Tailing pond monitoring architecture based on multi-modal sensing and cloud edge collaborative deep learning

Abstract

The invention relates to a tailing pond monitoring architecture based on multi-modal sensing and cloud edge collaborative deep learning, and the architecture comprises a multi-modal heterogeneous sensor network which employs a distributed deployment mode; a wireless sensor network communication and data acquisition unit, wherein the wireless sensor network communication and data acquisition unit is connected with the multi-mode heterogeneous sensor network; the multi-source heterogeneous data fusion and processing platform is in communication and data acquisition connection with the wireless sensor network; cloud edge collaborative adaptive deep learning intelligent early warning, wherein the design key point of the cloud edge collaborative adaptive deep learning intelligent early warning lies in collaborative work of an edge side and a cloud side; according to the integral structure provided by the embodiment of the invention, successful deployment and stable operation of the wet tailings pond under the high-altitude complex geological conditions are realized. The monitoring precision of the system reaches an expected target, the early warning accuracy is obviously improved, the system runs stably and reliably, and the safety risk of the tailing pond is effectively reduced.

Description

Technical Field

[0001] The present application relates to the field of tailings pond monitoring technology, and in particular to a tailings pond monitoring architecture based on multimodal sensing and cloud-edge collaborative deep learning. Background Art

[0002] The safe and stable operation of tailings dams is crucial for protecting people's lives and property and the ecological environment. However, tailings dams are typically large in scale, complex in structure, and long in service. They are also subject to the combined influence of multiple factors, including water, force, and chemistry, making them highly susceptible to instability and failure, leading to serious secondary disasters. Therefore, real-time, accurate, and intelligent safety monitoring and early warning for tailings dams are crucial.

[0003] In recent years, both domestic and international research on tailings dam safety monitoring has made progress. However, existing solutions still have numerous limitations. Traditional manual inspections rely on empirical judgment, making it difficult to achieve high-frequency, all-weather, and refined monitoring, and they are unable to obtain quantitative data. Early automated monitoring systems often used a single sensor type, which failed to fully reflect the multi-dimensional safety status of tailings dams and was prone to blind spots and misjudgments. Existing monitoring systems also lack effective multi-source heterogeneous data fusion algorithms, making it difficult to effectively integrate data from different sensor types and explore the correlations and complementarities between the data. This results in low utilization of monitoring data and hinders the formation of a holistic and consistent understanding of the safety status of tailings dams. Furthermore, early warning models lack a high level of intelligence and adaptive capabilities. Existing early warning models often use simple threshold alarms or shallow statistical models, which struggle to effectively handle the complex characteristics of tailings dam monitoring data, such as nonlinearity, time-varying characteristics, and uncertainty. Consequently, their accuracy and generalization capabilities are insufficient, and they are prone to false positives and missed negatives. These models have fixed parameters, making them difficult to adapt to the operating status of the tailings dam and environmental changes, and their performance may gradually decline over time. Furthermore, some existing systems utilize a centralized data processing architecture, which can easily lead to network congestion and overload central nodes. The system's low level of integration makes it difficult to flexibly expand monitoring scope and functionality. Finally, sensor performance degradation in harsh environments also impacts the long-term reliability of the system. Factors such as high humidity, severe corrosion, dust pollution, and electromagnetic interference can all cause traditional sensors to degrade in performance, affecting monitoring accuracy.

[0004] In summary, the existing tailings pond monitoring technology still has significant shortcomings in terms of monitoring comprehensiveness, data fusion depth, warning intelligence level, system architecture flexibility and long-term reliability. An innovative technical solution is urgently needed to solve the above problems. To this end, we proposed a tailings pond monitoring architecture based on multimodal sensing and cloud-edge collaborative deep learning. Summary of the Invention

[0005] This application provides a tailings pond monitoring architecture based on multimodal sensing and cloud-edge collaborative deep learning to solve the above-mentioned problems.

