Flood warning detection method and system based on GCN graph convolutional neural network

Abstract

The present invention provides a flood warning detection method and system based on a GCN graph convolutional neural network, the method comprising: S1, deploying multiple data acquisition devices in a target area; the data acquisition devices include online monitoring devices for real-time acquisition of original water flow data at the monitoring location; S2, preprocessing the original water flow data and processing outliers to obtain processed water flow data; S3, converting the processed water flow data into a vector form suitable for input into a neural network, and constructing a graph structure of the water flow data; S4, constructing a graph convolutional neural network architecture, training and optimizing the network; using an adaptive algorithm to dynamically adjust model parameters, and an output layer generates flood warning results; S5, training historical water flow data and real-time water flow data, and constructing a flood warning detection model; S6, using the flood warning detection model to perform flood warning prediction, and outputting the obtained real-time flood warning information to a user interface; the method can provide more accurate and reliable prediction results for flood warning.

Description

Technical Field

[0001] The present invention relates to the field of flood warning technology, and in particular to a flood warning detection method and system based on a GCN graph convolutional neural network. Background Art

[0002] In urban water management, real-time monitoring and early warning of water flow status are crucial for preventing flood events, protecting residents' lives, and maintaining the stable operation of infrastructure. However, traditional water flow monitoring systems often have the following shortcomings: First, the water flow network is a complex topological structure. However, traditional water flow monitoring systems generally collect discrete water flow data through sensor networks, and it is difficult for traditional water flow monitoring systems to accurately simulate complex network structures through local data; second, the water flow environment exhibits complex dynamic characteristics in time and space. However, traditional water flow monitoring systems usually use rules or simple models for modeling, which cannot capture complex spatiotemporal relationships, affecting their accurate understanding of the dynamic changes of water flow; third, water flow data is easily affected by various factors, such as pipeline damage, pressure fluctuations, etc., but traditional water flow monitoring systems have limited ability to identify and process abnormal values, which also greatly affects the reliability of traditional water flow monitoring systems; in addition, traditional water flow monitoring systems also have the problem of delayed response to dynamic changes in water flow, insufficient real-time performance, and in emergency situations, flood warnings are difficult to be promptly and effectively conveyed to relevant departments and residents. The above defects are what those skilled in the art expect to overcome. Summary of the invention

[0003] The purpose of the present invention is to overcome the above-mentioned defects in the prior art. The first aspect of the present invention provides a flood warning detection method based on GCN graph convolutional neural network.

[0004] Based on the first aspect of the present invention, the second aspect of the present invention also provides a flood warning detection system based on the GCN graph convolutional neural network.

[0005] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0006] The flood warning detection method based on GCN graph convolutional neural network includes the following steps:

[0007] S1. deploy multiple data acquisition devices in the target area; the data acquisition devices include online monitoring devices for acquiring original water flow data at the monitoring location in real time;

[0008] S2, preprocessing the original water flow data and performing outlier processing to obtain processed water flow data;

[0009] S3, converting the processed water flow data into a vector form suitable for input into a neural network, and constructing a graph structure of the water flow data;

[0010] S4. Build a graph convolutional neural network architecture, train and optimize the network, and use an adaptive algorithm to dynamically adjust model parameters. The output layer generates flood warning results.

[0011] S5. Based on the graph convolutional neural network architecture, historical water flow data and real-time water flow data are trained to build a flood warning detection model;

[0012] S6. Use the flood warning detection model to perform flood warning prediction, and output the obtained real-time flood warning information to the user interface.

[0013] Furthermore, the online monitoring equipment is a water flow sensor, a pressure sensor and a temperature sensor; the raw water flow data is raw flow velocity data, raw pressure data and raw temperature data. Through the coordinated work of multiple sensors, water flow data can be collected with high precision and high frequency.

[0014] Preferably, the data acquisition device also includes an image acquisition device for acquiring original image data of the water area in real time; the original image data includes original water surface status information, original water area boundary information, and original rainfall distribution information. By setting up the image acquisition device, more intuitive water area image data can be acquired, so that the actual situation of the water area can be acquired more comprehensively; at the same time, real-time updates can be achieved, and visual displays can be made, so that users can understand the actual status of the water area more timely and intuitively; in addition, when the sensor fails or errors occur, the image acquisition device can be used as a backup means to ensure the accuracy and reliability of the monitoring data.

