A new energy station stability risk edge intelligent perception system

Abstract

The application provides a new energy station stability risk edge intelligent perception system, comprising: obtaining local operation time sequence data of each unit, training a lightweight time sequence prediction model constructed locally by using the time sequence data to obtain local model parameters of each unit; receiving local model parameters of all units in the station, weighting and aggregating each local model parameter to obtain station-level aggregated model parameters; the station-level aggregated model parameters are sent to each unit to update the local model, and the station-level aggregated model parameters are uploaded to the cloud, and global federated learning is performed on the aggregated model parameters uploaded by each station in the cloud to obtain global model parameters and send the global model parameters to each station to update the station-level model; based on the updated station-level model or the global model, the operation state of the station is risk evaluated, and a stability risk index is output and a corresponding level warning is triggered. The application only transmits model parameters without uploading original data, which greatly reduces the communication bandwidth demand while protecting data privacy, and can effectively improve the real-time performance and accuracy of new energy station stability risk perception.

Description

Technical Field

[0001] This invention belongs to the field of intelligent monitoring technology for new energy power generation and power systems, and specifically relates to an intelligent sensing system for the edge of stability risks in new energy power plants. Background Technology

[0002] With the expansion of the scale and the improvement of the intelligence level of new energy power stations, a large number of sensors and intelligent devices are deployed inside the stations, generating massive amounts of operational data in real time. How to use this data to efficiently detect the stability risks of the power station (such as voltage over-limit, frequency fluctuations, oscillations, etc.) has become a key issue in ensuring the safety of new energy grid connection.

[0003] Traditional centralized monitoring solutions upload all data to the cloud for processing, but this solution has the following problems: First, massive data transmission consumes a large amount of bandwidth, resulting in high communication costs; Second, centralized data storage poses a risk of privacy breaches. Third, cloud processing suffers from significant latency, making it difficult to meet real-time early warning requirements. Edge computing offloads some computing tasks to edge nodes closer to the data source, reducing latency and bandwidth requirements. However, edge nodes have limited computing power and data volume, resulting in insufficient generalization ability of a single model.

[0004] Federated learning, as a distributed machine learning framework, allows multiple clients to train models locally and only upload model parameters to the server for aggregation, thus protecting data privacy. However, traditional federated learning typically only has a two-layer client-server architecture. For new energy power plant monitoring scenarios that include a three-layer "edge-cloud" structure, further layered optimization is needed. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent edge-sensing system for stability risks at new energy power plants, addressing the aforementioned shortcomings of existing technologies. This invention adopts a three-layer "end-edge-cloud" architecture: the end-side (generator unit) runs a lightweight model to extract features and trains locally; the edge-side (power plant centralized control) aggregates the end-side models to form a power plant-level model; and the cloud aggregates the models from each power plant to form a global model. This system, while protecting privacy, achieves a unity of distributed intelligence and centralized optimization, improving the real-time performance and accuracy of risk perception.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: In a first aspect, the present invention provides an intelligent sensing method for the edge of stability risk at new energy power stations, comprising the following steps: The time series data of each unit in each station is collected, and the lightweight time series prediction model constructed for each unit is trained using the time series data to obtain the end-side model parameters corresponding to each unit. The parameters of each end-side model are weighted and aggregated to obtain the site-level aggregated model parameters corresponding to each site, and the end-side model is updated using the site-level aggregated model parameters. Global federated learning is performed on the aggregated model parameters of each site to obtain global model parameters, and the site-level aggregated model is updated using the global model parameters. Based on the updated site-level aggregation model, a risk assessment of the site's operational status is conducted.

[0007] Preferably, the station-level aggregated model parameters for each station are obtained by weighted aggregation of the model parameters at each end using the following formula. The specific method is as follows:

[0008] in, These are the parameters for the station-level aggregation model; For the first The amount of data in the runtime sequence data of each unit; For the first The end-side model parameters corresponding to each unit.

[0009] Preferably, the lightweight time-series prediction model constructed for each unit is trained using the mini-batch stochastic gradient descent method.

[0010] Preferably, FedAvg or FedProx is used to perform global federated learning on the aggregated model parameters of each site to obtain global model parameters.

