A smart power plant management method and system based on distributed management

Abstract

This invention provides a smart power plant management method and system based on distributed management. The method includes the following steps: an initial allocation phase, in which the central node performs initial training on a first central model and a second central model based on an initial training dataset, and distributes the initially trained model parameters to each edge node; the edge nodes determine whether to perform manual inspection based on real-time collected environmental parameters and equipment operating parameters; and an iterative update phase, in which the central node obtains the inspection results of manual inspection, stores the inspection results of each monitoring location within a preset time period as an iterative training dataset, and the central node performs iterative training on the first central model and the second central model based on the iterative training dataset, and distributes the iteratively trained model parameters of the first central model and the second central model to each edge node.

Description

Technical Field

[0001] This invention relates to the field of smart power plant technology, and in particular to a smart power plant management method and system based on distributed management. Background Technology

[0002] The importance of smart power plants adopting a distributed architecture lies in their ability to significantly improve the operating efficiency of power plants, enhance the flexibility and reliability of the system, and promote the power plants to move towards intelligence and automation, thereby better adapting to the diversified and complex trends of future energy development.

[0003] First, the distributed architecture provides smart power plants with a high degree of flexibility. Under a distributed architecture, the various system components of a power plant can operate and be managed relatively independently, enabling the plant to quickly adjust system configurations and optimize resource allocation based on actual needs and operational conditions. Second, the distributed architecture significantly enhances the reliability of smart power plants. In a distributed system, components interact and collaborate through communication networks, which helps reduce the impact of single-point failures on the entire system. Furthermore, the distributed architecture supports redundant deployment and load balancing, further improving system stability and reliability. This reliability is crucial for power plants because any downtime or failure can severely impact the stable operation of the power grid and the power supply to users.

[0004] Smart power plants with distributed architectures can analyze the status of edge nodes to determine whether manual inspection is needed, thereby improving operational safety. Existing distributed architectures often have a central node that uniformly trains the model and distributes the trained model parameters to the edge nodes. However, the components in different locations of the power plant are different, so their normal operating standards also vary, making it difficult for the existing architecture to guarantee the accuracy of the judgment.

[0005] In view of this, the present invention is proposed. Summary of the Invention

[0006] The purpose of this invention is to provide a smart power plant management method and system based on distributed management. This solution sets up two judgment models at each edge node. On the one hand, the judgment accuracy is ensured by increasing the number of models. On the other hand, the judgment accuracy is ensured by adaptive training through separate training of the corresponding edge nodes.

[0007] This invention provides a smart power plant management method based on distributed management. The method is based on a distributed architecture, including a central node and edge nodes corresponding to each monitoring location. The central node is configured with a first central model based on environmental parameters and a second central model based on equipment operating parameters. The steps of the method include:

[0008] In the initial allocation phase, the central node performs initial training on the first central model and the second central model based on the initial training dataset, and distributes the model parameters of the first central model and the second central model after initial training to each edge node so that the edge nodes can construct the first edge model and the second edge model.

[0009] The edge nodes will construct a first vector and a second vector based on real-time collected environmental parameters and device operating parameters, and determine whether to perform manual detection based on the corresponding first edge model and second edge model.

[0010] In the iterative update phase, the central node obtains the inspection results of manual detection and stores the inspection results of each monitoring location within a preset time period as an iterative training dataset. The central node iteratively trains the first central model and the second central model based on the iterative training dataset and distributes the model parameters of the iteratively trained first central model and the second central model to each edge node so that the edge nodes update the first edge model and the second edge model.

[0011] The edge model is based on the updated first edge model and second edge model to re-determine in real time whether to perform manual detection, and is continuously iterated and updated.

[0012] Using the above scheme, the central node undergoes unified training in the initial stage, and the trained model parameters are uniformly distributed to each edge node. Each edge node builds an initial model based on the same model parameters. During the judgment process, since the two models are trained with different parameters, they may produce different results at the same time. Staff members conduct manual inspections based on manual inspection instructions issued by either model at the same monitoring location, and upload the numerical value of the necessity of inspection to the central node after the manual inspection. Therefore, the two models can correspond to each other and build a dataset that differs from their own judgment results, providing a more accurate basis for subsequent iterative updates. Moreover, this scheme can train each edge node separately to ensure processing accuracy.

