Automatic monitoring system and method for hydrological data and storage medium

Abstract

The invention discloses an automatic monitoring system and method for hydrological data and a storage medium. The automatic monitoring system comprises a multi-modal data acquisition module, an edge intelligent processing module and a cloud platform. The multi-modal data acquisition module acquires water level, water quality and meteorological data in real time through a radar water level gauge, a multi-parameter water quality meter and a rain gauge, and transmits the data to the edge intelligent processing module; the module extracts and aligns spatio-temporal features through a multi-scale spatio-temporal convolution unit, an uncertainty quantization unit is fused with Bayesian to calculate confidence, and complete data or abstract features are selectively transmitted to a cloud according to an abnormal classification judgment result; the cloud platform fuses multi-source features through a space-time attention mechanism, dynamically adjusts an early warning threshold based on a near-end strategy optimization algorithm, aggregates edge node model parameters through a federated learning framework in combination with a homomorphic encryption technology, updates a global model, and then synchronizes the global model to edge nodes to form closed-loop optimization.

Description

Technical Field

[0001] The present invention belongs to the field of hydrological data processing, and in particular relates to an automatic monitoring system, method and storage medium for hydrological data. Background Art

[0002] In the field of hydrological monitoring, accurate and timely acquisition of hydrological data is crucial for flood prevention and disaster reduction, water resources management, and ecological and environmental protection. With the advancement of science and technology, various hydrological monitoring technologies and systems have emerged. Traditional hydrological monitoring relies primarily on a single type of sensor, manually collecting data periodically or using simple data transmission methods to aggregate data to a monitoring center.

[0003] However, existing hydrological monitoring systems have numerous shortcomings. For one thing, data collection is limited in scope, often capturing only limited information such as water levels and flow rates, making it difficult to fully reflect the complex changes in the hydrological environment. For example, in water quality monitoring, traditional systems may be unable to detect changes in pollutants in the water in real time, resulting in delayed responses to water pollution incidents. Furthermore, data processing and analysis capabilities are limited, making it impossible to fully tap into the potential information behind the data. For massive amounts of hydrological data, traditional systems often rely on simple statistical analysis methods, making it impossible to accurately predict and warn of hydrological trends. Furthermore, existing systems lack effective adaptive mechanisms, making it difficult to make timely and accurate decisions when faced with complex and changing hydrological environments and emergencies.

[0004] These issues have impacted the accuracy and reliability of hydrological monitoring, hindering effective decision-making. In terms of flood prevention and disaster reduction, the inability to accurately and timely monitor water level changes and flood progression can lead to inadequate flood control measures, posing a serious threat to people's lives and property. In terms of water resource management, incomplete monitoring of water quality and quantity can lead to irrational development and utilization of water resources, impacting the balance of the ecological environment. Summary of the Invention

[0005] In response to one or more of the above-mentioned defects or improvement needs in the prior art, the present invention provides an automatic monitoring system, method and storage medium for hydrological data to solve the problems of the existing hydrological monitoring system, such as the single data collection dimension, limited data processing capability and lack of adaptive mechanism, so as to realize multimodal collection, intelligent processing and accurate situation prediction and decision support of hydrological data.

[0006] To achieve the above objectives, the present invention provides an automatic monitoring system for hydrological data, which includes a multimodal data acquisition module and an edge intelligent processing module in communication with the module, as well as a cloud platform;

[0007] The multimodal data acquisition module includes a radar water level gauge set in the water area to be measured, which is used to accurately collect water level data; a multi-parameter water quality meter set at different depths in the monitored water area, which is used to comprehensively collect multi-dimensional data related to water quality; and a rain gauge set around the monitored water area, which is used to accurately collect rainfall in meteorological data. This module is connected to the edge intelligent processing module through a data transmission interface.

