A reservoir water level prediction method based on a graph convolutional neural network

Abstract

This invention belongs to the field of hydrological forecasting and discloses a reservoir water level prediction method based on Graph Convolutional Neural Network (GCN). This method aims to solve the problem of reservoir water level prediction under rainfall conditions, providing effective support for reservoir operation. It mainly trains the GCN on rainfall image data and combines it with a Bidirectional Long Short-Term Memory (BiLSTM) network to extract the time-series features of reservoir water levels, thereby constructing a model capable of predicting reservoir water levels from rainfall images. Specifically, the GCN is used to extract the spatiotemporal features of rainfall images, and these features are fused using a multi-head attention mechanism. In BiLSTM, the input time-series data sequence is processed through two directional LSTM networks, thereby simultaneously capturing past and future information. By merging the hidden states from these two directions, the model can gain a deeper understanding of the inherent patterns in the time-series data, thus improving the prediction accuracy of reservoir water levels.

Description

Technical Field

[0001] This invention belongs to the field of hydrological forecasting, specifically relating to a method for predicting reservoir water levels based on graph convolutional neural networks. Technical Background

[0002] Reservoir operation is closely related to reservoir water levels, which in turn are inextricably linked to rainfall. Therefore, studying reservoir operation can begin with studying rainfall. Reservoir operation is influenced by various factors, including flood control, power generation, and farmland irrigation. Among these, flood control is the factor that puts the greatest pressure on reservoir operation, and floods are often triggered by extreme rainfall. Therefore, in the face of extreme rainfall, reservoirs need to make rapid adjustments to reduce the risk of levee breaches.

[0003] Reservoir water level is a basis for reservoir operation and scheduling, and changes in reservoir water level are closely related to rainfall. Analyzing past rainfall data can accurately grasp reservoir water level changes, which is crucial for reservoir prediction and scheduling. On the one hand, with the continuous improvement of hydrological technology and the maturity of water conservancy information technology, the hydrological industry is undergoing a transformation from traditional hydrology to modern hydrology. Hydrological data acquisition has gradually evolved from manual recording to automatic acquisition by hydrological stations, accumulating a wealth of historical water and rainfall data, providing new technical means and computing power support for reservoir water level prediction applications. Graph convolutional neural networks can effectively capture the spatial and temporal dependencies in hydrological data. By using graph structures to model watershed nodes, combining watershed rainfall images, and using bidirectional long short-term memory networks (BiLSTM) to extract time-series features of reservoir water levels, a model can be obtained that can extract the complex relationship between rainfall images and reservoir water level changes, effectively improving the accuracy of reservoir water level prediction. Summary of the Invention

[0004] The purpose of this invention is to provide a method for predicting reservoir water level data based on graph convolutional neural networks. By utilizing graph convolutional neural networks to extract spatiotemporal features and bidirectional long short-term memory networks to extract temporal features, this method provides technical support for reservoir water level prediction and effective support for reservoir scheduling.

[0005] To achieve the above objectives, the present invention provides the following solution:

[0006] A method for predicting reservoir water levels based on graph convolutional neural networks, characterized by the following steps:

[0007] Step 1: Collect rainfall image data and reservoir water level data of the target watershed for in-depth analysis and modeling;

[0008] Step 2: Perform data preprocessing on rainfall image data and reservoir water level data to ensure data quality consistency;

[0009] Step 3: Align the rainfall image data and reservoir water level data to ensure consistency in timestamps and data types, and perform accurate comparison and analysis.

[0010] Step 4: Train the rainfall image data using a graph convolutional neural network (GCN) and combine it with a bidirectional long short-term memory network (BiLSTM) to extract the time series features of the reservoir water level, thereby constructing a model that can predict the reservoir water level through rainfall images;

[0011] Step 5: Input the rainfall image data into the trained prediction model to predict the reservoir water level data;

[0012] Furthermore, step 1 includes the following steps:

[0013] Sub-step 1.1: Search for rainfall image data and reservoir water level data from the Ministry of Water Resources website and the water resources department website of the province or city to which the target river basin belongs;

[0014] Sub-step 1.2: After obtaining the initial data from the Ministry of Water Resources website and the provincial and municipal water resources department websites of the river basin, it is necessary to filter the data and select valuable data according to the requirements, such as rainfall image data and reservoir water level data of the specified river basin.

