Temporal enhanced deep learning flood forecasting method, device, equipment and medium

Abstract

The application discloses a time sequence enhanced deep learning flood forecasting method, device, equipment and medium; the method is: collecting hydrological and meteorological data; adopting principal component analysis to perform dimension reduction processing on original data in the data set; different flood prediction periods are divided; a time sequence enhanced deep learning model is constructed, the time sequence enhanced deep learning model is simulated for different flood flow prediction periods multiple times, a plurality of simulation prediction values of different prediction periods are obtained, and a determination coefficient R 2 The obtained simulation prediction values are evaluated; an error correction model is trained through the plurality of simulation prediction values, simulation errors of the time sequence enhanced deep learning model are predicted, and error correction values are obtained; the time sequence enhanced deep learning model and the error correction model are combined, an improved time sequence enhanced deep learning flood forecasting model is obtained, the flood forecasting model outputs a prediction result, and the prediction result is a sum of the error correction values and the model prediction values.

Description

Technical Field

[0001] This invention relates to the field of flood forecasting technology, specifically to a time-series augmented deep learning flood forecasting method, apparatus, equipment, and medium. Background Technology

[0002] Floods are considered one of the most common and destructive natural disasters, posing a significant threat to human life, infrastructure, and socio-economic conditions. In flood prevention and mitigation, flood forecasting, as an important non-engineering measure, faces both a research hotspot and a challenge in the field of hydrology, particularly how to achieve accurate predictions of flood events over long lead times.

[0003] In recent years, an increasing number of deep learning models have been applied to flood forecasting. These models mine information from historical data to establish relationships between flood flow, water level, and their influencing factors, and by adjusting model parameters to improve performance, they achieve the forecasting objective.

[0004] However, current research mainly verifies whether deep learning can be competent for flood forecasting, and the forecast lead time it provides is relatively short, making it difficult to achieve efficient and accurate simulation results. Moreover, most coupled deep learning models adopt a serial approach (e.g., training the output of a Transformer model first and then using it as the input of a TCN), which leads to many design constraints between modules, making it difficult to adjust the interaction between modules, and the training process is relatively independent, with no direct sharing of information during training. Summary of the Invention

[0005] In order to overcome the shortcomings of the existing technology, the present invention provides a time-series augmented deep learning flood forecasting method, which has better prediction accuracy over a long forecast period.

[0006] The second objective of this invention is to provide a temporal-enhanced deep learning flood forecasting device.

[0007] A third objective of this invention is to provide a corresponding electronic device.

[0008] A fourth objective of this invention is to provide a computer-readable storage medium.

[0009] The technical solution of the present invention to solve the above-mentioned technical problems is:

[0010] A time-series augmented deep learning flood forecasting method includes the following steps:

[0011] Step S1: Collect hydrological and meteorological data from hydrological stations along the main stream of the watershed, and preprocess the collected hydrological and meteorological data to obtain a dataset;

[0012] Step S2: Principal component analysis is used to reduce the dimensionality of the original data in the dataset to obtain the input data, and the input data is divided into training set and validation set;

[0013] Step S3: Extend the accurate flood forecast time and divide it into different flood forecast periods;

[0014] Step S4: Couple the Transformer module into the first temporal convolutional neural network model to construct a temporally enhanced deep learning model. The temporally enhanced deep learning model includes a first residual block, a Transformer module, a second residual block, a Flatten layer, a first Dense fully connected layer, and a second Dense fully connected layer. Input data is sequentially passed through the first residual block, the Transformer module, the second residual block, the Flatten layer, the first Dense fully connected layer, and the second Dense fully connected layer to obtain the model's predicted value.

[0015] Step S5: Set the hyperparameters of the temporal augmentation deep learning model, train the constructed temporal augmentation deep learning model using the training set, and then validate the temporal augmentation deep learning model using the validation set to obtain the trained temporal augmentation deep learning model.

[0016] Step S6: Simulate flood flow multiple times for different forecast periods using the trained time-series augmented deep learning model to obtain multiple simulated predicted values ​​for different forecast periods, and use the coefficient of determination R0. 2 The simulated predictions obtained were evaluated;

[0017] Step S7: Calculate the difference between the model prediction and the actual value at each time point in the multiple simulation results to obtain the error value at each time point; convert R0 to R0. 2 The error value in the highest simulated prediction is used as the output label, and the remaining error values ​​are used as input data. The training set and validation set are then divided again. The hyperparameters of the second temporal convolutional neural network model are set, and the second temporal convolutional neural network model is trained to obtain the error correction model. The simulated error of the temporal augmentation deep learning model is predicted using the trained error correction model to obtain the error correction value.

