Intelligent monitoring method and system for new energy intelligent station

Abstract

The invention discloses an intelligent monitoring method and system for a new energy intelligent station, and relates to the technical field of new energy station monitoring. The method comprises the following steps: an intelligent edge gateway of an edge node layer acquires multi-modal original data of heterogeneous equipment through a multi-protocol adaptation module; after the edge computing power module executes data cleaning, a real-time AI model is called according to the data type to generate an analysis result through reasoning; the key data is uploaded to a cloud platform layer through a key data channel; and if the result is a fault alarm, the cloud global task scheduling center is triggered to redistribute a calculation task, and the security policy center generates a reinforcement policy and issues a task instruction and a security policy to the edge node layer through the key channel. The technical problems that heterogeneous devices in the new energy station are difficult to access uniformly, data processing delay is large, and fault response is not appropriate are solved, and the technical effect that the stability and maintainability of the whole new energy intelligent station are improved through cooperative work of the intelligent edge gateway and the cloud platform is achieved.

Description

Technical Field

[0001] This invention relates to the field of monitoring technology for new energy power stations, and specifically to intelligent monitoring methods and systems for smart new energy power stations. Background Technology

[0002] With the rapid development of the new energy industry, the scale of new energy power plants such as wind power, photovoltaic, and electrochemical energy storage is constantly expanding, the number of equipment is growing exponentially, and the operating environment is becoming more complex, placing higher demands on the real-time performance and intelligence level of monitoring systems. Traditional monitoring methods for new energy power plants mainly rely on data acquisition methods using single protocols such as PLC and SCADA. Significant heterogeneity exists between different manufacturers, different protocols, and different types of equipment, making it difficult to unify data access and analysis. Furthermore, the large amount of multimodal data generated by equipment (such as vibration signals, temperature data, and image data) often needs to be transmitted to the cloud for centralized analysis, resulting in high transmission bandwidth pressure, high response latency, and delayed fault handling.

[0003] Meanwhile, renewable energy power plants are typically deployed in remote areas with limited communication networks and long cloud model update cycles, making it difficult to adapt to changes in on-site operating conditions in a timely manner. When a fault occurs, traditional systems often cannot achieve rapid judgment and coordinated handling at the edge, resulting in slow equipment response, high maintenance costs, and potentially affecting the stable operation of the entire power plant. Summary of the Invention

[0004] This application provides an intelligent monitoring method and system for smart new energy power stations, which solves the technical problems of difficulty in unified access of heterogeneous equipment, large data processing delay, and insufficient fault response in new energy power stations.

[0005] The first aspect of this application provides an intelligent monitoring method for smart power stations in the new energy sector, the method comprising: The first intelligent edge gateway device at the edge node layer collects the first multimodal raw data from the first heterogeneous device through the first multi-protocol adaptation module; the first edge computing power module receives and performs edge-side data cleaning on the first multimodal raw data to obtain the first multimodal cleaned data, and then calls the real-time AI analysis model to perform edge inference according to the data type of the first multimodal cleaned data to generate the first analysis result; the first analysis result is uploaded to the cloud platform layer through the key data channel of the communication network layer; when the first analysis result is a fault alarm, the cloud platform layer is triggered to: S1: reallocate the computing tasks of the edge node layer through the global task scheduling center and generate a task allocation instruction; S2: generate a security hardening policy through the security policy center; S3: send the task allocation instruction and security policy to the edge node layer through the key data channel.

[0006] A second aspect of this application provides an intelligent monitoring system for smart power stations in the new energy sector, the system comprising: The cloud platform layer is deployed in a remote data center; the edge node layer is deployed in a new energy power station; and the communication network layer connects the edge node layer and the cloud platform layer to achieve layered data transmission.

[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages: First, the intelligent edge gateway at the edge node layer collects multimodal raw data from heterogeneous devices through a multi-protocol adaptation module. Then, the edge computing module cleans the collected data at the edge and, based on the type of the cleaned data, calls the corresponding real-time AI analysis model for inference, generating analysis results. Key content from the analysis results is uploaded to the cloud platform through a critical data channel in the communication network layer. When the analysis result received by the cloud platform is a fault alarm, subsequent processing is automatically triggered: the global task scheduling center reallocates computing tasks to the edge nodes to generate task instructions; the security policy center generates corresponding security hardening policies; and finally, the task instructions and security policies are sent back to the edge nodes through the critical channel, achieving rapid response and security protection under edge-cloud collaboration. Attached Figure Description

[0008] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0009] Figure 1 This is a schematic diagram of the intelligent monitoring method for smart power stations for new energy, provided in an embodiment of this application.

