Cloud-side cooperative fault diagnosis method and system for intelligent substation equipment

Abstract

The invention discloses a cloud-side cooperative fault diagnosis method for intelligent substation equipment. The method comprises the following steps: acquiring equipment data information of each intelligent substation in a target power system and preprocessing the equipment data information to obtain a training data set; each intelligent substation constructs a local fault diagnosis model; uploading model parameters of the current local fault diagnosis model to a server; the cloud server obtains model parameters of the global fault diagnosis model based on a federated learning scheme and issues the model parameters to the intelligent substations; the intelligent substations train and update local fault diagnosis models according to the received model parameters; repeating the above steps until a set condition is reached, and obtaining a final local fault diagnosis model of each intelligent substation; and adopting the final local fault diagnosis model to carry out cloud-side cooperative fault diagnosis on the intelligent substation equipment. The invention also discloses a system for realizing the cloud-side collaborative fault diagnosis method for the intelligent substation equipment. The method is high in reliability, good in accuracy and high in efficiency.

Description

Technical Field

[0001] The present invention belongs to the field of electrical automation, and specifically relates to a cloud-edge collaborative fault diagnosis method and system for smart substation equipment. Background Art

[0002] With the development of economy and technology and the improvement of people's living standards, electricity has become an indispensable secondary energy source in people's production and life, bringing endless convenience to people's production and life. Therefore, ensuring the stable and reliable supply of electricity has become one of the most important tasks of the power system.

[0003] At present, the intelligent development of power system is in full swing. With the development of intelligent power system, the importance of smart substation in power system is becoming more and more important. Therefore, fault diagnosis of smart substation equipment is particularly important.

[0004] At present, the traditional fault diagnosis scheme of smart substations still adopts manual inspection and manual diagnosis. However, this method is not only time-consuming and labor-intensive, but also has poor reliability and accuracy. At present, although some researchers have proposed fault diagnosis schemes based on artificial intelligence algorithms, such schemes are relatively complex and have poor efficiency in the actual implementation of smart substations. Summary of the invention

[0005] One of the purposes of the present invention is to provide a cloud-edge collaborative fault diagnosis method for smart substation equipment with high reliability, good accuracy and high efficiency.

[0006] A second objective of the present invention is to provide a system for implementing the cloud-edge collaborative fault diagnosis method of the smart substation equipment.

[0007] The cloud-edge collaborative fault diagnosis method for the smart substation equipment provided by the present invention comprises the following steps:

[0008] S1. Obtain data information of equipment in each smart substation in the target power system;

[0009] S2. Preprocess the data information obtained in step S1 to obtain a training data set;

[0010] S3. Based on the RCNN network, knowledge graph and text generation model, each smart substation builds its own local fault diagnosis model;

[0011] S4. Each smart substation uploads the model parameters of its current local fault diagnosis model to the cloud server;

[0012] S5. The cloud server obtains the model parameters of the global fault diagnosis model based on the received model data and the federated learning solution, and sends them to each smart substation;

[0013] S6. Each smart substation uses the local training data set to train and update the local fault diagnosis model according to the received model parameters;

[0014] S7. Repeat steps S4 to S6 until the set conditions are met to obtain the final local fault diagnosis model of each smart substation;

[0015] S8. Each smart substation adopts the final local fault diagnosis model obtained in step S7 to perform cloud-edge collaborative fault diagnosis of smart substation equipment.

[0016] The step S1 of acquiring the data information of the equipment of each smart substation in the target power system specifically includes the following steps:

[0017] Obtain multimodal data information of the equipment in each smart substation in the target power system;

[0018] The multimodal data information includes text data information of the device, image data information of the device, video data information of the device, audio data information of the device and fault diagnosis data information of the device.

[0019] Step S2 preprocesses the data information obtained in step S1 to obtain a training data set, which specifically includes the following steps:

[0020] The preprocessing includes data filtering, data normalization and data enhancement;

[0021] Use the following formula to normalize the numerical data:

[0022]

[0023] In the formula, v(x i ) is the normalized data; x i is the data before normalization; min(X) is the data x before normalization i The minimum value of; max(X) is the data x before normalization i The maximum value of

[0024] Scale the image data to ensure that the ratio of the shortest side length in each patch to the preset value meets the set requirements; if the scaling cannot correspond to the longest side length in each patch image, fill it with 0 to ensure that the ratio meets the set requirements;

[0025] The following formula is used to enhance the image data:

[0026]

[0027] Where B H is the relative height of the marked box after data enhancement; B W is the relative width of the marked box after data enhancement; B h is the original height; B w is the original width; H is the height enhancement coefficient; h is the width enhancement coefficient.

