Novel multi-mode intelligent remote intelligent monitoring system and method
Through the edge AI analysis module and blockchain management, the problems of uneven site distribution, diverse alarm types and untimely processing in the base station monitoring system are solved, and intelligent identification of abnormal behavior of base stations and collaborative fault prediction are realized, which improves the efficiency and accuracy of the monitoring system.
Patent Information
- Application Number
- CN202510886608.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-30
- Publication Date
- 2025-10-28
AI Technical Summary
Traditional base station monitoring systems have problems such as uneven site distribution, diverse alarm types, untimely alarm processing, and lack of coordination between sites, resulting in difficult monitoring coverage, complex processing, and untimely response.
An edge AI analysis module is used for multimodal data acquisition, preprocessing, feature extraction, and anomaly identification. Combined with blockchain data management and auditing, it enables intelligent alarm processing and online incremental learning. A lightweight model is used for local fault prediction and collaborative analysis.
It enables timely identification and processing of abnormal base station behavior, reduces repeated alarms, improves fault location accuracy, and reduces processing complexity and response time.
Smart Images
Figure CN120856535A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of communication technology, specifically to a novel multimodal intelligent remote intelligent monitoring system and method. Background Technology
[0002] Traditional base station monitoring systems mainly rely on a centralized network management platform to centrally process and judge the operating parameters uploaded by each site, which has the following problems:
[0003] Uneven distribution of sites: The large number of base stations, their wide distribution and complex locations make it difficult to achieve monitoring coverage;
[0004] The alarm types are diverse, including power failure, signal abnormality, excessive ambient temperature and humidity, equipment aging or damage, and many other alarm scenarios;
[0005] Untimely alarm handling response: Traditional alarm handling algorithms are relatively simple and fixed. When dealing with complex or concurrent situations, the same person often has to handle multiple work orders. When an order is rejected and then reassigned or recalculated, a lot of time is spent, resulting in untimely alarm handling.
[0006] Lack of inter-site collaboration: Existing systems fail to effectively utilize the correlations between sites for joint analysis and prediction. Network administrators need a network monitoring and maintenance platform that is compatible with equipment from various manufacturers and provides unified monitoring of equipment failures. This would create a timely, accurate, and convenient closed-loop maintenance process encompassing multi-faceted monitoring, reporting, handling, and troubleshooting.
[0007] Therefore, site administrators need a new type of intelligent monitoring method that can effectively identify and analyze multi-source data collected from sites. Summary of the Invention
[0008] The purpose of this invention is to provide a novel multimodal intelligent remote intelligent monitoring system and method to solve the problems mentioned in the background art.
[0009] To achieve the above objectives, the present invention provides the following technical solution: a novel multimodal intelligent remote intelligent monitoring system, comprising an edge AI analysis module, wherein the edge AI analysis module includes:
[0010] The data acquisition layer is used to acquire multimodal raw data from sensor and radio frequency module devices. The data types cover temperature and humidity, voltage / current, GPS location information, RF signal strength, vibration, log events and site external information. Different data types have corresponding source devices, sampling frequencies and preprocessing methods.
[0011] The preprocessing layer cleans, standardizes, and reduces noise in the collected data.
[0012] The feature extraction layer uses CNN, LSTM, and GNN models to extract key features;
[0013] The AI inference layer runs a compressed and optimized neural network model to identify abnormal behavior of the base station.
[0014] The feedback learning layer supports online incremental learning and model updates.
[0015] The communication interface layer interacts with the network management system, blockchain module, and order dispatch module to exchange data and instructions. It weakens the order dispatch system WFM and directly connects with the corresponding operators through the monitoring system.
[0016] The preferred approach for designing and training the AI model for the edge AI analysis module is as follows:
[0017] The model selection and structure design adopt a hybrid model architecture, combining CNN, LSTM and graph neural network GNN. CNN is used to process image data, LSTM is used for time series modeling, GNN is used to model the topological relationship and cooperative behavior between multiple base stations, and the classifier uses a fully connected layer + Softmax to output the anomaly probability.
[0018] Lightweight model processing uses knowledge distillation to reduce model size, applies model pruning, quantization, and 8-bit integer conversion to reduce computing power requirements, and deploys optimized models on TensorRT and ONNXRuntime inference engines;
[0019] Training methods include offline training, which trains the initial model in the cloud using historical data; online incremental learning, which updates model parameters locally without uploading the original data using a federated learning framework; and transfer learning, which quickly adapts the existing model to new sites.
