CN117589444B

CN117589444B - Wind driven generator gear box fault diagnosis method based on federal learning

Info

Publication number: CN117589444B
Application number: CN202410072316.5A
Authority: CN
Inventors: 张萍; 陶洁; 赵志磊; 肖钊; 邱海文; 高贵兵; 邓杰文
Original assignee: Hunan University of Science and Technology
Current assignee: Hunan University of Science and Technology
Priority date: 2024-01-18
Filing date: 2024-01-18
Publication date: 2024-04-02
Anticipated expiration: 2044-01-18
Also published as: CN117589444A

Abstract

The application discloses a wind driven generator gear box fault diagnosis method based on federal learning, and belongs to the field of data management. The method comprises the following steps: collecting local data of each wind driven generator gear box, and performing local model training; receiving a global model transmitted by a central terminal, generating a fault diagnosis local model, testing, and transmitting to the central terminal if the fault diagnosis local model reaches a standard; receiving a fault diagnosis global model transmitted by a central terminal, monitoring the running condition of each wind driven generator gearbox in real time, and transmitting the determined wind driven generator gearbox identification and the determined abnormal error information to a control center if the abnormal error information is identified; and receiving fault sample data generated by the control center, and updating the fault diagnosis local model. The scheme can generate a fault diagnosis local model locally, can adapt to a new fault mode more quickly, and reduces the calculation pressure and communication cost of a central terminal; by updating the fault diagnosis local model, the accuracy and reliability of the model can be improved.

Description

Wind driven generator gear box fault diagnosis method based on federal learning

Technical Field

The application relates to the field of fault diagnosis, in particular to a wind driven generator gearbox fault diagnosis method based on federal learning.

Background

Wind turbine gearboxes are one of the key components in wind power systems, the proper operation of which is critical to the reliability of the system. The fault diagnosis can discover potential problems in time, reduce sudden faults and improve the reliability of the system.

In the prior art, a large amount of sensor data needs to be acquired, and then the characteristics in the sensor data are directly analyzed by using a signal processing method, so that whether the wind driven generator gearbox fails or not is judged.

In the prior art, however, signal processing methods are generally based on the definition of specific signal characteristics and may lack generalization capability. These methods may be difficult to accommodate when faced with complex and diverse operating environments and failure modes.

Disclosure of Invention

In order to overcome the defects, the embodiment of the application provides a wind driven generator gearbox fault diagnosis method based on federal learning, which solves the problems that in the prior art, a signal processing method is used for judging the fault of a wind driven generator gearbox, and the problem that the fault is difficult to test complex and various working environments and fault modes due to the fact that the fault is generally based on definition of specific signal characteristics and further lacks generalization capability.

In a first aspect, embodiments of the present application provide a method for diagnosing a failure of a gearbox of a wind turbine generator based on federal learning, the method being performed by a client, the method comprising:

collecting local data of each wind driven generator gearbox, performing local model training according to the local data to obtain a first local update model, and transmitting the first local update model to a central end for the central end to train according to the first local update model to obtain a global model;

receiving a global model transmitted by the central terminal, and generating a fault diagnosis local model according to the first local updating model, the global model and a preset fault diagnosis local model determining formula; the preset fault diagnosis local model determining formula is as follows:

；

wherein,representing the value of the fault diagnosis local model after the t+1st iteration; n represents the number of first locally updated models; i is the index of the first local update model; />The initial value of the fault diagnosis local model for the t+1st iteration is represented.

Testing the fault diagnosis local model according to the local data, and if the fault diagnosis local model reaches a preset fault diagnosis local model evaluation standard, transmitting the fault diagnosis local model to a central terminal for training by the central terminal according to the fault diagnosis local model to obtain a fault diagnosis global model;

Receiving a fault diagnosis global model transmitted by a central terminal, monitoring the running condition of each wind driven generator gearbox in real time according to the fault diagnosis global model, if abnormal error information is identified, determining a corresponding wind driven generator gearbox identifier according to the abnormal error information, and transmitting the wind driven generator gearbox identifier and the abnormal error information to a control center for the control center to generate alarm information so as to inform a worker of fault investigation;

and receiving fault sample data generated by the control center according to the abnormal error information and the troubleshooting information, and updating the fault diagnosis local model according to the fault sample data.