[0011] Secondly, the present invention provides an intelligent sensing system for the edge of stability risk at new energy power stations, comprising: The end-side intelligent unit is used to collect the runtime sequence data of each unit in each station, and use the time sequence data to train the lightweight time sequence prediction model constructed for each unit to obtain the end-side model parameters corresponding to each unit. The edge aggregation server is used to perform weighted aggregation of the parameters of each edge model to obtain the site-level aggregated model parameters corresponding to each site, and to update the edge model using the site-level aggregated model parameters. At the same time, risk assessment of the station's operational status is conducted based on the updated station-level aggregation model; The cloud-based global model server is used to perform global federated learning on the aggregated model parameters of each site to obtain global model parameters, and then use the global model parameters to update the site-level aggregated model.

[0012] Preferably, the end-layer intelligent unit includes: The data acquisition subunit is used to collect the runtime sequence data of each unit in each station; The local training subunit is used to train the lightweight time-series prediction model constructed for each unit using the time-series data, so as to obtain the end-side model parameters corresponding to each unit.

[0013] Preferably, the side-side aggregation server includes: The federated average aggregation sub-unit is used to perform weighted aggregation of the model parameters of each end side to obtain the site-level aggregated model parameters corresponding to each site. The parameters are distributed to the sub-units to update the end-side model using the parameters of the site-level aggregated model; The risk assessment unit is used to assess the risk of the station's operational status based on the updated station-level aggregation model.

[0014] Preferably, the cloud-based global model server includes: The global training subunit is used to perform global federated learning on the aggregated model parameters of each station to obtain the global model parameters. The model distribution sub-unit updates the site-level aggregated model using global model parameters.

[0015] Thirdly, the present invention provides an electronic device including a processor and a memory, wherein the memory stores computer instructions, and when the computer instructions are executed by the processor, the electronic device performs the method described thereon.

[0016] Fourthly, the present invention provides a computer program product, the computer program product including computer-executable instructions, which, when executed, implement the method described.

[0017] Compared with the prior art, the beneficial effects of the present invention are: This invention provides an intelligent edge sensing method for stability risks at new energy power plants. It employs a hierarchical collaborative architecture of "local training at the terminal side—aggregation at the power plant—global aggregation in the cloud." The original time-series data of each unit is always stored locally, with only model parameters uploaded. This fundamentally eliminates the privacy risks that may arise during centralized data storage and transmission, effectively protecting power plant operational data and trade secrets. Each unit only needs to upload lightweight model parameters instead of massive amounts of raw data, and the small number of model parameters significantly reduces the bandwidth usage and transmission costs within the power plant and between the power plant and the cloud. Simultaneously, the cloud only needs to process the aggregated model parameters instead of the full data, significantly reducing the computational overhead of global training. The aggregated model at the power plant can be deployed locally. At the power plant level, risk assessment and early warning decisions can be completed at the edge without waiting for results to be transmitted from the cloud, thus significantly reducing the early warning response delay and meeting the urgent needs of new energy power plants for real-time performance and rapid response. Each unit can be trained independently using local data, which can fully explore and adapt to the personalized operating characteristics of a single unit. The aggregation at the power plant level integrates the local knowledge of all units within the power plant, while the global aggregation in the cloud further integrates the diverse operating scenarios and experience across power plants. This allows models at all levels from the edge to the cloud to take into account both personalized adaptation and global generalization capabilities. This avoids the overfitting and insufficient generalization problems caused by the limited amount of data in a single edge model, and also avoids the shortcomings of a pure global model that ignores the individual differences of power plants and units.

[0018] Meanwhile, this method forms a continuous iterative closed loop of "local training - side aggregation - global optimization - feedback update". The models at each level can be continuously updated and optimized online as the operational data accumulates, enabling the perception model to adaptively track the dynamic changes in the operating status of new energy power stations and maintain a high risk identification accuracy in the long term. Attached Figure Description

[0019] Figure 1 This is a system layered architecture diagram related to the present invention.

[0020] Figure 2 This invention relates to a hierarchical federated learning flowchart.

