Thermal power plant knowledge sharing system and method based on federal learning
By using a federated learning-based knowledge-sharing system for thermal power plants, and employing a weighted average algorithm and adaptive clustering strategy, global and cluster models are constructed. This addresses the problem of insufficient model generalization ability caused by the large differences in data distribution among thermal power plants, and improves the robustness and generalization ability of the fault detection model.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- HUANENG POWER INT INC
- Filing Date
- 2026-01-07
- Publication Date
- 2026-05-15
AI Technical Summary
In thermal power plants, it is difficult for a single industrial entity to accumulate diverse and high-quality data, making it impossible to train sufficiently accurate and reliable deep learning models. Furthermore, the distribution of fault data varies greatly among different thermal power plants, and existing federated learning methods are unable to adapt to the generalization requirements of diverse data distributions.
A knowledge-sharing system for thermal power plants based on federated learning is adopted. The model parameters of thermal power plant clients are aggregated through a weighted average algorithm to construct global and cluster models. Data transmission is carried out using an encrypted communication protocol. The clients are divided into different clusters by an adaptive clustering strategy, and the model parameters are aggregated and updated separately for each cluster.
It effectively breaks down data silos, improves the generalization ability and robustness of fault detection models, is suitable for joint model training scenarios in the Industrial Internet of Things, and solves the knowledge sharing problem in non-independent and identically distributed scenarios.
Smart Images
Figure CN122047408A_ABST
Abstract
Description
Technical Field
[0001] The embodiments disclosed herein belong to the field of knowledge sharing technology for thermal power plants, specifically relating to a knowledge sharing system and method for thermal power plants based on federated learning. Background Technology
[0002] Artificial intelligence is playing an increasingly important role in fault detection in industrial sectors such as energy and power. However, for complex industrial control scenarios such as thermal power plants, a single industrial entity often struggles to accumulate diverse, high-quality data, making it impossible to train sufficiently accurate and reliable deep learning models.
[0003] Federated learning, as an emerging collaborative paradigm for artificial intelligence, enables multiple power plants to jointly participate in model training without sharing their local raw data. In this way, all parties can collaboratively build high-performance models while strictly ensuring data privacy and industrial information security.
[0004] However, real-world fault data often exhibits non-independent and identically distributed characteristics. A single global model is difficult to adapt to the generalization needs of diverse data distributions from all clients. Furthermore, thermal power plants may experience faults of different levels and types, leading to significant differences in the distribution of data samples among different participants. Summary of the Invention
[0005] The embodiments disclosed herein aim to at least address one of the technical problems existing in the prior art, and provide a knowledge-sharing system and method for thermal power plants based on federated learning.
[0006] One aspect of this disclosure provides a federated learning-based knowledge-sharing system for thermal power plants, the system comprising multiple thermal power plant clients and a knowledge-sharing platform; The multiple thermal power plant clients are configured to train fault detection models using local data and upload model parameters to the knowledge sharing platform, respectively. The knowledge-sharing platform is configured to aggregate the model parameters of each thermal power plant client using a weighted average algorithm, generate a global model and cluster models of different client clusters, and distribute them to the corresponding thermal power plant clients.
[0007] Furthermore, the system also includes a distributed data collaboration system configured to transmit data between the respective thermal power plant clients and the knowledge sharing platform based on an encrypted communication protocol.
[0008] Another aspect of this disclosure provides a federated learning-based knowledge-sharing method for thermal power plants, based on the federated learning-based knowledge-sharing system for thermal power plants described above, the method comprising: Fault detection models were trained using local data from each thermal power plant's client. Each fault detection model is divided into a global shared layer and a cluster shared layer. Based on the similarity of the gradient update vectors of the fault detection models in the cluster shared layer, an adaptive clustering strategy is adopted to divide each thermal power plant client into different clusters. The model parameters of the global shared layer and the cluster shared layer are aggregated and updated respectively. Specifically, the model parameters of the global shared layer are aggregated among all thermal power plant clients to generate a global model, and the model parameters of the cluster shared layer are aggregated within different clusters to generate multiple corresponding cluster models. The global model and each of the cluster models are distributed to the corresponding thermal power plant clients.
[0009] Furthermore, the gradient update vector is the difference between the initial model parameters and the locally trained model parameters; The similarity of the gradient update vectors is determined by the Euclidean distance or cosine similarity between them.