Further, after testing the fault diagnosis local model according to the local data, the method further comprises:

if the fault diagnosis local model does not reach the preset fault diagnosis local model evaluation standard, updating the fault diagnosis local model according to the local data and a preset fault diagnosis local model training formula until the fault diagnosis local model reaches the preset fault diagnosis local model evaluation standard, and transmitting the fault diagnosis local model to a central terminal for training by the central terminal according to the fault diagnosis local model to obtain a fault diagnosis global model; the fault diagnosis local model training formula comprises edge parameter entropy of a fault diagnosis local model which does not reach preset fault diagnosis local model evaluation standards.

Further, collecting local data of each wind driven generator gearbox, performing local model training according to the local data to obtain a first local update model, including:

dividing the local data into health sample data and fault sample data, and performing model training according to the health sample data to obtain a health model;

model training is carried out according to the fault sample data to obtain a fault model;

and carrying out local model training according to the health model and the fault model to obtain a first local updating model.

Further, dividing the local data into healthy sample data and fault sample data includes:

extracting the characteristics of the local data to obtain characteristic data, and comparing the characteristic data with preset health sample data determination standards;

and if the local data reach the preset healthy sample data determining standard, determining the local data as healthy sample data, otherwise, determining the local data as fault sample data.

Further, after receiving the fault diagnosis global model transmitted by the central terminal and monitoring the operation condition of each wind driven generator gearbox in real time according to the fault diagnosis global model, the method further comprises:

If the fault information is identified, determining a corresponding wind driven generator gear box identifier and a fault solution according to the fault information, and transmitting the wind driven generator gear box identifier and the fault solution to a control center for the control center to generate maintenance information so as to inform workers of fault maintenance.

Further, before receiving the global model transmitted by the central side and generating a fault diagnosis local model according to the first local update model, the global model and a preset fault diagnosis local model determination formula, the method further includes:

if an independent model generation instruction transmitted by a central terminal is received, determining identification information and first weight of a wind driven generator gearbox according to the independent model generation instruction;

acquiring local data corresponding to the wind driven generator gear box according to the identification information of the wind driven generator gear box, and performing local model training to obtain a second local updating model;

correspondingly, generating the fault diagnosis local model according to the first local update model, the global model and a preset fault diagnosis local model determining formula comprises the following steps:

Determining a second weight of the first local update model according to the total weight and the first weight;

generating a fault diagnosis local model according to the first local updating model, the second local updating model, the global model, the first weight, the second weight and a preset fault diagnosis local model determining formula.

Further, after generating the fault diagnosis local model according to the first local update model, the second local update model, the global model, the first weight, the second weight, and a preset fault diagnosis local model determination formula, the method further includes:

and testing the fault diagnosis local model according to the local data, and continuously updating the first weight and the second weight according to a preset weight adjustment strategy if the fault diagnosis local model does not reach a preset fault diagnosis local model evaluation standard until the fault diagnosis local model regenerated according to the first local update model, the second local update model, the global model, the first weight, the second weight and a preset fault diagnosis local model determination formula reaches the preset fault diagnosis local model evaluation standard.

Further, after monitoring the operation condition of each wind turbine gearbox in real time according to the fault diagnosis global model, the method further comprises:

collecting local data in real time, and carrying out timing test on the fault diagnosis global model according to the local data and a preset model test time interval;

if the fault diagnosis global model does not reach the preset fault diagnosis global model evaluation standard, continuously updating the fault diagnosis local model according to the local data and a preset fault diagnosis local model training formula until the fault diagnosis local model reaches the preset fault diagnosis local model evaluation standard, and transmitting the fault diagnosis local model to a central terminal for the central terminal to retrain according to the fault diagnosis local model to obtain the fault diagnosis global model; the fault diagnosis local model training formula comprises edge parameter entropy of a fault diagnosis local model before updating.