[0006] This application provides a tailings pond monitoring architecture based on multimodal sensing and cloud-edge collaborative deep learning, including:

[0007] A multimodal heterogeneous sensor network, which uses a distributed deployment approach to deploy multiple types of sensors in different areas of the tailings pond to collect multi-dimensional safety monitoring data of the tailings pond;

[0008] Wireless sensor network communication and data acquisition, the wireless sensor network communication and data acquisition is connected to the multimodal heterogeneous sensor network, the wireless sensor network communication and data acquisition is used to collect sensor data and transmit it wirelessly;

[0009] A multi-source heterogeneous data fusion and processing platform is deployed on a cloud server, communicates with the wireless sensor network and is connected to the data acquisition, and is used to receive, store, clean, and fuse monitoring data;

[0010] Cloud-edge collaborative adaptive deep learning intelligent early warning, the design of which focuses on the collaborative work between the edge and cloud sides to achieve rapid response and high-precision early warning;

[0011] A visual intelligent decision-making management platform is connected to the multi-source heterogeneous data fusion and processing platform and the cloud-edge collaborative adaptive deep learning intelligent early warning model to visualize monitoring data and early warning information and provide decision support functions.

[0012] Preferably, the multiple types of sensors include MEMS tilt array sensors, fiber grating (FBG) pressure / temperature / strain multi-parameter sensors, distributed fiber strain / temperature sensors, GNSS high-precision displacement monitoring stations, micro-weather stations, low-frequency vibration sensors, radar level gauges and industrial cameras.

[0013] Preferably, the wireless sensor network communication and data acquisition includes a low-power wide area network communication module, specifically a LoRaWAN or NB-IoT gateway, and also includes an edge computing node, which is deployed at the sensor node or the edge of the network for preprocessing, feature extraction and preliminary fusion of the collected data.

[0014] Preferably, the multi-source heterogeneous data fusion and processing platform includes:

[0015] Time series database, used to efficiently store massive monitoring data;

[0016] Data cleaning module, used for time-space synchronization, outlier removal, and improving data quality;

[0017] The multi-level deep fusion module uses a deep learning model to fuse data from different types of sensors and extract a comprehensive feature vector that reflects the safety status of the tailings pond.

[0018] Preferably, the deep learning model adopts one of LSTM, CNN or Transformer.

[0019] Preferably, the cloud-edge collaborative adaptive deep learning intelligent early warning includes:

[0020] The lightweight edge warning model is deployed on edge computing nodes and uses algorithms with low computational complexity to quickly identify risks and provide preliminary warnings.

[0021] The cloud-based high-precision early warning model is deployed on cloud servers and uses deep learning models to conduct refined risk assessment and early warning;

[0022] Adaptive learning and model update mechanism: the model can dynamically adjust model parameters and optimize model structure according to new monitoring data and actual operation conditions, thereby achieving online model updates and performance improvements.

[0023] Preferably, 7. the visual intelligent decision management platform includes:

[0024] The 3D GIS visualization interface integrates a 3D model of the tailings pond and overlays the model with sensor deployment locations, real-time monitoring data, and early warning information, providing an intuitive representation of the tailings pond's status.

[0025] Multi-dimensional data charts provide a variety of chart types to display the changing trends and analysis results of monitoring data, helping users to gain a deeper understanding of the safety status of the tailings pond;

[0026] Intelligent early warning push, push warning information through sound and light alarm, SMS, email, APP and other methods to ensure that relevant personnel can receive warning information in time;

[0027] Generate risk assessment reports: regularly generate tailings pond safety risk assessment reports to provide a basis for management decisions;

[0028] Remote control and linkage support remote control of on-site equipment, realize emergency disposal and linkage control, and improve emergency response efficiency.

[0029] Preferably, the visual intelligent decision management platform also supports multi-user collaborative operation, and different user roles have different permissions and information access scopes to ensure information security and operational specifications.