[0008] The edge intelligent processing module includes a multi-scale spatiotemporal convolution unit, an uncertainty quantification unit, and a feature transmission unit; the multi-scale spatiotemporal convolution unit is used to receive data from the multimodal data acquisition module and extract and analyze spatiotemporal features of different scales; the uncertainty quantification unit is connected to the multi-scale spatiotemporal convolution unit and the feature transmission unit respectively, and is used to quantify the uncertainty of the data and transmit the processed data to the cloud platform according to the results of the abnormality classification judgment;

[0009] The cloud-based self-evolving AI platform includes a multimodal data fusion module and a dynamic threshold optimization model that is communicatively connected to the multimodal data fusion module, as well as a federated learning framework; the multimodal data fusion module is communicatively connected to the edge intelligent processing module through an interface, and is used to receive the received edge features generated by the edge intelligent processing module and generate global fusion features. The dynamic threshold optimization model is used to calculate the warning threshold based on the fusion features and adjust the output threshold accordingly. The federated learning module updates the global model by aggregating edge model parameters through homomorphic encryption and feeds it back to the edge nodes to form a closed-loop optimization system.

[0010] As a further improvement of the present invention, the radar water level meter of the multimodal data acquisition module is connected to the edge intelligent processing module via an RS485 / Modbus interface;

[0011] The multi-parameter water quality meter is connected to the edge intelligent processing module through a 5G communication interface, and the tipping bucket rain gauge is connected to the edge intelligent processing module through a Wi-Fi communication interface.

[0012] As a further improvement of the present invention, when the multi-scale spatiotemporal convolution unit processes data, it first extracts the timestamp information therein, and uses the position encoding function to process the timestamp to obtain a time feature vector; at the same time, it extracts the position information in the original data, and processes the position information through a graph convolution network combined with an adjacency matrix to generate a spatial feature vector; finally, the time feature vector and the spatial feature vector are spliced in the last dimension to form an aligned feature vector.

[0013] As a further improvement of the present invention, when the uncertainty quantification unit quantifies the uncertainty of the data, the aligned feature vector is segmented according to each sample, and the Bayesian Dropout technique is applied to each segmented feature vector to obtain a feature vector list with uncertainty information;

[0014] Then, the feature vector list with uncertainty information is concatenated and input into the weight network;

[0015] Calculate the weight of each eigenvector according to its degree of uncertainty;

[0016] According to the calculated weights, the feature vectors with uncertainty information are weighted summed to obtain the fused feature vector.

[0017] The data fusion unit is used to receive the feature data of the edge intelligent processing module, the threshold optimization unit is connected to the data fusion unit and outputs the dynamic threshold, and the federated learning unit interacts with the edge intelligent processing module through an interface to update the model parameters

[0018] As a further improvement of the present invention, the abnormality classification judgment is specifically as follows: calculating the abnormality score of the fused feature vector and judging whether the abnormality score is greater than a set threshold; if the abnormality score is greater than the threshold, the complete original data is sent to the cloud platform; if the abnormality score is less than or equal to the threshold, only the processed summary features are sent to the cloud platform.

[0019] As a further improvement of the present invention, the multimodal data fusion module is communicatively connected to the edge intelligent processing module via an interface, is configured to receive the fusion feature data transmitted by the edge intelligent processing module, and is configured to perform the following operations:

[0020] The spatiotemporal attention mechanism is adopted to align the timestamps and spatial coordinates of multi-source data. Combined with the multi-scale convolutional network, multi-level spatiotemporal features are extracted from the fused features, external data sources are integrated, and a fused feature vector in a unified format is generated.

[0021] As a further improvement of the present invention, the dynamic threshold optimization model is trained based on the proximal strategy optimization algorithm, connected to the multimodal data fusion module, receives the fused feature data output by the multimodal data fusion module, and is configured to generate a dynamic threshold in the following manner:

[0022] Based on the formula Threshold = υ + α (b false alarm rate + y false alarm rate); where α, b, and y are adaptive weight coefficients;

[0023] Use reward function R = a·(1-false alarm rate) + b·(1-false alarm rate);

[0024] According to the historical false alarm rate, missed alarm rate and current data distribution entropy, the warning threshold is dynamically adjusted and output to the federated learning framework through the interface.