[0018] Step S8: Combine the time-series augmented deep learning model and the error correction model to obtain an improved time-series augmented deep learning flood forecasting model. The flood forecasting model outputs a prediction result, which is the sum of the error correction value and the model prediction value.

[0019] In a preferred embodiment of the present invention, in step S1, the hydrological and meteorological data of the main stream hydrological stations in the basin include multi-year water level, flow rate, rainfall, evaporation and temperature; the water level and flow rate data directly reflect the flood process of the main stream of the basin, the rainfall and evaporation data reflect the indirect impact on the flood process of the basin from the intensity of water vapor flow, and the temperature data reflects the potential impact on the flood process of the basin. These factors interact and jointly determine the formation and development of floods.

[0020] In a preferred embodiment of the present invention, the preprocessing in step S1 includes standardizing the hydrological and meteorological data.

[0021] In a preferred embodiment of the present invention, in step S2, principal component analysis is used to reduce the dimensionality of the original data in the dataset, and the principal components corresponding to the feature values ​​with a cumulative contribution rate greater than 80% are taken as the input data of the time-series augmented deep learning model. 80% of the input data is used as the training set of the time-series augmented deep learning model, and the other 20% is used as the validation set of the time-series augmented deep learning model.

[0022] In a preferred embodiment of the present invention, in step S3, the division of different flood forecast periods is as follows: the flood forecast period is divided into 3 days, 5 days, 7 days, and 15 days, wherein the 3-day forecast period uses the input data of the previous 3 days to predict the flow rate of the day; the 5-day forecast period uses the input data of the previous 5 days to predict the flow rate of the day; the 7-day forecast period uses the input data of the previous 7 days to predict the flow rate of the day; and the 15-day forecast period uses the input data of the previous 15 days to predict the flow rate of the day.

[0023] In a preferred embodiment of the present invention, in step S5, the hyperparameters of the temporally enhanced deep learning model include the number of neurons per layer, activation function, optimizer, loss function, learning rate, and dropout rate; wherein, the convolutional layers in the first residual block and the second residual block each have 128 neurons; the inner layers, model dimension, and number of multi-head attention heads of the Transformer module are set to 6, 16, and 8, respectively; the number of neurons in the first Dense fully connected layer and the second Dense fully connected layer are 64 and 1, respectively; the activation function is ReLU; the optimizer is Adam; the loss function is MSE; the learning rate is set to 0.001; and the dropout rate is set to 0.2.

[0024] In a preferred embodiment of the present invention, in step S7, the second temporal convolutional neural network model includes two residual blocks, one Flatten layer, and one Dense fully connected layer; wherein the hyperparameters of the second temporal convolutional neural network model are: each convolutional layer in the residual block has 64 neurons; the Dense fully connected layer has 1 neuron; the activation function is ReLU; the optimizer is Adam; the loss function is MSE; the learning rate is set to 0.001; and the dropout rate is set to 0.2.

[0025] A time-series augmented deep learning flood forecasting device includes:

[0026] The data acquisition module is used to collect hydrological and meteorological data from hydrological stations along the main stream of the river basin.

[0027] The data processing module is used to standardize the hydrological and meteorological data collected by the data acquisition module, and to perform dimensionality reduction processing on the standardized hydrological and meteorological data using principal component analysis.

[0028] The flood forecasting module includes a flood prediction module, an error correction module, and a result output module. The flood prediction module is used to predict flood flow for different forecast periods based on input hydrological and meteorological data and output model prediction values. The error correction module is used to perform error prediction on the model prediction values ​​output by the flood prediction module and output error correction values. The result output module is used to add the model prediction values ​​and the error correction values ​​to output the final flood forecast result.

[0029] An electronic device includes a central processing unit and a memory, the central processing unit being configured to invoke and run a computer program stored in the memory to perform the steps of the time-series augmented deep learning flood forecasting method.

[0030] A computer-readable storage medium stores, in the form of computer-readable instructions, a computer program implemented according to the time-series augmented deep learning flood forecasting method, which, when invoked by a computer, executes the steps included in the corresponding method.

