.e., each spatial location), the mean and standard deviation of the grid over time were calculated, and then these values ​​were used for normalization to ensure that the data were of consistent magnitude and to avoid the influence of features of different scales on model training.

[0104]

[0105] Where P(x,y,t) is the raw data value at position (x,y) at time t, μ(x,y) is the time mean at position (x,y), σ(x,y) is the time standard deviation at position (x,y), and P(x,y,t) is the time mean at position (x,y). ′ These are standardized data values. Spatial data resampling. This unifies spatial data of different resolutions to the same spatial resolution, ensuring the model can fully utilize spatial information.

[0106] like Figure 2 As shown, in some embodiments, the ConvLSTM model structure includes an encoding network and a prediction network, wherein the prediction network includes deconvolutional layers and ConvLSTM layers.

[0107] Environmental weather data is input into the coding network to obtain the corresponding spatiotemporal characteristics;

[0108] Deconvolutional layers are used to upsample spatiotemporal features to obtain features with the original spatial resolution.

[0109] The features of the original spatial resolution are input into the ConvLSTM layer for time series prediction to obtain the simulated snow water equivalent and the second simulated runoff value.

[0110] Specifically, in snowmelt runoff simulation, the goal is to predict runoff during the snowmelt process based on input data such as meteorological conditions, land use, vegetation cover, and satellite remote sensing. Furthermore, to achieve multi-step forecasting, accurate spatiotemporal series data needs to be generated, requiring the model to not only capture temporal changes but also understand the complexity of spatial distribution.

[0111] In some embodiments, the features of the original spatial resolution are input into a ConvLSTM layer for time series prediction to obtain simulated snowmelt equivalent and second simulated runoff values, specifically including:

[0112] The relevant parameters of the prediction network are determined according to the following formula:

[0113] f t = σ g (W xf *X t +W hf *H t-1 +W cf *C t-1 +b f )

[0114] i t = σ g (W xi *X t +W hi *H t-1 +W ci *C t-1 +b i )

[0115]

[0116] o t = σ g (W xo *X t +W ho *H t-1 +W co *C t +b o )

[0117] H t = o t °tanh(C t )

[0118] In the formula, σ g is a sigmoid activation function, tanh is a hyperbolic tangent activation function, X t is the input feature of the current time step, H t-1 is the hidden state of the previous time step, C t-1 is the memory cell of the previous step, f t , i t , C t , o t are the forget gate, input gate, memory cell and input gate respectively, and W and b are model parameters.

[0119] Determine the hidden state of the ConvLSTM layer according to the features of the original spatial resolution

[0120]

[0121] by the hidden state determining a simulated snow water equivalent and a second simulated runoff value:

[0122]

[0123] wherein E ConvLSTM is the second simulated runoff value, SWE ConvLSTM is the simulated snow water equivalent,

[0124] W ConvLSTM，o is a convolution kernel weight of the ConvLSTM model, and b ConvLSTM，o is a bias term of the ConvLSTM model. is the hidden state.

[0125] In some embodiments, the GRU model comprises an update gate and a reset gate;

[0126] determining a final hidden state of the GRU model according to functions corresponding to the update gate and the reset gate;

[0127] determining a simulated temperature value and a third simulated runoff value according to the final hidden state.

[0128] In some embodiments, determining the simulated temperature value and the third simulated runoff value according to the final hidden state specifically comprises:

[0129] determining the final hidden state h t of the GRU model according to the following formula:

[0130]

[0131] wherein is an element-wise multiplication, h t-1 is the hidden state of the previous time step;

[0132] determining the simulated temperature value and the third simulated runoff value according to the final hidden state h t

[0133] (E GRU , T GRU ) = W GRU，o *h t +b GRU，o

[0134] wherein E GRU is the third simulated runoff value, T GRU is the simulated temperature value, W GRU，o is a convolution kernel weight of the GRU model, and b GRU，o is a bias term of the GRU model.

[0135] ​A Transformer is mainly composed of two parts: Encoder and Decoder. In ensemble prediction, the Encoder part of the Transformer can be used to extract high-level features of time series data. The Transformer structure captures more extensive dependencies by weighting each part of the input through a self-attention mechanism. Unlike traditional recurrent neural networks, the Transformer does not rely on step-by-step iterative calculations, but processes all input data in parallel, making it more efficient in handling long time series and high-dimensional data. It includes:

[0136] 1) Self-attention mechanism allows the model to assign a weight to each input element, which represents the relationship between that element and other elements. Specifically, given an input sequence, the Transformer calculates the relationship (attention weight) between each input and all other inputs, and then weights all elements of the input sequence with these weights.