[0010] Furthermore, the adoption of an adaptive clustering strategy to divide each of the thermal power plant clients into different clusters includes: Construct a model similarity matrix based on the similarity of gradient-updated vectors; The cluster boundaries are determined by comparing the minimum inter-cluster distance with the maximum intra-cluster distance.
[0011] Furthermore, the aggregation and update of model parameters for the global shared layer and the cluster shared layer respectively includes: A weighted average algorithm is used to aggregate the model parameters of all fault detection models to generate a global model; A weighted average algorithm is used to aggregate the model parameters of all fault detection models in each cluster to generate a cluster model for each cluster.
[0012] Furthermore, the aggregation update of the model parameters in the global shared layer is expressed by the following formula:
[0013] in, For the updated global shared layer model parameters, For the client i Local data volume This is the sum of the local data volume of all clients. For the client i No. t The parameters of the globally shared layer model after each round of local training.
[0014] Furthermore, the step of aggregating the model parameters of all fault detection models using a weighted average algorithm to generate a global model includes: The near-term of the FedProx algorithm is introduced as a regularization factor into the objective function of local training, as shown in the following equation:
[0015] in, For the client i Local data volume This is the sum of the local data volume of all clients. For the client i The loss value for local training, The weighting coefficients for neighboring terms. For the client i Updated global shared layer model parameters These are the parameters of the current global shared layer model.
[0016] Furthermore, the aggregation update of the model parameters of the cluster shared layer is expressed by the following formula:
[0017] in, For the first t Round Cluster Cluster shared layer model parameters, For the client i Local data volume For cluster The sum of local data volume across all clients.
[0018] Another embodiment of this disclosure provides an electronic device, characterized in that it includes: At least one processor; and a memory communicatively connected to the at least one processor for storing one or more programs that, when executed by the at least one processor, enable the at least one processor to implement the federated learning-based knowledge-sharing method for thermal power plants described above.
[0019] This disclosure discloses a federated learning-based knowledge-sharing system and method for thermal power plants. It clusters clients with similar data distributions and builds a dedicated model for each cluster. By constructing a framework that supports efficient knowledge sharing between clusters, it trains a model suitable for users with highly heterogeneous data distributions. At the same time, it addresses the knowledge sharing problem between clusters in non-independent and identically distributed scenarios, avoiding overfitting caused by low and limited data diversity in smaller clusters. It effectively breaks down data silos and significantly improves the generalization ability and robustness of deep learning models such as fault detection. It is suitable for joint model training scenarios in the Industrial Internet of Things. Attached Figure Description
[0020] Figure 1This is a schematic diagram of the structure of a knowledge-sharing system for thermal power plants based on federated learning, according to an embodiment of this disclosure. Figure 2 This is a flowchart illustrating another embodiment of a knowledge-sharing method for thermal power plants based on federated learning. Figure 3 This is a schematic diagram of the structure of an electronic device according to another embodiment of the present disclosure. Detailed Implementation
[0021] The technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this disclosure, and not all of them. Based on the embodiments of this disclosure, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this disclosure.
[0022] Furthermore, the described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. Numerous specific details are provided in the following description to give a thorough understanding of embodiments of this disclosure. However, those skilled in the art will recognize that the technical solutions of this disclosure can be practiced without one or more of the specific details, or other methods, components, apparatuses, steps, etc., can be employed. In other instances, well-known methods, apparatuses, implementations, or operations are not shown or described in detail to avoid obscuring various aspects of this disclosure.
[0023] The flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all content and operations / steps, nor do they necessarily need to be performed in the described order. For example, some operations / steps can be broken down, while others can be combined or partially combined; therefore, the actual execution order may change depending on the specific circumstances.
[0024] It should be understood that although the terms first, second, third, etc., may be used in this disclosure to describe various components, these components should not be limited by these terms. These terms are used to distinguish one component from another. Therefore, the first component discussed below may be referred to as the second component without departing from the teachings of this disclosure. As used in this disclosure, the term "and / or" includes all combinations of any and more of the associated listed items.
[0025] Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of exemplary embodiments, and the modules or processes in the drawings are not necessarily necessary for implementing this disclosure, and therefore cannot be used to limit the scope of protection of this disclosure.