[0013] In some embodiments of the present invention, during the iterative update phase, the central node acquires the inspection results of manual detection and stores the inspection results of each monitoring location within a preset time period as an iterative training dataset:

[0014] Obtain the instruction records for manual detection in the first edge model and the second edge model respectively;

[0015] Obtain the inspection necessity parameters uploaded by the staff corresponding to each instruction record;

[0016] The inspection judgment parameters and corresponding inspection necessity parameters output by the first edge model or the second edge model when determining whether to perform manual inspection are constructed into a training data set;

[0017] The training data corresponding to the instruction records of the first edge model and the second edge model are then used to construct iterative training datasets.

[0018] In some embodiments of the present invention, during the iterative update phase, the step of the central node acquiring the inspection results of manual detection and storing the inspection results of each monitoring location within a preset time period as an iterative training dataset further includes:

[0019] Obtain the instruction records for manual detection in the first edge model and the second edge model respectively;

[0020] When manual inspection is performed on either the first edge model or the second edge model, it is determined whether manual inspection is performed simultaneously on the other first edge model or the second edge model.

[0021] If not, for the first or second edge model that has not been determined to perform manual inspection, obtain the inspection judgment parameters output at that moment, and obtain the instruction record of the other first or second edge model that has determined to perform manual inspection at that moment, as well as the inspection necessity parameters uploaded by the staff, to construct a training data.

[0022] The above-mentioned scheme sets up two decision models at each edge node. On the one hand, this increases the accuracy of the decision and avoids missed detections. On the other hand, if only one decision model is set up at the edge node, training data can only be generated when the model determines that manual inspection is required. Some of the generated training data are cases where the model determines that manual inspection is required but the inspection necessity parameters uploaded by the staff are low. In the process of continuous iterative training, the model will continuously shift towards not determining manual inspection, eventually leading to the loss of decision-making ability. However, this scheme sets up two decision models at the same time. When one decision model does not determine that manual inspection is required, the other decision model may determine that manual inspection is required. The other decision model generates training data in the opposite direction of the shift, ensuring that the decision is made and avoiding the occurrence of unidirectional shift.

[0023] In some embodiments of the present invention, in the step where the central node iteratively trains the first central model and the second central model based on the iterative training dataset:

[0024] The first center model is trained using the iterative training dataset corresponding to the first edge model;

[0025] The second center model is trained using the iterative training dataset corresponding to the second edge model.

[0026] In some embodiments of the present invention, the steps of the iterative update phase further include:

[0027] Each edge node is provided with a real-time storage unit, which stores the model parameters of the first edge model or the second edge model in that edge node;

[0028] When iteratively training an edge node, the model parameters in the corresponding real-time storage unit are called and applied to the first center model and the second center model, and the model is trained using the collected iterative training dataset.

[0029] By adopting the above scheme, each edge node is provided with a separate real-time storage unit, which can ensure that each edge node is trained separately, and that the training data is applicable to the monitoring position corresponding to the edge node, thereby ensuring that each edge node can perform the best judgment.

[0030] In some embodiments of the present invention, the steps of the iterative update phase further include:

[0031] At the beginning of each iteration update phase, the instruction records of the first edge model and the second edge model of each edge node when determining to conduct manual inspection are called, and the inspection judgment parameters output by the first edge model and the second edge model when determining to conduct manual inspection are compared with the inspection necessity parameters uploaded by the staff.

[0032] Determine whether the inspection judgment parameters of each instruction record match the corresponding inspection necessity parameters to obtain the matching degree of each edge node;

[0033] The edge nodes are sorted based on the matching degree, and each edge node is iteratively trained based on the sorting order.

[0034] In some embodiments of the present invention, in the step of determining whether the inspection judgment parameters of each instruction record match the corresponding inspection necessity parameters to obtain the matching degree of each edge node:

[0035] Calculate the difference between the inspection judgment parameter and the corresponding inspection necessity parameter, and determine whether they match based on the difference;

[0036] The ratio of the number of mismatched instruction records to the total number of instruction records is calculated as the matching degree of the edge node.

[0037] In some embodiments of the present invention, in the steps of sorting the edge nodes based on the matching degree and iteratively training each edge node based on the sorting order, the edge nodes are arranged in descending order based on the matching degree, and iterative training is performed on each of the edge nodes arranged in descending order.