[0028] In step S3, each smart substation builds its own local fault diagnosis model based on the RCNN network, knowledge graph and text generation model, which specifically includes the following steps:

[0029] Based on the RCNN network, knowledge graph and text generation model, each smart substation builds its own local fault diagnosis model;

[0030] The local fault diagnosis model includes a fault classification module, a fault derivation module and a fault diagnosis module;

[0031] A fault classification module is constructed based on the RCNN network; the fault classification module is used to extract feature information and perform fault classification based on the input image data information; the fault classification module includes a feature extraction submodule, an RPN submodule, a RoIPooling submodule and an RCNN submodule; the feature extraction submodule, the RPN submodule, the RoIPooling submodule and the RCNN submodule are connected in series in sequence; the feature extraction submodule includes 13 convolutional layers, 13 Relu layers and 4 pooling layers connected in series in sequence, the feature extraction submodule is used to extract the feature map of the input image and upload it to the RPN submodule; the RPN submodule is used to select the target candidate area; the RoI Pooling submodule is used to integrate the feature map output by the feature extraction submodule and the target candidate area output by the RPN submodule to obtain a fixed-size regional feature map and upload it to the RCNN submodule; the RCNN submodule is used to classify the uploaded regional feature map through the softmax function, and correct the state of the object through bounding box regression;

[0032] A fault deduction module is constructed based on the knowledge graph; the fault deduction module is used to construct a multimodal knowledge graph and perform fault deduction based on the output of the fault classification module and the acquired text data information, video data information and audio data information; the constructed multimodal knowledge graph includes information representation, knowledge processing and knowledge update; in the reasoning stage, the multimodal knowledge graph obtains an entity information vector based on the obtained feature vector; after obtaining the text feature vector representing the test information in the target scenario, the text feature vector is mapped to the entity, attribute and relationship in the knowledge base, and the feature vector obtained by the fault classification module is mapped to the entity-relationship-attribute vector in the knowledge graph using cosine similarity; the cosine similarity cosθ is calculated as Where cosθ is the cosine similarity; A i is the component of vector A; B i is the component of vector B; n is the total number of components; matching is performed based on cosine similarity to obtain entity-text vectors corresponding to entities and attribute values ​​in the multimodal knowledge graph; triples are formed using the primary device entity, secondary device entity, and fault phenomenon attribute value in the knowledge graph; the fault entity corresponding to the secondary device attribute value is determined through semantic matching templates, path query, and query reasoning;

[0033] A fault diagnosis module is constructed based on the text generation model; the fault diagnosis module is used for the final fault diagnosis; after retrieving the query node from the knowledge graph, the answer text is generated using the feature vector obtained by visual object detection using the node name; the detection results are sorted according to the confidence level to obtain a semi-ordered state of the text feature vector; the text feature vector is semantically matched with the name text of the node in the knowledge graph; the sorting operation is performed based on the matched node attributes to generate an ordered text vector; the text generation model uses the generated text vector to generate an interpretable fault text, thereby completing the final fault diagnosis.

[0034] The cloud server in step S5 obtains the model parameters of the global fault diagnosis model based on the received model data and the federated learning scheme, and sends them to each smart substation, which specifically includes the following steps:

[0035] Based on the received model data and the FedAvg algorithm, the cloud server uses the following formula to perform weighted aggregation on the received model data of each smart substation:

[0036]

[0037] Where w t+1 is the model parameter of the global model updated in the t+1th round; w t is the model parameter of the global model updated in the tth round; λ is the learning rate; h kis the ratio of the number of samples in the local training data set of the kth smart substation to the number of samples in all training data sets of all smart substations; is the model parameter of the local fault diagnosis model of the k-th smart substation during the t-th round of training update;

[0038] After the weighted aggregation is completed, the cloud server sends the model parameters of the global fault diagnosis model obtained through collaborative training to each smart substation.