[0026] like Figure 1As shown, one embodiment of this disclosure provides a knowledge-sharing system for thermal power plants based on federated learning. The system includes multiple thermal power plant clients 100 and a knowledge-sharing platform 200. The multiple thermal power plant clients 100 are configured to train fault detection models using local data and upload the model parameters to the knowledge-sharing platform 200. The knowledge-sharing platform 200 is configured to aggregate the model parameters of each thermal power plant client 100 using a weighted average algorithm, generate a global model and cluster models for different client clusters, and distribute them to the corresponding thermal power plant clients 100.
[0027] Specifically, each thermal power plant acts as both a data contributor and a model user, serving as a federated learning client. It trains its fault detection model using its local data (such as network traffic, DCS system parameters, and equipment parameters). This local data is labeled as normal or specific fault types (such as boiler tube leaks). The thermal power plant client 100 first performs preliminary model training locally, and then uploads the updated model parameters, rather than the original model parameter data. Simultaneously, the thermal power plant client 100 also acts as an end user, receiving and deploying the enhanced fault detection model distributed by the knowledge-sharing platform 200, thereby improving its local safety protection and predictive maintenance capabilities.
[0028] The knowledge-sharing platform 200 is the core of the system, acting as a federated learning server. It aggregates all local model parameters and uses them to update the generalized model, which is then deployed to each thermal power plant client 100. Its specific responsibilities include: initializing a standard deep learning model (such as LSTM or CNN) as the initial global model; using a weighted average algorithm to aggregate the model parameters uploaded by all thermal power plant clients 100 to generate a more powerful global model; and simultaneously, based on the similarity between the gradient update vectors of the fault detection models of each thermal power plant client 100, dividing each thermal power plant client 100 into different clusters, initializing several other standard deep learning models as the initial cluster model for each cluster, and using a weighted average algorithm to aggregate the model parameters uploaded by the thermal power plant clients 100 within each cluster to generate a more powerful cluster model.
[0029] The knowledge-sharing platform 200 also builds and maintains a standardized verification dataset covering a variety of typical fault scenarios. After each round of updates to the global model and cluster model and before distribution, the fault detection model is rigorously evaluated using this dataset to ensure that key indicators such as recognition accuracy and recall rate reach the preset reliability thresholds before the verified global model and cluster model are distributed to the corresponding thermal power plant clients 100 participating in the collaboration.
[0030] For example, such as Figure 1As shown, the federated learning-based thermal power plant knowledge sharing system also includes a distributed data collaboration system 300, which is configured to transmit data between each thermal power plant client 100 and the knowledge sharing platform 200 based on an encrypted communication protocol.
[0031] Specifically, the distributed data collaboration system 300 is a technical framework connecting the thermal power plant client 100 and the knowledge sharing platform 200. It is built based on encrypted communication protocols such as HTTPS / SSL to ensure the confidentiality, integrity and efficiency of model parameters, update instructions and verification results transmitted in the wide area network, and to resist security threats such as man-in-the-middle attacks.
[0032] This disclosure discloses a federated learning-based knowledge-sharing system for thermal power plants. It clusters clients with similar data distributions and builds a dedicated model for each cluster. By constructing a framework that supports efficient knowledge sharing between clusters, it trains models suitable for users with highly heterogeneous data distributions. At the same time, it addresses the knowledge sharing problem between clusters in non-independent and identically distributed scenarios, avoiding overfitting caused by low and limited data diversity in smaller clusters. It effectively breaks down data silos and significantly improves the generalization ability and robustness of deep learning models such as fault detection. It is suitable for joint model training scenarios in the Industrial Internet of Things.
[0033] like Figure 2 As shown, another embodiment of this disclosure provides a knowledge-sharing method for thermal power plants based on federated learning, which is based on the federated learning-based knowledge-sharing system for thermal power plants described above. The method includes: Step S1: Train the fault detection model using local data from each thermal power plant client.
[0034] Specifically, all thermal power plant clients i All are equipped with the same initial model, and each round of training is defined as a local iteration. (Power plant client) i Have local dataset , Represents local dataset The amount of data.
[0035] Step S2: Divide each fault detection model into a global shared layer and a cluster shared layer. Based on the similarity of the gradient update vectors of the fault detection models in the cluster shared layer, use an adaptive clustering strategy to divide each thermal power plant client into different clusters.