[0015] Furthermore, the raw water flow data is preprocessed and outlier processing is performed to obtain processed water flow data, including:

[0016] S21, using data cleaning technology to remove noise and invalid information in the original water flow data;

[0017] S22, using an outlier detection algorithm to identify and process outliers in the original water flow data;

[0018] S23, normalizing the original water flow data to obtain processed water flow data.

[0019] It can quickly and accurately identify and process outliers in the original water flow data, so that the processed water flow data has the advantages of high quality and good consistency, ensuring the data quality of the input model.

[0020] Furthermore, the processed water flow data is converted into a vector form suitable for input into a neural network, and a graph structure of the water flow data is constructed, including:

[0021] S31, converting the processed water flow data into an input vector, each online monitoring device corresponds to a feature of the vector, to form a feature vector of the input layer;

[0022] S32. Based on the association relationship between the online monitoring devices, a graph structure of water flow data is constructed; and the online monitoring devices are used as nodes of the graph, and the relationship between the nodes is represented by the edges between the nodes.

[0023] The relationship between online monitoring devices can be better described and analyzed, providing more accurate monitoring and decision support. This setup helps improve the management and operation of water flow systems and improve the reliability and efficiency of this detection method.

[0024] Furthermore, the construction of the graph convolutional neural network architecture, training and optimizing the network, and using an adaptive algorithm to dynamically adjust model parameters, and the output layer generates flood warning results, include:

[0025] S41, using the graph model structure to generate an adjacency matrix, and combining the sensor features with the adjacency matrix to generate an input for the graph convolutional neural network;

[0026] S42, performing a graph convolution operation, considering the information transfer between the node and the adjacent nodes, aggregating the features of the node through the adjacency matrix, and updating the node representation;

[0027] S43, using activation function, the output of each layer is introduced into nonlinearity through the activation function to enhance the expression ability of the network;

[0028] S44, initializing the weight and bias parameters of the graph convolutional neural network, and introducing an adaptive learning algorithm to achieve real-time adjustment of the model parameters;

[0029] S45. Define a loss function, calculate the gradient of the loss function to the model parameters through the back propagation algorithm, use the adaptive learning algorithm to adjust the learning rate or other parameters of the gradient descent, update the weight and bias parameters of the model, and monitor the performance of the model on the training data in real time;

[0030] S46, dynamically adjust the model parameters by repeatedly iterating the adaptive learning process, and introduce an online learning mechanism so that the model can respond to new data in a timely manner;

[0031] S47, passing the processed water flow data from the input layer to the fully connected layer through the weight matrix, and performing weighted summation on the passed processed water flow data, considering the weight relationship between different features;

[0032] S48. According to actual needs and performance monitoring results, the parameters of the output layer are dynamically adjusted to adapt to different prediction targets; the output layer generates flood warning results.

[0033] This method can perceive and adapt to the dynamic changes of the water flow environment in real time, and adjust the model parameters according to actual conditions to better adapt to different water flow states and scenarios.

[0034] Furthermore, the graph convolutional neural network architecture is used to train historical water flow data and real-time water flow data, and to construct a flood warning detection model, including:

[0035] S51, using an adaptive learning algorithm to adjust model parameters, and perform training and cross-validation;

[0036] S52, performing anomaly detection on real-time water flow data to identify abnormal conditions; at the same time, combining the strategy of adaptively adjusting the threshold, dynamically adjusting the boundary for distinguishing abnormalities;

[0037] S53. Use the LSTM model (long short-term memory network model) to analyze historical water flow data and establish a flood warning detection model.

[0038] This method can flexibly adjust the criteria for distinguishing anomalies in a dynamic water flow environment, so as to better adapt to changes in different scenarios; through the LSTM model, it can more accurately predict future water flow trends and provide more accurate and reliable prediction results for flood warnings.

[0039] Preferably, the method of performing anomaly detection on real-time water flow data to identify abnormal conditions; and combining a strategy of adaptively adjusting thresholds to dynamically adjust the boundaries for distinguishing abnormalities, includes:

[0040] S521, online monitoring of real-time data of water flow environment, and anomaly detection through graph convolutional neural network;

[0041] S522. Evaluate the degree of abnormality according to the abnormality detection result; and adjust the parameters of the graph convolutional neural network using an adaptive learning algorithm based on the degree of abnormality;

[0042] S523, updating the adjusted parameters online to the graph convolutional neural network to adapt to the dynamic changes of the current water flow environment.