[0039] Each smart substation described in step S6 uses a local training data set to train and update a local fault diagnosis model according to the received model parameters, which specifically includes the following steps:

[0040] Each smart substation trains its own local fault diagnosis model using the local training data set based on the received model parameters;

[0041] During training, the following formula is used as the loss function:

[0042]

[0043] Where L k is the local training loss function value of the kth smart substation; m k is the number of samples in the local training data set of the kth smart substation; c is the number of sample categories; y ij To indicate that sample i belongs to category j; p(x ij ) is the predicted probability that sample i belongs to category j;

[0044] During the training process, the stochastic gradient descent algorithm is used for training.

[0045] Step S7 repeats steps S4 to S6 until the set condition is met, and specifically includes the following steps:

[0046] Repeat steps S4 to S6 until the following objective function is achieved:

[0047]

[0048] Where nn is the total number of smart substations.

[0049] The present invention also provides a system for implementing the cloud-edge collaborative fault diagnosis method for the smart substation equipment, including a data acquisition module, a data processing module, a model construction module, a parameter upload module, a parameter fusion module, a local training module, a collaborative training module and a fault diagnosis module; the data acquisition module, the data processing module, the model construction module, the parameter upload module, the parameter fusion module, the local training module, the collaborative training module and the fault diagnosis module are connected in series in sequence; the data acquisition module is used to acquire data information of equipment of each smart substation in the target power system, and upload the data information to the data processing module; the data processing module is used to pre-process the acquired data information according to the received data information to obtain a training data set, and upload the data information to the model construction module; the model construction module is used to generate a model based on the RCNN network, the knowledge graph and the text according to the received data information, and each smart substation builds its own local fault diagnosis model, and uploads the data information to the parameter upload module; the parameter upload module is used to generate a model based on the received data information, and each smart substation builds its own local fault diagnosis model according to the received data information, and uploads the data information to the parameter upload module; The model parameters of the current local fault diagnosis model are uploaded to the cloud server, and the data information is uploaded to the parameter fusion module; the parameter fusion module is used to obtain the model parameters of the global fault diagnosis model based on the received data information, and the cloud server obtains the model data based on the federated learning scheme, and sends it to each smart substation, and uploads the data information to the local training module; the local training module is used to train and update the local fault diagnosis model based on the received data information, and each smart substation uses the local training data set according to the received model parameters, and uploads the data information to the collaborative training module; the collaborative training module is used to repeat the work of the parameter upload module, the parameter fusion module and the local training module according to the received data information until the set conditions are met, and obtain the final local fault diagnosis model of each smart substation, and upload the data information to the fault diagnosis module; the fault diagnosis module is used to control each smart substation to adopt the final local fault diagnosis model according to the received data information, and perform cloud-edge collaborative fault diagnosis of smart substation equipment.

[0050] The cloud-edge collaborative fault diagnosis method and system for smart substation equipment provided by the present invention constructs a local fault diagnosis model of the smart substation through an RCNN network, a knowledge graph and a text generation model, and performs cloud-edge collaborative training of the local fault diagnosis models of each smart substation through a cloud server. Therefore, the present invention can not only realize cloud-edge collaborative fault diagnosis of smart substation equipment, but also has higher reliability, better accuracy and higher efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 The figure is a schematic diagram of the method flow of the present invention.

[0052] Figure 2 Schematic diagram of the functional modules of the system of the present invention. DETAILED DESCRIPTION

[0053] like Figure 1 The method flow diagram of the method of the present invention is shown as follows: the cloud-edge collaborative fault diagnosis method for smart substation equipment provided by the present invention comprises the following steps:

[0054] S1. Obtaining data information of equipment in each smart substation in the target power system; specifically including the following steps:

[0055] Obtain multimodal data information of the equipment in each smart substation in the target power system;

[0056] The multimodal data information includes text data information of the device, image data information of the device, video data information of the device, audio data information of the device and fault diagnosis data information of the device;

[0057] S2. Preprocess the data information obtained in step S1 to obtain a training data set; specifically, the steps include:

[0058] The preprocessing includes data filtering, data normalization and data enhancement;

[0059] Use the following formula to normalize the numerical data:

[0060]