[0036] Specifically, the model parameters trained locally by each thermal power plant client are uploaded to the knowledge sharing platform, where a more powerful global model and cluster models for different client clusters are generated.
[0037] In fault detection models, higher-level weights are more task-relevant than lower-level weights. When non-independent, identically distributed (i.i.d.) data exists, the differences between model parameters are greater than in models trained on i.i.d. data. Therefore, the model is divided into a global sharing layer and a cluster sharing layer. The global sharing layer refers to the model's front-end basic feature extraction layer, used to learn common feature patterns across all thermal power plant clients (such as basic physical laws of equipment operation, common fault symptoms, etc.), and is updated by the knowledge-sharing platform through federated learning. The cluster sharing layer refers to the deep network at the model's back end, closer to the output, which is trained individually for thermal power plant client clusters with similar data distributions (such as the same unit type and similar operating environment) to adapt to the differentiated needs of specific clusters.
[0038] Simply transmitting the entire model's parameters is inefficient. Since gradient values reflect the changes in model parameters after local training, and gradients trained on similarly distributed data have higher similarity, clients are divided into different clusters based on gradient values. Communication rounds When the global shared layer model parameters are represented as The model parameters of the cluster shared layer are represented as follows: .in and These represent the number of layers in the global shared layer and the cluster shared layer, respectively. (Power plant client) exist The model parameters after local training in each round are represented as follows: Gradient update vector Define as client Initial model parameters Compared with locally trained model parameters The difference between .
[0039] Due to a lack of prior knowledge about the data distribution of thermal power plants, optimal settings are pre-defined. k Finding the optimal value is quite difficult. This embodiment proposes an adaptive clustering strategy that eliminates the need for pre-setting or trial-and-error determination. k Instead of calculating the value, the clustering structure is extracted by analyzing the distance between the model gradient update vectors of the clients. Based on the important criterion for evaluating clustering—the minimum inter-cluster distance should not be less than the maximum intra-cluster distance—cluster boundaries are automatically detected and clients are grouped into the corresponding clusters.
[0040] Specifically, using the model similarity matrix The clients are grouped, and this matrix is calculated from the stable local models of all clients after several rounds of local training with a preset number of iterations. For example, the elements in the matrix... express The gradient update distance between any two clients in each round. ,in It is the dimension of the gradient vector. Represents the gradient vector of the first element. Values. Matrix The OK Indicates the client The distance between the gradient update vectors of each client and all other clients. i Treating each client as an independent cluster, with a maximum distance of 0 within the cluster. First, merge the two closest clients to initialize the first non-single-point cluster. Then, compare them sequentially. The algorithm iterates through the distance between two clusters and the maximum intra-cluster distance. If the distance between two clusters is less than the larger of the maximum intra-cluster distances, the two clusters are merged, and the intra-cluster distances (distances between the two farthest points within the cluster) and the distances to other clusters (shortest distances between all points in the two clusters) of each cluster are updated. The algorithm continues iteratively until no cluster pairs satisfy the merging condition are found, thus automatically determining the final number of clusters and the cluster boundaries.
[0041] Step S2 above performs clustering by updating vectors via gradients rather than model parameters. This allows the direction of locally optimal parameters to be deduced from the initial parameters, enabling more accurate capture of the characteristics of each client's data. Furthermore, only a few rounds of efficient communication between the power plant client and the knowledge-sharing platform are required to build a stable client cluster in a flexible and adaptive manner.
[0042] Step S3: Perform aggregated updates on the model parameters of the global shared layer and the cluster shared layer respectively.
[0043] Specifically, the model parameters of the global shared layer are aggregated across all thermal power plant clients to generate a global model, while the model parameters of the cluster shared layer are aggregated within different clusters to generate multiple corresponding cluster models.
[0044] During the cluster knowledge sharing phase, assuming that after several iterations, clients are assigned to specific clusters, and through average cluster... The cluster-specific model is updated by sharing the cluster-specific model parameters among all clients within the cluster, as shown in the following formula:
[0045] in, For the first t Round Cluster Cluster shared layer model parameters, For the client i Local data volume For cluster The sum of local data volume across all clients in the cluster. After averaging the parameters within the cluster, the updated cluster model is distributed to all thermal power plant clients within the cluster, synchronizing the latest shared knowledge.