[0038] Using the above scheme, during the iterative update phase, the edge nodes are sorted by matching degree. The matching degree is determined by the difference between the inspection judgment parameter and the corresponding inspection necessity parameter, which can represent the difference between the model's output and the actual situation. This scheme prioritizes edge nodes with large differences between the model's judgment and the actual situation by sorting the edge nodes, thus ensuring the accuracy of subsequent judgments.

[0039] In some embodiments of the present invention, in the step of re-determining in real time whether to perform manual detection based on the updated first edge model and second edge model, the first edge model and the second edge model are determined at intervals of time:

[0040] The first edge model constructs a first vector from the environmental parameters collected in real time as input and outputs inspection judgment parameters. The inspection judgment parameters are compared with a preset judgment threshold to determine whether manual inspection should be performed.

[0041] The second edge model constructs a second vector from the real-time collected device operating parameters as input and outputs inspection judgment parameters. The inspection judgment parameters are compared with a preset judgment threshold to determine whether manual inspection should be performed.

[0042] Another aspect of the present invention relates to a smart power plant management system based on distributed management. The system includes a computer device, which includes a processor and a memory. The memory stores computer instructions, and the processor is used to execute the computer instructions stored in the memory. When the computer instructions are executed by the processor, the system implements the steps of the method.

[0043] In summary, the present invention has the following beneficial effects:

[0044] 1. In the initial stage, the central node of this solution undergoes unified training, and the trained model parameters are uniformly distributed to each edge node. Each edge node builds an initial model based on the same model parameters. During the judgment process, since the two models are trained with different parameters, they may produce different results at the same time. Staff members conduct manual inspections based on manual inspection instructions issued by either model at the same monitoring location, and upload the numerical value of the necessity of inspection to the central node after the manual inspection. Therefore, the two models can correspond to each other and build a dataset that differs from their own judgment results, providing a more accurate basis for subsequent iterative updates. In addition, this solution can train each edge node separately to ensure processing accuracy.

[0045] 2. This solution sets up two decision models at each edge node. On the one hand, this increases the accuracy of decision-making and avoids missed detections. On the other hand, if only one decision model is set up at an edge node, training data can only be generated when that model determines that manual inspection is required. Some of the generated training data consists of cases where the model determines that manual inspection is required but the inspection necessity parameters uploaded by the staff are low. As the training continues to iterate, the model will continuously shift in the direction of not determining manual inspection, eventually leading to a loss of decision-making ability. This solution sets up two decision models simultaneously. When one decision model does not determine that manual inspection is required, the other decision model may determine that manual inspection is required. The other decision model generates training data in the opposite direction of the shift, ensuring that the decision is made and avoiding unidirectional shift.

[0046] 3. This solution has a separate real-time storage unit for each edge node, which can ensure that each edge node is trained independently and that the training data is applicable to the monitoring location corresponding to the edge node, thereby ensuring that each edge node can achieve the best judgment effect.

[0047] 4. In the iterative update phase, this scheme sorts the edge nodes by matching degree. The matching degree is determined by the difference between the inspection judgment parameter and the corresponding inspection necessity parameter, which can represent the difference between the model's output result and the actual situation. This scheme prioritizes edge nodes with large differences between the model's judgment and the actual situation by sorting the edge nodes, thus ensuring the accuracy of subsequent judgments. Attached Figure Description

[0048] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the 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.

[0049] Figure 1 This is a schematic diagram of the first embodiment of the smart power plant management method based on distributed management of the present invention;

[0050] Figure 2 This is a schematic diagram illustrating one method for constructing the iterative training dataset during the iterative update phase of the smart power plant management method based on distributed management according to the present invention.

[0051] Figure 3 This is a schematic diagram illustrating another method for constructing the iterative training dataset in the iterative update phase of the smart power plant management method based on distributed management according to the present invention. Detailed Implementation

[0052] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Rather, they are merely examples of systems and methods consistent with some aspects of the invention as detailed in the appended claims.

[0053] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The singular forms “a,” “the,” and “the” used in this invention and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0054] like Figure 1 As shown, this invention provides a smart power plant management method based on distributed management. The method is based on a distributed architecture, including a central node and edge nodes corresponding to each monitoring location. The central node is configured with a first central model based on environmental parameters and a second central model based on equipment operating parameters. The steps of the method include:

[0055] In the specific implementation process, the first center model and the second center model of the central node and the first edge model and the second edge model of each edge node have the same model structure and can all be convolutional neural network models.