[0010] Figure 2 This is a schematic diagram of the intelligent monitoring system structure for smart power stations for new energy, provided in an embodiment of this application.

[0011] Figure labeling: Cloud platform layer 11, edge node layer 12, communication network layer 13. Detailed Implementation

[0012] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0013] Example 1, as Figure 1As shown, this application provides an intelligent monitoring method for smart power stations in the new energy sector, the method including: The first intelligent edge gateway device at the edge node layer collects the first multimodal raw data from the first heterogeneous device through the first multi-protocol adaptation module.

[0014] In this embodiment, the first intelligent edge gateway device deployed at the edge node layer is used to uniformly collect data from various heterogeneous devices within the new energy power station. Since the devices in the power station come from different manufacturers and use different communication protocols (such as HTTP, Modbus, etc.) and data formats, the intelligent edge gateway integrates a first multi-protocol adaptation module to achieve automatic identification, parsing, and compatibility with multiple industrial protocols. Through this multi-protocol adaptation module, the gateway can simultaneously access multiple types of devices such as vibration sensors, temperature sensors, cameras, inverters, and energy storage battery management units, and collect corresponding multimodal raw data from these devices in real time, including but not limited to vibration waveforms, temperature values, voltage and current data, operation logs, device images, or video frames. The multi-protocol adaptation module performs data format conversion, data field mapping, and verification according to the device protocol type, ensuring that data output from different devices can be uniformly received and parsed. Finally, the intelligent edge gateway packages the protocol-adapted raw multimodal data of each device into a standardized data structure, which serves as the first multimodal raw data of the first heterogeneous device, and transmits it to the subsequent edge computing module, providing reliable data input for edge-side AI analysis and inference.

[0015] The first edge computing module receives and performs edge-side data cleaning on the first multimodal raw data to obtain the first multimodal cleaned data. Then, based on the data type of the first multimodal cleaned data, it calls the real-time AI analysis model to perform edge inference and generate the first analysis result.

[0016] In one embodiment, the first edge computing module is used to perform local intelligent processing on the first multimodal raw data from the multi-protocol adaptation module. This module first performs edge-side data cleaning operations on the received raw data, including outlier removal, missing data imputation, noise filtering, timestamp alignment, and data format standardization, to ensure the quality and consistency of the input data. Outlier removal can be performed by detecting outliers based on threshold rules or the three-standard-deviation method based on statistics; missing data imputation can be performed by linear interpolation, moving average, or historical data from similar devices; noise filtering can be performed by signal processing methods such as low-pass filtering, Kalman filtering, or wavelet denoising; timestamp alignment can be performed by using a unified clock source and time window resampling; and data format standardization can be performed by uniformly encoding and formatting fields according to a preset data structure template. The cleaned first multimodal data will be classified according to its data type, such as vibration signals, temperature data, electrical operating parameters, and image data. Subsequently, multiple built-in lightweight real-time AI analysis models are activated, automatically selecting the corresponding analysis model for inference based on the current data type. For example, vibration data calls the fault prediction model, and temperature data calls the overheat warning model. Each real-time AI analysis model runs locally on the edge device in inference mode, generating analysis results in milliseconds or seconds, thus outputting the first analysis result. This first analysis result can include equipment status assessment, fault prediction value, risk level, image recognition results, or alarm information, providing a basis for subsequent uploading and cloud processing.

[0017] Furthermore, the method of invoking a real-time AI analysis model to perform edge inference based on the data type of the first multimodal cleaned data to generate a first analysis result includes: If the first multimodal cleaning data is a vibration signal, then the fault prediction model is invoked to predict and output the first analysis result; if the first multimodal cleaning data is temperature data, then the overheating early warning model is invoked to provide an overheating early warning and output the first analysis result; if the first multimodal cleaning data is image data, then the equipment status recognition model is used to identify and output the first analysis result.