[0031] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0032] 1. The temporal augmentation deep learning flood forecasting method of this invention can further optimize the model structure of deep learning in the field of flood forecasting, so as to improve its prediction accuracy in the long-term forecast period. By combining the local feature extraction capability of the temporal convolutional neural network model (TCN) and the global temporal modeling capability of the Transformer module, the performance of the flood forecasting model is improved.

[0033] 2. The temporal enhancement deep learning model in this invention adopts a model structure that couples the Transformer module in the first temporal convolutional neural network model. This model structure can capture both local short-term dependencies and global long-term dependencies when processing time-series data, and performs better in long-series flood forecasting tasks, thereby extending the flood forecast lead time.

[0034] 3. The flood forecasting model in this invention incorporates pre- and post-processing techniques to enhance the model's understanding of the flood evolution process, thereby more accurately predicting future flood events. Therefore, this invention has profound significance for achieving accurate flood forecasting with a long lead time and promoting the construction of smart water conservancy projects.

[0035] 4. The improved time-series augmented deep learning flood forecasting model of this invention, namely the PRE-TCN-T-POST model, combines time-series augmentation techniques with pre-processing and post-processing techniques to improve the efficiency and accuracy of processing long-term, large-scale, multivariate flood data. In the pre-processing part of the PRE-TCN-T-POST model, principal component analysis is used to reduce the dimensionality of the data. Subsequently, a time-series augmented deep learning model (TCN-T) is selected as the main body of the PRE-TCN-T-POST model. Through repeated interaction of local features and global dependency information, the ability to extract temporal information is enhanced. In the post-processing stage of the PRE-TCN-T-POST model, an error correction model is adopted to further improve the accuracy of the simulation. Compared with traditional CNN, LSTM and t-TCN models, the PRE-TCN-T-POST model in this invention demonstrates significant improvements in efficiency and accuracy when processing long-term, large-scale, multivariate flood data by integrating the data dimensionality reduction capabilities of principal component analysis, the multi-feature and long-sequence data processing advantages of the time-enhanced deep learning model (TCN-T), and the fine-tuning function of the error correction model. Attached Figure Description

[0036] Figure 1 This is a flowchart illustrating the temporal augmentation deep learning flood forecasting method of the present invention.

[0037] Figure 2 This is a map showing the location of the Yichang hydrological station in the middle reaches of the Yangtze River.

[0038] Figure 3 This is a diagram showing the main structure of the PRE-TCN-T-POST model for the Yichang Hydrological Station.

[0039] Figure 4 This is a diagram showing the simulated flood flow results for the 3-day forecast period using the PRE-TCN-T-POST model at the Yichang hydrological station.

[0040] Figure 5This is a graph showing the simulated flood flow results for the 5-day forecast period using the PRE-TCN-T-POST model at the Yichang hydrological station.

[0041] Figure 6 This is a graph showing the simulated flood flow results for the 7-day forecast period using the PRE-TCN-T-POST model at the Yichang hydrological station.

[0042] Figure 7 This is a graph showing the simulated flood flow results for the 15-day forecast period using the PRE-TCN-T-POST model at the Yichang hydrological station.

[0043] Figure 8 This is a diagram showing the main structure of the CNN forecasting model for the Yichang hydrological station.

[0044] Figure 9 This is a diagram showing the main structure of the LSTM forecasting model for the Yichang hydrological station.

[0045] Figure 10 This is a diagram showing the main structure of the t-TCN model for Yichang Station.

[0046] Figure 11 The image shows the simulation results of flood flow over a 3-day forecast period using CNN, LSTM, and t-TCN models at Yichang Station.

[0047] Figure 12 The image shows the simulation results of flood flow over a 5-day forecast period using CNN, LSTM, and t-TCN models at Yichang Station.

[0048] Figure 13 The image shows the simulated flood flow results for the 7-day forecast period using CNN, LSTM, and t-TCN models at Yichang Station.

[0049] Figure 14 The image shows the simulated flood flow results for the 7-day forecast period using CNN, LSTM, and t-TCN models at Yichang Station. Detailed Implementation

[0050] The present invention will be further described in detail below with reference to the embodiments and accompanying drawings, but the embodiments of the present invention are not limited thereto.