[0030] Preferably, the visual intelligent decision-making management platform can be connected to the tailings pond emergency command system to achieve rapid transmission of early warning information and unified scheduling of emergency resources.

[0031] The above technical solution provided by the embodiment of the present application has the following advantages compared with the prior art:

[0032] The overall structure provided by the embodiments of the present application realizes:

[0033] 1. Significantly improved monitoring accuracy and comprehensiveness: Through a multimodal, heterogeneous sensor network, accurate perception of the multi-dimensional safety status of the tailings pond is achieved, overcoming the limitations of traditional single-sensor monitoring, which suffers from limited information dimensions and inaccuracy. Monitoring accuracy has increased by over 30%, and monitoring coverage has reached 100%.

[0034] 2. Significantly improved warning intelligence and adaptive capabilities: Through cloud-edge collaborative adaptive deep learning warning models, the deep features and nonlinear relationships in monitoring data are fully explored, improving the adaptability of the warning model to complex environments and multi-factor coupling. Warning accuracy is increased by more than 25%, and false alarm and missed alarm rates are reduced by more than 60%.

[0035] 3. Significantly enhanced system response speed and reliability: Edge computing reduces data transmission volume, reduces central server load, and improves system response speed. The application of LPWAN wireless communication technology and a self-organizing mesh network of sensor nodes enhances the system's long-term stable operation capabilities in the harsh tailings pond environment, increasing the system's mean time between failures (MTBF) by more than three times.

[0036] 4. Operation and maintenance costs are effectively reduced: By adopting low-power sensors, wireless transmission technology, edge computing and remote management platforms, the system wiring, maintenance and manual inspection costs are reduced, monitoring efficiency is improved, and the overall operation and maintenance costs of the system are reduced by more than 50%.

[0037] 5. Significantly enhanced decision-making support capabilities: The visual intelligent management platform provides intuitive three-dimensional GIS display, rich chart analysis, intelligent early warning information push and risk assessment reports, etc., providing tailings pond safety managers with comprehensive, convenient and efficient decision-making support tools, thereby improving the intelligence and information level of tailings pond safety management. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0039] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0040] Figure 1 It is the overall principle diagram of the present invention;

[0041] Figure 2 This is the interface diagram of the tailings pond monitoring architecture based on multimodal sensing and cloud-edge collaborative deep learning of the present invention. DETAILED DESCRIPTION

[0042] To make the purpose, technical solutions, and advantages of the embodiments of this application more clear, the technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the drawings in the embodiments of this application. Obviously, the described embodiments are part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0043] The various embodiments of the present application may be presented in the form of a range. It should be understood that the description in the form of a range is merely for convenience and brevity and should not be construed as a rigid limitation on the scope of the present application. Therefore, it should be considered that the range description has specifically disclosed all possible sub-ranges and single numerical values within the range. For example, it should be considered that the range description from 1 to 6 has specifically disclosed sub-ranges, such as from 1 to 3, from 1 to 4, from 1 to 5, from 2 to 4, from 2 to 6, from 3 to 6, etc., as well as single numbers within the range, such as 1, 2, 3, 4, 5 and 6, regardless of the range. In addition, whenever a numerical range is indicated in this application, it is intended to include any quoted number (fraction or integer) within the indicated range. Unless otherwise specified, the various raw materials, reagents, instruments and equipment used in this application are all commercially available or can be prepared using existing equipment.

[0044] In this application, unless otherwise specified, the directional words used, such as "upper" and "lower", specifically refer to the directions of the drawings in the accompanying drawings. In addition, in this application, the terms "including", "comprising", etc. mean "including but not limited to". In this application, relational terms such as "first" and "second" are merely used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. In this application, "and / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. Wherein A and B can be singular or plural. In this application, "at least one" means one or more, and "plurality" means two or more. "At least one", "at least one of the following" or similar expressions refer to any combination of these items, including any combination of singular or plural items. For example, "at least one of a, b, or c" or "at least one of a, b and c" can both mean: a, b, c, ab, i.e. a and b, ac, bc or abc, where a, b, c can be single or multiple.