[0018] Preferably, after the first edge computing module identifies the type of the first multimodal cleaning data, it will trigger the corresponding AI analysis model to perform edge-side inference based on different data types. If the first multimodal cleaning data is a vibration signal, a preset fault prediction model will be automatically invoked. This fault prediction model analyzes the vibration sequence based on feature extraction (such as time-domain features, frequency-domain features, wavelet packet energy features, etc.) and deep learning prediction networks (such as LSTM, 1D-CNN, or temporal residual networks) to determine the possible abnormal modes or early fault trends of the equipment, and outputs the corresponding prediction results and risk levels as the first analysis result. If the first multimodal cleaning data is temperature data, an overheat warning model will be invoked. This overheat warning model performs trend analysis on the temperature change curve and makes a comprehensive judgment based on the equipment operating threshold, rate change characteristics, and scene rules. When the temperature is detected to exceed the preset safety range or show a continuous upward abnormal trend, the model will give an overheat risk warning and output the corresponding warning level and suggested handling measures as the first analysis result. If the first multimodal cleaning data is image data, the image is processed by an equipment status recognition model. This model, based on a lightweight convolutional neural network or object detection network, identifies key equipment components in the image, including conditions such as insulation damage, component loosening, abnormal liquid level, and foreign object obstruction. The identified equipment status information is further parsed into structured recognition results and output as the first analysis result. Through the above classification and reasoning mechanism, it can be ensured that edge nodes can quickly and accurately perform local intelligent analysis for different types of data, achieving real-time fault warning and status assessment capabilities.

[0019] Furthermore, the first multimodal cleaning data is compressed to 1 / K of the original data volume using model compression technology, and then loaded into the first edge computing module for edge inference.

[0020] Preferably, to reduce the storage and computational overhead of edge nodes and improve real-time inference efficiency, the first multimodal cleaning data will be reduced in size using model compression technology before entering the edge inference stage. Specifically, the system adopts corresponding compression strategies according to different data types. For example, for vibration signals, eigenvector compression, principal component analysis (PCA) dimensionality reduction, or wavelet coefficient truncation are used; for image data, lightweight image encoding, regional feature extraction, or deep feature vector compression are used; and for time-series data such as temperature, time window sampling, key point extraction, or trend line fitting are used. Through these techniques, the original cleaning data can be compressed to 1 / K (K≥2) of the original data volume while preserving as much key information related to AI inference as possible, significantly reducing the data volume. The compressed data is loaded into the first edge computing module in a unified format, which calls the corresponding lightweight AI analysis model to perform edge-side inference. Due to the significant reduction in the input data size, the inference speed of the model is improved, while reducing the bandwidth and computing power consumption of edge devices, thereby ensuring that the edge inference process is more efficient and stable and meets the real-time monitoring needs of new energy power plants.

[0021] The first analysis result is uploaded to the cloud platform layer through the key data channel of the communication network layer.

[0022] In one embodiment, to ensure that critical analytical data can be received by the cloud platform in a timely manner and used in subsequent decision-making, the system sets up an independent critical data channel in the communication network layer. This critical data channel features high reliability, high priority, and low latency transmission characteristics, and employs a dedicated link quality monitoring and dynamic bandwidth allocation mechanism to guarantee the real-time transmission of critical data. After the first edge computing module generates the first analytical result, the intelligent edge gateway at the edge node layer encapsulates the analytical result, including adding metadata such as data type identifiers, timestamps, device numbers, and verification information. Subsequently, the encapsulated analytical result is sent to the critical data channel via an encrypted transmission protocol, such as TLS or a dedicated industrial encryption channel. This critical data channel prioritizes and forwards such data packets, bypassing queuing for non-critical business processes, ensuring that the cloud platform receives the analytical result with the shortest possible latency. Upon receiving the result, the cloud platform layer triggers subsequent processing flows such as fault assessment, task scheduling, or security policy generation, realizing a rapid response mechanism under edge-cloud collaboration.

[0023] Furthermore, the first multimodal cleaning data is asynchronously uploaded to the cloud platform layer through the non-critical data channel of the communication network layer.