[0061] In the formula, v(x i ) is the normalized data; x i is the data before normalization; min(X) is the data x before normalization i The minimum value of; max(X) is the data x before normalization i The maximum value of

[0062] Scale the image data to ensure that the ratio of the shortest side length in each patch to the preset value meets the set requirements; if the scaling cannot correspond to the longest side length in each patch image, fill it with 0 to ensure that the ratio meets the set requirements;

[0063] The following formula is used to enhance the image data:

[0064]

[0065] Where B H is the relative height of the marked box after data enhancement; B W is the relative width of the marked box after data enhancement; B h is the original height; B wis the original width; H is the height enhancement coefficient; h is the width enhancement coefficient;

[0066] S3. Based on the RCNN network, knowledge graph and text generation model, each smart substation builds its own local fault diagnosis model; specifically, the following steps are included:

[0067] Based on the RCNN network, knowledge graph and text generation model, each smart substation builds its own local fault diagnosis model;

[0068] The local fault diagnosis model includes a fault classification module, a fault derivation module and a fault diagnosis module;

[0069] A fault classification module is constructed based on the RCNN network; the fault classification module is used to extract feature information and perform fault classification based on the input image data information; the fault classification module includes a feature extraction submodule, an RPN submodule, a RoIPooling submodule and an RCNN submodule; the feature extraction submodule, the RPN submodule, the RoIPooling submodule and the RCNN submodule are connected in series in sequence; the feature extraction submodule includes 13 convolutional layers, 13 Relu layers and 4 pooling layers connected in series in sequence, the feature extraction submodule is used to extract the feature map of the input image and upload it to the RPN submodule; the RPN submodule is used to select the target candidate area; the RoI Pooling submodule is used to integrate the feature map output by the feature extraction submodule and the target candidate area output by the RPN submodule to obtain a fixed-size regional feature map and upload it to the RCNN submodule; the RCNN submodule is used to classify the uploaded regional feature map through the softmax function, and correct the state of the object through bounding box regression;

[0070] A fault deduction module is constructed based on the knowledge graph; the fault deduction module is used to construct a multimodal knowledge graph and perform fault deduction based on the output of the fault classification module and the acquired text data information, video data information and audio data information; the constructed multimodal knowledge graph includes information representation, knowledge processing and knowledge update; in the reasoning stage, the multimodal knowledge graph obtains an entity information vector based on the obtained feature vector; after obtaining the text feature vector representing the test information in the target scenario, the text feature vector is mapped to the entity, attribute and relationship in the knowledge base, and the feature vector obtained by the fault classification module is mapped to the entity-relationship-attribute vector in the knowledge graph using cosine similarity; the cosine similarity cosθ is calculated as Where cosθ is the cosine similarity; A i is the component of vector A; B iis the component of vector B; n is the total number of components; matching is performed based on cosine similarity to obtain entity-text vectors corresponding to entities and attribute values ​​in the multimodal knowledge graph; triples are formed using the primary device entity, secondary device entity, and fault phenomenon attribute value in the knowledge graph; the fault entity corresponding to the secondary device attribute value is determined through semantic matching templates, path query, and query reasoning;

[0071] A fault diagnosis module is constructed based on the text generation model; the fault diagnosis module is used for the final fault diagnosis; after retrieving the query node from the knowledge graph, the answer text is generated using the feature vector obtained by visual object detection using the node name; the detection results are sorted according to the confidence level to obtain a semi-ordered state of the text feature vector; the text feature vector is semantically matched with the name text of the node in the knowledge graph; the sorting operation is performed based on the matched node attributes to generate an ordered text vector; the text generation model uses the generated text vector to generate an interpretable fault text, thereby completing the final fault diagnosis;

[0072] In specific implementation, the fault diagnosis module is built based on the BART text generation model; BART is a denoising autoencoder built using a sequence-to-sequence model that can be applied to various text generation tasks; it uses a standard Transformer-based neural machine translation architecture that corrupts the input text using a noise function and learns to reconstruct the original text into a sequence-to-sequence model; this approach can flexibly process the original input text and effectively learn text reconstruction;

[0073] S4. Each smart substation uploads the model parameters of its current local fault diagnosis model to the cloud server;

