Hydrological forecasting method based on deep integrated learning and physical mechanism

Abstract

The invention provides a hydrological forecasting method based on deep ensemble learning and a physical mechanism, and belongs to the technical field of hydrological forecasting, and the method comprises the steps: collecting remote sensing data and time series data, respectively extracting features through a CNN and an LSTM, and carrying out the fusion through a double-cross attention module. Thirdly, combining the hydrological physical model SWAT with the deep learning model CNN to obtain enhanced features; and constructing a plurality of deep learning models as basic learners, performing training by using the training set data, and forming a final prediction result through an integration strategy. According to the hydrological forecasting method based on the deep integrated learning and the physical mechanism, the advantages of the physical mechanism and the deep learning are combined, and the accuracy and the stability of hydrological forecasting are improved. The method is of great significance to water resource management and flood control and disaster reduction.

Description

Technical Field

[0001] The present invention relates to the technical field of hydrological forecasting, and in particular to a hydrological forecasting method based on deep integrated learning and physical mechanism. Background Art

[0002] As an important tool for water resources management and flood prevention and disaster reduction, hydrological forecasting is of great importance. With water resources becoming increasingly scarce and flood disasters occurring frequently, accurate hydrological forecasting is of great significance for protecting people's lives and property and making rational use of water resources. However, traditional hydrological forecasting methods face many challenges, which limits their effectiveness in practical applications.

[0003] Traditional hydrological forecasting methods mainly rely on hydrological physical models. These models simulate and predict hydrological processes based on physical mechanisms and can reflect the operating laws of hydrological systems to a certain extent. Among them, representative models such as the SWAT model have been widely used in the fields of water resources management and environmental science research. They simulate and predict the hydrological dynamics of the basin by simulating hydrological processes such as rainfall, runoff, and evapotranspiration. However, when dealing with complex hydrological processes, these models are often limited by various factors such as data acquisition, model parameter settings, and computational efficiency.

[0004] First, data acquisition is a major problem in the application of hydrological physical models. Due to the complexity and uncertainty of the hydrological system, it is often very difficult to obtain comprehensive and accurate hydrological data. In addition, the natural environment and socio-economic conditions in different regions vary greatly, resulting in uneven temporal and spatial distribution of hydrological data, which further increases the difficulty of data acquisition. Incomplete and inaccurate data will directly affect the simulation effect and prediction accuracy of the hydrological physical model.

[0005] Secondly, model parameter setting is also an important factor affecting the prediction accuracy of hydrological and physical models. Hydrological and physical models usually contain a large number of parameters, and the values ​​of these parameters will directly affect the simulation effect and prediction results of the model. However, due to the complexity and uncertainty of the hydrological system, it is often very difficult to determine the model parameters. In practical applications, empirical formulas, historical data fitting and other methods are usually used to determine model parameters, but these methods often have certain subjectivity and uncertainty, resulting in inaccurate and unstable values ​​of model parameters.

[0006] Furthermore, computational efficiency is another important factor that limits the widespread application of hydrological physics models. Since hydrological physics models usually contain a large number of physical processes and computational steps, the computational workload of the models is huge and the computational efficiency is low. Especially in the simulation of large-scale watersheds or complex hydrological systems, the computational efficiency of hydrological physics models is more prominent. This not only increases the cost and time of model application, but also limits the scope and effect of the models in practical applications.

[0007] In response to the problems existing in traditional hydrological forecasting methods, machine learning algorithms such as deep learning have provided new ideas and methods for hydrological forecasting. Deep learning models, such as convolutional neural networks (CNN) and long short-term memory neural networks (LSTM), can automatically learn features from large amounts of data and make predictions, and have strong nonlinear fitting and adaptive capabilities. These models perform well in processing complex, nonlinear, and high-dimensional data, and have therefore received widespread attention and application in the field of hydrological forecasting.

[0008] However, pure deep learning models also have certain limitations. Due to the lack of physical mechanism constraints, deep learning models are prone to produce unreasonable prediction results when data is sparse or abnormal. In addition, the prediction results of deep learning models are often difficult to explain and understand, which brings certain difficulties to the application and verification of the model.