[0024] Preferably, to reduce the load on critical business links and facilitate long-term trend analysis and model training on the cloud platform, the system sets up an independent non-critical data channel in the communication network layer for transmitting data that does not require real-time processing. After the first edge computing module completes the cleaning of the multimodal raw data, it generates structured first multimodal cleaned data. The intelligent edge gateway at the edge node layer, after confirming that this data is not critical data requiring immediate processing, places it in the data buffer queue of the non-critical channel, and the scheduler uploads it asynchronously according to network status and bandwidth availability. During the upload process, the system can employ data batch packaging, compression encoding, and recoverable transmission mechanisms to further reduce communication load and improve transmission stability. The transmission priority of the non-critical data channel is lower than that of the critical channel, and it will not occupy transmission resources for important data such as fault alarms. After receiving this cleaned data, the cloud platform layer can use it for background processing tasks such as historical archive construction, AI model training, device health assessment, or strategy optimization, thereby achieving comprehensive management and in-depth utilization of edge-side data.

[0025] When the first analysis result is a fault alarm, the cloud platform layer is triggered as follows: S1: The computing tasks of the edge node layer are reallocated through the global task scheduling center, and a task allocation instruction is generated; S2: A security hardening policy is generated through the security policy center; S3: The task allocation instruction and the security policy are sent to the edge node layer through the critical data channel.

[0026] In one embodiment, when the first analysis result uploaded by the first edge computing module is determined to be a fault alarm by the cloud platform layer, the cloud platform layer will immediately initiate a preset emergency handling process to achieve rapid response and coordinated control of the edge nodes. Specifically, firstly, the global task scheduling center of the cloud platform layer will obtain the current computing resource usage, model running load, and task priority of multiple edge nodes in real time, and re-plan the allocation scheme of computing tasks according to the status of the edge node where the faulty device is located. For example, it may adjust the inference task load of the node, schedule other nodes for collaborative processing, or migrate some high-load models to nodes with more computing resources. After completing the optimization, the task scheduling center generates clear task allocation instructions. Subsequently, the security policy center of the cloud platform layer will dynamically generate a security hardening policy for the node from the policy library based on the fault type, risk level, and current security status of the edge node. For example, it may restrict the data upload permissions of the abnormal device, enable enhanced intrusion detection rules, strengthen firmware integrity verification, or enable security isolation mode. This security hardening policy aims to ensure the operational security and data trustworthiness of the edge node during the fault occurrence. After generating task allocation instructions and security hardening policies, the cloud platform synchronously sends both to the target edge node layer through the critical data channel of the communication network layer. This critical data channel features high priority and low latency, ensuring that instructions and policies reach the edge gateway device in the shortest possible time. Upon receiving the instructions, the edge node adjusts its local computing process according to the task allocation and immediately executes the security hardening measures issued by the security policy center, thereby achieving rapid response and security protection through cloud-edge collaboration.

[0027] Furthermore, when the first analysis result is a fault alarm, the AI ​​model training center of the cloud platform layer is triggered to update the lightweight model parameters, generate an updated AI model, and then send it to the edge node layer.

[0028] Preferably, when the cloud platform layer receives the first analysis result from the edge node and confirms it as a fault alarm, in addition to triggering task scheduling and security hardening measures, it will also activate the AI ​​model training center in the cloud to improve the accuracy and adaptability of subsequent edge inference. Specifically, the AI ​​model training center first retrieves the multimodal cleaned data, historical operating data, and past state data of similar devices corresponding to this fault alarm, and performs feature extraction, sample labeling, and data augmentation on this data to ensure that the dataset used for training has sufficient representativeness and quality. Subsequently, the model training center performs incremental training or fine-tuning operations on the lightweight model used for edge inference based on a pre-built algorithm framework, such as lightweight CNN, MobileNet, Tiny-YOLO, LSTM, or Transformer. During the training process, techniques such as transfer learning, distillation learning, or parameter pruning are used to ensure that the updated model maintains high accuracy while its structural scale is still adapted to the computing power of the edge device. After training is completed, the system will generate an updated lightweight AI model and perform format conversion, quantization processing, and integrity verification on the model to ensure that the model can run stably on various types of intelligent edge gateway devices. After the model update is completed, the cloud platform distributes the updated AI model to the corresponding edge node layer through key data channels. Upon receiving the model file, the edge node automatically performs model version comparison, validity verification, and model replacement, and then reloads the model locally to execute the new inference task. Through this mechanism, dynamic optimization of the edge-side AI model is achieved in the cloud, enabling edge inference capabilities to continuously adapt to changes in device status and the evolution of fault modes, thereby further improving the accuracy and intelligence level of the overall monitoring system.