[0046] During the global knowledge sharing phase, set Indicates the client In the The global shared layer parameters are updated through local training in each round. The aggregation formula on the knowledge-sharing platform is the same as the classic FedAvg algorithm, calculated by averaging the values of all participating clients in the first round. The model parameters in the wheel are updated as shown in the following formula:
[0047] in, For the updated global shared layer model parameters, For the client i Local data volume This is the sum of the local data volume of all clients. For the client i No. t The parameters of the globally shared layer model after each round of local training.
[0048] In some embodiments, to significantly enhance the global robustness of the model during local training, considering the non-independent and identically distributed data distribution and the performance differences of devices in the network, the FedProx algorithm can be introduced to improve the local objective function during updates by the power plant client. Specifically, an L2 regularized proximal term is introduced into the client's objective function, as shown in the following equation:
[0049] in, For the client i Local data volume This is the sum of the local data volume of all clients. For the client i The loss value for local training, The weighting coefficients for neighboring terms. For the client i Updated global shared layer model parameters These are the parameters of the current global shared layer model.
[0050] It is a squared term of the L2 norm (i.e., Euclidean distance), used to measure the client. i In the t Global shared layer model parameters updated after each training round The model parameters of the current global model at this layer The difference between them. This term, as a regularization factor, encourages local updates to more closely resemble the global model, thereby mitigating bias caused by uneven distribution of local data. When When set to zero, the regularization effect is eliminated, making local updates entirely based on the local loss function and no longer directly affected by the global model parameters.
[0051] Step S3 above solves the systematic bias and variance problems introduced by non-independent and identically distributed data by directly integrating FedProx regularization into the local training process.
[0052] Step S4: Distribute the global model and each cluster model to the corresponding thermal power plant client.
[0053] Specifically, the knowledge-sharing platform distributes the generated, higher-performance global model and cluster models for different client clusters to the corresponding thermal power plant clients for download. Each upload and download process between the thermal power plant client and the knowledge-sharing platform is defined as a communication round. The total communication overhead of federated learning comprises two key elements: the number of communication rounds per session. t The amount of shared data and the number of communication rounds required for model convergence. t The total communication overhead can be reduced by decreasing the amount of shared data in each iteration and reducing the number of iterations.
[0054] Each thermal power plant client communicates with the knowledge sharing platform through a distributed data collaboration system built on encrypted communication protocols such as HTTPS / SSL. This ensures the confidentiality, integrity, and efficiency of model parameters, update instructions, and verification results transmitted over the wide area network, and protects against security threats such as man-in-the-middle attacks.
[0055] This disclosure discloses a federated learning-based knowledge-sharing method for thermal power plants. It clusters clients with similar data distributions and builds a dedicated model for each cluster. By constructing a framework that supports efficient knowledge sharing between clusters, it trains a model suitable for users with highly heterogeneous data distributions. At the same time, it addresses the knowledge sharing problem between clusters in non-independent and identically distributed scenarios, avoiding overfitting caused by low and limited data diversity in smaller clusters. It effectively breaks down data silos and significantly improves the generalization ability and robustness of deep learning models such as fault detection. It is suitable for joint model training scenarios in the Industrial Internet of Things.
[0056] like Figure 3 As shown, another embodiment of this disclosure provides an electronic device, including: At least one processor 301; and a memory 302 communicatively connected to the at least one processor 301 for storing one or more programs that, when executed by the at least one processor 301, enable the at least one processor 301 to implement the federated learning-based knowledge-sharing method for thermal power plants described above.
[0057] The memory 302 and processor 301 are connected via a bus, which can include any number of interconnecting buses and bridges. The bus connects various circuits of one or more processors 301 and memory 302 together. The bus can also connect various other circuits, such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and therefore will not be described further herein. A bus interface provides an interface between the bus and the transceiver. The transceiver can be a single element or multiple elements, such as multiple receivers and transmitters, providing a unit for communicating with various other devices over a transmission medium. Data processed by processor 301 is transmitted over a wireless medium via an antenna, which further receives data and transmits it to processor 301.
[0058] Processor 301 is responsible for managing the bus and general processing, and can also provide various functions, including timing, peripheral interfaces, voltage regulation, power management, and other control functions. Memory 302 can be used to store data used by processor 301 during operation.