[0020] Preferably, the edge AI analysis module includes key algorithm examples, wherein:
[0021] The algorithm used to predict battery life and temperature rise trends adopts the Sequential model, adds LSTM and Dense layers, and is compiled using the Adam optimizer and mean squared error loss function.
[0022] The algorithm used to model the influence relationship between base stations and improve the accuracy of fault location is based on the PyTorch framework. It constructs a GNNModel class, which contains two GCNConv layers, defines the feature dimensions and defines the propagation logic for the relationship between stations, and finally outputs the probability processed by Softmax.
[0023] Preferably, it includes a blockchain data management and auditing module, wherein the blockchain data management and auditing module includes:
[0024] Node roles include carrier nodes, responsible for uploading alarm logs; maintenance provider nodes, which execute remote control commands; and audit nodes, which verify data integrity and compliance.
[0025] The consensus mechanism adopts Practical Byzantine Fault Tolerance (PBFT), which is suitable for scenarios with limited trust.
[0026] Data storage: Each block contains one or more alarm event summaries and operation logs;
[0027] The data auditing and traceability function records all data changes on the blockchain to form an immutable operation track. Users can query the historical alarms and operation records of specific base stations through API or web interface. It supports exporting complete audit reports for internal review or regulatory reporting, and setting automatic alarm policies to trigger warnings when abnormal access behavior is detected.
[0028] Preferably, the system includes an intelligent alarm handling solution, in which the monitoring system and alarm handling personnel can interact directly. Before an alarm is generated, AI intelligently sorts the personnel based on their abilities, location, department, workload, health status, and other factors, and sends the alarm information and the sorting to the top-ranked personnel. The personnel can decide in advance whether to accept or reject the alarm. When an alarm is actually generated, the latest information is automatically synchronized to the personnel. If there is any related information from a related station or other related information, it is also automatically synchronized.
[0029] A novel method for a multimodal intelligent remote monitoring system includes an edge AI analysis step, specifically:
[0030] A lightweight AI model is deployed in the monitoring system to acquire multimodal raw data from sensors and radio frequency modules through the data acquisition layer. The data types include temperature and humidity, voltage / current, GPS location information, RF signal strength, vibration, log events, and external information of the site. Different data types have corresponding source devices, sampling frequencies, and preprocessing methods.
[0031] The preprocessing layer is used to clean, standardize, and reduce noise in the collected data.
[0032] Key features are extracted using CNN, LSTM, and GNN models through a feature extraction layer;
[0033] The compressed and optimized neural network model is run in the AI inference layer to complete the identification of abnormal behavior of the base station;
[0034] The feedback learning layer supports online incremental learning and model updates;
[0035] Through the communication interface layer, the WFM (Weakened Work Order Manpower Dispatch System) interacts with the network management system, blockchain module, and order dispatch module to exchange data and instructions, enabling direct connection between the monitoring system and the corresponding operators.
[0036] The preferred approach for designing and training AI models in the edge AI analysis step is as follows:
[0037] The model selection and structure design adopt a hybrid model architecture, combining CNN, LSTM and graph neural network GNN. CNN is used to process image data, LSTM is used for time series modeling, GNN is used to model the topological relationship and cooperative behavior between multiple base stations, and the classifier uses a fully connected layer + Softmax to output the anomaly probability.
[0038] Lightweight model processing uses knowledge distillation to reduce model size, applies model pruning, quantization, and 8-bit integer conversion to reduce computing power requirements, and deploys optimized models on TensorRT and ONNXRuntime inference engines;
[0039] Training methods include offline training, which trains the initial model in the cloud using historical data; online incremental learning, which updates model parameters locally without uploading the original data using a federated learning framework; and transfer learning, which quickly adapts the existing model to new sites.
[0040] Preferably, the edge AI analysis step includes key algorithm examples, wherein:
[0041] The algorithm used to predict battery life and temperature rise trends adopts the Sequential model, adds LSTM and Dense layers, and is compiled using the Adam optimizer and mean squared error loss function.
[0042] The algorithm used to model the influence relationship between base stations and improve the accuracy of fault location is based on the PyTorch framework. It constructs a GNNModel class, which contains two GCNConv layers, defines the feature dimensions and defines the propagation logic for the relationship between stations, and finally outputs the probability processed by Softmax.