[0029] In summary, the embodiments of this application have at least the following technical effects: First, the first intelligent edge gateway device at the edge node layer collects the first multimodal raw data from the first heterogeneous device through the first multi-protocol adaptation module. Then, the first edge computing module receives and performs edge-side data cleaning on the first multimodal raw data to obtain the first multimodal cleaned data. Based on the data type of the first multimodal cleaned data, it calls a real-time AI analysis model to perform edge inference and generate a first analysis result. Next, the first analysis result is uploaded to the cloud platform layer through the key data channel of the communication network layer. Then, when the first analysis result is a fault alarm, the cloud platform layer is triggered. Then, the computing tasks of the edge node layer are reallocated through the global task scheduling center, generating task allocation instructions. Finally, a security hardening strategy is generated through the security policy center, and the task allocation instructions and security strategy are distributed to the edge node layer through the key data channel. This solves the technical problems of difficult unified access for heterogeneous devices, large data processing latency, and insufficient fault response in new energy power stations, achieving the technical effect of improving the stability and maintainability of the entire new energy smart power station through the collaborative work of the intelligent edge gateway and the cloud platform.

[0030] Example 2, based on the same inventive concept as the intelligent monitoring method for smart power stations in the foregoing examples, such as... Figure 2 As shown, this application provides an intelligent monitoring system for smart power stations in the new energy sector. The system includes: The cloud platform layer is deployed in a remote data center; the edge node layer is deployed in a new energy power station; and the communication network layer connects the edge node layer and the cloud platform layer to achieve layered data transmission.

[0031] In one embodiment, the entire intelligent monitoring system adopts a three-layer architecture: a cloud platform layer, an edge node layer, and a communication network layer, to achieve coordinated operation of distributed computing and centralized management. The cloud platform layer is deployed in a remote data center, relying on a cloud server cluster to provide large-scale computing power, model training capabilities, global scheduling capabilities, and unified security management capabilities. As the central control and analysis core of the system, the cloud platform is responsible for receiving key data from each edge node, performing fault diagnosis, issuing computing tasks and security policies, and maintaining overall system stability. The edge node layer is deployed inside the actual new energy power station, typically near wind turbines, photovoltaic blocks, energy storage units, or station control systems. Each edge node consists of an intelligent edge gateway device, possessing functions such as on-site data acquisition, edge-side data cleaning, local AI inference, and security protection. Edge nodes can load models and switch tasks according to cloud instructions, and process multimodal data locally in real time, thereby ensuring rapid analysis and timely response to equipment status. The communication network layer connects the cloud platform layer and the edge node layer, providing a stable and secure data transmission channel between them. This layer can consist of dedicated communication lines, wireless networks, industrial Ethernet, or other data links, and is divided into critical data channels and non-critical data channels according to service type, to carry data streams of different priorities, such as alarm information, inference results, cleaned data, and background training data. Through the layered transmission mechanism of the communication network layer, the cloud platform can receive analysis results from edge nodes in real time, and simultaneously quickly distribute updated models, scheduling instructions, and security policies to edge nodes, achieving an efficient edge-cloud collaborative working mode.

[0032] Furthermore, the cloud platform layer includes: A global task scheduling center is used to allocate computing tasks to the multiple intelligent edge gateway devices in the edge node layer through the communication network layer; an AI model training center is used to generate AI models and distribute them to the edge node layer through the communication network layer; and a security policy center is used to generate protection policies and synchronize them to the edge node layer through the communication network layer.