[0043] Preferably, the use of the LSTM model to analyze historical water flow data includes:

[0044] S531. Preprocess historical water flow data, design and build LSTM model;

[0045] S532. Use historical water flow data to train the LSTM model; use the trained LSTM model to predict the water flow state at a future time point.

[0046] Secondly, a flood warning detection system based on GCN graph convolutional neural network is provided, including the following modules:

[0047] Data acquisition module, used to monitor and obtain the original water flow data and original image data of the target area in real time;

[0048] A data preprocessing module is used to preprocess and process the outliers of the original water flow data acquired by the data acquisition module to obtain processed water flow data; and convert the processed water flow data into a vector form suitable for input into a neural network;

[0049] The feature extraction module uses graph convolutional neural network as a feature extraction tool to extract spatial features and temporal features from the original pipeline data;

[0050] A model training module is used to use historical water flow data to learn the features extracted by the feature extraction module to build a flood warning detection model; use an adaptive learning algorithm to adjust model parameters, and perform training and cross-validation;

[0051] Anomaly detection module, which is used to analyze real-time water flow data to detect abnormal situations in time;

[0052] An adaptive threshold adjustment module is used to dynamically adjust the threshold of anomaly detection based on real-time water flow data and real-time anomaly detection results;

[0053] The historical data analysis module uses the LSTM model to analyze historical water flow data and establish a flood warning detection model;

[0054] An image monitoring module is used to pre-process the original image data acquired by the data acquisition module to obtain processed image data; analyze the processed image data and integrate the historical water flow data to fully understand the dynamic changes of the water flow environment;

[0055] Mobile application integration module to provide real-time flood warning information to users.

[0056] Compared with the prior art, the present invention has the following beneficial effects:

[0057] The flood warning detection method and system based on the GCN graph convolutional neural network provided by the present invention acquires the original water flow data at the monitoring location in real time by deploying multiple data acquisition devices, timely understands the water flow situation, and improves the flexibility and accuracy of the method; by preprocessing and outlier processing the original water flow data, noise and abnormal data can be removed to obtain accurate water flow data after processing; the processed water flow data is converted into a vector form suitable for inputting into a neural network, and a graph structure of the water flow data is constructed, and the graph convolutional neural network is used for training and optimization, and the graph convolutional neural network can successfully capture the temporal and spatial correlation of water flow data and can more accurately model the dynamic water flow environment; by effectively integrating topological relationships and node features, the comprehensive understanding of water flow temporal and spatial information can be improved, and the accuracy and efficiency of prediction can be improved; in addition, an adaptive algorithm is used to dynamically adjust the model parameters, and the model can be optimized according to the actual situation, so that the prediction results are more accurate and reliable; at the same time, historical water flow data and real-time water flow data are trained based on the graph convolutional neural network architecture to construct a flood warning detection model, which can comprehensively consider the influence of historical data and real-time data, and further improve the accuracy and reliability of prediction. By outputting real-time flood warning information to the user interface, users can timely understand the flood warning situation and take corresponding protective measures to protect people’s lives and property. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 It is a flow chart of a flood warning detection method based on a GCN graph convolutional neural network provided by the present invention;

[0059] Figure 2 This is a structural diagram of the graph convolutional neural network architecture in the flood warning detection method based on the GCN graph convolutional neural network provided by the present invention. DETAILED DESCRIPTION

[0060] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0061] See also Figure 1 to Figure 2 , an embodiment of the present invention provides a flood warning detection method based on a GCN graph convolutional neural network, the method comprising the following steps:

[0062] S1. Multiple data acquisition devices are deployed in the target area; the data acquisition devices include online monitoring devices and image acquisition devices; wherein the online monitoring devices are used to obtain the original water flow data at the monitoring location in real time; the image acquisition devices are used to obtain the original image data of the water area in real time.

[0063] In this embodiment, the online monitoring device uses a plurality of water flow sensors, pressure sensors and temperature sensors. Therefore, in this embodiment, the raw water flow data acquired in real time is raw flow velocity data, raw pressure data and raw temperature data. Through the collaborative work of multiple sensors, water flow data can be collected with high precision and high frequency. Specifically, the above sensors realize remote monitoring and real-time data transmission through the Internet of Things technology to improve flexibility.