[0025] As a further improvement of the present invention, the federated learning framework interacts with the edge intelligent processing module through the gRPC interface, receives the local model parameters uploaded by it, and connects with the dynamic threshold optimization model to provide model parameter update support and is configured to perform the following operations:

[0026] Homomorphic encryption technology is used to aggregate edge node model parameters without exposing the original data; the global model is updated based on the federated learning algorithm and synchronized to the edge intelligent processing module to improve prediction accuracy; abnormal data reported by edge nodes is labeled and analyzed to optimize the model's ability to recognize abnormal patterns.

[0027] On this basis, the present invention also provides a method for automatic monitoring of hydrological data, which includes the following steps:

[0028] S1. Collecting original multimodal data through the multimodal data acquisition module and transmitting it to the edge intelligent processing module;

[0029] S2. The multi-scale spatiotemporal convolution unit of the edge intelligent processing module processes the raw data to generate alignment features; the uncertainty quantification unit calculates the confidence score and transmits the processed data to the cloud platform according to the abnormality classification judgment result;

[0030] S3, the multimodal data fusion module of the cloud platform integrates the features of multiple edge nodes to generate a global fusion feature; the dynamic threshold optimization model calculates the warning threshold based on the fusion feature; the federated learning framework aggregates the edge model parameters and updates the global model;

[0031] S4. The cloud platform synchronizes the optimized model parameters to the edge nodes to form a closed-loop optimization system.

[0032] On this basis, the present invention further provides a storage medium, wherein the storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned automatic monitoring method is executed.

[0033] The above-mentioned improved technical features can be combined with each other as long as they do not conflict with each other.

[0034] In general, the above technical solutions conceived by the present invention have the following beneficial effects compared with the prior art:

[0035] The automatic monitoring system for hydrological data of the present invention comprehensively and accurately collects multi-dimensional hydrological data such as water level, water quality, and meteorology through a multimodal data acquisition module, and effectively extracts spatiotemporal features in the data through an edge intelligent processing module to manage data uncertainty; finally, the dynamic threshold optimization model of the cloud-based self-evolving AI platform can dynamically adjust the warning threshold according to actual conditions. The federated learning framework can achieve collaborative optimization of the model while ensuring data privacy, giving the system stronger adaptive decision-making capabilities and better practical value and application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 is a block diagram of a system for automatically monitoring hydrological data according to an embodiment of the present invention;

[0037] Figure 2 4 is a flow chart of a method for automatically monitoring hydrological data in an embodiment of the present invention. DETAILED DESCRIPTION

[0038] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and embodiments.

[0039] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0040] See also Figure 1 ,like Figure 1As shown in , the hydrological data monitoring system in the preferred embodiment of the present invention includes a multimodal data acquisition module including a radar water level meter arranged in the water area to be measured, for accurately collecting water level data; a multi-parameter water quality meter arranged at different depths in the monitored water area, for comprehensively collecting multi-dimensional data related to water quality; a rain gauge arranged around the monitored water area, for accurately collecting rainfall in meteorological data; the module is connected to the edge intelligent processing module through a data transmission interface; the edge intelligent processing module includes a multi-scale spatiotemporal convolution unit, an uncertainty quantification unit, and a feature transmission unit; the multi-scale spatiotemporal convolution unit is used to receive data from the multimodal data acquisition module, and extract and analyze spatiotemporal features of different scales; the uncertainty quantification unit is respectively connected to the multi-scale spatiotemporal convolution unit and the edge intelligent processing module. The element is connected to the feature transmission unit to quantify the uncertainty of the data, and transmits the processed data to the cloud platform according to the results of the abnormality classification judgment; the cloud self-evolving AI platform includes a multimodal data fusion module and a dynamic threshold optimization model communicated with the multimodal data fusion module, and a federated learning framework; the multimodal data fusion module is communicated with the edge intelligent processing module through an interface, and is used to receive the receiving edge features generated by the edge intelligent processing module and generate global fusion features. The dynamic threshold optimization model is used to calculate the warning threshold based on the fusion feature and adjust the output threshold accordingly. The federated learning module updates the global model through homomorphic encryption to aggregate the edge model parameters and feeds it back to the edge node to form a closed-loop optimization system.