[0009] Therefore, how to combine the physical mechanism of hydrological physics models and the data-driven capabilities of deep learning models to improve the accuracy and stability of hydrological forecasts has become a hot topic and difficulty in current research on hydrological forecasting technology. Summary of the invention

[0010] The purpose of the present invention is to provide a hydrological forecasting method based on deep integrated learning and physical mechanism, which combines the advantages of physical mechanism and deep learning and improves the accuracy and stability of hydrological forecasting. This method is of great significance for water resources management and flood prevention and disaster reduction.

[0011] To achieve the above object, the present invention provides a hydrological forecasting method based on deep integrated learning and physical mechanism, comprising the following steps: S1. Collect and preprocess remote sensing data and time series data, construct a data set, and divide the data set into a training set, a validation set, and a test set; S2. On the training set and the validation set, the convolutional neural network (CNN) is used to extract remote sensing data features from the remote sensing data, denoted as , using the long short-term memory neural network LSTM to extract time series features from time series data, denoted as ; S3, yes and Normalize the remote sensing features using a double cross attention module , normalized time series features Perform fusion to obtain fusion features; S4, combining the hydrological physics model with the deep learning model to obtain enhanced features; S5. Construct multiple deep learning models as basic learners, use the training set data to train the basic learners, and integrate the prediction results of the basic learners through the integration strategy to form the final prediction results. Then use the validation set data to verify the model to obtain the optimized deep integration learning model. S6. Input the test set data into the optimized deep ensemble learning model to generate flood forecast results.

[0012] Preferably, in step S1, the time series data includes hydrological data and meteorological data, the hydrological data includes water level, water temperature, water evaporation, flow, groundwater level, ice conditions, the meteorological data includes temperature, precipitation, air pressure, wind speed, and the remote sensing data includes MODIS data products and meteorological satellite data.

[0013] Preferably, in step S1, preprocessing includes data cleaning and normalization; The normalization formula is as follows: ; in, is the original pixel value or the original time series data value; is the normalized pixel value or the normalized time series data value; It is the minimum value of pixel value or the minimum value of time series data value; It is the maximum value of pixel value or the maximum value of time series data.

[0014] Preferably, in step S2, the convolutional neural network CNN includes a convolution layer, an activation function and a pooling layer, and the long short-term memory neural network LSTM includes a forget gate, an input gate, a unit state candidate, a unit state update, an output gate and a hidden state update; Among them, the convolution layer: ; in, It is the convolution operation at position Output: is the convolution kernel; is the input remote sensing image; Represents the convolution operation; The input image is The pixel value at ; is the position of the convolution kernel relative to the output feature map Offset The value at , It is the position coordinate on the output feature map; , is the position coordinate on the convolution kernel; The activation function is the ReLU activation function; The forget gate, input gate, cell state candidate, cell state update, output gate, and hidden state update are as follows: ; ; ; ; ; ; in, is the output of the forget gate; is the sigmoid function; is the weight matrix of the forget gate; represents the vector formed by concatenating the hidden state of the previous time step and the input of the current time step; is the hidden state of the previous time step; is the input of the current time step; is the bias term of the forget gate; is the output of the input gate; is the weight matrix of the input gate; is the bias term of the input gate; Candidate for unit status; is the updated unit status; is the bias term of the cell state candidate; is the updated unit status; is the cell state at the previous time step; represents element-wise multiplication; is the output of the output gate; is the weight matrix of the output gate; is the bias term of the output gate; is the updated hidden state.

[0015] Preferably, in step S3, the dual cross attention module includes two parts: channel attention and spatial attention. The channel attention calculation and Channel attention is achieved through global average pooling and fully connected layers; Spatial Attention Computation and The spatial attention is achieved through global average pooling and fully connected layers.

[0016] Preferably, in step S3, a double cross attention module is used to fuse the normalized remote sensing features and the normalized time series features, and the specific operation is as follows: Channel Fusion: ; in, is the channel fusion feature; and They are and The channel attention weight of Spatial Fusion: ; in, It is the spatial fusion feature; and They are and The spatial attention weights of Fusion features: ; in, For fusion features; Using principal component analysis Perform dimensionality reduction processing.