[0056] This solution is applied to actual power plants, where multiple monitoring locations are set up, each equipped with a sensor to collect environmental parameters and equipment operating parameters.

[0057] Step S100, initial allocation stage: In the initial allocation stage, the central node performs initial training on the first central model and the second central model based on the initial training dataset, and distributes the model parameters of the first central model and the second central model after initial training to each edge node, so that the edge nodes can construct the first edge model and the second edge model.

[0058] In the specific implementation process, during the initial allocation phase, the first center model and the second center model are trained by pre-collecting environmental parameter training datasets and equipment operation parameter training datasets from various monitoring locations. During the training process, the mean squared error, cross-entropy, or Smooth L1 loss function can be used to calculate the loss and train the first center model and the second center model.

[0059] In some embodiments of the present invention, if the mean square error is used to calculate the loss, it is calculated based on the following formula:

[0060]

[0061] in, This represents the value of the mean square error. For the sample size, Indicates the first The true value of each sample Indicates the first The predicted value for each sample.

[0062] During the initial allocation phase, the same model parameters are distributed to each edge node, the same model parameters are used for the first edge model constructed by each edge node, and the same model parameters are used for the second edge model constructed by each edge node.

[0063] In step S200, the edge node will construct a first vector and a second vector based on the real-time collected environmental parameters and device operating parameters, and determine whether to perform manual detection based on the corresponding first edge model and second edge model.

[0064] In the specific implementation process, the environmental parameters and equipment operating parameters collected in real time are encoded to obtain a first vector and a second vector. The first edge model and the second edge model take the first vector and the second vector as inputs respectively and output inspection judgment parameters.

[0065] In the specific implementation process, the environmental parameters include ambient temperature, humidity, sound intensity, sound frequency, and concentration of harmful gases; the equipment operating parameters include equipment operating temperature and equipment vibration frequency.

[0066] In the specific implementation process, environmental parameters and equipment operating parameters are encoded into a preset length using a pre-set encoding method to obtain the first vector and the second vector.

[0067] Step S300, Iterative Update Stage: In the iterative update stage, the central node obtains the inspection results of manual detection and stores the inspection results of each monitoring location within a preset time period as an iterative training dataset. The central node iteratively trains the first central model and the second central model based on the iterative training dataset and distributes the model parameters of the iteratively trained first central model and the second central model to each edge node so that the edge nodes update the first edge model and the second edge model.

[0068] In the specific implementation process, during the iterative update phase, when iteratively training any edge node, the model parameters of the center node are updated to the current model parameters of that edge node before training is performed, ensuring that the training corresponds to the edge node.

[0069] In step S400, the edge model re-determines in real time whether to perform manual detection based on the updated first edge model and second edge model, and continues to iteratively update.

[0070] In practice, the frequency of the iterative update can be preset or activated at any time based on the activation command of the staff.

[0071] By adopting the above scheme, this scheme sets up two decision models at an edge node to ensure the safety of the corresponding monitoring location in the power plant;

[0072] Furthermore, the central node of this scheme undergoes unified training in the initial stage, and the trained model parameters are uniformly distributed to each edge node. Each edge node builds an initial model based on the same model parameters. During the judgment process, since the two models are trained with different parameters, they may produce different results at the same time. Staff members conduct manual inspections based on manual inspection instructions issued by either model at the same monitoring location, and upload the numerical value of the necessity of inspection to the central node after the manual inspection. Therefore, the two models can correspond to each other and build a dataset that differs from their own judgment results, providing a more accurate basis for subsequent iterative updates. Moreover, this scheme can train each edge node separately to ensure processing accuracy.

[0073] like Figure 2As shown, in some embodiments of the present invention, during the iterative update phase, the central node acquires the inspection results of manual detection and stores the inspection results of each monitoring location within a preset time period as an iterative training dataset:

[0074] Step S310: Obtain the instruction records of the first edge model and the second edge model in determining whether to perform manual detection;

[0075] In the specific implementation process, after completing a manual inspection, the staff uploads the inspection data and the necessary parameters for evaluating the manual inspection to the storage center of the central node, and the central node can access the data in the storage center.