[0043] Preferably, it includes blockchain data management and auditing steps, specifically:
[0044] Configure node roles, including operator nodes responsible for uploading alarm logs, maintenance provider nodes that execute remote control commands, and audit nodes that verify data integrity and compliance;
[0045] It adopts the Practical Byzantine Fault-Tolerant (PBFT) consensus mechanism, which is suitable for scenarios with limited trust.
[0046] Each block is configured to contain one or more alarm event summaries and operation logs for data storage;
[0047] It enables data auditing and traceability functions, recording all data changes on the blockchain to form an immutable operation track. Users can query the historical alarms and operation records of specific base stations through API or web interface. It supports exporting complete audit reports for internal review or regulatory reporting, and setting automatic alarm policies to trigger warnings when abnormal access behavior is detected.
[0048] Preferably, it includes intelligent alarm processing steps, specifically:
[0049] The monitoring system interacts directly with alarm handlers. Before generating an alarm, AI intelligently sorts the handlers based on factors such as their abilities, location, department, workload, health status, and other factors.
[0050] When an alarm is generated, it is preprocessed first, and the alarm information and its order are sent to the top-ranked handlers. The handlers can decide in advance whether to accept or reject it.
[0051] When an alarm is actually generated, the latest information is automatically synchronized to the person in charge. If any related stations or other information related to the alarm occurs, the information is also automatically synchronized.
[0052] Compared with the prior art, the beneficial effects of the present invention are:
[0053] The novel multimodal intelligent remote intelligent monitoring system and method proposed in this invention addresses the current problems of diverse site alarm scenarios, frequent alarm generation, numerous alarms with recurring root causes, complex alarm processing, and missed alarms or unprocessed alarms. By implementing multi-source data fusion, local intelligent identification, low-bandwidth transmission, site collaborative early warning, and intelligent dispatching, it can effectively solve the problems of excessive and difficult-to-handle recurring alarms, missed alarms, and untimely alarm processing responses that lead to greater impacts during site management. Attached Figure Description
[0054] Figure 1 This is a system block diagram of the present invention. Detailed Implementation
[0055] To make the objectives, technical solutions, and advantages of the present invention clear and complete, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only some, not all, embodiments of the present invention, and are merely illustrative of the embodiments of the present invention. They are not intended to limit the embodiments of the present invention. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0056] For example 1, please refer to Figure 1This invention provides a technical solution: a novel multimodal intelligent remote intelligent monitoring system, comprising:
[0057] 1. Edge AI Analysis Module
[0058] Function Description
[0059] Deploy lightweight AI models in the monitoring system to identify abnormal behavior and predict faults of base stations locally, and continuously improve identification and prediction capabilities through an online incremental learning mechanism.
[0060] Architecture Composition
[0061] Data acquisition layer: Acquires multimodal raw data from sensors and radio frequency modules.
[0062] Preprocessing layer: Cleaning, standardizing, and noise reduction of the collected data.
[0063] Feature extraction layer: Uses models such as CNN, LSTM, and GNN to extract key features.
[0064] AI Inference Layer: Runs a compressed and optimized neural network model to complete anomaly identification.
[0065] Feedback learning layer: Supports online incremental learning and model updates.
[0066] Communication interface layer: interacts with the network management system, blockchain module, and order dispatch module to exchange data and instructions, weakens the order dispatch system (WFM), and directly connects with the corresponding operators through the monitoring system.
[0067] Multimodal data acquisition and preprocessing
[0068]
[0069]
[0070] AI Model Design and Training
[0071] (1) Model selection and structural design
[0072] A hybrid model architecture is adopted, combining CNN, LSTM and graph neural network (GNN) to address the problems of multidimensional heterogeneous data and inter-site correlation.
[0073] CNN: Used for processing image-based data (such as RF heatmaps and vibration spectrum maps);
[0074] LSTM: Used for time series modeling (such as temperature and voltage change trends);
[0075] GNN: Used to model the topological relationships and cooperative behavior among multiple base stations;
[0076] Classifier: Fully connected layer + Softmax outputs anomaly probability.
[0077] (2) Model lightweighting
[0078] Use knowledge distillation to reduce the model size;
[0079] The computational requirements are reduced by applying model pruning, quantization, and 8-bit integer conversion.