[0051] Example 1

[0052] The flood characteristics of the middle reaches of the Yangtze River are influenced by a combination of factors such as climate, topography and human activities. The hydrological situation is complex and floods and other disasters are gradually emerging, which bring new challenges to local water resource management and agricultural production.

[0053] Therefore, this invention selects the Yichang hydrological station in the middle reaches of the Yangtze River as a research example (e.g., Figure 2As shown, an improved temporal augmentation deep learning flood forecasting model (PRE-TCN-T-POST) and traditional CNN, LSTM and TCN forecasting models were constructed to simulate flood flows at different forecast periods at the Yichang hydrological station.

[0054] See Figure 1 The specific steps of the temporal augmentation deep learning flood forecasting method of the present invention are as follows:

[0055] Step S1: Collect hydrological and meteorological data from hydrological stations along the main stream of the watershed, and preprocess the collected hydrological and meteorological data to obtain a dataset;

[0056] In this embodiment, flood data from the Yichang Hydrological Station in the middle reaches of the Yangtze River from 2008 to 2020 were collected and organized, including water level, flow rate, rainfall, evaporation, and temperature. These data were then standardized. The water level and flow rate data directly reflect the flood process at the Yichang Hydrological Station, while the rainfall and evaporation data reflect the indirect impact on the flood process at the Yichang Hydrological Station from the perspective of water vapor flow intensity. The temperature data reflects the potential impact on the flood process at the Yichang Hydrological Station.

[0057] Step S2: Use principal component analysis to reduce the dimensionality of the original data in the dataset to obtain the input data; divide the input data into training set and validation set;

[0058] In this embodiment, the standardized water level, flow rate, rainfall, evaporation, and temperature data from the Yichang hydrological station are used as input data x∈R T×D (R is the dataset, T is the length of the data sequence, and D is the feature dimension, which here consists of five dimensions: water level, flow rate, rainfall, evaporation, and temperature.) Principal component analysis (PRE) is used as the preprocessing method, and the principal components corresponding to the eigenvalues ​​with a cumulative contribution rate greater than 80% are selected. (d pca The reduced feature dimensions are used as the input data for the constructed temporal augmentation deep learning model. 80% of the input data is used as the training set for the temporal augmentation deep learning model, and the other 20% is used as the validation set for the temporal augmentation deep learning model.

[0059] In this embodiment, data from the Yichang Hydrological Station from 2008 to 2018 is used as the training set for the time-series augmented deep learning model, and data from 2019 to 2020 is used as the validation set for the time-series augmented deep learning model.

[0060] Step S3: Extend the accurate flood forecast time and divide it into different flood forecast periods;

[0061] In this embodiment, different flood forecast periods are defined as follows: the flood forecast period is divided into 3 days, 5 days, 7 days, and 15 days. The 3-day forecast period uses the input data from the previous 3 days to predict the flow rate of the day; the 5-day forecast period uses the input data from the previous 5 days to predict the flow rate of the day; the 7-day forecast period uses the input data from the previous 7 days to predict the flow rate of the day; and the 15-day forecast period uses the input data from the previous 15 days to predict the flow rate of the day.

[0062] Step S4: Couple the Transformer module into the first temporal convolutional neural network model (TCN1) to enhance the extraction strength of temporal information from the data, and establish a temporally enhanced deep learning model (TCN-T). The temporally enhanced deep learning model (TCN-T) includes a first residual block, a Transformer module, a second residual block, a Flatten layer, a first Dense fully connected layer, and a second Dense fully connected layer. Both the first and second residual blocks contain two convolutional layers. The input data sequentially passes through the first residual block, the Transformer module, the second residual block, the Flatten layer, the first Dense fully connected layer, and the second Dense fully connected layer to obtain the model's predicted value.

[0063] Because Temporal Convolutional Neural Networks (TCNs) are both efficient and stable when processing multi-feature and time-series data, the multi-head self-attention mechanism in the Transformer module can dynamically weight the input at each time step, thereby more accurately capturing the correlation and importance between different time steps. The specific input data processing flow is as follows:

[0064] First, the first residual block (i.e., residual block 1) of the first temporal convolutional neural network model (TCN1) is used to extract the features and local temporal dependencies of the input data. The main processing formula can be expressed as follows:

[0065]

[0066] y TCN1 =Relu(y1)+x1;

[0067] In the formula: y1 is the output of the convolution operation, w k For the convolution kernel, x t-d×k The elements are input to the convolutional layer, d is the dilation rate, and y is the dilation rate. TCN1 The output of the first residual block is given by ReLU, where ReLU is the activation function and x1 is the element input to the residual block.