[0059] It is understood that the above embodiments are merely exemplary embodiments used to illustrate the principles of this disclosure, and this disclosure is not limited thereto. For those skilled in the art, various modifications and improvements can be made without departing from the spirit and substance of this disclosure, and these modifications and improvements are also considered to be within the scope of protection of this disclosure.
Claims
1. A knowledge-sharing system for thermal power plants based on federated learning, characterized in that, The system includes multiple thermal power plant clients and a knowledge-sharing platform; The multiple thermal power plant clients are configured to train fault detection models using local data and upload model parameters to the knowledge sharing platform, respectively. The knowledge-sharing platform is configured to aggregate the model parameters of each thermal power plant client using a weighted average algorithm, generate a global model and cluster models of different client clusters, and distribute them to the corresponding thermal power plant clients.
2. The knowledge-sharing system for thermal power plants based on federated learning according to claim 1, characterized in that, The system also includes a distributed data collaboration system, configured to transmit data between the respective thermal power plant clients and the knowledge sharing platform based on an encrypted communication protocol.
3. A knowledge-sharing method for thermal power plants based on federated learning, based on the knowledge-sharing system for thermal power plants based on federated learning as described in claim 1 or 2, characterized in that, The method includes: Fault detection models were trained using local data from each thermal power plant's client. Each fault detection model is divided into a global shared layer and a cluster shared layer. Based on the similarity of the gradient update vectors of the fault detection models in the cluster shared layer, an adaptive clustering strategy is adopted to divide each thermal power plant client into different clusters. The model parameters of the global shared layer and the cluster shared layer are aggregated and updated respectively. Specifically, the model parameters of the global shared layer are aggregated among all thermal power plant clients to generate a global model, and the model parameters of the cluster shared layer are aggregated within different clusters to generate multiple corresponding cluster models. The global model and each of the cluster models are distributed to the corresponding thermal power plant clients.
4. The knowledge-sharing method for thermal power plants based on federated learning according to claim 3, characterized in that, The gradient update vector is the difference between the initial model parameters and the locally trained model parameters; The similarity of the gradient update vectors is determined by the Euclidean distance or cosine similarity between them.
5. The knowledge-sharing method for thermal power plants based on federated learning according to claim 4, characterized in that, The adoption of an adaptive clustering strategy to divide the various thermal power plant clients into different clusters includes: Construct a model similarity matrix based on the similarity of gradient-updated vectors; The cluster boundaries are determined by comparing the minimum inter-cluster distance with the maximum intra-cluster distance.
6. The knowledge-sharing method for thermal power plants based on federated learning according to claim 3, characterized in that, The aggregation and update of model parameters for the global shared layer and the cluster shared layer respectively includes: A weighted average algorithm is used to aggregate the model parameters of all fault detection models to generate a global model; A weighted average algorithm is used to aggregate the model parameters of all fault detection models in each cluster to generate a cluster model for each cluster.
7. The knowledge-sharing method for thermal power plants based on federated learning according to claim 6, characterized in that, The aggregation update of model parameters in the global shared layer is expressed by the following formula: in, For the updated global shared layer model parameters, For the client i Local data volume This is the sum of the local data volume of all clients. For the client i No. t The parameters of the globally shared layer model after each round of local training.
8. The knowledge-sharing method for thermal power plants based on federated learning according to claim 7, characterized in that, The method of aggregating model parameters from all fault detection models using a weighted average algorithm to generate a global model includes: The near-term of the FedProx algorithm is introduced as a regularization factor into the objective function of local training, as shown in the following equation: in, For the client i Local data volume This is the sum of the local data volume of all clients. For the client i The loss value for local training, The weighting coefficients for neighboring terms. For the client i Updated global shared layer model parameters These are the parameters of the current global shared layer model.
9. The knowledge-sharing method for thermal power plants based on federated learning according to claim 7, characterized in that, The aggregation update of model parameters in the cluster shared layer is expressed by the following formula: in, For the first t Round Cluster Cluster shared layer model parameters, For the client i Local data volume For cluster The sum of local data volume across all clients.
10. An electronic device, characterized in that, include: At least one processor; And a memory communicatively connected to the at least one processor for storing one or more programs that, when executed by the at least one processor, enable the at least one processor to implement the federated learning-based knowledge-sharing method for thermal power plants according to any one of claims 3 to 9.