[0021] Figure 3 This is a block diagram of the end-side intelligent unit structure involved in the present invention. Detailed Implementation

[0022] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0023] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0024] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0025] As used in this application specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if detected [the described condition or event]" may be interpreted, depending on the context, as meaning "once determined," "in response to determination," "once detected [the described condition or event]," or "in response to detection [the described condition or event]."

[0026] Furthermore, in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0027] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0028] Example 1 This embodiment provides a method for intelligent sensing of stability risks at new energy power stations, including the following steps: The time series data of each unit in each station is collected, and the lightweight time series prediction model constructed for each unit is trained using the time series data to obtain the end-side model parameters corresponding to each unit. The parameters of each end-side model are weighted and aggregated to obtain the site-level aggregated model parameters corresponding to each site, and the end-side model is updated using the site-level aggregated model parameters. Global federated learning is performed on the aggregated model parameters of each site to obtain global model parameters, and the site-level aggregated model is updated using the global model parameters. Based on the updated site-level aggregation model, a risk assessment of the site's operational status is conducted.

[0029] Example 2 This embodiment provides an intelligent edge sensing system for stability risks at new energy power stations, comprising an edge-side intelligent unit, an edge-side aggregation server, a cloud-based global model server, and a risk early warning unit, wherein: End-side intelligent unit: Deployed in each unit (wind turbine or photovoltaic inverter) to collect the operating sequence data of each unit in each site, and use the time sequence data to train the lightweight time sequence prediction model built for each unit to obtain the end-side model parameters corresponding to each unit.

[0030] In this embodiment, the edge-side intelligent unit includes: Data acquisition subunit: Collects the operating sequence data of each unit in each station. The operating sequence data includes electrical quantities (voltage, current, power, frequency) and equipment status quantities (temperature, vibration, speed, etc.). The sampling frequency can be set according to requirements (e.g., 10Hz-1kHz), and the data is organized through a sliding window (e.g., a 10-second window).

[0031] Model building subunit: Construct a lightweight LSTM model. The input of the lightweight LSTM model is time series data within a sliding window, and the output is a low-dimensional feature vector (e.g., 16-dimensional) that reflects the unit's operating status.

[0032] The lightweight LSTM model employs a single-layer structure with hidden layer dimensions of 32-64 and fewer than 10k parameters, making it suitable for operation on embedded devices (such as the ARM Cortex-M series). This feature vector serves as the input feature for the subsequent risk assessment model.

[0033] Local training subunit: Uses time-series data output from the data acquisition subunit to incrementally train the lightweight LSTM model and update the model parameters.

[0034] In this embodiment, the lightweight LSTM model is trained using mini-batch stochastic gradient descent.

[0035] The loss function of the lightweight LSTM model is set according to the task (such as the mean squared error of predicting the voltage value at the next time step, or the cross-entropy loss for classifying risk states). Local training epochs are usually 1-5 epochs to avoid overfitting.

[0036] Side-side aggregation server: Deployed in the site control center, usually an industrial control computer or server, it is used to perform weighted aggregation of the parameters of each side-side model to obtain the site-level aggregation model parameters corresponding to each site, and to update the side-side model using the site-level aggregation model parameters; at the same time, based on the updated site-level aggregation model, risk assessment of the site's operating status is performed.

[0037] In this embodiment, the edge aggregation server specifically includes: Model receiving subunit: Used to receive edge model parameters (usually neural network weights and biases) uploaded by all edge-layer intelligent units in this site.

[0038] FedAvg algorithm is used to weight and aggregate the model parameters of each end to obtain the site-level aggregated model parameters corresponding to each site.

[0039] In this embodiment, the parameters of each end-side model are weighted and aggregated using the following formula:

[0040] in, These are the parameters for the station-level aggregation model; For the first The amount of data in the runtime sequence data of each unit; For the first The end-side model parameters corresponding to each unit.

[0041] Parameter distribution sub-unit: Distributes the parameters of the station-level aggregated model back to each end side for updating the end side model.

[0042] Side-side model storage sub-unit: Stores the side-side aggregated model parameters (i.e. aggregated weights) of this site, serving as the basic model for site-level risk assessment.

[0043] Edge-Cloud Communication Subunit: Exchanges models with the global model server in the cloud. Specifically, it uploads the edge-aggregated model parameters of this site to the cloud and receives the global model parameters sent by the cloud.