[0080] Deploy optimized models on inference engines such as TensorRT and ONNX Runtime.
[0081] (3) Training methods
[0082] Offline training: Training the initial model in the cloud using historical data;
[0083] Online incremental learning: Updating model parameters locally without uploading the original data using a federated learning framework;
[0084] Transfer learning: Quickly adapting existing models to new websites.
[0085] Key Algorithm Examples
[0086] (1) Used to predict indicators such as battery life and temperature rise trend.
[0087]
[0088] Used to model the influence relationships between base stations, improving the accuracy of fault location.
[0089]
[0090] 2. Implementation Scheme of Blockchain Data Management and Audit Module
[0091] Architecture Composition
[0092] Node roles:
[0093] Carrier node: Responsible for uploading alarm logs;
[0094] Maintenance node: Executes remote control commands;
[0095] Audit point: Verify data integrity and compliance.
[0096] Consensus mechanism: Practical Byzantine Fault Tolerance (PBFT), suitable for scenarios with limited trust;
[0097] Data storage: Each block contains one or more alarm event summaries and operation logs.
[0098] Data auditing and traceability
[0099] All data changes are recorded on the blockchain, forming an immutable operation track;
[0100] Users can query the historical alarms and operation records of a specific base station through API or Web interface;
[0101] Supports exporting complete audit reports for internal review or regulatory reporting;
[0102] Configure an automatic alarm policy to trigger an alert when abnormal access behavior is detected.
[0103] 3. Intelligent alarm handling solution
[0104] The monitoring system and alarm handlers can interact directly. All processing occurs before alarms are sent. AI intelligently prioritizes personnel based on factors such as ability, location, department, workload, and health. When an alarm is generated, it first performs preprocessing and sends the alarm information and its ranking to the top-ranked handlers. The handlers can decide in advance whether to accept or reject the alarm. When an alarm is actually generated, the latest information is automatically synchronized to the handler. If related stations or other related information occur, synchronization will also be automatic.
[0105] Example 2, based on Example 1, proposes a novel method for a multimodal intelligent remote intelligent monitoring system, which includes an edge AI analysis step, specifically as follows:
[0106] A lightweight AI model is deployed in the monitoring system. Through the data acquisition layer, multimodal raw data is acquired from sensors and RF modules. Data types include temperature and humidity, voltage / current, GPS location information, RF signal strength, vibration, log events, and external site information. Different data types have corresponding source devices, sampling frequencies, and preprocessing methods. The preprocessing layer cleans, standardizes, and reduces noise in the acquired data. A feature extraction layer uses CNN, LSTM, and GNN models to extract key features. A compressed and optimized neural network model is run in the AI inference layer to identify abnormal base station behavior. An online incremental learning and model update layer supports this. Through the communication interface layer, data and instructions are exchanged with the network management system, blockchain module, and dispatch module, weakening the manual dispatch system (WFM) to achieve direct connection between the monitoring system and corresponding operators.
[0107] The AI model design and training methods in the edge AI analysis steps are as follows: Model selection and structure design adopt a hybrid model architecture, combining CNN, LSTM, and Graph Neural Network (GNN). CNN is used to process image data, LSTM is used for time series modeling, and GNN is used to model the topological relationships and collaborative behaviors between multiple base stations. The classifier uses a fully connected layer + Softmax to output anomaly probabilities. Model lightweighting uses knowledge distillation to reduce the model size, and applies model pruning, quantization, and 8-bit integer conversion to reduce computing power requirements. The optimized model is deployed on TensorRT and ONNXRuntime inference engines. Training methods include offline training, training the initial model in the cloud using historical data; online incremental learning, updating model parameters locally without uploading the original data through a federated learning framework; and transfer learning, quickly adapting the existing model for new sites.
[0108] The edge AI analysis steps include key algorithm examples, including: the algorithm for predicting battery life and temperature rise trends uses a Sequential model with added LSTM and Dense layers, and is compiled using the Adam optimizer and mean squared error loss function; the algorithm for modeling the influence relationship between base stations and improving fault location accuracy is based on the PyTorch framework, constructs a GNNModel class containing two GCNConv layers, defines feature dimensions and propagation logic for the relationship between stations, and finally outputs a probability processed by Softmax.