If business change information transmitted by a central terminal is received, updating the fault diagnosis local model according to the business change information and a preset fault diagnosis local model training formula until the fault diagnosis local model reaches a preset fault diagnosis local model evaluation standard, transmitting the fault diagnosis local model to the central terminal, and retraining by the central terminal according to the fault diagnosis local model to obtain a fault diagnosis global model; the fault diagnosis local model training formula comprises edge parameter entropy of a fault diagnosis local model before updating.

if a model optimization suggestion transmitted by a central terminal is received, updating the fault diagnosis local model according to the model optimization suggestion and a preset fault diagnosis local model training formula until the fault diagnosis local model reaches a preset fault diagnosis local model evaluation standard, transmitting the fault diagnosis local model to the central terminal, and retraining by the central terminal according to the fault diagnosis local model to obtain a fault diagnosis global model; the fault diagnosis local model training formula comprises edge parameter entropy of a fault diagnosis local model before updating.

In the embodiment of the application, local data of each wind driven generator gearbox are collected, local model training is carried out according to the local data to obtain a first local update model, the first local update model is transmitted to a central end, and the central end carries out training according to the first local update model to obtain a global model; receiving a global model transmitted by the central terminal, and generating a fault diagnosis local model according to the first local updating model, the global model and a preset fault diagnosis local model determining formula; testing the fault diagnosis local model according to the local data, and if the fault diagnosis local model reaches a preset fault diagnosis local model evaluation standard, transmitting the fault diagnosis local model to a central terminal for training by the central terminal according to the fault diagnosis local model to obtain a fault diagnosis global model; receiving a fault diagnosis global model transmitted by a central terminal, monitoring the running condition of each wind driven generator gearbox in real time according to the fault diagnosis global model, if abnormal error information is identified, determining a corresponding wind driven generator gearbox identifier according to the abnormal error information, and transmitting the wind driven generator gearbox identifier and the abnormal error information to a control center for the control center to generate alarm information so as to inform a worker of fault investigation; and receiving fault sample data generated by the control center according to the abnormal error information and the troubleshooting information, and updating the fault diagnosis local model according to the fault sample data. By the aid of the wind driven generator gearbox fault diagnosis method based on federal learning, a fault diagnosis local model is generated locally, so that the method can adapt to a new fault mode more quickly, and the calculation pressure and communication cost of a central end are reduced; and the fault diagnosis local model is updated according to the fault sample data, so that the accuracy and the reliability of the model can be improved.

Compared with the prior art, the invention has the following beneficial effects:

(1) The fault diagnosis global model can monitor the running condition of each wind driven generator gearbox in real time; once abnormal error information is found, the system can timely identify the problem and take corresponding measures, so that the response speed to potential faults can be improved;

(2) When the fault diagnosis global model identifies abnormal error information, the system can determine a corresponding wind driven generator gear box standard according to the information, can accurately position a fault source, and reduces time and energy expenditure of maintenance personnel in the fault investigation process;

(3) Once the system identifies the abnormality, the identification of the gearbox of the wind driven generator and the abnormal error information are transmitted to the control center, and alarm information can be generated, so that a worker can quickly acquire the fault condition, and take necessary maintenance measures, thereby reducing the downtime caused by the fault.

Drawings

FIG. 1 is a schematic flow chart of a method for diagnosing a failure of a gearbox of a wind turbine based on federal learning according to an embodiment of the present application;

FIG. 2 is a schematic flow chart of a method for diagnosing a failure of a gearbox of a wind turbine based on federal learning according to a second embodiment of the present application;

FIG. 3 is a schematic flow chart of a method for diagnosing a failure of a wind turbine gearbox based on federal learning according to a third embodiment of the present application;

fig. 4 is a flowchart of a method for diagnosing a fault of a gearbox of a wind turbine based on federal learning according to a fourth embodiment of the present application.

Detailed Description

For the purpose of making the objects, technical solutions and advantages of the present application more apparent, the following detailed description of specific embodiments thereof is given with reference to the accompanying drawings. It is to be understood that the specific embodiments described herein are merely illustrative of the application and not limiting thereof. It should be further noted that, for convenience of description, only some, but not all of the matters related to the present application are shown in the accompanying drawings. Before discussing exemplary embodiments in more detail, it should be mentioned that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although a flowchart depicts operations (or steps) as a sequential process, many of the operations can be performed in parallel, concurrently, or at the same time. Furthermore, the order of the operations may be rearranged. The process may be terminated when its operations are completed, but may have additional steps not included in the figures. The processes may correspond to methods, functions, procedures, subroutines, and the like.