[0076] Step S321: Obtain the inspection necessity parameters uploaded by the staff for each instruction record;

[0077] Step S331: Construct a training dataset by combining the inspection judgment parameters and corresponding inspection necessity parameters output by the first edge model or the second edge model when determining whether to perform manual inspection.

[0078] Step S341, and construct the training data of the instruction records corresponding to the first edge model and the second edge model into iterative training datasets respectively.

[0079] like Figure 3 As shown, in some embodiments of the present invention, during the iterative update phase, the step of the central node acquiring the inspection results of manual detection and storing the inspection results of each monitoring location within a preset time period as an iterative training dataset further includes:

[0080] Obtain the instruction records for manual detection in the first edge model and the second edge model respectively;

[0081] Step S322: Determine whether, when either the first edge model or the second edge model is manually inspected, the other first edge model or the second edge model is simultaneously manually inspected.

[0082] Step S332: If not, for the first edge model or the second edge model that has not been determined to be manually inspected, obtain the inspection judgment parameters output at this moment, and obtain the instruction record of the other first edge model or the second edge model at this moment that determines to be manually inspected, the inspection necessity parameters uploaded by the staff, and construct a training data.

[0083] The above-mentioned scheme sets up two decision models at each edge node. On the one hand, this increases the accuracy of the decision and avoids missed detections. On the other hand, if only one decision model is set up at the edge node, training data can only be generated when the model determines that manual inspection is required. Some of the generated training data are cases where the model determines that manual inspection is required but the inspection necessity parameters uploaded by the staff are low. In the process of continuous iterative training, the model will continuously shift towards not determining manual inspection, eventually leading to the loss of decision-making ability. However, this scheme sets up two decision models at the same time. When one decision model does not determine that manual inspection is required, the other decision model may determine that manual inspection is required. The other decision model generates training data in the opposite direction of the shift, ensuring that the decision is made and avoiding the occurrence of unidirectional shift.

[0084] In some embodiments of the present invention, in the step where the central node iteratively trains the first central model and the second central model based on the iterative training dataset:

[0085] The first center model is trained using the iterative training dataset corresponding to the first edge model;

[0086] The second center model is trained using the iterative training dataset corresponding to the second edge model.

[0087] In practice, the loss function used in the iterative update phase is the same as the loss function used in the initial allocation phase.

[0088] In some embodiments of the present invention, the steps of the iterative update phase further include:

[0089] Each edge node is provided with a real-time storage unit, which stores the model parameters of the first edge model or the second edge model in that edge node;

[0090] When iteratively training an edge node, the model parameters in the corresponding real-time storage unit are called and applied to the first center model and the second center model, and the model is trained using the collected iterative training dataset.

[0091] In the specific implementation process, the model parameters of the first edge model and the second edge model of each edge node can be constructed into parameter matrices and stored in the corresponding real-time storage unit.

[0092] By adopting the above scheme, each edge node is provided with a separate real-time storage unit, which can ensure that each edge node is trained separately, and that the training data is applicable to the monitoring position corresponding to the edge node, thereby ensuring that each edge node can perform the best judgment.

[0093] In some embodiments of the present invention, the steps of the iterative update phase further include:

[0094] At the beginning of each iteration update phase, the instruction records of the first edge model and the second edge model of each edge node when determining to conduct manual inspection are called, and the inspection judgment parameters output by the first edge model and the second edge model when determining to conduct manual inspection are compared with the inspection necessity parameters uploaded by the staff.

[0095] Determine whether the inspection judgment parameters of each instruction record match the corresponding inspection necessity parameters to obtain the matching degree of each edge node;

[0096] The edge nodes are sorted based on the matching degree, and each edge node is iteratively trained based on the sorting order.

[0097] In some embodiments of the present invention, in the step of determining whether the inspection judgment parameters of each instruction record match the corresponding inspection necessity parameters to obtain the matching degree of each edge node:

[0098] Calculate the difference between the inspection judgment parameter and the corresponding inspection necessity parameter, and determine whether they match based on the difference;

[0099] The ratio of the number of mismatched instruction records to the total number of instruction records is calculated as the matching degree of the edge node.