[0068] The output shape after processing by the first residual block is: Where, d TCNThis represents the feature dimension after processing the first residual block.

[0069] Then, using the output of the first residual block As input to the Transformer module, the calculation of global temporal dependencies, and the main processing formula can be expressed as follows:

[0070]

[0071] Where Q, L, and V are the query, key, and value matrices, respectively, obtained by linear transformation of the input spatiotemporal sequence; d l The dimension of the key;

[0072] The output shape after processing by the Transformer module is: d t The feature dimension after processing by the Transformer module, where d t With d TCN same.

[0073] Then the output of the Transformer The input data is passed to the second residual block (i.e., residual block 2). The first temporal convolutional neural network model (TCN1) will continue to process the input data and enhance its local temporal information again. The main processing process is the same as that of the first residual block (residual block 1). The output of the second residual block is processed by the Flatten layer, the first Dense fully connected layer and the second Dense fully connected layer to obtain the model prediction value of the temporally enhanced deep learning model (TCN-T).

[0074] Step S5: Set the hyperparameters of the Temporal Augmentation Deep Learning Model (TCN-T), train the constructed Temporal Augmentation Deep Learning Model using the training set, and then validate the Temporal Augmentation Deep Learning Model using the validation set to obtain the trained Temporal Augmentation Deep Learning Model.

[0075] In this embodiment, the hyperparameters of the temporally enhanced deep learning model (TCN-T) include the number of neurons per layer, activation function, optimizer, loss function, learning rate, and dropout rate. Specifically, the convolutional layers in both the first and second residual blocks have 128 neurons. The number of inner layers, model dimension, and multi-head attention heads in the Transformer module are set to 6, 16, and 8, respectively. The first and second Dense fully connected layers have 64 and 1 neurons, respectively. The activation function is ReLU; the optimizer is Adam; the loss function is MSE; the learning rate is set to 0.001; and the dropout rate is set to 0.2.

[0076] Step S6: Simulate flood flow multiple times for different forecast periods using the trained time-series augmented deep learning model to obtain multiple simulated predicted values ​​for different forecast periods, and use the coefficient of determination R0. 2 The simulated predictions obtained were evaluated;

[0077] Step S7: Calculate the difference between the model prediction and the actual value at each time point in the multiple simulation results to obtain the error value at each time point; convert R0 to R0. 2 The error value in the highest simulated prediction is used as the output label, and the remaining error values ​​are used as input data. The training set and validation set are divided again, the hyperparameters of the second temporal convolutional neural network model (TCN2) are set, and the second temporal convolutional neural network model (TCN2) is trained to obtain the error correction model. The simulated error of the temporal augmentation deep learning model (TCN-T) is predicted by the trained error correction model to obtain the error correction value.

[0078] In this embodiment, the second temporal convolutional neural network model (TCN2) includes two residual blocks, one flattened layer, and one dense fully connected layer. The hyperparameters of the second temporal convolutional neural network model (TCN2) are as follows: each convolutional layer in the residual block has 64 neurons; each dense fully connected layer has 1 neuron; the activation function is ReLU; the optimizer is Adam; the loss function is MSE; the learning rate is set to 0.001; and the dropout rate is set to 0.2.

[0079] Step S8: Combine the temporal augmentation deep learning model (TCN-T) and the error correction model to obtain an improved temporal augmentation deep learning flood forecasting model (PRE-TCN-T-POST). The flood forecasting model (PRE-TCN-T-POST) outputs a prediction result, which is the sum of the error correction value and the model prediction value.