Technical solutions in the embodiments of the present application will be clearly described below with reference to the drawings in the embodiments of the present application, and it is apparent that the described embodiments are some embodiments of the present application, but not all embodiments. All other embodiments obtained by a person of ordinary skill in the art based on the embodiments in the present application are within the scope of the protection of the present application.

The terms first, second and the like in the description and in the claims, are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used may be interchanged, as appropriate, such that embodiments of the present application may be implemented in sequences other than those illustrated or described herein, and that the objects identified by "first," "second," etc. are generally of a type and not limited to the number of objects, e.g., the first object may be one or more. Furthermore, in the description and claims, "and/or" means at least one of the connected objects, and the character "/", generally means that the associated object is an "or" relationship.

The method for diagnosing the fault of the wind driven generator gearbox based on federal learning, which is provided by the embodiment of the application, is described in detail below by means of specific embodiments and application scenes thereof with reference to the accompanying drawings.

Embodiment one: FIG. 1 is a schematic flow chart of a method for diagnosing a failure of a gearbox of a wind turbine based on federal learning according to an embodiment of the present application. As shown in fig. 1, the method specifically comprises the following steps:

s101, collecting local data of each wind driven generator gear box, performing local model training according to the local data to obtain a first local update model, and transmitting the first local update model to a central terminal for training by the central terminal according to the first local update model to obtain a global model.

Firstly, the use scenario of the scheme may be a scenario in which after a client generates a fault diagnosis local model, the fault diagnosis local model is evaluated, the fault diagnosis local model reaching a standard is sent to a central end, the fault diagnosis global model sent by the central end is used for monitoring the operation condition of each wind driven generator gearbox, and after abnormal error information is identified, the abnormal error information and corresponding wind driven generator gearbox identification are sent to a control center.

Based on the above usage scenario, it can be appreciated that the execution subject of the present application may be a client, which is not limited herein.

In this scenario, the method is performed by a client.

The clients may be stand-alone devices or entities involved in training, such as mobile devices, sensors, or nodes in other distributed systems. Each client holds local data and performs model updates locally.

The local data of the wind turbine gearbox may comprise sensor data relating to the operational state of the gearbox, in particular vibration data, temperature data, rotational speed data, oil temperature and pressure data and current data. The vibration data can be vibration signals acquired through equipment such as an acceleration sensor and the like and are used for analyzing vibration characteristics inside the gearbox. The temperature data may be temperatures of various parts within the gearbox measured by temperature sensors for monitoring temperature anomalies. The rotational speed data may be gearbox rotational speed information obtained by a rotational speed sensor that facilitates detection of gear operating conditions. The oil temperature and oil pressure data may be operating parameters of the gearbox lubrication system for monitoring lubrication conditions. The current data may be used to analyze motor operating conditions.

The first local update model may be an update model that is independently generated by each client based on local data of each wind turbine gearbox, and in particular, the model may include unique features and local information of each generator.

Data related to the running states of the gearboxes of the wind driven generators can be collected, and then the collected data are divided into a training set and a testing set according to a certain proportion. The training set is used for training of the model and the test set is used for evaluating the performance of the model. And then, performing model training by using the collected local training data, and specifically, performing model training by adopting a machine learning or deep learning method to obtain a first local updated model. And finally, the first local update model can be transmitted to the central end through a secure channel.

S102, receiving a global model transmitted by the central terminal, and generating a fault diagnosis local model according to the first local update model, the global model and a preset fault diagnosis local model determination formula; the preset fault diagnosis local model determining formula is as follows:

；

wherein,representing the value of the fault diagnosis local model after the t+1st iteration; n represents the number of first locally updated models; i is the index of the first local update model；/>The initial value of the fault diagnosis local model for the t+1st iteration is represented.

The global model may be a model generated by the central side from the first locally updated model uploaded by the respective client side. The method integrates the information of all clients and is a global model of the whole system.

The preset fault diagnosis local model determining formula may be a calculation process defined at the client, and is used for generating the fault diagnosis local model by combining the first local update model, the global model and other relevant information.