[0137]

[0138] where Q is the query vector, K is the key vector, V is the value vector, d k is the dimension of the key vector, the normalization factor, the attention weight is generated by calculating the similarity between Q and K, and then the weight is applied to V.

[0139] 2) Feedforward neural network. After each self-attention layer, the Transformer further processes information through a feedforward neural network. These neural networks consist of two linear transformations and an activation function (usually ReLU).

[0140] 3) Residual connection and layer normalization. The Transformer uses a residual connection to avoid the problem of gradient vanishing and uses layer normalization to stabilize the training process.

[0141] 4) Model output prediction value: After multiple layers of Encoder and Decoder structure, the hidden state output by the Transformer model:

[0142]

[0143] where X 1:t represents the input sequence at time step t, is the corresponding hidden representation.

[0144] In some embodiments, environmental weather data, simulated snow water equivalent, and simulated temperature values are input into the Transformer model to obtain a first simulated runoff value, specifically including:

[0145] Inputting the environmental weather data, simulated snow water equivalent and simulated temperature value into the Transformer model to obtain a hidden state of the Transformer model

[0146] According to the hidden state of the Transformer model Determine a first simulated runoff:

[0147]

[0148] In the formula, W Transformer，o is a weight of the Transformer model, b Transformer，o is a bias term of the Transformer model, E Transformer is the first simulated runoff.

[0149] Preferably, the ConLSTM simulated runoff value, the GRU simulated runoff value and the Transformer simulated runoff value are weighted and summed to obtain a snowmelt runoff phenomenon prediction result, specifically comprising:

[0150] P final = ω1(t)·E ConvLSTM + ω2(t)·E GRU + ω3(t)·E Transformer

[0151] In the formula, P final is a snowmelt runoff phenomenon prediction value, E ConvLSTM , E GRU and E Transformer are simulated runoff values of the ConvLSTM, GRU and Transformer models respectively, and ω1, ω2 and ω3 are weight coefficients.

[0152] Specifically, the calculation formula of the weight coefficient is as follows:

[0153]

[0154] Wherein, NSE i is the Nash efficiency coefficient of the i-th model in the training period, and the weight coefficient ω i is inversely proportional to the model prediction error.

[0155] For example Figure 3ConvLSTM model simulation of the average snow water equivalent in the test set and the spatial distribution of the remote sensing measured value are shown, the snow water equivalent (SWE) and runoff simulation are carried out by using the ConvLSTM model, the catchment area above the Lijiang Hydropower Station of the Jinsha River is taken as the research area, the historical measured data from 2010 to 2017 is used for training, and the data from 2018 to 2023 is used for verification. The input variables include precipitation (P), air temperature (T), remote sensing snow depth (SD), snow water equivalent (SWE), normalized vegetation index (NDVI) and flow (Q). The spatial resolution of the model is set to 5km*5km, and the time step is 1 day. In terms of snow water equivalent simulation, the NSE coefficient of the ConvLSTM model on the training set is 0.87, indicating that the model can well capture the spatio-temporal variation characteristics of SWE. On the test set, the NSE decreases to 0.81, but still maintains high simulation accuracy. In terms of snowmelt runoff simulation, based on the predicted SWE of ConvLSTM as input, combined with precipitation, air temperature and other variables, runoff simulation is further carried out. The NSE coefficient on the training set is 0.83, and the NSE on the test set decreases to 0.76, still with strong generalization ability.

[0156] As shown in Figure 4 the comparison chart of the simulated temperature and the measured temperature of the GRU model in the test set over time, the temperature (T) and runoff simulation are carried out by using the GRU model, the catchment area above the Lijiang Hydropower Station of the Jinsha River is taken as the research area, the historical measured data from 2010 to 2017 is used for training, and the data from 2018 to 2023 is used for verification. The input variables are basin average precipitation, temperature, normalized vegetation index, snow depth and snow water equivalent. The model input and output are time series, and the time step is 1 day.

[0157] In terms of temperature simulation, the NSE coefficient of the GRU model on the training set is 0.88, indicating that the model can well capture the time series variation characteristics of temperature. On the test set, the NSE decreases to 0.85, but still maintains high simulation accuracy. In terms of snowmelt runoff simulation, based on the predicted T of GRU as input, combined with precipitation, normalized vegetation index and other variables, runoff simulation is further carried out. The NSE coefficient on the training set is 0.79, and the NSE on the test set decreases to 0.74, still with certain generalization ability.