[0045] like Figure 1 and Figure 2 As shown: This embodiment of the application provides a tailings pond monitoring architecture based on multimodal sensing and cloud-edge collaborative deep learning, including:

[0046] A multimodal heterogeneous sensor network, which uses a distributed deployment approach to deploy multiple types of sensors in different areas of the tailings pond to collect multi-dimensional safety monitoring data of the tailings pond;

[0047] Wireless sensor network communication and data acquisition, the wireless sensor network communication and data acquisition is connected to the multimodal heterogeneous sensor network, the wireless sensor network communication and data acquisition is used to collect sensor data and transmit it wirelessly;

[0048] A multi-source heterogeneous data fusion and processing platform is deployed on a cloud server, communicates with the wireless sensor network and is connected to the data acquisition, and is used to receive, store, clean, and fuse monitoring data;

[0049] Cloud-edge collaborative adaptive deep learning intelligent early warning, the design of which focuses on the collaborative work between the edge and cloud sides to achieve rapid response and high-precision early warning;

[0050] A visual intelligent decision-making management platform is connected to the multi-source heterogeneous data fusion and processing platform and the cloud-edge collaborative adaptive deep learning intelligent early warning model to visualize monitoring data and early warning information and provide decision support functions.

[0051] The various types of sensors include MEMS tilt array sensors, fiber Bragg grating (FBG) pressure / temperature / strain multi-parameter sensors, distributed fiber strain / temperature sensors, GNSS high-precision displacement monitoring stations, micro weather stations, low-frequency vibration sensors, radar level gauges and industrial cameras.

[0052] The wireless sensor network communication and data acquisition includes a low-power wide area network communication module, specifically a LoRaWAN or NB-IoT gateway, and also includes an edge computing node. The edge computing node is deployed at the sensor node or the edge of the network for preprocessing, feature extraction and preliminary fusion of the collected data.

[0053] The multi-source heterogeneous data fusion and processing platform includes:

[0054] Time series database, used to efficiently store massive monitoring data;

[0055] Data cleaning module, used for time-space synchronization, outlier removal, and improving data quality;

[0056] The multi-level deep fusion module uses a deep learning model to fuse data from different types of sensors and extract a comprehensive feature vector that reflects the safety status of the tailings pond.

[0057] The deep learning model adopts one of LSTM, CNN or Transformer.

[0058] The cloud-edge collaborative adaptive deep learning intelligent warning includes:

[0059] The lightweight edge warning model is deployed on edge computing nodes and uses algorithms with low computational complexity to quickly identify risks and provide preliminary warnings.

[0060] The cloud-based high-precision early warning model is deployed on cloud servers and uses deep learning models to conduct refined risk assessment and early warning;

[0061] Adaptive learning and model update mechanism: the model can dynamically adjust model parameters and optimize model structure according to new monitoring data and actual operation conditions, thereby achieving online model updates and performance improvements.

[0062] The visual intelligent decision management platform includes:

[0063] The 3D GIS visualization interface integrates a 3D model of the tailings pond and overlays the model with sensor deployment locations, real-time monitoring data, and early warning information, providing an intuitive representation of the tailings pond's status.

[0064] Multi-dimensional data charts provide a variety of chart types to display the changing trends and analysis results of monitoring data, helping users to gain a deeper understanding of the safety status of the tailings pond;

[0065] Intelligent early warning push, push warning information through sound and light alarm, SMS, email, APP and other methods to ensure that relevant personnel can receive warning information in time;

[0066] Generate risk assessment reports: regularly generate tailings pond safety risk assessment reports to provide a basis for management decisions;

[0067] Remote control and linkage support remote control of on-site equipment, realize emergency disposal and linkage control, and improve emergency response efficiency.