[0017] Preferably, in step S4, the hydrological physics model is a SWAT model, and the deep learning model is a CNN; Quantify the SWAT model output: ; in, It is the quantitative output of the SWAT model; It is the actual hydrological data; is a quantization function; Fuse the quantized output of the SWAT model with the fused features of step S3: ; in, It is the fusion feature; is the fusion function; Use the fused features to train the CNN model, and the loss function is as follows: ; in, is the loss function; is the output predicted by the model; It is the actual hydrological data; Use the back-propagation algorithm to optimize the weights of the CNN model: ; in, is the updated weight; is the old weight; is the learning rate; is the loss function Weight The partial derivative of Integrate the physical process constraints provided by the SWAT model into the CNN model: ; in, is the final prediction output, is the adjustment function considering physical constraints; Use the Monte Carlo method to estimate the conditional probability distribution of flood forecast results: ; in, The features after fusion The final prediction output is probability; is the indicator function; is the number of simulations; Each simulation , use the CNN model to make predictions and get the prediction output; For the Parameter samples for the simulation.

[0018] Preferably, in step S5, multiple deep learning models are constructed as basic learners, and the basic learners are trained using the training set data. The specific operations are as follows: Construct multiple base learners and record them as ; Train the base learner: ; in, It is a basic learner The optimal parameters of is the input data; is the true label; is the training set data; Is to use parameters The base learner For input data prediction.

[0019] Therefore, the present invention adopts the above-mentioned hydrological forecasting method based on deep integrated learning and physical mechanism, and the beneficial technical effects are as follows: The hydrological forecasting method based on deep integrated learning and physical mechanism provided by the present invention has significant technical advantages over traditional hydrological forecasting methods.

[0020] First, by introducing deep learning models, especially convolutional neural networks (CNN) and long short-term memory neural networks (LSTM), the present invention can fully utilize the rich information in remote sensing data and time series data, automatically learn and extract key features, thereby improving the accuracy of hydrological forecasts. At the same time, the remote sensing features and time series features are fused through the dual cross-attention module, further enhancing the model's ability to capture complex hydrological processes.

[0021] Secondly, the present invention combines hydrological physical models (such as SWAT models) with deep learning models, and realizes the complementarity of physical mechanisms and data-driven capabilities by quantifying the output of the SWAT model and fusing it with fusion features. This not only improves the rationality of the forecast results, but also enhances the robustness of the model in cases of sparse or abnormal data.

[0022] In addition, the present invention adopts a deep ensemble learning strategy, constructs multiple deep learning models as basic learners, and fuses their prediction results through an ensemble strategy. This strategy can effectively reduce the prediction risk of a single model and improve the stability and reliability of the overall forecast results.

[0023] In summary, the hydrological forecasting method based on deep integrated learning and physical mechanism provided by the present invention has significant beneficial technical effects in improving forecast accuracy, enhancing model robustness, and improving the stability of forecast results, and provides new ideas and methods for the further development of hydrological forecasting technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 This is a flow chart of a hydrological forecasting method based on deep integrated learning and physical mechanism of the present invention; Figure 2 Flowchart for remote sensing and time series data feature fusion; Figure 3 This is the architecture diagram of the deep ensemble learning model. DETAILED DESCRIPTION

[0025] The technical solution of the present invention is further described below through the accompanying drawings and embodiments.

[0026] Unless otherwise defined, technical or scientific terms used in the present invention shall have the common meanings understood by one having ordinary skills in the field to which the present invention belongs.

[0027] Embodiment 1 like Figure 1 As shown, it is a flow chart of a hydrological forecasting method based on deep integrated learning and physical mechanism of the present invention, which specifically includes the following steps: Step S1, collect remote sensing data and time series data and perform preprocessing, construct a data set, and divide the data set into a training set, a validation set, and a test set.