[0015] Furthermore, step 2 includes the following steps:

[0016] Sub-step 2.1: Check the collected rainfall image data and reservoir water level data to confirm whether there are any missing data. Depending on the specific situation, supplement or delete the missing data as appropriate.

[0017] Sub-step 2.2 involves preprocessing the rainfall image data to ensure that each rainfall image is the same size, so that batch processing can be performed.

[0018] Furthermore, step 3 includes the following steps:

[0019] Sub-step 3.1: Unify the time unit for rainfall image data and reservoir water level data, and convert the string timestamp into a standard date and time format;

[0020] Sub-step 3.2 ensures that all data related to the reservoir water level are of the same data type for subsequent analysis and processing;

[0021] Furthermore, step 4 includes the following steps:

[0022] Sub-step 4.1: Divide the corresponding rainfall image data and reservoir water level data into training set, validation set and test set according to the ratio of 7:1.5:1.5;

[0023] Sub-step 4.2 divides the graph convolutional neural network into five layers: data preprocessing layer, feature encoder layer, feature decoder layer, water level prediction layer, and output layer. The data preprocessing layer is mainly responsible for preprocessing the hydrological data of the target watershed and analyzing the correlation between historical and future data. The feature encoder layer and the feature decoder layer are responsible for mining and fusing hydrological data features through the graph convolutional neural network (GCN). The water level prediction layer mines the time-series hydrological data after feature fusion based on BiLSTM to predict the reservoir water level data at future times.

[0024] Sub-step 4.3 selects rainfall image data as the key factor affecting reservoir water level changes. The correlation coefficient between historical rainfall image data and future reservoir water level data is calculated, and the cross-correlation diagram between historical rainfall maps and future reservoir water level data is analyzed to determine the magnitude of the influencing factor's data volume. The Pearson correlation coefficient is used to measure the strength of the linear relationship between the two time series of rainfall image data and reservoir water level data. The calculation formula is as follows:

[0025]

[0026] In the formula, x i and y i These are the i-th data points in sequences X and Y, respectively; and Let X and Y represent the average values ​​of sequences X and Y, respectively.

[0027] Sub-step 4.4 involves training the prediction model. Suitable data is selected from the training set for model training. An encoder-decoder structure is used, and a convolutional neural network (GCN) is employed to extract spatiotemporal features from rainfall images. These features are then fused using a multi-head attention mechanism. Finally, a BiLSTM is used to predict reservoir water level data. During training, the correlation coefficient between historical rainfall image data and future reservoir water level data is calculated, and the cross-correlation graph between these two data points is analyzed to determine the amount of data influencing factors. The cross-correlation function is as follows:

[0028] R = E(X) s *Y t )

[0029] In the formula, X s Y represents the value of time series X at time s. t R represents the value of the time series Y at time t, and R represents the degree of correlation between the two.

[0030] Data processed by an encoder-decoder and fused using a multi-head attention mechanism is used as input to a Bidirectional Long Short-Term Memory (BiLSTM) network for final reservoir water level prediction. In BiLSTM, the input time-series data sequence is processed through two directional LSTM networks to capture past and future information. By merging the hidden states from these two directions, the model can better understand the inherent patterns in the time-series data, thereby improving prediction accuracy. The prediction function is as follows:

[0031] y ′ =f BiLSTM (X ′ )

[0032] Where X' represents the fused rainfall image data, and y' represents the predicted reservoir water level.

[0033] Sub-step 4.5 trains a prediction model based on a graph convolutional neural network (GCN) by calculating the loss between the predicted value y' and the true value y. The loss function consists of the rainfall image data error (PE) and the mean squared error (MSE).

[0034] The formula for calculating the error of rainfall image data is as follows:

[0035] PE = y max -y max ′

[0036] The formula for calculating the mean squared error is as follows:

[0037]

[0038] The formula for calculating the loss function is as follows:

[0039] Loss = PE + MSE

[0040] Furthermore, step 5 includes the following steps:

[0041] Sub-step 5.1 involves performing an inverse normalization process on the prediction output of the prediction model, converting the predicted values ​​back to the original data range, enhancing the interpretability and intuitiveness of the results, more accurately verifying the prediction performance of the model, and finally outputting the prediction results of the reservoir water level data.