[0068] The visual intelligent decision-making management platform also supports multi-user collaborative operations. Different user roles have different permissions and information access scopes to ensure information security and operational standards.

[0069] The visual intelligent decision-making management platform can be connected to the tailings pond emergency command system to achieve rapid transmission of early warning information and unified dispatch of emergency resources. Specific embodiment:

[0071] Selection and deployment of various types of sensors:

[0072] 1. Monitoring dam surface deformation: MEMS tilt array sensor nodes were deployed at key locations, including the dam crest, dam slope, and dam foot, at intervals of 20 meters horizontally along the dam axis and 15 meters vertically perpendicular to the dam axis. Each node contained nine MEMS tilt sensors (model: ADIS16227), forming a 3×3 array, for a total of 300 nodes. Furthermore, high-precision GNSS displacement monitoring stations (using RTK technology) were deployed at five characteristic points on the dam crest. Trimble R10 sensors were used for these monitoring stations.

[0073] For internal dam condition monitoring, 500 fiber Bragg grating (FBG) multi-parameter sensors for pressure, temperature, and strain (model: FBG-PP-100) and eight distributed fiber strain / temperature sensors (model: DTS-800) were embedded in a grid pattern at various depths and locations within the dam (along the cross-section and longitudinal sections of the dam). Distributed fiber sensing utilizes OTDR demodulation, and the LUNA ODiSI-B is used as the main unit.

[0074] For reservoir water level and meteorological environment monitoring: 8 radar level gauges are installed on the reservoir water surface, and 3 micro-meteorological stations are set up around the tailings pond. The micro-meteorological stations include sensors such as wind speed, wind direction, temperature, humidity, and rainfall. The sensor model is Vaisala WXT520.

[0075] For vibration monitoring of the surrounding environment: 30 low-frequency vibration sensors, model PCB 393B04, are deployed in sensitive areas around the tailings pond.

[0076] For safety monitoring: 15 high-definition industrial cameras with a resolution of 4K and night vision function were installed at key locations such as the dam top, dam slope, and flood drainage facilities. The brand is Hikvision.

[0077] All sensor nodes are powered by a combination of solar and wind energy and integrate LoRaWAN communication modules. The LoRaWAN modules use the Semtech SX1276 chip.

[0078] 2. Wireless sensor network communication and data acquisition:

[0079] Sensor nodes wirelessly transmit monitoring data via the LoRaWAN protocol to a LoRaWAN gateway deployed at the tailings pond monitoring center. The gateway model is the Industrial Smart Gateway P68. The LoRaWAN gateway transmits the data to a cloud server via a 4G / 5G network.

[0080] Edge computing nodes are deployed in areas with concentrated sensor nodes. These nodes utilize the NVIDIA Jetson Nano development board, pre-installed with data preprocessing (Kalman filtering and wavelet transform), feature extraction (statistical and frequency domain features), and a lightweight early warning model (random forest algorithm), enabling edge data processing and preliminary early warning. The edge computing nodes run the Ubuntu 18.04 operating system and utilize Python 3.6 as the development language. Related algorithm libraries include Scikit-learn, NumPy, and SciPy.

[0081] 3. Multi-source heterogeneous data fusion and processing platform:

[0082] The IoTDB time series database is used to store massive monitoring data. The server configuration is 8-core CPU, 32GB memory, and 1TB SSD hard drive.

[0083] A data fusion and processing module based on Python language was developed, and a spatiotemporal synchronization algorithm was used to correct data timestamps and spatial coordinates. An improved Kalman filter algorithm was used for data cleaning, and wavelet packet decomposition, empirical mode decomposition (EMD) and other methods were used for feature extraction.

[0084] A deep learning-based data fusion model was constructed, using the Transformer model to fuse multimodal sensor data and extract a comprehensive feature vector reflecting the safety status of the tailings pond. The Transformer model was implemented using the PyTorch framework and consists of a six-layer encoder and a six-layer decoder.