[0109] The system includes blockchain data management and auditing steps, specifically: setting up node roles, including operator nodes responsible for uploading alarm logs, maintenance node nodes that execute remote control commands, and audit nodes that verify data integrity and compliance; adopting the Practical Byzantine Fault Tolerance (PBFT) consensus mechanism, suitable for limited trust scenarios; setting each block to contain one or more alarm event summaries and operation logs for data storage; implementing data auditing and traceability functions, recording all data changes on the chain to form an immutable operation trajectory; allowing users to query historical alarms and operation records of specific base stations via API or web interface; supporting the export of complete audit reports for internal review or regulatory reporting; and setting up automatic alarm policies to trigger alerts when abnormal access behavior is detected.
[0110] The system includes intelligent alarm handling steps, specifically: the monitoring system interacts directly with alarm handlers. Before generating an alarm, AI intelligently ranks handlers based on factors such as ability, location, department, workload, health, and other factors. When an alarm is generated, preprocessing is performed, sending the alarm information and ranking to the top-ranked handlers, who can decide in advance whether to accept or reject it. When an alarm is actually generated, the latest information is automatically synchronized to the handler, and synchronization is also automatic if related stations or other relevant information occur.
[0111] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.
Claims
1. A novel multimodal intelligent remote intelligent monitoring system, characterized in that: It includes an edge AI analysis module, which comprises: The data acquisition layer is used to acquire multimodal raw data from sensor and radio frequency module devices. The data types cover temperature and humidity, voltage / current, GPS location information, RF signal strength, vibration, log events and site external information. Different data types have corresponding source devices, sampling frequencies and preprocessing methods. The preprocessing layer cleans, standardizes, and reduces noise in the collected data. The feature extraction layer uses CNN, LSTM, and GNN models to extract key features; The AI inference layer runs a compressed and optimized neural network model to identify abnormal behavior of the base station. The feedback learning layer supports online incremental learning and model updates. The communication interface layer interacts with the network management system, blockchain module, and order dispatch module to exchange data and instructions. It weakens the order dispatch system WFM and directly connects with the corresponding operators through the monitoring system.
2. The novel multimodal intelligent remote intelligent monitoring system according to claim 1, characterized in that: The AI model design and training method for the edge AI analysis module are as follows: The model selection and structure design adopt a hybrid model architecture, combining CNN, LSTM and graph neural network GNN. CNN is used to process image data, LSTM is used for time series modeling, GNN is used to model the topological relationship and cooperative behavior between multiple base stations, and the classifier uses a fully connected layer + Softmax to output the anomaly probability. Lightweight model processing uses knowledge distillation to reduce model size, applies model pruning, quantization, and 8-bit integer conversion to reduce computing power requirements, and deploys optimized models on TensorRT and ONNXRuntime inference engines; Training methods include offline training, which uses historical data to train the initial model in the cloud; Online incremental learning updates model parameters locally without uploading the original data through a federated learning framework; Transfer learning enables the rapid adaptation of existing models to new websites.
3. The novel multimodal intelligent remote intelligent monitoring system according to claim 2, characterized in that: The edge AI analysis module includes key algorithm examples, among which: The algorithm used to predict battery life and temperature rise trends adopts the Sequential model, adds LSTM and Dense layers, and is compiled using the Adam optimizer and mean squared error loss function. The algorithm used to model the influence relationship between base stations and improve the accuracy of fault location is based on the PyTorch framework. It constructs a GNNModel class, which contains two GCNConv layers, defines the feature dimensions and defines the propagation logic for the relationship between stations, and finally outputs the probability processed by Softmax.
4. The novel multimodal intelligent remote intelligent monitoring system according to claim 3, characterized in that: It includes a blockchain data management and auditing module, which comprises: Node roles include carrier nodes, responsible for uploading alarm logs; maintenance provider nodes, which execute remote control commands; and audit nodes, which verify data integrity and compliance. The consensus mechanism adopts Practical Byzantine Fault Tolerance (PBFT), which is suitable for scenarios with limited trust. Data storage: Each block contains one or more alarm event summaries and operation logs; The data auditing and traceability function records all data changes on the blockchain to form an immutable operation track. Users can query the historical alarms and operation records of specific base stations through API or web interface. It supports exporting complete audit reports for internal review or regulatory reporting, and setting automatic alarm policies to trigger warnings when abnormal access behavior is detected.