[0042] Compared with the prior art, the present invention has the following beneficial effects:

[0043] This invention proposes a reservoir water level prediction method based on graph convolutional neural networks (GNNs). GNNs effectively capture the spatial and temporal dependencies in rainfall image data, and a bidirectional long short-term memory (LSTM) network is combined to further extract the temporal relationships of reservoir water level changes. This allows for a deeper understanding of the impact of rainfall image data on reservoir water levels, enabling water level prediction based on rainfall image data. This method can predict subsequent reservoir water level changes in a timely manner during rainfall, providing a scientific basis for developing response measures. This not only helps optimize reservoir scheduling strategies but also effectively supports flood prevention efforts, providing strong technical and decision support for water resource management. Attached Figure Description

[0044] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0045] Figure 1 This is a basic implementation flowchart of the present invention.

[0046] Figure 2 This is a diagram illustrating the construction of a convolutional neural network.

[0047] Figure 3 This is a diagram illustrating the implementation of a bidirectional long short-term memory network.

[0048] Figure 4 This is a diagram illustrating the construction of the prediction model for this invention.

[0049] Figure 5 This is a flowchart illustrating the implementation of model prediction. Detailed implementation method:

[0050] The technical solution of the present invention will be described in detail below, but the scope of protection of the present invention is not limited to the embodiments described.

[0051] This embodiment presents a reservoir water level prediction method based on graph convolutional neural networks, such as... Figure 1 The steps shown are as follows:

[0052] Step 1: Collect rainfall image data and reservoir water level data of the target watershed for in-depth analysis and modeling;

[0053] Furthermore, step 1 includes the following steps:

[0054] Sub-step 1.1: Search for rainfall image data and reservoir water level data from the Ministry of Water Resources website and the water resources department website of the province or city to which the target river basin belongs;

[0055] Sub-step 1.2: Determine the range of rainfall image data and reservoir water level data that need to be obtained from the Ministry of Water Resources website and the provincial and municipal water resources department websites of the target river basin, and filter the initial data within the range to select the valuable data according to the requirements.

[0056] Step 2: Perform data preprocessing on rainfall image data and reservoir water level data to ensure data quality consistency;

[0057] Furthermore, step 2 includes the following steps:

[0058] Sub-step 2.1: After collecting the data, check the completeness of the collected rainfall image data and reservoir water level data to confirm whether there is any missing data. Depending on the specific situation, supplement or delete the missing data appropriately.

[0059] Sub-step 2.2 involves preprocessing the rainfall image data to ensure that each rainfall image is the same size, so that batch processing can be performed.

[0060] Step 3: Align the rainfall image data and reservoir water level data to ensure consistency in timestamps and data types, and perform accurate comparison and analysis.

[0061] Furthermore, step 3 includes the following steps:

[0062] Sub-step 3.1 unifies the time units of rainfall image data and reservoir water level data, converting string timestamps into a standard date and time format so that the time of rainfall image data can correspond to the time of reservoir water level data;

[0063] Sub-step 3.2 ensures that all data related to the reservoir water level are of the same data type for subsequent analysis and processing;

[0064] Step 4: Train the graph convolutional neural network (GCN) on the rainfall image data and combine it with... Figure 3 The bidirectional long short-term memory network (BiLSTM) shown extracts the time series features of the reservoir water level, thereby constructing a model that can predict the reservoir water level through rainfall images;

[0065] Furthermore, step 4 includes the following steps:

[0066] Sub-step 4.1: Divide the corresponding rainfall image data and reservoir water level data into training set, validation set and test set according to the ratio of 7:1.5:1.5;

[0067] Sub-step 4.2, the graph convolutional neural network is as follows: Figure 2The structure is divided into five layers: data preprocessing layer, feature encoder layer, feature decoder layer, water level prediction layer, and output layer. The data preprocessing layer is mainly responsible for preprocessing the hydrological data of the target watershed and analyzing the correlation between historical and future data. The feature encoder and feature decoder layers are responsible for mining and fusing hydrological data features through graph convolutional neural networks (GCN). The water level prediction layer mines the time-series hydrological data after feature fusion based on BiLSTM to predict the reservoir water level data at future times.