[0044] Cloud-based global model server: Deployed in the group's data center, it is used to perform global federated learning on the aggregated model parameters of each site to obtain global model parameters, and then use the global model parameters to update the site-level aggregated model.

[0045] In this embodiment, the cloud-based global model server specifically includes: Model collection subunit: Used to receive aggregated model parameters uploaded by the aggregation servers at each site.

[0046] Global training subunit: Federated learning is performed on the side-side aggregated model parameters of each station to obtain the global model parameters.

[0047] In this embodiment, FedAvg or FedProx is used to perform global federated learning on the aggregated model parameters of each site to obtain global model parameters.

[0048] The number of global training rounds can be set as needed (e.g., once a month or once a quarter).

[0049] Model distribution subunit: Periodically distributes global model parameters to each side aggregation server for initializing or updating the side aggregation server.

[0050] Global model storage sub-unit: Used to store global model parameters for comparative analysis and model rollback.

[0051] Risk warning unit: Can be deployed on the edge or in the cloud to conduct risk assessment of the station's operating status based on the updated station-level aggregation model.

[0052] In this embodiment, the risk warning unit specifically includes: Data acquisition interface: Obtain real-time features from the edge-side intelligent unit (if deployed on the edge, it can directly receive features uploaded from the edge) or read historical features from the database.

[0053] Risk assessment model: Input real-time features into the updated edge aggregation server and output a stable risk indicator. In this embodiment, the stability risk indicators include: Voltage stability margin prediction (range 0-1, the lower the value, the more dangerous). Frequency stability margin prediction Oscillation Risk Index (0-1, the higher the value, the more dangerous) Equipment Health Index Early warning rule engine: Based on the risk assessment results and the comparison with preset thresholds, different levels of early warnings (such as red, orange, yellow and blue) are triggered.

[0054] Visual interface: Displays risk trend curves, current risk distribution, and a list of early warning events, and supports overlaying risk heat maps onto site wiring diagrams.

[0055] The core workflow of this system is as follows: The edge-side intelligent units train lightweight LSTM models using local data to obtain edge-side model parameters reflecting the unit's operating status. All edge-side intelligent units upload these model parameters to the edge-side aggregation server, which then uses federated averaging to obtain site-level aggregated model parameters. The edge-side aggregation server sends the aggregated model parameters back to the edge-side intelligent units for model updates and periodically uploads them to the cloud-based global model server. The cloud-based global model server performs global federated learning on the aggregated model parameters from each site to obtain global model parameters. The cloud-based global model server then sends these global model parameters back to the edge-side aggregation servers for initialization or updates. During stability risk assessment, real-time feature vectors extracted from the edge-side can be directly input into the edge-side aggregation model (or the cloud-based global model) to output risk indicators.

[0056] The beneficial effects of this sensing system include: 1) Protect data privacy; raw data does not leave the device. 2) Reduce communication bandwidth, only transmit model parameters (reducing weekly data volume from a centralized 10TB to 0.5TB). 3) Reduce early warning delay and process warnings on the nearest side (delay < 1 second); 4) Enhance model generalization ability by integrating multi-site knowledge into a cloud-based global model; 5) Supports site personalization, while retaining local characteristics in the side model.

[0057] Example 3 Reference Figure 1 This embodiment provides a new energy power station stability risk edge intelligent perception system, including: an edge intelligent unit, an edge aggregation server, and a cloud global model server.

[0058] End-side intelligent unit: A certain offshore wind farm consists of 30 wind turbines, and each turbine is equipped with an end-side intelligent unit 101. The end-side unit adopts an ARM Cortex-M7 processor and integrates a data acquisition module (acquiring voltage, current, speed, vibration, etc.), a lightweight LSTM model, a local training module, and an NB-IoT communication module.

[0059] Edge Layer: The site's centralized control center deploys edge aggregation server 102, which is an x86 server running a Linux system. It includes model aggregation services, an edge model library, and risk warning applications. The edge server communicates with all end-side devices via a fiber optic ring network.

[0060] Cloud layer: The group's data center deploys a global model server 103 in the cloud, which adopts a GPU server cluster and runs a global federated learning framework, including model collection, global training and model distribution services.