[0080] To verify the performance of the flood forecasting model (PRE-TCN-T-POST) in this invention, this embodiment constructs traditional CNN, LSTM, and t-TCN models to simulate flood discharges at the Yichang hydrological station for different forecast periods; the specific steps are as follows:

[0081] Step 1: Use the standardized water level, flow rate, rainfall, evaporation and temperature of Yichang Hydrological Station as the input data of the model, and the measured flow rate data of Yichang Hydrological Station as the label to establish the input and output structure of the model;

[0082] Step 2: Construct CNN prediction models, LSTM prediction models, and t-TCN prediction models, where,

[0083] The CNN prediction model consists of 3 convolutional layers, 3 max-pooling layers, 1 flattened layer, and 2 dense fully connected layers. The number of neurons in the convolutional layers is 256, 128, and 64, respectively; the number of neurons in the dense fully connected layers is 64 and 1, respectively. The activation function is ReLU, the optimizer is Adam, the loss function is MSE, the learning rate is 0.001, and the dropout rate is 0.2. The main structure of the CNN prediction model is as follows: Figure 8 As shown;

[0084] The LSTM prediction model consists of two LSTM layers and one Dense fully connected layer; the LSTM layers have 256 and 128 neurons respectively; the Dense fully connected layer has 1 neuron; the activation function is ReLU, the optimizer is Adam, the loss function is MSE, the learning rate is 0.001, and the dropout rate is 0.2; the main structure of the LSTM prediction model is as follows. Figure 9 As shown;

[0085] The t-TCN prediction model uses a Transformer and a TCN neural network model connected in series, without altering the original structure of the TCN neural network model. It includes one Transformer module, two residual blocks (each containing two convolutional layers), one Flatten layer, and two Dense fully connected layers. The Transformer module has 6 inner layers, 16 model dimensions, and 8 multi-head attention heads. Each convolutional layer has 128 neurons. The Dense fully connected layers have 64 and 1 neurons respectively. The activation function is ReLU, the optimizer is Adam, the loss function is MSE, the learning rate is 0.001, and the dropout rate is 0.2. The main structure of the t-TCN prediction model is as follows: Figure 10 As shown;

[0086] Step 3: Train the model based on the input and output data of the traditional model, and simulate the flood discharge at the Yichang hydrological station for different forecast periods. The simulation results of the CNN forecast model, LSTM forecast model, and t-TCN forecast model are obtained respectively. The simulation results are as follows: Figures 11-14 As shown;

[0087] Step 4: By selecting appropriate evaluation indicators, the simulation results of the flood forecasting model (PRE-TCN-T-POST) in this invention are compared and analyzed with those of traditional CNN forecasting models, LSTM forecasting models, and t-TCN forecasting models to obtain the simulation advantages of the flood forecasting model (PRE-TCN-T-POST) in this invention at different lead times; wherein, the selected evaluation indicators include the Nash efficiency coefficient (NSE) and the coefficient of determination (R²). 2), root mean square error (RMSE) and mean absolute error (MAE), using NSE and R 2 The accuracy of the model simulation results is evaluated; the closer the value is to 1, the better the simulation effect. The RMSE and MAE values ​​are used to evaluate the error of the model simulation results; the smaller the values, the smaller the error of the model results. The specific formulas are as follows:

[0088]

[0089]

[0090]

[0091]

[0092] Among them, Y i With P i These are the actual value and the simulated value, respectively; Y' i With P' i These are the average of the actual values ​​and the average of the simulated values, respectively.

[0093] Through NSE, R 2 The simulation results of the flood forecasting model (PRE-TCN-T-POST) in this invention were compared and analyzed with those of traditional CNN forecasting models, LSTM forecasting models and t-TCN forecasting models using RMSE and MAE. The evaluation results are shown in Table 1.

[0094] Table 1. Results of evaluation indicators for Yichang Hydrological Station under different forecast periods using four models.

[0095]

[0096] As shown in Table 1, the flood forecasting model (PRE-TCN-T-POST) of this invention has NSE and R values ​​at forecast times of 3 days, 5 days, 7 days, and 15 days. 2 All scores are above 0.9 and are greater than the NSE and R-values ​​of traditional CNN prediction models, LSTM prediction models, and t-TCN prediction models. 2 This indicates that the flood forecasting model of the present invention (PRE-TCN-T-POST) with enhanced time series and pre- and post-processing has the highest simulation accuracy in long-term flood forecasting simulation. In addition, the RMSE and MAE of the flood forecasting model of the present invention (PRE-TCN-T-POST) at 3-day, 5-day, 7-day, and 15-day forecast periods are lower than those of traditional CNN forecasting models, LSTM forecasting models, and t-TCN forecasting models, indicating that the flood forecasting model of the present invention (PRE-TCN-T-POST) has the lowest simulation error in long-term flood forecasting simulation.