[0028] Data source overview: Remote sensing data: Collect MODIS data products, high-resolution satellite data, meteorological satellite data, etc. to ensure data diversity and integrity, covering the target basin and its surrounding areas. These data will be used to extract key information such as surface water changes, vegetation cover, soil type, etc.

[0029] Time series data: Collect hydrological data including water level, water temperature, evaporation, flow, groundwater level, ice conditions, etc., as well as meteorological data such as temperature, precipitation, air pressure, wind speed, etc. These data will be used to capture the dynamic changes of hydrological and meteorological processes in the basin.

[0030] Data preprocessing: Data cleaning: Data cleaning techniques such as interpolation, outlier detection and replacement, etc., ensure the accuracy and completeness of the data.

[0031] Normalization: Normalize the collected data so that its value is in the range of [0,1] to improve model training efficiency and prediction accuracy.

[0032] The normalization formula is as follows: ; in, is the original pixel value or the original time series data value; is the normalized pixel value or the normalized time series data value; It is the minimum value of pixel value or the minimum value of time series data value; It is the maximum value of pixel value or the maximum value of time series data.

[0033] Dataset division: Divide the preprocessed dataset into training set, validation set, and test set to ensure that the model has sufficient data support during the training, validation, and testing stages.

[0034] Step S2: On the training set and the validation set, the convolutional neural network (CNN) is used to extract remote sensing data features from the remote sensing data, denoted as , using the long short-term memory neural network LSTM to extract time series features from time series data, denoted as .

[0035] CNN feature extraction: Use convolutional neural networks (CNN) to extract spatial distribution features such as surface water changes, vegetation cover, soil types, etc. from remote sensing data. Optimize feature extraction effects by adjusting the architecture and parameters of CNN.

[0036] LSTM feature extraction: Long short-term memory neural network (LSTM) is used to extract dynamic features such as water level changes, flow changes, and temperature fluctuations from time series data. The cyclic structure of LSTM enables it to capture long-term dependencies in time series data.

[0037] Step S3: and Normalize the remote sensing features using a double cross attention module , normalized time series features The fusion is performed to obtain the fusion features.

[0038] Normalization: The features extracted by CNN and LSTM are normalized again to ensure that the dimensions of the features are consistent.

[0039] Double cross attention module: Double cross attention module is introduced to fuse the normalized remote sensing features and time series features. This module realizes effective fusion and complementarity between features by calculating channel attention and spatial attention.

[0040] The dual cross attention module includes two parts: channel attention and spatial attention. The channel attention calculation and The channel attention is implemented through global average pooling and fully connected layers.

[0041] Spatial Attention Computation and The spatial attention is achieved through global average pooling and fully connected layers.

[0042] like Figure 2 The figure shows the flow chart of remote sensing and time series data feature fusion.

[0043] Channel Fusion: ; in, is the channel fusion feature; and They are and The channel attention weights of .

[0044] Spatial Fusion: ; in, It is the spatial fusion feature; and They are and The spatial attention weights.

[0045] Fusion features: ; in, For fusion features; Dimensionality reduction: Use principal component analysis (PCA) and other dimensionality reduction techniques to reduce the fusion features and reduce the complexity and computational complexity of the model.

[0046] Step S4: Combine the hydrological physics model with the deep learning model to obtain enhanced features.

[0047] SWAT model application: Run the hydrological physics model SWAT to simulate the hydrological process in the basin and obtain hydrological physics parameters such as soil moisture and surface runoff.

[0048] Feature fusion: The quantized output of the SWAT model is fused with the fusion features to achieve the complementarity of physical mechanisms and data-driven capabilities. By quantizing the output of the SWAT model, it is converted into a format compatible with the fusion features for subsequent model training and prediction.

[0049] Specifically: Quantify the SWAT model output: ; in, It is the quantitative output of the SWAT model; It is the actual hydrological data; is a quantization function.

[0050] Fuse the quantized output of the SWAT model with the fused features of step S3: ; in, It is the fusion feature; is the fusion function.

[0051] Use the fused features to train the CNN model, and the loss function is as follows: ; in, is the loss function; is the output predicted by the model; It is the actual hydrological data.