[0158] Fig. 5(A) shows a probability distribution diagram of exceeding based on linear conversion of site simulation and measured flow, Fig. 5(B) shows a probability distribution diagram of exceeding based on logarithmic conversion of site simulation and measured flow, according to the snow water equivalent data simulated by ConvLSTM, the temperature data simulated by GRU, combined with other meteorological, underlying surface and hydrological data to train the Transformer model, the snowmelt runoff value simulated by the model is obtained, the NSE coefficient on the training set is 0.88, and the NSE on the verification set is 0.83. The final snowmelt runoff value is obtained by using the weighted average of the simulated runoff values of the ConvLSTM model, the GRU model and the Transformer model, the NSE coefficient on the training period is 0.92, and the NSE coefficient on the verification period is 0.89, the simulation accuracy is high, and the system can provide effective support for water power station inflow prediction.

[0159] The application provides a snowmelt runoff phenomenon prediction system, which comprises a data acquisition unit, a ConvLSTM model unit, a GRU model unit, a Transformer model unit and a judgment unit.

[0160] The data acquisition unit is used for acquiring environmental weather data of a target region.

[0161] The ConvLSTM model unit is used for inputting the environmental weather data into a ConvLSTM model to obtain simulated snow water equivalent.

[0162] The GRU model unit is used for inputting the environmental weather data into a GRU model,

[0163] to obtain a simulated temperature value.

[0164] The Transformer model unit is used for inputting the environmental weather data, the simulated snow water equivalent and the simulated temperature value into a Transformer model to obtain a first simulated runoff value.

[0165] The judgment unit is used for determining a snowmelt runoff phenomenon prediction result according to the first simulated runoff value.

[0166] The application provides a readable storage medium, which stores a computer program, and the computer program is executed by a processor to make the processor execute the following steps:

[0167] Acquiring environmental weather data of a target region.

[0168] Inputting the environmental weather data into a ConvLSTM model to obtain simulated snow water equivalent.

[0169] Inputting the environmental weather data into a GRU model to obtain a simulated temperature value.

[0170] Inputting the environmental weather data, the simulated snow water equivalent and the simulated temperature value into a Transformer model to obtain a first simulated runoff value.

[0171] a model, to obtain a first simulated runoff value;

[0172] determine a snowmelt runoff phenomenon prediction result according to the first simulated runoff value.

[0173] The application provides a computer device, including a memory and a processor, the memory stores a computer program, and the computer program is executed by the processor to perform the following steps:

[0174] collect environmental weather data of a target area;

[0175] input the environmental weather data into a ConvLSTM model to obtain a simulated snow water equivalent;

[0176] input the environmental weather data into a GRU model to obtain a simulated temperature value;

[0177] input the environmental weather data, the simulated snow water equivalent and the simulated temperature value into a Transformer model to obtain a first simulated runoff value;

[0178] a model, to obtain a first simulated runoff value;

[0179] determine a snowmelt runoff phenomenon prediction result according to the first simulated runoff value.

[0180] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by a computer program instructing related hardware, and the program can be stored in a non-volatile computer readable storage medium. When the program is executed, the program can include the processes of the above-mentioned embodiments. Any reference to a memory, storage, database or other medium used in the embodiments provided by the application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM) and memory bus dynamic RAM (RDRAM).

[0181] Any technical features in the above-described embodiments can be combined in any manner, and for the sake of brevity, not all possible combinations are described, however, as long as there is no conflict in combining the technical features, the combinations shall be considered within the scope of the present disclosure.

[0182] The above-described embodiments are merely illustrative for the present application and are described in more specific and detailed manner, but should not be construed as limiting the scope of the present application. It should be noted that, for those skilled in the art, some modifications and improvements can be made without departing from the concept of the present application, and these are all within the scope of the present application. The above disclosure is merely the preferred embodiments of the present application, and of course cannot limit the scope of the present application. Therefore, any equivalent changes made according to the claims of the present application shall be considered within the scope of the present application.

Claims

1. A method of predicting a snowmelt runoff phenomenon, characterized by, The method comprises: collecting environmental weather data of a target area; inputting the environmental weather data into a ConvLSTM model to obtain simulated snow water equivalent; inputting the environmental weather data into a GRU model to obtain simulated temperature values; inputting the environmental weather data, simulated snow water equivalent, and simulated temperature values into a Transformer model to obtain first simulated runoff values; determining a snowmelt runoff phenomenon prediction result according to the first simulated runoff values.