[0097] In summary, this invention proposes an improved time-series augmented deep learning flood forecasting method, namely, constructing a PRE-TCN-T-POST flood forecasting model. This model combines time-series augmentation techniques with pre-processing and post-processing techniques to improve the efficiency and accuracy of processing long-term, large-scale, multivariate flood data. The pre-processing part of the PRE-TCN-T-POST flood forecasting model uses principal component analysis to reduce the dimensionality of the data. Subsequently, a TCN-T network is selected as the main body of the PRE-TCN-T-POST flood forecasting model. Through repeated interaction of local features and global dependency information, the ability to extract temporal information is enhanced. In the post-processing stage of the PRE-TCN-T-POST flood forecasting model, an error correction model is adopted to further improve the accuracy of the simulation. Compared with traditional CNN, LSTM and t-TCN models, the PRE-TCN-T-POST flood forecasting model demonstrates significant improvements in efficiency and accuracy when processing long-term, large-scale, multivariate flood data by integrating the data dimensionality reduction capabilities of principal component analysis, the advantages of TCN-T in processing multi-feature and long-sequence data, and the fine-tuning capabilities of the error correction model.

[0098] Example 2

[0099] The temporal augmentation deep learning flood forecasting device of the present invention includes:

[0100] The data acquisition module is used to collect hydrological and meteorological data from hydrological stations along the main stream of the river basin.

[0101] The data processing module is used to standardize the hydrological and meteorological data collected by the data acquisition module, and to perform dimensionality reduction processing on the standardized hydrological and meteorological data using principal component analysis.

[0102] The flood forecasting module includes a flood prediction module, an error correction module, and a result output module. The flood prediction module is used to predict flood flow for different forecast periods based on input hydrological and meteorological data and output model prediction values. The error correction module is used to perform error prediction on the model prediction values ​​output by the flood prediction module and output error correction values. The result output module is used to add the model prediction values ​​and the error correction values ​​to output the final flood forecast result.

[0103] Example 3

[0104] The electronic device of the present invention includes a central processing unit and a memory, wherein the central processing unit is used to invoke and run a computer program stored in the memory to perform the steps of the time-series augmented deep learning flood forecasting method in Embodiment 1.

[0105] Example 4

[0106] The electronic device of the present invention has a computer-readable storage medium that stores, in the form of computer-readable instructions, a computer program implemented according to the time-series enhanced deep learning flood forecasting method in Embodiment 1. When the computer program is invoked and executed by a computer, it performs the steps included in the corresponding method.

[0107] The above are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above content. Any changes, modifications, substitutions, combinations, or simplifications made without departing from the spirit and principle of the present invention shall be considered equivalent substitutions and shall be included within the protection scope of the present invention.

Claims

1. A temporal augmentation deep learning flood forecasting method, characterized in that, Includes the following steps: Step S1: Collect hydrological and meteorological data from hydrological stations along the main stream of the watershed, and preprocess the collected hydrological and meteorological data to obtain a dataset; Step S2: Principal component analysis is used to reduce the dimensionality of the original data in the dataset to obtain the input data, and the input data is divided into training set and validation set; Step S3: Extend the accurate flood forecast time and divide it into different flood forecast periods; Step S4: Couple the Transformer module into the first temporal convolutional neural network model to construct a temporally enhanced deep learning model. The temporally enhanced deep learning model includes a first residual block, a Transformer module, a second residual block, a Flatten layer, a first Dense fully connected layer, and a second Dense fully connected layer. Input data is sequentially passed through the first residual block, the Transformer module, the second residual block, the Flatten layer, the first Dense fully connected layer, and the second Dense fully connected layer to obtain the model's predicted value. Step S5: Set the hyperparameters of the temporal augmentation deep learning model, train the constructed temporal augmentation deep learning model using the training set, and then validate the temporal augmentation deep learning model using the validation set to obtain the trained temporal augmentation deep learning model. Step S6: Simulate flood flow multiple times for different forecast periods using the trained time-series augmented deep learning model to obtain multiple simulated predicted values ​​for different forecast periods, and use the coefficient of determination R0. 2 The simulated predictions obtained were evaluated; Step S7: Calculate the difference between the model prediction and the actual value at each time point in the multiple simulation results to obtain the error value at each time point; convert R0 to R0. 2 The error value in the highest simulated prediction is used as the output label, and the remaining error values ​​are used as input data. The training set and validation set are then divided again. The hyperparameters of the second temporal convolutional neural network model are set, and the second temporal convolutional neural network model is trained to obtain the error correction model. The simulated error of the temporal augmentation deep learning model is predicted using the trained error correction model to obtain the error correction value. Step S8: Combine the time-series augmented deep learning model and the error correction model to obtain an improved time-series augmented deep learning flood forecasting model. The flood forecasting model outputs a prediction result, which is the sum of the error correction value and the model prediction value.