[0064] Image acquisition equipment is used to obtain the original image data of the water area in real time; specifically, the original image data includes the original water surface status information, the original water area boundary information, and the original rainfall distribution information. In addition, the image acquisition equipment is also used to monitor the pipeline surface image; specifically, computer vision technology is used to monitor and analyze the pipeline surface image in real time to obtain more comprehensive pipeline status information. This sophisticated image analysis not only helps to discover potential pipeline problems, but also identifies factors that may cause water flow abnormalities at an early stage. The use of computer vision technology also further improves the water flow monitoring system's perception accuracy of pipeline status, enabling it to have a more comprehensive understanding of the real-time situation of the water flow environment.

[0065] Image acquisition equipment can obtain the actual situation of the water area more comprehensively by collecting more intuitive water area image data. At the same time, it can also be updated in real time and displayed visually, so that users can understand the actual status of the water area more timely and intuitively. In addition, when the sensor fails or errors occur, the image acquisition equipment can be used as a backup means to ensure the accuracy and reliability of the monitoring data.

[0066] S2. Preprocess the original water flow data and perform outlier processing to obtain processed water flow data.

[0067] Specifically, the following steps are included:

[0068] S21. Using data cleaning technology to remove noise and invalid information from the original water flow data.

[0069] S22: using an outlier detection algorithm to identify and process outliers in the original water flow data.

[0070] S23, normalizing the original water flow data to obtain processed water flow data.

[0071] Among them, filtering or smoothing techniques are used to process the noise in the original water flow data to reduce the interference of noise on the model. The above steps can quickly and accurately identify and process the outliers in the original water flow data, so that the processed water flow data has the advantages of high quality and good consistency, ensuring the data quality of the input model.

[0072] S3. Convert the processed water flow data into a vector form suitable for input into a neural network, and construct a graph structure of the water flow data.

[0073] Specifically, the following steps are included:

[0074] S31, converting the processed water flow data into an input vector, each online monitoring device corresponds to a feature of the vector, to form a feature vector of the input layer.

[0075] S32. Based on the association relationship between the online monitoring devices, a graph structure of water flow data is constructed; and the online monitoring devices are used as nodes of the graph, and the relationship between the nodes is represented by the edges between the nodes.

[0076] Taking the above steps can better describe and analyze the relationship between online monitoring devices, provide more accurate monitoring and decision support. This setup helps improve the management and operation of water flow systems and improve the reliability and efficiency of this detection method.

[0077] S4. Build a graph convolutional neural network architecture, train and optimize the network, and use an adaptive algorithm to dynamically adjust model parameters. The output layer generates flood warning results.

[0078] Specifically, the following steps are included:

[0079] S41. Generate an adjacency matrix using the graph model structure, and combine the sensor features with the adjacency matrix to generate the input of the graph convolutional neural network.

[0080] S42. Perform graph convolution operation, consider the information transfer between nodes and adjacent nodes, aggregate the features of nodes through the adjacency matrix, and update the node representation.

[0081] S43. Use activation function to introduce nonlinearity into the output of each layer through the activation function to enhance the network's expressive power.

[0082] S44. Initialize the weights and bias parameters of the graph convolutional neural network and introduce an adaptive learning algorithm to achieve real-time adjustment of the model parameters.

[0083] S45. Define the loss function, calculate the gradient of the loss function with respect to the model parameters through the back-propagation algorithm, use the adaptive learning algorithm to adjust the learning rate or other parameters of the gradient descent, update the weight and bias parameters of the model, and monitor the performance of the model on the training data in real time.

[0084] S46. Perform an adaptive learning process through repeated iterations, dynamically adjust model parameters, and introduce an online learning mechanism so that the model can respond to new data in a timely manner.

[0085] S47, passing the processed water flow data from the input layer to the fully connected layer through the weight matrix, and performing weighted summation on the passed processed water flow data, considering the weight relationship between different features.

[0086] S48. According to actual needs and performance monitoring results, the parameters of the output layer are dynamically adjusted to adapt to different prediction targets; the output layer generates flood warning results.

[0087] Taking the above steps can perceive and adapt to the dynamic changes of the water flow environment in real time, and adjust the model parameters according to the actual situation to better adapt to different water flow states and scenarios.

[0088] S5. Based on the graph convolutional neural network architecture, historical water flow data and real-time water flow data are trained to build a flood warning detection model.