[0074] S5. The cloud server obtains the model parameters of the global fault diagnosis model based on the received model data and the federated learning solution, and sends them to each smart substation; specifically, the steps include:

[0075] Based on the received model data and the FedAvg algorithm, the cloud server uses the following formula to perform weighted aggregation on the received model data of each smart substation:

[0076]

[0077] Where w t+1 is the model parameter of the global model updated in the t+1th round; w t is the model parameter of the global model updated in the tth round; λ is the learning rate; h k is the ratio of the number of samples in the local training data set of the kth smart substation to the number of samples in all training data sets of all smart substations; is the model parameter of the local fault diagnosis model of the k-th smart substation during the t-th round of training update;

[0078] After the weighted aggregation is completed, the cloud server sends the model parameters of the global fault diagnosis model obtained through collaborative training to each smart substation;

[0079] S6. Each smart substation uses the local training data set to train and update the local fault diagnosis model according to the received model parameters; specifically, the steps include:

[0080] Each smart substation trains its own local fault diagnosis model using the local training data set based on the received model parameters;

[0081] During training, the following formula is used as the loss function:

[0082]

[0083] Where L k is the local training loss function value of the kth smart substation; m k is the number of samples in the local training data set of the kth smart substation; c is the number of sample categories; y ij To indicate that sample i belongs to category j; p(x ij ) is the predicted probability that sample i belongs to category j;

[0084] During the training process, the stochastic gradient descent algorithm is used for training;

[0085] S7. Repeat steps S4 to S6 until the set conditions are met to obtain the final local fault diagnosis model of each smart substation; specifically comprising the following steps:

[0086] Repeat steps S4 to S6 until the following objective function is achieved:

[0087]

[0088] Where nn is the total number of smart substations;

[0089] S8. Each smart substation adopts the final local fault diagnosis model obtained in step S7 to perform cloud-edge collaborative fault diagnosis of smart substation equipment.

[0090] like Figure 2The figure shows a schematic diagram of the functional modules of the system of the present invention: the system disclosed in the present invention for realizing the cloud-edge collaborative fault diagnosis method of the smart substation equipment comprises a data acquisition module, a data processing module, a model construction module, a parameter upload module, a parameter fusion module, a local training module, a collaborative training module and a fault diagnosis module; the data acquisition module, the data processing module, the model construction module, the parameter upload module, the parameter fusion module, the local training module, the collaborative training module and the fault diagnosis module are connected in series in sequence; the data acquisition module is used to acquire the data information of the equipment of each smart substation in the target power system, and upload the data information to the data processing module; the data processing module is used to pre-process the acquired data information according to the received data information, obtain the training data set, and upload the data information to the model construction module; the model construction module is used to generate a model based on the received data information, each smart substation builds its own local fault diagnosis model, and uploads the data information to the parameter upload module; the parameter upload module is used to generate a model based on the received data information, Each smart substation uploads the model parameters of its current local fault diagnosis model to the cloud server, and uploads the data information to the parameter fusion module; the parameter fusion module is used to obtain the model parameters of the global fault diagnosis model based on the received data information, and the cloud server obtains the model data based on the federated learning scheme, and sends it to each smart substation, and uploads the data information to the local training module; the local training module is used to train and update the local fault diagnosis model based on the received data information, and each smart substation uses the local training data set according to the received model parameters, and uploads the data information to the collaborative training module; the collaborative training module is used to repeat the work of the parameter upload module, the parameter fusion module and the local training module according to the received data information until the set conditions are met, and obtain the final local fault diagnosis model of each smart substation, and upload the data information to the fault diagnosis module; the fault diagnosis module is used to control each smart substation to adopt the final local fault diagnosis model according to the received data information, and perform cloud-edge collaborative fault diagnosis of smart substation equipment.

Claims

1. A cloud-edge collaborative fault diagnosis method for smart substation equipment, comprising the following steps: S1. Obtain data information of equipment in each smart substation in the target power system; S2. Preprocess the data information obtained in step S1 to obtain a training data set; S3. Based on the RCNN network, knowledge graph and text generation model, each smart substation builds its own local fault diagnosis model; S4. Each smart substation uploads the model parameters of its current local fault diagnosis model to the cloud server; S5. The cloud server obtains the model parameters of the global fault diagnosis model based on the received model data and the federated learning solution, and sends them to each smart substation; S6. Each smart substation uses the local training data set to train and update the local fault diagnosis model according to the received model parameters; S7. Repeat steps S4 to S6 until the set conditions are met to obtain the final local fault diagnosis model of each smart substation; S8. Each smart substation adopts the final local fault diagnosis model obtained in step S7 to perform cloud-edge collaborative fault diagnosis of smart substation equipment.