[0100] In the specific implementation process, in the step of calculating the difference between the inspection judgment parameter and the corresponding inspection necessity parameter, and determining whether they match based on the difference, the absolute value of the difference is compared with a preset difference threshold. If it is greater than the difference threshold, then they are determined to be mismatched.

[0101] In some embodiments of the present invention, in the steps of sorting the edge nodes based on the matching degree and iteratively training each edge node based on the sorting order, the edge nodes are arranged in descending order based on the matching degree, and iterative training is performed on each of the edge nodes arranged in descending order.

[0102] Using the above scheme, during the iterative update phase, the edge nodes are sorted by matching degree. The matching degree is determined by the difference between the inspection judgment parameter and the corresponding inspection necessity parameter, which can represent the difference between the model's output and the actual situation. This scheme prioritizes edge nodes with large differences between the model's judgment and the actual situation by sorting the edge nodes, thus ensuring the accuracy of subsequent judgments.

[0103] In some embodiments of the present invention, in the step of re-determining in real time whether to perform manual detection based on the updated first edge model and second edge model, the first edge model and the second edge model are determined at intervals of time:

[0104] The first edge model constructs a first vector from the environmental parameters collected in real time as input and outputs inspection judgment parameters. The inspection judgment parameters are compared with a preset judgment threshold to determine whether manual inspection should be performed.

[0105] The second edge model constructs a second vector from the real-time collected device operating parameters as input and outputs inspection judgment parameters. The inspection judgment parameters are compared with a preset judgment threshold to determine whether manual inspection should be performed.

[0106] In some embodiments of the present invention, the method further includes the following steps:

[0107] After completing a preset number of iterations, the model parameters of any two edge nodes are used to construct a decision group, and the difference between the model parameters of the two edge nodes in the decision group is calculated.

[0108] Based on the difference, determine whether two edge nodes are related edge nodes;

[0109] If two edge nodes are related edge nodes, when manual detection is performed at the monitoring location corresponding to one edge node, the monitoring location corresponding to the other related edge node will be simultaneously released to the staff.

[0110] In the specific implementation process, the cosine distance can be used to calculate the difference in model parameters between two edge nodes in the judgment group.

[0111] Using the above scheme, after completing a preset number of iterations, the differences in model parameters of each edge node will theoretically become larger and larger. However, if the differences in model parameters between two edge nodes are small, it indicates that there is a certain correlation between them. Therefore, when manually inspecting the monitoring location corresponding to one edge node, this scheme will simultaneously release the monitoring location corresponding to another related edge node to the staff. Based on the correlation, the staff will conduct synchronous inspections to further ensure the safety of the power plant.

[0112] Another aspect of the present invention relates to a smart power plant management system based on distributed management. The system includes a computer device, which includes a processor and a memory. The memory stores computer instructions, and the processor is used to execute the computer instructions stored in the memory. When the computer instructions are executed by the processor, the system implements the steps of the method.

[0113] This invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the aforementioned smart power plant management method based on distributed management. The computer-readable storage medium can be a tangible storage medium, such as random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, register, floppy disk, hard disk, removable storage disk, CD-ROM, or any other form of storage medium known in the art.

[0114] Those skilled in the art will understand that the exemplary components, systems, and methods described in conjunction with the embodiments disclosed herein can be implemented in hardware, software, or a combination of both. Whether implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this invention. When implemented in hardware, it can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this invention are programs or code segments used to perform the desired tasks. The programs or code segments can be stored in a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried in a carrier wave.

[0115] It should be clarified that the present invention is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present invention is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of the present invention.

[0116] In this invention, features described and / or illustrated for one embodiment may be used in the same or similar manner in one or more other embodiments, and / or combined with or in place of features of other embodiments.