[0061] Risk warning applications can be deployed at the edge to achieve real-time warnings (such as voltage over-limit warnings), or deployed in the cloud to achieve cross-site risk analysis (such as regional oscillation risk).

[0062] Reference Figure 2 This embodiment describes the hierarchical federated learning process of a photovoltaic power station.

[0063] Step S401: Initialization. An initial LSTM model (random weights) is generated in the cloud and distributed to the edge side, which then distributes it to all inverter end sides.

[0064] Step S402: Local Training. For each inverter, the model is trained once using local data from the most recent hour (10-second sliding window, 360 samples in total) with the Adam optimizer and a learning rate of 0.001. The loss function is the mean square error of the predicted voltage deviation.

[0065] Step S403: Edge aggregation. The terminal uploads model parameters (approximately 8KB) to the edge. After receiving 30 models, the edge performs a weighted average based on the data volume of each terminal to obtain the site aggregated model.

[0066] Step S404: Side-side application. The side-side uses an aggregated model to perform risk assessment on the real-time data of the entire field, and outputs the predicted voltage stability margin value of each inverter. When the margin is lower than 0.2, an early warning is triggered.

[0067] Step S405: Cloud Aggregation. The edge device uploads the aggregated site model to the cloud (once a week). After receiving the models from 10 sites, the cloud performs a global federated average to obtain the global model.

[0068] Step S406: Model Distribution. The cloud distributes the global model to each edge, and the edge then distributes it to the endpoint for the next round of training. Simultaneously, the cloud saves historical model versions.

[0069] This embodiment includes a risk warning application example. In a wind farm, a risk warning application is deployed on a side-side server to perform real-time risk assessment based on the latest aggregation model. The model input is the real-time characteristics (16 dimensions) of each wind turbine, and the output is an oscillation risk index (0-1). Thresholds are set: above 0.7 is high risk (red), 0.4-0.7 is medium risk (orange), 0.2-0.4 is low risk (yellow), and below 0.2 is normal (green).

[0070] At a certain moment, the oscillation risk index of wind turbine No. 5 rose to 0.82, triggering a red alert. The alert application automatically recorded the event, flashed the image of wind turbine No. 5 on the monitoring interface, and pushed the alarm information to the operators' mobile phones. The operators reviewed the historical characteristic curves of the wind turbine and found abnormal vibration amplitude, so they arranged for maintenance. Upon inspection, it was found to be a gearbox failure, and the proactive handling prevented damage to the unit.

[0071] Application performance comparison in this embodiment This system was deployed at 10 renewable energy power plants in a certain region, and compared with a centralized solution:

[0072] The results show that this application has significant advantages in terms of communication, latency, privacy, and scalability.

[0073] Example 4 This embodiment also provides a computing device. The computing device includes a bus, a processor, a memory, and a communication interface. The processor, memory, and communication interface communicate with each other via the bus. The computing device can be a server or a terminal device. It should be understood that this application does not limit the number of processors and memory in the computing device.

[0074] A bus can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of representation, a bus can include a path for transmitting information between various components of a computing device (e.g., memory, processor, communication interfaces).

[0075] The processor may include any one or more of the following: central processing unit (CPU), graphics processing unit (GPU), tensor processing unit (TPU), application specific integrated circuit (ASIC), field-programmable gate array (FPGA), microprocessor (MP), or digital signal processor (DSP).

[0076] The memory may include volatile memory, such as random access memory (RAM). The processor may also include non-volatile memory, such as read-only memory (ROM), flash memory, hard disk drive (HDD), or solid state drive (SSD).

[0077] The memory stores executable program code, which the processor executes to implement the functions of the aforementioned units, thereby achieving, for example, the method described in Embodiment 1. That is, the memory may store instructions for the methods and functions relating to the computing device in any of the above embodiments.

[0078] The communication interface uses transceiver modules such as, but not limited to, network interface cards and transceivers to enable communication between computing devices and other devices or communication networks.

[0079] Example 5 This embodiment also provides a computer-readable storage medium storing computer instructions that, when executed by a processor, cause the processor to perform the methods and functions of the computing device involved in any of the above embodiments.