2. The time-series augmented deep learning flood forecasting method according to claim 1, characterized in that, In step S1, the hydrological and meteorological data of the main stream hydrological stations in the basin include multi-year water level, flow rate, rainfall, evaporation and temperature.

3. The time-series augmented deep learning flood forecasting method according to claim 2, characterized in that, In step S1, the preprocessing includes standardizing the hydrological and meteorological data.

4. The time-series augmented deep learning flood forecasting method according to claim 1, characterized in that, In step S2, principal component analysis is used to reduce the dimensionality of the original data in the dataset. The principal components corresponding to the feature values ​​with a cumulative contribution rate greater than 80% are taken as the input data of the time-series augmented deep learning model. 80% of the input data is used as the training set of the time-series augmented deep learning model, and the other 20% is used as the validation set of the time-series augmented deep learning model.

5. The time-series augmented deep learning flood forecasting method according to claim 1, characterized in that, In step S3, the different flood forecast periods are divided as follows: the flood forecast period is divided into 3 days, 5 days, 7 days, and 15 days. The 3-day forecast period uses the input data of the previous 3 days to predict the flow rate of the day; the 5-day forecast period uses the input data of the previous 5 days to predict the flow rate of the day; the 7-day forecast period uses the input data of the previous 7 days to predict the flow rate of the day; and the 15-day forecast period uses the input data of the previous 15 days to predict the flow rate of the day.

6. The time-series augmented deep learning flood forecasting method according to claim 1, characterized in that, In step S5, the hyperparameters of the temporally enhanced deep learning model include the number of neurons per layer, activation function, optimizer, loss function, learning rate, and dropout rate; wherein, the convolutional layers in the first residual block and the second residual block each have 128 neurons; the number of inner layers, model dimension, and number of multi-head attention heads in the Transformer module are set to 6, 16, and 8, respectively; the number of neurons in the first Dense fully connected layer and the second Dense fully connected layer are 64 and 1, respectively; the activation function is set to ReLU; the optimizer is set to Adam; the loss function is MSE; the learning rate is set to 0.001; and the dropout rate is set to 0.

2.

7. The time-series augmented deep learning flood forecasting method according to claim 1, characterized in that, In step S7, the second temporal convolutional neural network model includes two residual blocks, one Flatten layer, and one Dense fully connected layer. The hyperparameters of the second temporal convolutional neural network model are as follows: each convolutional layer in the residual block has 64 neurons; each Dense fully connected layer has 1 neuron; the activation function is set to ReLU; the optimizer is set to Adam; the loss function is set to MSE; the learning rate is set to 0.001; and the dropout rate is set to 0.

2.

8. A time-series augmented deep learning flood forecasting device employing the time-series augmented deep learning flood forecasting method according to any one of claims 1 to 7, characterized in that, include: The data acquisition module is used to collect hydrological and meteorological data from hydrological stations along the main stream of the river basin. The data processing module is used to standardize the hydrological and meteorological data collected by the data acquisition module, and to perform dimensionality reduction on the standardized hydrological and meteorological data using principal component analysis. The flood forecasting module includes a flood prediction module, an error correction module, and a result output module. The flood prediction module is used to predict flood flow for different forecast periods based on input hydrological and meteorological data and output model prediction values. The error correction module is used to perform error prediction on the model prediction values ​​output by the flood prediction module and output error correction values. The result output module is used to add the model prediction value and the error correction value to output the final flood prediction result.

9. An electronic device comprising a central processing unit and a memory, characterized in that, The central processing unit is used to invoke and run a computer program stored in the memory to perform the steps of the time-series augmented deep learning flood forecasting method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, It stores, in the form of computer-readable instructions, a computer program implementing the time-series augmented deep learning flood forecasting method according to any one of claims 1 to 7, which, when called by a computer, executes the steps included in the corresponding method.

Images