[0041] During actual use, the multimodal data acquisition module collects water level, water quality and meteorological data in real time through radar water level gauges, multi-parameter water quality meters and rain gauges, and transmits them to the edge intelligent processing module via RS485 / 5G / Wi-Fi; this module extracts and aligns spatiotemporal features through a multi-scale spatiotemporal convolution unit, and the uncertainty quantification unit integrates Bayesian Dropout technology to calculate the confidence level, and selectively transmits complete data or summary features to the cloud based on the abnormal classification judgment results; the multimodal data fusion module of the cloud platform integrates multi-source features, and the dynamic threshold optimization model generates dynamic thresholds based on the PPO algorithm. The federated learning framework aggregates edge node model parameters through homomorphic encryption to update the global model; finally, the cloud synchronizes the optimized model parameters to the edge node, forming a closed-loop monitoring system that includes data collection, edge intelligent processing, and cloud collaborative optimization, to achieve accurate prediction and intelligent decision-making of hydrological conditions.

[0042] In a specific embodiment, the radar water level meter of the multimodal data acquisition module is connected to the edge intelligent processing module through the RS485 / Modbus interface; the multi-parameter water quality meter is connected to the edge intelligent processing module through the 5G communication interface, and the tipping bucket rain gauge is connected to the edge intelligent processing module through the Wi-Fi communication interface.

[0043] Specifically, when the multi-scale spatiotemporal convolution unit processes data, it first extracts the timestamp information and applies a position encoding function to the timestamp to obtain a temporal feature vector. It also extracts the position information from the original data and processes it using a graph convolutional network combined with an adjacency matrix to generate a spatial feature vector. Finally, the temporal and spatial feature vectors are concatenated along the last dimension to form an aligned feature vector. When the uncertainty quantification unit quantifies the uncertainty of the data, it segments the aligned feature vectors into sample units and applies the Bayesian Dropout technique to each segmented feature vector to obtain a list of feature vectors with uncertainty information. These lists of feature vectors with uncertainty information are then concatenated and input into a weighted network. The weight of each feature vector is calculated based on its degree of uncertainty. Based on the calculated weights, the feature vectors with uncertainty information are weighted summed to obtain a fused feature vector.

[0044] Furthermore, in a preferred embodiment of the present invention, the data fusion unit in the edge intelligent processing module is used to receive feature data from the edge intelligent processing module, the threshold optimization unit is connected to the data fusion unit and outputs a dynamic threshold, and the federated learning unit interacts with the edge intelligent processing module via an interface to update model parameters. The abnormality classification judgment is specifically as follows: the abnormality score of the fused feature vector is calculated and whether the abnormality score is greater than the set threshold; if the abnormality score is greater than the threshold, the complete original data is sent to the cloud platform; if the abnormality score is less than or equal to the threshold, only the processed summary features are sent to the cloud platform.

[0045] In more detail, the multimodal data fusion module in the preferred embodiment of the present invention is connected to the edge intelligent processing module via an interface, is used to receive the fused feature data transmitted by the edge intelligent processing module, and is configured to perform the following operations: using a spatiotemporal attention mechanism to align the timestamps and spatial coordinates of multi-source data, combining a multi-scale convolutional network to extract multi-level spatiotemporal features from the fused features, integrating external data sources, and generating a fused feature vector in a unified format. The dynamic threshold optimization model is trained based on a proximal strategy optimization algorithm, connected to the multimodal data fusion module, receives the fused feature data output by it, and is configured to generate a dynamic threshold in the following manner: based on the formula:

[0046] Threshold = υ + α (b false alarm rate + y false alarm rate); where α, b, and y are adaptive weight coefficients;

[0047] Use reward function R = a·(1-false alarm rate) + b·(1-false alarm rate);

[0048] At the same time, based on the historical false alarm rate, missed alarm rate and current data distribution entropy, the warning threshold is dynamically adjusted and output to the federated learning framework through the interface.

[0049] In more detail, the federated learning framework interacts with the edge intelligent processing module through the gRPC interface, receives the local model parameters uploaded by it, and connects to the dynamic threshold optimization model to provide model parameter update support. It is configured to perform the following operations: using homomorphic encryption technology to aggregate edge node model parameters without exposing the original data; updating the global model based on the federated learning algorithm and synchronizing it to the edge intelligent processing module to improve prediction accuracy; annotating and analyzing abnormal data reported by edge nodes to optimize the model's ability to recognize abnormal patterns.