[0085] 4. Cloud-edge collaborative adaptive deep learning intelligent early warning model:

[0086] a) Lightweight Edge Warning Model: A lightweight edge warning model is built using the random forest algorithm for rapid risk identification. The random forest model consists of 100 decision trees.

[0087] b) Cloud-based high-precision early warning model: We use the LSTM-Attention model to build a high-precision cloud-based early warning model. The LSTM network processes time series data, and the Attention mechanism enhances the model's focus on key features, enabling refined risk assessment and early warning. The model structure is as follows:

[0088] ① Input layer: receives the fused multi-source feature vector.

[0089] ②LSTM layer: 2 layers, 128 neurons per layer.

[0090] ③Attention layer: uses Bahdanau Attention mechanism.

[0091] ④Output layer: fully connected layer, outputs warning level.

[0092] c) Model training: Historical monitoring data, tailings pond engineering geological data, meteorological data, numerical simulation results, dam break case data, etc. are used for model training. The cross entropy loss function is used as the loss function, and the Adam optimizer is used.

[0093] d) Model adaptive update: Using online learning and incremental learning mechanisms, model parameters and structure are regularly updated based on new monitoring data and model performance feedback.

[0094] e) Warning level: Set four warning levels: normal state (green), concern state (blue), warning state (yellow), emergency state (red), and formulate corresponding emergency response measures.

[0095] 5. Visual intelligent management and decision support platform:

[0096] a) Use CesiumJS 3D GIS engine to build a 3D visualization platform for the tailings pond, integrate the BIM model, and realize the refined 3D display of the tailings pond.

[0097] b) The platform interface provides functional modules such as real-time data monitoring, historical data query, trend analysis, early warning information management, risk assessment report generation, and remote control.

[0098] c) The platform supports multi-user collaborative operations, with different user roles having different permissions and information access scopes. User permission management adopts the RBAC model.

[0099] d) The platform is connected to the tailings pond emergency command system to achieve rapid transmission of early warning information and unified dispatch of emergency resources, and uses Modbus TCP protocol for data exchange.

[0100] Through the above-mentioned specific implementation, the tailings pond intelligent monitoring and early warning system proposed by the present invention was successfully deployed and stably operated in a high-altitude wet tailings pond with complex geological conditions. Field testing demonstrated that the system's monitoring accuracy met expectations, with significantly improved early warning accuracy. The system operated stably and reliably, effectively reducing the safety risks of the tailings pond and providing strong technical support for safe production and emergency management of the tailings pond.

[0101] The interface of the tailings pond monitoring architecture based on multimodal sensing and cloud-edge collaborative deep learning is as follows: Figure 2 shown.

[0102] The foregoing is merely a list of specific embodiments of the present application, intended to enable those skilled in the art to understand or implement the present application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application is not limited to the embodiments shown herein, but rather is intended to conform to the broadest scope consistent with the principles and novel features claimed herein.

Claims

1. Tailings pond monitoring architecture based on multimodal sensing and cloud-edge collaborative deep learning, characterized by: include: A multimodal heterogeneous sensor network, which uses a distributed deployment approach to deploy multiple types of sensors in different areas of the tailings pond to collect multi-dimensional safety monitoring data of the tailings pond; Wireless sensor network communication and data acquisition, the wireless sensor network communication and data acquisition is connected to the multimodal heterogeneous sensor network, the wireless sensor network communication and data acquisition is used to collect sensor data and transmit it wirelessly; A multi-source heterogeneous data fusion and processing platform is deployed on a cloud server, communicates with the wireless sensor network and is connected to the data acquisition, and is used to receive, store, clean, and fuse monitoring data; Cloud-edge collaborative adaptive deep learning intelligent early warning, the design of which focuses on the collaborative work between the edge and cloud sides to achieve rapid response and high-precision early warning; A visual intelligent decision-making management platform is connected to the multi-source heterogeneous data fusion and processing platform and the cloud-edge collaborative adaptive deep learning intelligent early warning model to visualize monitoring data and early warning information and provide decision support functions.