[0089] Specifically, the following steps are included:

[0090] S51. Use adaptive learning algorithm to adjust model parameters, and perform training and cross-validation.

[0091] S52, perform anomaly detection on real-time water flow data to identify abnormal conditions; at the same time, combine the strategy of adaptively adjusting the threshold value to dynamically adjust the boundary of abnormality identification. Specifically, the following steps are also included:

[0092] S521, online monitoring of real-time data of water flow environment, and anomaly detection through graph convolutional neural network.

[0093] S522. Evaluate the degree of abnormality according to the abnormality detection result; and adjust the parameters of the graph convolutional neural network using an adaptive learning algorithm based on the degree of abnormality.

[0094] S523, updating the adjusted parameters online to the graph convolutional neural network to adapt to the dynamic changes of the current water flow environment.

[0095] S53, using the LSTM model (Long Short-Term Memory Network Model) to analyze historical water flow data and establish a flood warning detection model. Specifically, the following steps are also included:

[0096] S531. Preprocess historical water flow data, design and build an LSTM model.

[0097] S532. Use historical water flow data to train the LSTM model; use the trained LSTM model to predict the water flow state at a future time point.

[0098] Taking the above steps can flexibly adjust the criteria for judging anomalies in a dynamic water flow environment, so as to better adapt to changes in different scenarios; through the LSTM model, future water flow trends can be predicted more accurately, providing more accurate and reliable prediction results for flood warnings.

[0099] S6. Use the flood warning detection model to perform flood warning prediction, and output the obtained real-time flood warning information to the user interface.

[0100] The flood warning detection method provided in this embodiment provides a user-friendly interface through real-time monitoring, efficient and accurate prediction, adaptive adjustment, and integrated historical and real-time data. It can improve the accuracy, timeliness and reliability of flood warnings, and is of great significance for preventing flood disasters and ensuring public safety.

[0101] An embodiment of the present invention also provides a flood warning detection system based on a GCN graph convolutional neural network, which includes the following modules:

[0102] The data acquisition module is used to monitor and obtain the original water flow data and original image data of the target area in real time.

[0103] The data preprocessing module is used to preprocess and process the outliers of the original water flow data acquired by the data acquisition module to obtain processed water flow data; and convert the processed water flow data into a vector form suitable for input into a neural network.

[0104] The feature extraction module uses graph convolutional neural network as a feature extraction tool to extract spatial features and temporal features from the original water flow data. Specifically, the graph convolutional neural network adopts a multi-layer convolutional layer and graph convolutional layer structure, which can more effectively capture the complex spatiotemporal characteristics of the water flow network.

[0105] The model training module is used to use historical water flow data to learn the features extracted by the feature extraction module to build a flood warning detection model; use an adaptive learning algorithm to adjust model parameters, and perform training and cross-validation. Among them, the model training module combines a graph convolutional neural network with an adaptive learning algorithm, and uses an adaptive learning algorithm to adjust model parameters, perform training and cross-validation to improve the generalization ability of the model. A semi-supervised learning method is also used to improve the generalization performance of the model using labeled and unlabeled data.

[0106] The anomaly detection module is used to analyze the real-time water flow data in real time to detect abnormal situations in time. Specifically, the anomaly detection module is combined with the time window technology to more accurately capture the instantaneous changes in water flow state; and combined with the strategy of adaptively adjusting the threshold, the system's adaptability to different working conditions is improved by dynamically adjusting the boundaries for distinguishing abnormalities. Preferably, in this embodiment, the anomaly detection module also has a log recording function to record the results of each anomaly detection. Among them, the anomaly detection results include information such as anomaly type and timestamp to facilitate historical data analysis and system performance optimization.

[0107] The adaptive threshold adjustment module is used to dynamically adjust the threshold of anomaly detection according to real-time water flow data and real-time anomaly detection results.

[0108] The historical data analysis module uses the LSTM model to analyze historical water flow data and establish a flood warning detection model. Specifically, the historical data analysis module stores the data from the real-time anomaly detection module and other related data in a queryable database to analyze historical data; uses time series analysis methods to analyze the historical trends of water flow data to predict future water flow states and identify possible periodic changes; tracks historical abnormal events, analyzes causes, impacts, and system responses to improve real-time anomaly detection algorithms and improve system reliability.