[0050] On this basis, the present invention also provides a method for automatic monitoring of hydrological data, such as Figure 2 As shown in , the method includes the following steps:

[0051] S1, collects original multimodal data through the multimodal data acquisition module and transmits it to the edge intelligent processing module;

[0052] S2, the edge intelligent processing module's multi-scale spatiotemporal convolution unit processes raw data and generates alignment features; the uncertainty quantification unit calculates confidence scores and transmits the processed data to the cloud platform based on the anomaly classification results;

[0053] S3, the cloud platform's multimodal data fusion module integrates features from multiple edge nodes to generate global fusion features; the dynamic threshold optimization model calculates warning thresholds based on the fusion features; the federated learning framework aggregates edge model parameters and updates the global model;

[0054] S4. The cloud platform synchronizes the optimized model parameters to the edge nodes to form a closed-loop optimization system.

[0055] On this basis, the present invention also provides a storage medium, which stores a computer program. When the computer program is executed by the processor, the steps in the above-mentioned automatic monitoring method are executed: the original multimodal data is collected by the multimodal data acquisition module and transmitted to the edge intelligent processing module; the multi-scale spatiotemporal convolution unit of the edge intelligent processing module processes the original data to generate alignment features; the uncertainty quantification unit calculates the confidence score and transmits the processed data to the cloud platform according to the abnormality classification judgment result; the multimodal data fusion module of the cloud platform integrates the features of multiple edge nodes to generate global fusion features; the dynamic threshold optimization model calculates the warning threshold based on the fusion features; the federated learning framework aggregates the edge model parameters and updates the global model.

[0056] It will be easily understood by those skilled in the art that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. An automatic monitoring system for hydrological data, characterized in that: It includes a multimodal data acquisition module, an edge intelligent processing module communicating with it, and a cloud platform; The multimodal data acquisition module includes a radar water level gauge set in the water area to be measured, which is used to accurately collect water level data; a multi-parameter water quality meter set at different depths in the monitored water area, which is used to comprehensively collect multi-dimensional data related to water quality; and a rain gauge set around the monitored water area, which is used to accurately collect rainfall in meteorological data. This module is connected to the edge intelligent processing module through a data transmission interface. The edge intelligent processing module includes a multi-scale spatiotemporal convolution unit, an uncertainty quantification unit, and a feature transmission unit; the multi-scale spatiotemporal convolution unit is used to receive data from the multimodal data acquisition module and extract and analyze spatiotemporal features of different scales; the uncertainty quantification unit is connected to the multi-scale spatiotemporal convolution unit and the feature transmission unit respectively, and is used to quantify the uncertainty of the data and transmit the processed data to the cloud platform according to the results of the abnormality classification judgment; The cloud-based self-evolving AI platform includes a multimodal data fusion module and a dynamic threshold optimization model that is communicatively connected to the multimodal data fusion module, as well as a federated learning framework; the multimodal data fusion module is communicatively connected to the edge intelligent processing module through an interface, and is used to receive the received edge features generated by the edge intelligent processing module and generate global fusion features. The dynamic threshold optimization model is used to calculate the warning threshold based on the fusion features and adjust the output threshold accordingly. The federated learning module updates the global model by aggregating edge model parameters through homomorphic encryption and feeds it back to the edge nodes to form a closed-loop optimization system.

2. The automatic monitoring system for hydrological data according to claim 1, wherein: The radar water level meter of the multimodal data acquisition module is connected to the edge intelligent processing module via the RS485 / Modbus interface; The multi-parameter water quality meter is connected to the edge intelligent processing module through a 5G communication interface, and the tipping bucket rain gauge is connected to the edge intelligent processing module through a Wi-Fi communication interface.