[0080] Generally, the various embodiments of this disclosure can be implemented in hardware or dedicated circuitry, software, logic, or any combination thereof. Some aspects can be implemented in hardware, while others can be implemented in firmware or software, which can be executed by a controller, microprocessor, or other computing device. Although various aspects of the embodiments of this disclosure are shown and described as block diagrams, flowcharts, or represented using some other illustration, it should be understood that the blocks, apparatuses, systems, techniques, or methods described herein can be implemented as, as non-limiting examples, in hardware, software, firmware, dedicated circuitry or logic, general-purpose hardware or controllers or other computing devices, or some combination thereof.

[0081] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A method for intelligent sensing of stability risk edges in new energy power plants, characterized in that, Includes the following steps: The time series data of each unit in each station is collected, and the lightweight time series prediction model constructed for each unit is trained using the time series data to obtain the end-side model parameters corresponding to each unit. The parameters of each end-side model are weighted and aggregated to obtain the site-level aggregated model parameters corresponding to each site, and the end-side model is updated using the site-level aggregated model parameters. Global federated learning is performed on the aggregated model parameters of each site to obtain global model parameters, and the site-level aggregated model is updated using the global model parameters. Based on the updated site-level aggregation model, a risk assessment of the site's operational status is conducted.

2. The intelligent sensing method for stability risk edge of a new energy power station according to claim 1, characterized in that, The station-level aggregated model parameters for each station are obtained by weighting and aggregating the parameters of each end-side model using the following formula: in, These are the parameters for the station-level aggregation model; For the first The amount of data in the runtime sequence data of each unit; For the first The end-side model parameters corresponding to each unit.

3. The intelligent sensing method for stability risk edge of a new energy power station according to claim 1, characterized in that, The lightweight time series prediction model for each unit was trained using the mini-batch stochastic gradient descent method.

4. The intelligent sensing method for stability risk edge of a new energy power station according to claim 1, characterized in that, FedAvg or FedProx is used to perform global federated learning on the aggregated model parameters of each site to obtain global model parameters.

5. A smart sensing system for edge stability risks at new energy power stations, characterized in that, include: The end-side intelligent unit is used to collect the runtime sequence data of each unit in each station, and use the time sequence data to train the lightweight time sequence prediction model constructed for each unit to obtain the end-side model parameters corresponding to each unit. The edge aggregation server is used to perform weighted aggregation of the parameters of each edge model to obtain the site-level aggregated model parameters corresponding to each site, and to update the edge model using the site-level aggregated model parameters. At the same time, risk assessment of the station's operational status is conducted based on the updated station-level aggregation model; The cloud-based global model server is used to perform global federated learning on the aggregated model parameters of each site to obtain global model parameters, and then use the global model parameters to update the site-level aggregated model.

6. The intelligent sensing system for the edge of stability risk at a new energy power station according to claim 5, characterized in that, The end-layer intelligent unit includes: The data acquisition subunit is used to collect the runtime sequence data of each unit in each station; The local training subunit is used to train the lightweight time-series prediction model constructed for each unit using the time-series data, so as to obtain the end-side model parameters corresponding to each unit.

7. The intelligent sensing system for edge stability risk of a new energy power station according to claim 5, characterized in that, The side-side aggregation server includes: The federated average aggregation sub-unit is used to perform weighted aggregation of the model parameters of each end side to obtain the site-level aggregated model parameters corresponding to each site. The parameters are distributed to the sub-units to update the end-side model using the parameters of the site-level aggregated model; The risk assessment unit is used to assess the risk of the station's operational status based on the updated station-level aggregation model.

8. The intelligent sensing system for the edge of stability risk at a new energy power station according to claim 5, characterized in that, The cloud-based global model server includes: The global training subunit is used to perform global federated learning on the aggregated model parameters of each station to obtain the global model parameters. The model distribution sub-unit updates the site-level aggregated model using global model parameters.

9. An electronic device, characterized in that, It includes a processor and a memory, the memory storing computer instructions that, when executed by the processor, cause the electronic device to perform the method of any one of claims 1 to 4.

10. A computer program product, characterized in that, The computer program product includes computer-executable instructions that, when executed, implement the method of any one of claims 1 to 4.

Images