2. The tailings pond monitoring architecture based on multimodal sensing and cloud-edge collaborative deep learning according to claim 1 is characterized by: The various types of sensors include MEMS tilt array sensors, fiber Bragg grating (FBG) pressure / temperature / strain multi-parameter sensors, distributed fiber strain / temperature sensors, GNSS high-precision displacement monitoring stations, micro weather stations, low-frequency vibration sensors, radar level gauges and industrial cameras.

3. The tailings pond monitoring architecture based on multimodal sensing and cloud-edge collaborative deep learning according to claim 1 is characterized by: The wireless sensor network communication and data acquisition includes a low-power wide area network communication module, specifically a LoRaWAN or NB-IoT gateway, and also includes an edge computing node. The edge computing node is deployed at the sensor node or the edge of the network for preprocessing, feature extraction and preliminary fusion of the collected data.

4. The tailings pond monitoring architecture based on multimodal sensing and cloud-edge collaborative deep learning according to claim 1 is characterized by: The multi-source heterogeneous data fusion and processing platform includes: Time series database, used to efficiently store massive monitoring data; Data cleaning module, used for time-space synchronization, outlier removal, and improving data quality; The multi-level deep fusion module uses a deep learning model to fuse data from different types of sensors and extract a comprehensive feature vector that reflects the safety status of the tailings pond.

5. The tailings pond monitoring architecture based on multimodal sensing and cloud-edge collaborative deep learning according to claim 4 is characterized by: The deep learning model adopts one of LSTM, CNN or Transformer.

6. The tailings pond monitoring architecture based on multimodal sensing and cloud-edge collaborative deep learning according to claim 1 is characterized by: The cloud-edge collaborative adaptive deep learning intelligent warning includes: The lightweight edge warning model is deployed on edge computing nodes and uses algorithms with low computational complexity to quickly identify risks and provide preliminary warnings. The cloud-based high-precision early warning model is deployed on cloud servers and uses deep learning models to conduct refined risk assessment and early warning; Adaptive learning and model update mechanism: the model can dynamically adjust model parameters and optimize model structure according to new monitoring data and actual operation conditions, thereby achieving online model updates and performance improvements.

7. The tailings pond monitoring architecture based on multimodal sensing and cloud-edge collaborative deep learning according to claim 1 is characterized by: The visual intelligent decision management platform includes: The 3D GIS visualization interface integrates a 3D model of the tailings pond and overlays the model with sensor deployment locations, real-time monitoring data, and early warning information, providing an intuitive representation of the tailings pond's status. Multi-dimensional data charts provide a variety of chart types to display the changing trends and analysis results of monitoring data, helping users to gain a deeper understanding of the safety status of the tailings pond; Intelligent early warning push, push warning information through sound and light alarm, SMS, email, APP and other methods to ensure that relevant personnel can receive warning information in time; Generate risk assessment reports: regularly generate tailings pond safety risk assessment reports to provide a basis for management decisions; Remote control and linkage support remote control of on-site equipment, realize emergency disposal and linkage control, and improve emergency response efficiency.

8. The tailings pond monitoring architecture based on multimodal sensing and cloud-edge collaborative deep learning according to claim 7 is characterized by: The visual intelligent decision-making management platform also supports multi-user collaborative operations. Different user roles have different permissions and information access scopes to ensure information security and operational standards.

9. The tailings pond monitoring architecture based on multimodal sensing and cloud-edge collaborative deep learning according to claim 8 is characterized by: The visual intelligent decision-making management platform can be connected to the tailings pond emergency command system to achieve rapid transmission of early warning information and unified dispatch of emergency resources.

Images