[0068] Sub-step 4.3 selects rainfall image data as the key factor affecting reservoir water level changes. The correlation coefficient between historical rainfall image data and future reservoir water level data is calculated, and the cross-correlation diagram between historical rainfall maps and future reservoir water level data is analyzed to determine the magnitude of the influencing factor's data volume. The Pearson correlation coefficient is used to measure the strength of the linear relationship between the two time series of rainfall image data and reservoir water level data. The calculation formula is as follows:

[0069]

[0070] In the formula, x i and y i These are the i-th data points in sequences X and Y, respectively; and Let X and Y represent the average values ​​of sequences X and Y, respectively.

[0071] Sub-step 4.4: Train the prediction model. Select suitable data from the training set for model training, adopt an encoder-decoder structure, extract spatiotemporal features of rainfall images through a convolutional neural network (GCN), fuse these features using a multi-head attention mechanism, and finally use... Figure 3 The BiLSTM model shown is used to predict reservoir water level data. The prediction model structure is as follows: Figure 4 As shown.

[0072] During training, the correlation coefficient between historical rainfall image data and future reservoir water level data is calculated, and the cross-correlation plot between these two data points is analyzed to determine the amount of data influencing the factors. The cross-correlation function is as follows:

[0073] R = E(X) s *Y t )

[0074] In the formula, X s Y represents the value of time series X at time s. t R represents the value of the time series Y at time t, and R represents the degree of correlation between the two.

[0075] Data processed by an encoder-decoder and fused using a multi-head attention mechanism is used as input to a Bidirectional Long Short-Term Memory (BiLSTM) network for final reservoir water level prediction. In BiLSTM, the input time-series data sequence is processed through two directional LSTM networks to capture past and future information. By merging the hidden states from these two directions, the model can better understand the inherent patterns in the time-series data, thereby improving prediction accuracy. The prediction function is as follows:

[0076] y ′ =f BiLSTM (X ′ )

[0077] Where X' represents the fused rainfall image data, and y' represents the predicted reservoir water level.

[0078] Sub-step 4.5 trains a prediction model based on a graph convolutional neural network (GCN) by calculating the loss between the predicted value y' and the true value y. The loss function consists of the rainfall image data error (PE) and the mean squared error (MSE).

[0079] The formula for calculating the error of rainfall image data is as follows:

[0080] PE = y max -y max ′

[0081] The formula for calculating the mean squared error is as follows:

[0082]

[0083] The formula for calculating the loss function is as follows:

[0084] Loss = PE + MSE

[0085] Step 5: Input the rainfall image data into the trained prediction model to predict the reservoir water level data;

[0086] Furthermore, step 5 includes the following steps:

[0087] Sub-step 5.2 involves performing an inverse normalization process on the prediction model's output, converting the predicted values ​​back to the original data range. This enhances the interpretability and intuitiveness of the results, more accurately verifies the model's predictive performance, and finally outputs the predicted results for the reservoir water level data. The model prediction process in this embodiment is as follows: Figure 5 As shown.

Claims

1. A method for predicting reservoir water levels based on graph convolutional neural networks, comprising the following steps: Step 1: Collect rainfall image data and reservoir water level data for the target watershed in order to conduct in-depth analysis and modeling; Step 2: Perform data quality control on rainfall image data and reservoir water level data to ensure data quality consistency; Step 3: Align the rainfall image data and reservoir water level data to ensure consistency in timestamps and data types, and perform accurate comparison and analysis. Step 4: Based on the data fusion in Step 3, a Graph Convolutional Neural Network (GCN) is used to encode and decode the input rainfall image training data, extract the spatiotemporal characteristics of the rainfall image dataset, and fuse them using a multi-head attention mechanism. A Bidirectional Long Short-Term Memory (BiLSTM) network is then used to mine and extract the time-series features of the fused reservoir water level, thereby constructing a reservoir water level model that integrates the GCN and BiLSTM networks to predict future reservoir water levels. The reservoir water level prediction model constructed in step 4 is divided into five layers: data preprocessing layer, feature encoder layer, feature decoder layer, water level prediction layer, and output layer. In step 4, the data preprocessing layer preprocesses the hydrological data of the target watershed and analyzes the correlation between historical and future data. The feature encoder and feature decoder layers mine and fuse hydrological data features through graph convolutional neural networks (GCNs). The water level prediction layer mines the time-series hydrological data after feature fusion based on BiLSTM to predict reservoir water level data at future times. The output layer outputs the final prediction results. Step 5: Input the rainfall image data into the trained prediction model to predict the reservoir water level data.