[0052] Use the back-propagation algorithm to optimize the weights of the CNN model: ; in, is the updated weight; is the old weight; is the learning rate; is the partial derivative of the loss function with respect to the weights.

[0053] Integrate the physical process constraints provided by the SWAT model into the CNN model: ; in, is the final prediction output, is an adjustment function that takes physical constraints into account.

[0054] Use the Monte Carlo method to estimate the conditional probability distribution of flood forecast results: ; in, The features after fusion The final prediction output is probability; is the indicator function; is the number of simulations; Each simulation , use the CNN model to make predictions and get the prediction output; For the Parameter samples for the simulation.

[0055] Step S5: deep ensemble learning model construction and training.

[0056] Model construction: Build multiple deep learning models as basic learners, including CNN and LSTM with different architectures. Optimize the performance of the model by adjusting the model's architecture, parameters, and training strategy.

[0057] like Figure 3 As shown in the figure, it is a deep ensemble learning model architecture diagram, specifically: construct multiple basic learners and record them as .

[0058] Train the base learner: ; in, It is a basic learner The optimal parameters of is the input data; is the true label; is the training set data; Is to use parameters The base learner For input data prediction.

[0059] Model training: Use the training set data to train each basic learner, and integrate the prediction results of the basic learners through integration strategies (such as weighted average method, voting method, etc.). The selection of integration strategy should be determined according to the characteristics and needs of the specific problem.

[0060] Model validation: Use validation set data to validate the model and evaluate its performance. Optimize the model’s prediction accuracy and stability by adjusting the model’s parameters and integration strategy.

[0061] Step S6: Model verification and prediction.

[0062] Test set prediction: The test set data is input into the optimized deep ensemble learning model to generate flood forecast results. During the prediction process, the accuracy and completeness of the input data should be ensured to improve the reliability of the prediction results.

[0063] Result analysis: Compare the model prediction results with the actual hydrological data, and calculate indicators such as prediction accuracy, mean square error (MSE), and mean absolute error (MAE). By analyzing these indicators, the prediction performance of the model is evaluated and possible improvement directions are found.

[0064] It is worth noting that the contents not elaborated in detail in the present invention are all prior art and are well known to those skilled in the art.

[0065] Therefore, the present invention adopts the above-mentioned hydrological forecasting method based on deep integrated learning and physical mechanism, combines the advantages of physical mechanism and deep learning, and improves the accuracy and stability of hydrological forecasting. This method is of great significance for water resources management and flood prevention and disaster reduction.

[0066] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A hydrological forecasting method based on deep ensemble learning and physical mechanism, characterized in that: The following steps are involved: S1. Collect and preprocess remote sensing data and time series data, construct a data set, and divide the data set into a training set, a validation set, and a test set; S2. On the training set and the validation set, the convolutional neural network (CNN) is used to extract remote sensing data features from the remote sensing data, denoted as , using the long short-term memory neural network LSTM to extract time series features from time series data, denoted as ; S3, yes and Normalize the remote sensing features using a double cross attention module , normalized time series features Perform fusion to obtain fusion features; S4, combining the hydrological physics model with the deep learning model to obtain enhanced features; S5. Construct multiple deep learning models as basic learners, use the training set data to train the basic learners, and integrate the prediction results of the basic learners through the integration strategy to form the final prediction results. Then use the validation set data to verify the model to obtain the optimized deep integration learning model. S6. Input the test set data into the optimized deep ensemble learning model to generate flood forecast results.

2. A hydrological forecasting method based on deep integrated learning and physical mechanism according to claim 1, characterized in that: In step S1, the time series data includes hydrological data and meteorological data. The hydrological data includes water level, water temperature, water evaporation, flow, groundwater level, and ice conditions. The meteorological data includes temperature, precipitation, air pressure, and wind speed. The remote sensing data includes MODIS data products and meteorological satellite data.