2. The cloud-edge collaborative fault diagnosis method for smart substation equipment according to claim 1 is characterized in that The step S1 of acquiring the data information of the equipment of each smart substation in the target power system specifically includes the following steps: Obtain multimodal data information of the equipment in each smart substation in the target power system; The multimodal data information includes text data information of the device, image data information of the device, video data information of the device, audio data information of the device and fault diagnosis data information of the device.

3. The cloud-edge collaborative fault diagnosis method for smart substation equipment according to claim 2 is characterized in that Step S2 preprocesses the data information obtained in step S1 to obtain a training data set, which specifically includes the following steps: The preprocessing includes data filtering, data normalization and data enhancement; Use the following formula to normalize the numerical data: In the formula, v(x i ) is the normalized data; x i is the data before normalization; min(X) is the data x before normalization i The minimum value of; max(X) is the data x before normalization i The maximum value of Scale the image data to ensure that the ratio of the shortest side length in each patch to the preset value meets the set requirements; if the scaling cannot correspond to the longest side length in each patch image, fill it with 0 to ensure that the ratio meets the set requirements; The following formula is used to enhance the image data: Where B H is the relative height of the marked box after data enhancement; B W is the relative width of the marked box after data enhancement; B h is the original height; B w is the original width; H is the height enhancement coefficient; h is the width enhancement coefficient.

4. The cloud-edge collaborative fault diagnosis method for smart substation equipment according to claim 3 is characterized in that In step S3, each smart substation builds its own local fault diagnosis model based on the RCNN network, knowledge graph and text generation model, which specifically includes the following steps: Based on the RCNN network, knowledge graph and text generation model, each smart substation builds its own local fault diagnosis model; The local fault diagnosis model includes a fault classification module, a fault derivation module and a fault diagnosis module; Build a fault classification module based on the RCNN network; The fault classification module is used to extract feature information and perform fault classification based on the input image data information; the fault classification module includes a feature extraction submodule, an RPN submodule, a RoIPooling submodule and an RCNN submodule; the feature extraction submodule, the RPN submodule, the RoIPooling submodule and the RCNN submodule are connected in series in sequence; the feature extraction submodule includes 13 convolutional layers, 13 Relu layers and 4 pooling layers connected in series in sequence, the feature extraction submodule is used to extract the feature map of the input image and upload it to the RPN submodule; the RPN submodule is used to select the target candidate area; the RoI Pooling submodule is used to integrate the feature map output by the feature extraction submodule and the target candidate area output by the RPN submodule to obtain a fixed-size regional feature map, and upload it to the RCNN submodule; The RCNN submodule is used to classify the uploaded regional feature maps through the softmax function and correct the state of the object through bounding box regression; Build a fault deduction module based on the knowledge graph; The fault deduction module is used to construct a multimodal knowledge graph and perform fault deduction based on the output of the fault classification module and the acquired text data information, video data information and audio data information; the constructed multimodal knowledge graph includes information representation, knowledge processing and knowledge update; in the reasoning stage, the multimodal knowledge graph obtains the entity information vector based on the obtained feature vector; after obtaining the text feature vector representing the test information in the target scene, the text feature vector is mapped with the entity, attribute and relationship in the knowledge base, and the feature vector obtained by the fault classification module is mapped to the entity-relationship-attribute vector in the knowledge graph using cosine similarity; the cosine similarity cosθ is calculated as Where cosθ is the cosine similarity; A i is the component of vector A; B i is the component of vector B; n is the total number of components; matching is performed based on cosine similarity to obtain entity-text vectors corresponding to entities and attribute values ​​in the multimodal knowledge graph; triples are formed using the primary device entity, secondary device entity, and fault phenomenon attribute value in the knowledge graph; the fault entity corresponding to the secondary device attribute value is determined through semantic matching templates, path query, and query reasoning; A fault diagnosis module is constructed based on the text generation model; the fault diagnosis module is used for the final fault diagnosis; after retrieving the query node from the knowledge graph, the answer text is generated using the feature vector obtained by visual object detection using the node name; the detection results are sorted according to the confidence level to obtain a semi-ordered state of the text feature vector; the text feature vector is semantically matched with the name text of the node in the knowledge graph; the sorting operation is performed based on the matched node attributes to generate an ordered text vector; the text generation model uses the generated text vector to generate an interpretable fault text, thereby completing the final fault diagnosis.