Specifically, the global model may be used as an initialization model of the local fault diagnosis model, and the local fault diagnosis model of this time is calculated by using the following formula through the first local update model and the global model which are trained last time. The initial value of the fault diagnosis local model is W0, namely the global model trained last time. If there are N first local update models, the initial value of the ith first local update model is recorded as. The fault diagnosis local model after the first aggregation can be represented as W1, and the aggregation calculation process is as follows:

after the first aggregation, the first local update model is:

setting the iteration aggregation frequency of the model as t, and setting the initial value of the model of the ith fan and the (t+1) th iteration as follows:

utilizing local data pairsUpdate to get->And will update +.>Uploading to a central terminal, and carrying out t+1st aggregation on the fault diagnosis local model:

when W is _t+1 －W _t ＝At this point, the iteration ends.

The fault diagnosis local model may be a model generated by the client according to a preset fault diagnosis local model determination formula. It is a model for fault diagnosis locally, combining local information and global information. The fault diagnosis local model may be a model trained by using local data of a certain number of specific wind turbine gearboxes. These data can be used to teach the model how to identify normal operation and various potential failure modes. The model, by learning local data from multiple gearboxes, can identify different failure modes, such as gear wear, bearing failure, lubrication system problems, and the like. And the wind driven generator model is a local model, has more specificity and adaptability, and can be better adapted to the operation characteristics and environmental conditions of each wind driven generator. The fault diagnosis local model can fuse information of the first local update model to keep adaptability of the initial operation stage to each wind driven generator.

The first local update model and the global model can be used as training parameters, and the fault diagnosis local model is obtained by continuously training in combination with a preset fault diagnosis local model determination formula.

And S103, testing the fault diagnosis local model according to the local data, and if the fault diagnosis local model reaches a preset fault diagnosis local model evaluation standard, transmitting the fault diagnosis local model to a central terminal for training by the central terminal according to the fault diagnosis local model to obtain a fault diagnosis global model.

The pre-set fault diagnosis local model evaluation criterion may be a pre-defined index for evaluating the performance of a locally generated fault diagnosis model. For example, accuracy, recall, etc. may be included.

The central end may be a central server or a central node responsible for the global model. It receives local model updates from the various clients and builds a global model from these updates in a step-by-step manner.

The client can use the test set of the local data to test the locally generated fault diagnosis local model and evaluate whether the performance of the locally generated fault diagnosis local model meets the preset standard. If the fault diagnosis local model meets the standard, the client can transmit the model to the central terminal through the secure channel. After the central terminal receives the fault diagnosis local models from a plurality of clients, the models can be used for training the global model, specifically, a federal learning method can be adopted to integrate the fault diagnosis local models, and a fault diagnosis global model is obtained through training and is fused with the information of each client.

The fault diagnosis global model may be a model generated by integrating fault diagnosis local models of a plurality of wind turbine gearboxes. This global model integrates the information of the individual wind turbines in the system to provide fault diagnosis and monitoring of the overall wind turbine system. The fault diagnosis global model is obtained by collecting and integrating fault diagnosis local models from each wind turbine gearbox. Each local model is focused on a specific number of wind turbines, so the global model can cover the whole system. By integrating the information of the plurality of local models, the fault diagnosis global model can provide more global and system-level fault diagnosis, so that the system can more comprehensively know and respond to the running condition of the whole wind power generator group. The overall view angle of the fault diagnosis global model enables the fault diagnosis global model to better capture various fault modes possibly existing in the system and improves generalization capability of the fresh air power generator. Once the construction of the fault diagnosis global model is completed, the fault diagnosis global model can be used for monitoring the running condition of the whole wind driven generator system in real time and simultaneously carrying out fault diagnosis at a system level, and the model can identify possible faults or abnormal conditions in the system.

On the basis of the above technical solution, optionally, after testing the fault diagnosis local model according to the local data, the method further includes:

In this scheme, the preset fault diagnosis local model training formula may be a mathematical expression including a loss function and a possible regularization term, and is used to measure the fitting degree of the model to the training data. This formula may be dependent on specific problems and requirements, and may contain data fitting terms, regularization terms, and other custom terms.

The edge parameter entropy may be an indicator for measuring uncertainty or variability of model parameters between individual clients. It may reflect the degree of variation of the model in the distributed learning, and the edge parameter entropy may be a feature of the fault diagnosis local model that does not reach the evaluation criterion.