5. A novel multimodal intelligent remote intelligent monitoring system according to claim 4, characterized in that: It includes an intelligent alarm handling solution, enabling direct interaction between the monitoring system and alarm handlers. Before generating an alarm, AI intelligently ranks the handlers based on their abilities, location, department, workload, health, and other factors, and sends the alarm information and ranking to the top-ranked handlers. The handlers can decide in advance whether to accept or reject the alarm. When an alarm is actually generated, the latest information is automatically synchronized to the handler. If there are related stations or other related information, synchronization is also automatic.
6. A method for the novel multimodal intelligent remote intelligent monitoring system according to claim 5, characterized in that: It includes edge AI analysis steps, specifically: A lightweight AI model is deployed in the monitoring system to acquire multimodal raw data from sensors and radio frequency modules through the data acquisition layer. The data types include temperature and humidity, voltage / current, GPS location information, RF signal strength, vibration, log events, and external information of the site. Different data types have corresponding source devices, sampling frequencies, and preprocessing methods. The preprocessing layer is used to clean, standardize, and reduce noise in the collected data. Key features are extracted using CNN, LSTM, and GNN models through a feature extraction layer; The compressed and optimized neural network model is run in the AI inference layer to complete the identification of abnormal behavior of the base station; The feedback learning layer supports online incremental learning and model updates; Through the communication interface layer, the WFM (Weakened Work Order Manpower Dispatch System) interacts with the network management system, blockchain module, and order dispatch module to exchange data and instructions, enabling direct connection between the monitoring system and the corresponding operators.
7. A method according to claim 6, characterized in that: The AI model design and training methods in the edge AI analysis process are as follows: The model selection and structure design adopt a hybrid model architecture, combining CNN, LSTM and graph neural network GNN. CNN is used to process image data, LSTM is used for time series modeling, GNN is used to model the topological relationship and cooperative behavior between multiple base stations, and the classifier uses a fully connected layer + Softmax to output the anomaly probability. Lightweight model processing uses knowledge distillation to reduce model size, applies model pruning, quantization, and 8-bit integer conversion to reduce computing power requirements, and deploys optimized models on TensorRT and ONNXRuntime inference engines; Training methods include offline training, which uses historical data to train the initial model in the cloud; Online incremental learning updates model parameters locally without uploading the original data through a federated learning framework; Transfer learning enables the rapid adaptation of existing models to new websites.
8. A method according to claim 6, characterized in that: The edge AI analytics steps include examples of key algorithms, among which: The algorithm used to predict battery life and temperature rise trends adopts the Sequential model, adds LSTM and Dense layers, and is compiled using the Adam optimizer and mean squared error loss function. The algorithm used to model the influence relationship between base stations and improve the accuracy of fault location is based on the PyTorch framework. It constructs a GNNModel class, which contains two GCNConv layers, defines the feature dimensions and defines the propagation logic for the relationship between stations, and finally outputs the probability processed by Softmax.
9. A method according to claim 6, characterized in that: It includes blockchain data management and auditing steps, specifically: Configure node roles, including operator nodes responsible for uploading alarm logs, maintenance provider nodes that execute remote control commands, and audit nodes that verify data integrity and compliance; It adopts the Practical Byzantine Fault-Tolerant (PBFT) consensus mechanism, which is suitable for scenarios with limited trust. Each block is configured to contain one or more alarm event summaries and operation logs for data storage; It enables data auditing and traceability functions, recording all data changes on the blockchain to form an immutable operation track. Users can query the historical alarms and operation records of specific base stations through API or web interface. It supports exporting complete audit reports for internal review or regulatory reporting, and setting automatic alarm policies to trigger warnings when abnormal access behavior is detected.
10. A method according to claim 6, characterized in that: Includes intelligent alarm processing steps, specifically: The monitoring system interacts directly with alarm handlers. Before generating an alarm, AI intelligently sorts the handlers based on factors such as their abilities, location, department, workload, health status, and other factors. When an alarm is generated, it is preprocessed first, and the alarm information and its order are sent to the top-ranked handlers. The handlers can decide in advance whether to accept or reject it. When an alarm is actually generated, the latest information is automatically synchronized to the person in charge. If any related stations or other information related to the alarm occurs, the information is also automatically synchronized.