[0109] The image monitoring module is used to pre-process the raw image data acquired by the data acquisition module to obtain processed image data; analyze the processed image data and integrate the historical water flow data to fully understand the dynamic changes of the water flow environment. The image monitoring module uses computer vision technology to perform a detailed analysis of the image on the surface of the pipeline to improve the perception accuracy of the water flow monitoring system on the pipeline status, so that it can more comprehensively understand the real-time situation of the water flow environment. Specifically, the pre-processing of the raw image data includes denoising, adjusting brightness and contrast, image enhancement, etc., to ensure that the image quality meets the requirements of subsequent analysis.

[0110] The mobile application integration module is used to provide users with real-time flood warning information. Preferably, the mobile application integration module also supports the user's geographic location service, so that the user can quickly obtain real-time flood information.

[0111] The flood warning detection system based on the GCN graph convolutional neural network provided in this embodiment adopts a graph convolutional neural network to capture the correlation of water flow data in time and space, so that the warning detection system can more accurately model the dynamic water flow environment; the topological structure and node features are integrated to improve the comprehensive understanding of the water flow spatiotemporal information of the warning detection system, which not only makes the warning detection system more adaptable and flexible, but also improves the accuracy and reliability of water flow monitoring. The warning detection system can not only perceive and adapt to the dynamic changes of the water flow environment in real time, but also adjust the model parameters according to actual conditions to better adapt to different water flow states and scenes, thereby improving the reliability of the warning detection system. The warning detection system can accurately identify and filter out abnormal values, improve the accuracy and credibility of the water flow data input into the system, enhance the adaptability of the warning detection system to complex water flow environments, and also help improve the performance of the warning detection system. At the same time, the warning detection system organically combines a friendly user interface, mobile applications and geographic location services, which facilitates users to obtain real-time flood information and improves user experience.

[0112] The above embodiments are preferred implementation modes of the present invention, but the implementation modes of the present invention are not limited to the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications that do not deviate from the spirit and principles of the present invention should be equivalent replacement methods and are included in the protection scope of the present invention.

Claims

1. A flood warning detection method based on GCN graph convolutional neural network, characterized in that: A flood warning detection system based on GCN graph convolutional neural network is included, including the following modules: Data acquisition module, used to monitor and obtain the original water flow data and original image data of the target area in real time; A data preprocessing module is used to preprocess and process the outliers of the original water flow data acquired by the data acquisition module to obtain processed water flow data; and convert the processed water flow data into a vector form suitable for input into a neural network; The feature extraction module uses graph convolutional neural network as a feature extraction tool to extract spatial features and temporal features from the original pipeline data; A model training module is used to use historical water flow data to learn the features extracted by the feature extraction module to build a flood warning detection model; use an adaptive learning algorithm to adjust model parameters, and perform training and cross-validation; Anomaly detection module, which is used to analyze real-time water flow data to detect abnormal situations in time; An adaptive threshold adjustment module is used to dynamically adjust the threshold of anomaly detection based on real-time water flow data and real-time anomaly detection results; The historical data analysis module uses the LSTM model to analyze historical water flow data and establish a flood warning detection model; An image monitoring module, used for preprocessing the original image data acquired by the data acquisition module to obtain processed image data; Analyze the processed image data and integrate historical water flow data to fully understand the dynamic changes of the water flow environment; Mobile application integration module to provide real-time flood warning information to users; The following steps are also included: S1. deploy multiple data acquisition devices in the target area; the data acquisition devices include online monitoring devices for acquiring original water flow data at the monitoring location in real time; S2, preprocessing the original water flow data and performing outlier processing to obtain processed water flow data; S3, converting the processed water flow data into a vector form suitable for input into a neural network, and constructing a graph structure of the water flow data; S4. Build a graph convolutional neural network architecture, train and optimize the network, and use an adaptive algorithm to dynamically adjust model parameters. The output layer generates flood warning results. S5. Based on the graph convolutional neural network architecture, historical water flow data and real-time water flow data are trained to build a flood warning detection model; S6. Use the flood warning detection model to perform flood warning prediction, and output the obtained real-time flood warning information to the user interface.

2. The flood warning detection method based on GCN graph convolutional neural network according to claim 1 is characterized in that: The online monitoring equipment includes a water flow sensor, a pressure sensor and a temperature sensor; the original water flow data includes original flow velocity data, original pressure data and original temperature data.