3. A hydrological forecasting method based on deep integrated learning and physical mechanism according to claim 2, characterized in that: In step S1, preprocessing includes data cleaning and normalization; The normalization formula is as follows: ； in, is the original pixel value or the original time series data value; is the normalized pixel value or the normalized time series data value; It is the minimum value of pixel value or the minimum value of time series data value; It is the maximum value of pixel value or the maximum value of time series data.

4. A hydrological forecasting method based on deep integrated learning and physical mechanism according to claim 3, characterized in that: In step S2, the convolutional neural network CNN includes a convolution layer, an activation function and a pooling layer, and the long short-term memory neural network LSTM includes a forget gate, an input gate, a unit state candidate, a unit state update, an output gate and a hidden state update; Among them, the convolution layer: ； in, It is the convolution operation at position Output: is the convolution kernel; is the input remote sensing image; Represents the convolution operation; The input image is The pixel value at ; is the position of the convolution kernel relative to the output feature map Offset The value at , It is the position coordinate on the output feature map; , is the position coordinate on the convolution kernel; The activation function is the ReLU activation function; The forget gate, input gate, cell state candidate, cell state update, output gate, and hidden state update are as follows: ； ； ； ； ； ； in, is the output of the forget gate; is the sigmoid function; is the weight matrix of the forget gate; represents the vector formed by concatenating the hidden state of the previous time step and the input of the current time step; is the hidden state of the previous time step; is the input of the current time step; is the bias term of the forget gate; is the output of the input gate; is the weight matrix of the input gate; is the bias term of the input gate; Candidate for unit status; is the updated unit status; is the bias term of the cell state candidate; is the updated unit status; is the cell state at the previous time step; represents element-wise multiplication; is the output of the output gate; is the weight matrix of the output gate; is the bias term of the output gate; is the updated hidden state.

5. A hydrological forecasting method based on deep integrated learning and physical mechanism according to claim 4, characterized in that: In step S3, the dual cross attention module includes two parts: channel attention and spatial attention. The channel attention calculation and Channel attention is achieved through global average pooling and fully connected layers; Spatial Attention Computation and The spatial attention is achieved through global average pooling and fully connected layers.

6. A hydrological forecasting method based on deep integrated learning and physical mechanism according to claim 5, characterized in that: In step S3, a double cross attention module is used to fuse the normalized remote sensing features and the normalized time series features. The specific operations are as follows: Channel Fusion: ； in, is the channel fusion feature; and They are and The channel attention weight of Spatial Fusion: ； in, It is the spatial fusion feature; and They are and The spatial attention weights of Fusion features: ； in, For fusion features; Using principal component analysis Perform dimensionality reduction processing.

7. A hydrological forecasting method based on deep integrated learning and physical mechanism according to claim 6, characterized in that: In step S4, the hydrological physics model is the SWAT model, and the deep learning model is the CNN; Quantify the SWAT model output: ； in, It is the quantitative output of the SWAT model; It is the actual hydrological data; is a quantization function; Fuse the quantized output of the SWAT model with the fused features of step S3: ； in, It is the fusion feature; is the fusion function; Use the fused features to train the CNN model, and the loss function is as follows: ； in, is the loss function; is the output predicted by the model; It is the actual hydrological data; Use the back-propagation algorithm to optimize the weights of the CNN model: ； in, is the updated weight; is the old weight; is the learning rate; is the loss function Weight The partial derivative of Integrate the physical process constraints provided by the SWAT model into the CNN model: ； in, is the final prediction output, is the adjustment function considering physical constraints; Use the Monte Carlo method to estimate the conditional probability distribution of flood forecast results: ； in, The features after fusion The final prediction output is probability; is the indicator function; is the number of simulations; Each simulation , use the CNN model to make predictions and get the prediction output; For the Parameter samples for the simulation.

8. The hydrological forecasting method based on deep integrated learning and physical mechanism according to claim 7, characterized in that: In step S5, multiple deep learning models are constructed as basic learners, and the basic learners are trained using the training set data. The specific operations are as follows: Construct multiple base learners and record them as ; Train the base learner: ； in, It is a basic learner The optimal parameters of is the input data; is the true label; is the training set data; Is to use parameters The base learner For input data prediction.

Images