5. The cloud-edge collaborative fault diagnosis method for smart substation equipment according to claim 4 is characterized in that The cloud server in step S5 obtains the model parameters of the global fault diagnosis model based on the received model data and the federated learning scheme, and sends them to each smart substation, which specifically includes the following steps: Based on the received model data and the FedAvg algorithm, the cloud server uses the following formula to perform weighted aggregation on the received model data of each smart substation: Where w t+1 is the model parameter of the global model updated in the t+1th round; w t is the model parameter of the global model updated in the tth round; λ is the learning rate; h k is the ratio of the number of samples in the local training data set of the kth smart substation to the number of samples in all training data sets of all smart substations; are the model parameters of the local fault diagnosis model of the k-th smart substation during the t-th round of training update; After the weighted aggregation is completed, the cloud server sends the model parameters of the global fault diagnosis model obtained through collaborative training to each smart substation.

6. The cloud-edge collaborative fault diagnosis method for smart substation equipment according to claim 5 is characterized in that Each smart substation described in step S6 uses a local training data set to train and update a local fault diagnosis model according to the received model parameters, which specifically includes the following steps: Each smart substation trains its own local fault diagnosis model using the local training data set based on the received model parameters; During training, the following formula is used as the loss function: Where L k is the local training loss function value of the kth smart substation; m k is the number of samples in the local training data set of the kth smart substation; c is the number of sample categories; y ij To indicate that sample i belongs to category j; p(x ij ) is the predicted probability that sample i belongs to category j; During the training process, the stochastic gradient descent algorithm is used for training.

7. The cloud-edge collaborative fault diagnosis method for smart substation equipment according to claim 6 is characterized in that Step S7 repeats steps S4 to S6 until the set condition is met, and specifically includes the following steps: Repeat steps S4 to S6 until the following objective function is achieved: Where nn is the total number of smart substations.

8. A system for implementing the cloud-edge collaborative fault diagnosis method for smart substation equipment according to any one of claims 1 to 7, characterized in that It includes a data acquisition module, a data processing module, a model building module, a parameter upload module, a parameter fusion module, a local training module, a collaborative training module and a fault diagnosis module; the data acquisition module, the data processing module, the model building module, the parameter upload module, the parameter fusion module, the local training module, the collaborative training module and the fault diagnosis module are connected in series in sequence; the data acquisition module is used to obtain the data information of the equipment of each smart substation in the target power system, and upload the data information to the data processing module; The data processing module is used to pre-process the acquired data information according to the received data information, obtain the training data set, and upload the data information to the model construction module; The model building module is used to generate models based on the RCNN network, knowledge graph and text according to the received data information. Each smart substation builds its own local fault diagnosis model and uploads the data information to the parameter uploading module; The parameter upload module is used for each smart substation to upload the model parameters of its current local fault diagnosis model to the cloud server according to the received data information, and upload the data information to the parameter fusion module; The parameter fusion module is used to obtain the model parameters of the global fault diagnosis model based on the received data information and the model data received by the cloud server based on the federated learning solution, and send them to each smart substation, and upload the data information to the local training module; The local training module is used to train and update the local fault diagnosis model based on the received data information and the model parameters received by each smart substation using the local training data set, and upload the data information to the collaborative training module; The collaborative training module is used to repeat the work of the parameter upload module, parameter fusion module and local training module according to the received data information until the set conditions are met, obtain the final local fault diagnosis model of each smart substation, and upload the data information to the fault diagnosis module; the fault diagnosis module is used to control each smart substation to adopt the final local fault diagnosis model according to the received data information, and perform cloud-edge collaborative fault diagnosis of smart substation equipment.

Images