2. The snowmelt runoff phenomenon prediction method according to claim 1, characterized by, After determining the snowmelt runoff phenomenon prediction result according to the first simulated runoff values, the method further comprises: inputting the environmental weather data into the ConvLSTM model to obtain second simulated runoff values; inputting the environmental weather data into the GRU model to obtain third simulated runoff values; performing weighted summation on the first simulated runoff values, second simulated runoff values, and third simulated runoff values to determine the snowmelt runoff phenomenon prediction result.

3. The snowmelt runoff phenomenon prediction method according to claim 2, characterized by, The ConvLSTM model structure comprises an encoding network and a prediction network, and the prediction network comprises a deconvolution layer and a ConvLSTM layer; inputting the environmental weather data into the encoding network to obtain corresponding spatio-temporal features; performing up-sampling on the spatio-temporal features by using the deconvolution layer to obtain features of an original spatial resolution; inputting the features of the original spatial resolution into the ConvLSTM layer to perform time series prediction and obtain simulated snow water equivalent and second simulated runoff values.

4. The snowmelt runoff phenomenon prediction method according to claim 3, characterized by, The inputting of the features of the original spatial resolution into the ConvLSTM layer to perform time series prediction and obtain simulated snow water equivalent and second simulated runoff values specifically comprises: determining relevant parameters of the prediction network according to the following formula: where σ g is a sigmoid activation function, tanh is a hyperbolic tangent activation function, X t is the input feature of the current time step, H t-1 is the hidden state of the previous time step, C t-1 is the memory cell of the previous step, f t , i t , C t , o t are the forget gate, input gate, memory cell and input gate respectively, and W and b are model parameters; determining a hidden state of a ConvLSTM layer as a function of the original spatial resolution by the hidden state determining a simulated snow water equivalent and a second simulated runoff value: In the formula, E ConvLSTM is a second simulated runoff value, SWE ConvLSTM is a simulated snow water equivalent, W ConvLSTM，o is a convolution kernel weight of the ConvLSTM model, b ConvLSTM，o is a bias term of the ConvLSTM model, is a hidden state.

5. The snowmelt runoff phenomenon prediction method according to claim 2, characterized by, The GRU model comprises an update gate and a reset gate; determining a final hidden state of the GRU model according to functions corresponding to the update gate and the reset gate; determining simulated temperature values and third simulated runoff values according to the final hidden state.

6. The snowmelt runoff phenomenon prediction method according to claim 5, characterized by, The determination of the simulated temperature values and the third simulated runoff values according to the final hidden state specifically comprises: The final hidden state h of the GRU model is determined according to the following formula t : where is the element-wise multiplication, h t-1 is the hidden state of the previous time step. According to the final hidden state h t determining an analog temperature value and a third analog runoff value: (E GRU ,T GRU )=W GRU，o *h t +b GRU，o wherein E GRU is a third simulated runoff value, T GRU is a simulated temperature value, W GRU，o is a convolution kernel weight of the GRU model, b GRU，o is a bias term of the GRU model.

7. The snowmelt runoff phenomenon prediction method according to claim 1, characterized by, The inputting of the environmental weather data, simulated snow water equivalent, and simulated temperature values into the Transformer model to obtain first simulated runoff values specifically comprises: inputting the environmental weather data, the simulated snow water equivalent and the simulated temperature value into a Transformer model to obtain a hidden state of the Transformer model According to the Transformer model hidden state determining a first simulated runoff: where W Transformer，o are weights of the Transformer model, b Transformer，o are bias terms of the Transformer model, E Transformer is the first simulated runoff.

8. A snowmelt runoff phenomenon prediction system characterized by, The system comprises a data collection unit, a ConvLSTM model unit, a GRU model unit, a Transformer model unit, and a judgment unit. The data collection unit is configured to collect environmental weather data of a target area. The ConvLSTM model unit is configured to input the environmental weather data into a ConvLSTM model to obtain simulated snow water equivalent. The GRU model unit is configured to input the environmental weather data into a GRU model to obtain simulated temperature values. The Transformer model unit is configured to input the environmental weather data, simulated snow water equivalent, and simulated temperature values into a Transformer model to obtain first simulated runoff values. The judgment unit is configured to determine a snowmelt runoff phenomenon prediction result according to the first simulated runoff values. 9.A readable storage medium, storing a computer program, wherein the computer program is executed by a processor to make the processor perform the steps of the method in any one of claims 1 to 7. 10.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer program is executed by the processor to make the processor perform the steps of the method in any one of claims 1 to 7.

Images