2. The method for predicting reservoir water levels based on graph convolutional neural networks according to claim 1, characterized in that, Step 1 includes the following sub-steps: Sub-step 1.1: Search for rainfall image data and reservoir water level data from the Ministry of Water Resources website and the water resources department website of the province or city to which the target river basin belongs; Sub-step 1.2: After obtaining the initial data from the Ministry of Water Resources website and the provincial and municipal water resources department websites of the basin, it is necessary to filter the data and select the valuable rainfall image data and reservoir water level data of the basin according to the requirements.

3. The method for predicting reservoir water levels based on graph convolutional neural networks according to claim 2, characterized in that, Step 2 includes the following sub-steps: Sub-step 2.1: Check the collected rainfall image data and reservoir water level data to confirm whether there are any missing data. Depending on the specific situation, supplement or delete the missing data as appropriate. Sub-step 2.2 involves preprocessing the rainfall image data to ensure that each rainfall image is the same size so that it can be processed in batches.

4. The method for predicting reservoir water levels based on graph convolutional neural networks according to claim 3, characterized in that, Step 3 includes the following sub-steps: Sub-step 3.1: Unify the time unit for rainfall image data and reservoir water level data, and convert the string timestamp into a standard date and time format; Sub-step 3.2 ensures that all data related to the reservoir water level are of the same data type for subsequent analysis and processing.

5. The method for predicting reservoir water levels based on graph convolutional neural networks according to claim 4, characterized in that, Step 4 includes the following sub-steps: Sub-step 4.1: Divide the corresponding rainfall image data and reservoir water level data into training set, validation set and test set according to the ratio of 7:1.5:1.5; Sub-step 4.2 involves calculating the correlation coefficient between historical rainfall image data and future reservoir water level data, and analyzing the cross-correlation diagram between historical rainfall map data and future reservoir water level data to identify key factors influencing changes in reservoir water level data. The Pearson correlation coefficient is used to measure the strength of the linear relationship between these two time series of rainfall image data and reservoir water level data. The calculation formula is as follows: ； In the formula, x i and y i These are the i-th data points of the rainfall image sequence X and the reservoir water level sequence Y, respectively. and Let X and Y represent the average values ​​of sequences X and Y, respectively. Sub-step 4.3 adopts an encoder-decoder structure, extracts the spatiotemporal features of the rainfall image sequence through a convolutional neural network GCN, and fuses the features using a multi-head attention mechanism; BiLSTM is used to mine the time-series hydrological data after feature fusion to achieve the prediction of reservoir water level data. During training, the correlation coefficient between historical rainfall image data and future reservoir water level data is calculated, and the cross-correlation plot between historical rainfall image data and future reservoir water level data is analyzed to determine the amount of data affecting the factors. The cross-correlation function R is calculated as follows: ； In the formula, This represents the value of the rainfall image sequence X at time s. Let Y represent the value of the reservoir water level sequence at time t, and R represent the degree of correlation between the two. Data processed by an encoder-decoder and fused using a multi-head attention mechanism is used as input to a bidirectional long short-term memory (BiLSTM) network. The input time-series data sequence is processed by LSTM networks in two directions to capture information from the past and future. By merging the hidden states from these two directions, the model can better understand the inherent patterns of the time-series data, thereby improving prediction accuracy. The prediction function is as follows: ； Where X' is the fused rainfall image time series, and y′ is the predicted reservoir water level; Sub-step 4.4 trains a prediction model based on graph convolutional neural network by calculating the loss between the predicted value y′ and the true value y. The loss function consists of the error PE and the mean squared error MSE. The formula for calculating the error of rainfall image data is as follows: ； The formula for calculating the mean squared error is as follows: ； Where n is the total number of predictions; The formula for calculating the loss function is as follows: 。 6. The method for predicting reservoir water levels based on graph convolutional neural networks according to claim 5, characterized in that, Step 5 includes the following sub-steps: Sub-step 5.1 involves performing an inverse normalization process on the prediction output of the prediction model, converting the predicted values ​​back to the original data range, enhancing the interpretability and intuitiveness of the results, more accurately verifying the prediction performance of the model, and finally outputting the prediction results of the reservoir water level data.

Images