[0033] Preferably, the cloud platform layer integrates a global task scheduling center, an AI model training center, and a security policy center to achieve unified management and intelligent collaboration of multiple edge nodes. The global task scheduling center is responsible for formulating the optimal computing task allocation scheme based on the real-time operating status, computing resource usage, model inference load, and device health information of multiple intelligent edge gateway devices. When an edge node experiences a fault alarm or resource shortage, the scheduling center will re-plan the task distribution based on factors such as task priority, available computing power of the node, and network latency. This involves migrating some inference computing tasks to other edge nodes with more sufficient computing power or switching tasks. The generated task allocation instructions are sent to the corresponding edge nodes in real time through the communication network layer, enabling them to execute monitoring and inference tasks according to the latest allocation scheme. The AI ​​model training center is used to build and update lightweight AI models suitable for edge inference. The training center collects historical cleaned data, alarm data, and device operation logs from multiple edge nodes. After preprocessing, this data is used for model training, parameter fine-tuning, or incremental learning. After model training is completed, the system quantizes, compresses, and converts the model's format to adapt to the computing power limitations of edge devices. Finally, the training center distributes the updated AI model to the target edge nodes via the communication network layer, enabling them to load and use the latest model for inference tasks, thus improving prediction accuracy and robustness. The security policy center generates dynamic security protection policies for edge nodes. This center analyzes fault types, risk levels, and the current security status of edge nodes, and based on this, generates corresponding security hardening measures using a policy library. These measures include adjusting communication encryption levels, restricting access permissions, isolating abnormal data sources, and enhancing intrusion detection rules. The generated protection policies are synchronously distributed to edge nodes via the communication network layer, and the edge nodes execute them immediately upon receipt, thereby improving the security and stability of the site monitoring system under fault conditions. Through the coordinated operation of these three centers, comprehensive cloud-based management of edge nodes in terms of task allocation, intelligent inference, and security protection is achieved, constructing an efficient, intelligent, and evolvable smart new energy site monitoring system.

[0034] Furthermore, the edge node layer includes: Multiple intelligent edge gateway devices are provided, each integrating a multi-protocol adaptation module, an edge computing power module, and a local security protection module. These multiple intelligent edge gateway devices connect to multiple heterogeneous devices in the new energy power station through multiple multi-protocol adaptation modules.

[0035] Preferably, the edge node layer is configured with multiple intelligent edge gateway devices for unified access and local intelligent processing of various types of equipment within the new energy power station. Each intelligent edge gateway device integrates a multi-protocol adaptation module, an edge computing power module, and a local security protection module to ensure that protocol parsing, data processing, and security protection tasks can be completed simultaneously on-site. Specifically, the multi-protocol adaptation module supports various industrial communication protocols, including HTTP, Modbus, OPC UA, etc., enabling unified access to heterogeneous devices such as wind turbines, photovoltaic inverters, battery management systems (BMS), energy storage converters, temperature sensors, vibration sensors, and video acquisition equipment. Each gateway can be configured with multiple multi-protocol adaptation modules according to actual deployment needs to achieve parallel acquisition from different data sources. During the access process, the module automatically identifies the device type and communication protocol, and parses, maps, validates, and converts the acquired data to generate standardized data for subsequent processing. Multiple intelligent edge gateway devices collectively form a distributed data acquisition and processing system at the edge node layer. Deployed in different areas of the new energy power station, they connect to corresponding heterogeneous devices through their respective multi-protocol adaptation modules, achieving broad coverage of multimodal data. After data acquisition, each gateway transmits the data to its local edge computing module for cleaning, analysis, and inference. Simultaneously, a local security module monitors and protects all data access and transmission processes, including communication encryption, intrusion detection, unauthorized access blocking, and firmware integrity verification, ensuring the security and reliability of the entire edge node layer. Through this mechanism, multiple intelligent edge gateway devices achieve highly compatible, low-latency, and scalable data acquisition capabilities for various heterogeneous devices in the new energy power station, laying a data foundation for subsequent edge intelligent analysis and cloud-based collaborative processing.

[0036] Furthermore, the communication network layer includes critical data channels and non-critical data channels.