3. The automatic monitoring system for hydrological data according to claim 1, wherein: When the multi-scale spatiotemporal convolution unit processes data, it first extracts the timestamp information therein, and uses the position encoding function to process the timestamp to obtain a time feature vector; at the same time, it extracts the position information in the original data, and processes the position information through a graph convolution network combined with an adjacency matrix to generate a spatial feature vector; finally, the time feature vector and the spatial feature vector are spliced in the last dimension to form an aligned feature vector.

4. The automatic monitoring system for hydrological data according to claim 1, wherein: When the uncertainty quantification unit quantifies the uncertainty of the data, the aligned feature vector is segmented into units of each sample, and the Bayesian Dropout technique is applied to each segmented feature vector to obtain a feature vector list with uncertainty information; Then, the feature vector list with uncertainty information is concatenated and input into the weight network; Calculate the weight of each eigenvector according to its degree of uncertainty; According to the calculated weights, the feature vectors with uncertainty information are weighted summed to obtain the fused feature vector. The data fusion unit is used to receive feature data of the edge intelligent processing module, the threshold optimization unit is connected to the data fusion unit and outputs a dynamic threshold, and the federated learning unit interacts with the edge intelligent processing module through an interface to update the model parameters.

5. The automatic monitoring system for hydrological data according to claim 4, wherein: The abnormality classification judgment is specifically as follows: calculating the abnormality score of the fused feature vector and judging whether the abnormality score is greater than the set threshold; if the abnormality score is greater than the threshold, the complete original data is sent to the cloud platform; if the abnormality score is less than or equal to the threshold, only the processed summary features are sent to the cloud platform.

6. The automatic monitoring system for hydrological data according to claim 1, wherein: The multimodal data fusion module is in communication with the edge intelligent processing module via an interface, is configured to receive the fusion feature data transmitted by the edge intelligent processing module, and is configured to perform the following operations: The spatiotemporal attention mechanism is adopted to align the timestamps and spatial coordinates of multi-source data. Combined with the multi-scale convolutional network, multi-level spatiotemporal features are extracted from the fused features, external data sources are integrated, and a fused feature vector in a unified format is generated.

7. The automatic monitoring system for hydrological data according to claim 6, wherein: The dynamic threshold optimization model is trained based on the proximal strategy optimization algorithm, connected to the multimodal data fusion module, receives the fused feature data output by the module, and is configured to generate a dynamic threshold in the following manner: Based on the formula Threshold = υ + α (b false alarm rate + y false alarm rate); where α, b, and y are adaptive weight coefficients; Use reward function R = a·(1-false alarm rate) + b·(1-false alarm rate); According to the historical false alarm rate, missed alarm rate and current data distribution entropy, the warning threshold is dynamically adjusted and output to the federated learning framework through the interface.

8. The automatic monitoring system for hydrological data according to claim 7, wherein: The federated learning framework interacts with the edge intelligent processing module through the gRPC interface, receives the local model parameters uploaded by it, and connects to the dynamic threshold optimization model to provide model parameter update support. It is configured to perform the following operations: Homomorphic encryption technology is used to aggregate edge node model parameters without exposing the original data; the global model is updated based on the federated learning algorithm and synchronized to the edge intelligent processing module to improve prediction accuracy; abnormal data reported by edge nodes is labeled and analyzed to optimize the model's ability to recognize abnormal patterns.

9. A method for automatic monitoring of hydrological data, characterized in that: The method is implemented based on the automatic monitoring system according to any one of claims 1 to 8, and comprises the following steps: S1. Collecting original multimodal data through the multimodal data acquisition module and transmitting it to the edge intelligent processing module; S2: The multi-scale spatiotemporal convolution unit of the edge intelligent processing module processes the raw data to generate alignment features; the uncertainty quantification unit calculates the confidence score and transmits the processed data to the cloud platform according to the abnormality classification judgment results; S3, the multimodal data fusion module of the cloud platform integrates the features of multiple edge nodes to generate a global fusion feature; the dynamic threshold optimization model calculates the warning threshold based on the fusion feature; the federated learning framework aggregates the edge model parameters and updates the global model; S4. The cloud platform synchronizes the optimized model parameters to the edge nodes to form a closed-loop optimization system.

10. A storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by the processor, the steps of the automatic monitoring method according to claim 9 are executed.

Images