[0117] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, various modifications and variations of the embodiments of the present invention are possible. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A smart power plant management method based on distributed management, characterized in that, The method is based on a distributed architecture, including a central node and edge nodes corresponding one-to-one with each monitoring location. The central node is configured with a first central model based on environmental parameters and a second central model based on device operating parameters. The steps of the method include: In the initial allocation phase, the central node performs initial training on the first central model and the second central model based on the initial training dataset, and distributes the model parameters of the first central model and the second central model after initial training to each edge node so that the edge nodes can construct the first edge model and the second edge model. The edge nodes will construct a first vector and a second vector based on real-time collected environmental parameters and device operating parameters, and determine whether to perform manual detection based on the corresponding first edge model and second edge model. In the iterative update phase, the central node acquires the inspection results of manual inspections and stores the inspection results of each monitoring location within a preset time period as an iterative training dataset. It also acquires the instruction records of the first edge model and the second edge model when determining to conduct manual inspections; acquires the inspection necessity parameters uploaded by the staff corresponding to each instruction record; constructs a training dataset by combining the inspection judgment parameters output by the first edge model or the second edge model when determining to conduct manual inspections with the corresponding inspection necessity parameters; and constructs iterative training datasets by combining the training data corresponding to the instruction records of the first edge model and the second edge model; and determines the judgment of either the first edge model or the second edge model. During manual inspection, it is determined whether another first edge model or second edge model simultaneously performs manual inspection. If not, the inspection judgment parameters output at that moment are obtained for the first edge model or second edge model that did not perform manual inspection, and the instruction record of the other first edge model or second edge model that performed manual inspection at that moment is obtained, along with the inspection necessity parameters uploaded by the staff, to construct a training dataset. The central node iteratively trains the first central model and the second central model based on the iterative training dataset, and distributes the model parameters of the iteratively trained first central model and the second central model to each edge node so that the edge nodes update the first edge model and the second edge model. The edge model is based on the updated first edge model and second edge model to re-determine in real time whether to perform manual detection, and is continuously iterated and updated.

2. The smart power plant management method based on distributed management according to claim 1, characterized in that, In the step where the central node iteratively trains the first central model and the second central model based on the iterative training dataset: The first center model is trained using the iterative training dataset corresponding to the first edge model; The second center model is trained using the iterative training dataset corresponding to the second edge model.

3. The smart power plant management method based on distributed management according to claim 1, characterized in that, The steps of the iterative update phase also include: Each edge node is provided with a real-time storage unit, which stores the model parameters of the first edge model or the second edge model in that edge node; When iteratively training an edge node, the model parameters in the corresponding real-time storage unit are called and applied to the first center model and the second center model, and the model is trained using the collected iterative training dataset.

4. The smart power plant management method based on distributed management according to claim 1, characterized in that, The steps of the iterative update phase also include: At the beginning of each iteration update phase, the instruction records of the first edge model and the second edge model of each edge node when determining to conduct manual inspection are called, and the inspection judgment parameters output by the first edge model and the second edge model when determining to conduct manual inspection are compared with the inspection necessity parameters uploaded by the staff. Determine whether the inspection judgment parameters of each instruction record match the corresponding inspection necessity parameters to obtain the matching degree of each edge node; The edge nodes are sorted based on the matching degree; Iterative training is performed on each edge node based on the sorting order.

5. The smart power plant management method based on distributed management according to claim 4, characterized in that, In the step of determining whether the inspection judgment parameters of each instruction record match the corresponding inspection necessity parameters to obtain the matching degree of each edge node: Calculate the difference between the inspection judgment parameter and the corresponding inspection necessity parameter, and determine whether they match based on the difference; The ratio of the number of mismatched instruction records to the total number of instruction records is calculated as the matching degree of the edge node.

6. The smart power plant management method based on distributed management according to claim 4, characterized in that, In the steps of sorting the edge nodes based on the matching degree and iteratively training each edge node based on the sorting order, the edge nodes are arranged in descending order based on the matching degree, and iterative training is performed on each of the edge nodes arranged in descending order.

7. The smart power plant management method based on distributed management according to any one of claims 1-6, characterized in that, The method further includes the following steps: After completing a preset number of iterations, the model parameters of any two edge nodes are used to construct a decision group, and the difference between the model parameters of the two edge nodes in the decision group is calculated. Based on the difference, determine whether two edge nodes are related edge nodes; If two edge nodes are related edge nodes, when manual detection is performed at the monitoring location corresponding to one edge node, the monitoring location corresponding to the other related edge node will be simultaneously released to the staff.

8. A smart power plant management system based on distributed management, characterized in that: The system includes a computer device, which includes a processor and a memory. The memory stores computer instructions, and the processor executes the computer instructions stored in the memory. When the computer instructions are executed by the processor, the system implements the steps of the method according to any one of claims 1-7.

Images