[0037] Preferably, the communication network layer connects the cloud platform layer and the edge node layer, and undertakes the transmission tasks of different types of data. To balance the system's real-time requirements and bandwidth resource management, the communication network layer is divided into two categories: critical data channels and non-critical data channels. The critical data channel is mainly used to transmit data that requires real-time processing or has high priority, such as fault alarm information generated by edge inference, device anomaly analysis results, task allocation instructions and security hardening strategies issued by the cloud, etc. This channel adopts a high-priority scheduling mechanism and is configured with stricter link reliability assurance strategies to ensure that critical data is transmitted to the cloud or edge nodes with the shortest possible latency. Correspondingly, the non-critical data channel is used to transmit data with low real-time requirements, including multimodal cleaned data generated on the edge side and historical operation records, etc. This channel uploads data asynchronously, effectively reducing network load through strategies such as batch packaging, data compression, and buffer queue scheduling. The two types of channels operate collaboratively, ensuring low-latency transmission of important services while enabling large-scale data to be efficiently synchronized to the cloud in the background, achieving overall system communication stability, flexibility, and efficiency.

[0038] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. An intelligent monitoring method for smart power stations in new energy sources, characterized in that, The method includes: The first intelligent edge gateway device at the edge node layer collects the first multimodal raw data from the first heterogeneous device through the first multi-protocol adaptation module; The first edge computing module receives and performs edge-side data cleaning on the first multimodal raw data to obtain the first multimodal cleaned data. Then, based on the data type of the first multimodal cleaned data, it calls the real-time AI analysis model to perform edge inference and generate the first analysis result. The first analysis result is uploaded to the cloud platform layer through the key data channel of the communication network layer; When the first analysis result is a fault alarm, the cloud platform layer is triggered: S1: The computing tasks of the edge node layer are reallocated through the global task scheduling center, and task allocation instructions are generated; S2: Generate security hardening policies through the security policy center; S3: Send the task allocation instructions and security policies to the edge node layer through the critical data channel.

2. The intelligent monitoring method for smart power stations for new energy as described in claim 1, characterized in that, When the first analysis result is a fault alarm, the AI ​​model training center of the cloud platform layer is triggered to update the lightweight model parameters, generate an updated AI model, and then send it to the edge node layer.

3. The intelligent monitoring method for new energy smart power stations as described in claim 1, characterized in that, The method includes calling a real-time AI analysis model to perform edge inference based on the data type of the first multimodal cleaned data, and generating a first analysis result. If the first multimodal cleaning data is a vibration signal, then the fault prediction model is invoked to predict and output the first analysis result; If the first multimodal cleaning data is temperature data, then the overheating early warning model is invoked to issue an overheating early warning and the first analysis result is output. If the first multimodal cleaning data is image data, then the equipment status recognition model is used to identify and output the first analysis result.

4. The intelligent monitoring method for new energy smart power stations as described in claim 3, characterized in that, The first multimodal cleaning data is compressed to 1 / K of the original data volume using model compression technology and then loaded into the first edge computing module for edge inference.

5. The intelligent monitoring method for new energy smart power stations as described in claim 1, characterized in that, The first multimodal cleaning data is asynchronously uploaded to the cloud platform layer through the non-critical data channel of the communication network layer.

6. An intelligent monitoring system for smart power stations in new energy, characterized in that, The system is used to execute the intelligent monitoring method for smart power stations for new energy as described in any one of claims 1-5, and the system includes: A cloud platform layer, which is deployed in a remote data center; An edge node layer, which is deployed in new energy power stations; The edge node layer and the cloud platform layer are connected through the communication network layer to achieve layered data transmission.

7. The intelligent monitoring system for new energy smart power stations as described in claim 6, characterized in that, The edge node layer includes: Multiple intelligent edge gateway devices, wherein each intelligent edge gateway device integrates a multi-protocol adaptation module, an edge computing power module, and a local security protection module; The multiple intelligent edge gateway devices are connected to multiple heterogeneous devices in the new energy power station through multiple multi-protocol adaptation modules.

8. The intelligent monitoring system for new energy smart power stations as described in claim 7, characterized in that, The cloud platform layer includes: A global task scheduling center is used to allocate computing tasks to the multiple intelligent edge gateway devices of the edge node layer through the communication network layer; The AI ​​model training center is used to generate AI models and distribute them to the edge node layer through the communication network layer. The security policy center is used to generate protection policies and synchronize them to the edge node layer through the communication network layer.

9. The intelligent monitoring system for new energy smart power stations as described in claim 6, characterized in that, The communication network layer includes critical data channels and non-critical data channels.

Images