3. The flood warning detection method based on GCN graph convolutional neural network according to claim 2 is characterized in that: The data acquisition device also includes an image acquisition device for acquiring original image data of the water area in real time; the original image data includes original water surface state information, original water area boundary information and original rainfall distribution information.

4. The flood warning detection method based on GCN graph convolutional neural network according to claim 1 is characterized in that: The raw water flow data is preprocessed and outlier processing is performed to obtain processed water flow data, including: S21, using data cleaning technology to remove noise and invalid information in the original water flow data; S22, using an outlier detection algorithm to identify and process outliers in the original water flow data; S23, normalizing the original water flow data to obtain processed water flow data.

5. The flood warning detection method based on GCN graph convolutional neural network according to claim 1 is characterized in that: The process of converting the processed water flow data into a vector form suitable for inputting into a neural network and constructing a graph structure of the water flow data includes: S31, converting the processed water flow data into an input vector, each online monitoring device corresponds to a feature of the vector, to form a feature vector of the input layer; S32. Based on the association relationship between the online monitoring devices, a graph structure of water flow data is constructed; and the online monitoring devices are used as nodes of the graph, and the relationship between the nodes is represented by the edges between the nodes.

6. The flood warning detection method based on GCN graph convolutional neural network according to claim 1 is characterized in that: The graph convolutional neural network architecture is constructed, the network is trained and optimized, and an adaptive algorithm is used to dynamically adjust the model parameters, and the output layer generates flood warning results, including: S41, using the graph model structure to generate an adjacency matrix, and combining the sensor features with the adjacency matrix to generate an input for the graph convolutional neural network; S42, performing a graph convolution operation, considering the information transfer between the node and the adjacent nodes, aggregating the features of the node through the adjacency matrix, and updating the node representation; S43, using activation function, the output of each layer is introduced into nonlinearity through the activation function to enhance the expression ability of the network; S44, initializing the weight and bias parameters of the graph convolutional neural network, and introducing an adaptive learning algorithm to achieve real-time adjustment of the model parameters; S45. Define a loss function, calculate the gradient of the loss function to the model parameters through the back propagation algorithm, use the adaptive learning algorithm to adjust the learning rate or other parameters of the gradient descent, update the weight and bias parameters of the model, and monitor the performance of the model on the training data in real time; S46, dynamically adjust the model parameters by repeatedly iterating the adaptive learning process, and introduce an online learning mechanism so that the model can respond to new data in a timely manner; S47, passing the processed water flow data from the input layer to the fully connected layer through the weight matrix, and performing weighted summation on the passed processed water flow data, considering the weight relationship between different features; S48. According to actual needs and performance monitoring results, the parameters of the output layer are dynamically adjusted to adapt to different prediction targets; the output layer generates flood warning results.

7. The flood warning detection method based on GCN graph convolutional neural network according to claim 1 is characterized in that: The graph convolutional neural network architecture is used to train historical water flow data and real-time water flow data, and to build a flood warning detection model, including: S51, using an adaptive learning algorithm to adjust model parameters, and perform training and cross-validation; S52, performing anomaly detection on real-time water flow data to identify abnormal conditions; at the same time, combining the strategy of adaptively adjusting the threshold, dynamically adjusting the boundary for distinguishing abnormalities; S53. Use the LSTM model (Long Short-Term Memory Network model) to analyze historical water flow data and establish a flood warning detection model.

8. The flood warning detection method based on GCN graph convolutional neural network according to claim 7 is characterized in that: The real-time water flow data is subjected to anomaly detection to identify abnormal conditions; At the same time, combined with the strategy of adaptively adjusting the threshold, the boundary for distinguishing abnormalities is dynamically adjusted, including: S521, online monitoring of real-time data of water flow environment, and anomaly detection through graph convolutional neural network; S522. Evaluate the degree of abnormality according to the abnormality detection result; and adjust the parameters of the graph convolutional neural network using an adaptive learning algorithm based on the degree of abnormality; S523, updating the adjusted parameters online to the graph convolutional neural network to adapt to the dynamic changes of the current water flow environment.

9. The flood warning detection method based on GCN graph convolutional neural network according to claim 7 is characterized in that: The use of the LSTM model to analyze historical water flow data includes: S531. Preprocess historical water flow data, design and build LSTM model; S532. Use historical water flow data to train the LSTM model; use the trained LSTM model to predict the water flow state at a future time point.

Images