If the fault diagnosis local model does not reach the preset fault diagnosis local model evaluation standard, the fault diagnosis local model can be updated by using local data, and the gradient of the loss function is optimized to reduce the gap between the predicted value and the actual value. The edge parameter entropy can be used as a regularization term and added into the training process to balance the fitting and stability of the model, and specifically, the regularization super-parameter can be adjusted to achieve the effect. In the training process, the performance of the fault diagnosis local model is continuously monitored, and the fault diagnosis local model is continuously and iteratively updated until the preset fault diagnosis local model evaluation standard is met.

In the scheme, the accuracy of generating the fault diagnosis global model by the follow-up center terminal according to the fault diagnosis local model can be improved by continuously performing fault diagnosis local model training until the evaluation standard is reached. Meanwhile, edge parameter entropy is added into a training formula, so that overfitting can be prevented, and the stability of the fault diagnosis local model is improved.

And S104, receiving a fault diagnosis global model transmitted by a central terminal, monitoring the running condition of each wind driven generator gearbox in real time according to the fault diagnosis global model, if abnormal error information is identified, determining a corresponding wind driven generator gearbox identifier according to the abnormal error information, and transmitting the wind driven generator gearbox identifier and the abnormal error information to a control center for the control center to generate alarm information so as to inform a worker of fault investigation.

The abnormal error information may be error information which is not recorded by the fault diagnosis global model, and is a new fault or abnormal condition.

The wind turbine gearbox identification may be an identification of the wind turbine gearbox identified as having abnormal error information to facilitate subsequent troubleshooting and repair.

The control center can be a central control node which is responsible for the coordination and management of the whole system, can be responsible for receiving abnormal information and generating alarm information, and informs staff to perform corresponding operation.

The client can monitor the running condition of each wind driven generator gearbox in real time by using a fault diagnosis global model. If the model identifies abnormal error information, namely error information which is not recorded by the model, the identification of the gearbox of the wind driven generator and the abnormal error information are transmitted to a control center through network communication. After the control center receives the abnormal information, corresponding alarm information is generated, and specifically, the method can comprise the operations of notifying related staff, recording abnormal error information and the like.

On the basis of the technical scheme, optionally, after receiving the fault diagnosis global model transmitted by the central end and monitoring the running condition of each wind driven generator gearbox in real time according to the fault diagnosis global model, the method further comprises the following steps:

In the scheme, the fault information can be information such as error codes, fault types, possible reasons and the like of the wind driven generator gearbox predicted by the model to be in a fault or abnormal operation mode.

The real-time data of each wind driven generator gearbox can be monitored by using a fault diagnosis global model, whether an abnormal condition exists or not is detected by using the model, and if the abnormal condition exists, relevant fault information is extracted from model output. And determining a corresponding wind driven generator gearbox identifier according to the monitored fault information. The corresponding fault solution can be queried according to the fault information, and in particular, the corresponding fault solution can be queried according to the error code in the fault information. And finally, transmitting the wind driven generator gearbox identification and the fault solution to a control center through a wireless communication technology.

In this scheme, through the running condition of each aerogenerator gear box of real-time supervision to inquiry solution and transmission to control center when breaking down can make the staff carry out the trouble maintenance fast, improves maintenance efficiency, reduces equipment downtime, and then reduces the loss.

On the basis of the above technical solution, optionally, after monitoring the operation condition of each wind turbine gearbox in real time according to the fault diagnosis global model, the method further includes:

In this scheme, the preset model test time interval may be a preset model test time interval, which indicates how often to perform the test of the fault diagnosis global model.

The real-time local data can be continuously collected from the wind driven generator gearbox, and the fault diagnosis global model is tested at regular time by utilizing the collected local data according to a preset time interval. And then checking the test result, and judging whether the fault diagnosis global model meets the preset fault diagnosis global model evaluation standard. And if the fault diagnosis global model does not reach the evaluation standard, continuously updating the fault diagnosis local model according to the collected local data and a preset fault diagnosis local model training formula. The fault diagnosis local model training formula comprises edge parameter entropy of the fault diagnosis local model before updating, so that information of the edge parameter entropy is guaranteed to be fully considered in updating. Once the local model meets the preset fault diagnosis local model evaluation standard, the updated fault diagnosis local model is transmitted to the central terminal through a wireless communication technology.

In the scheme, the local data are collected in real time, the fault diagnosis global model is tested at fixed time, the change of the model performance can be perceived in time in a continuously changing environment, potential problems and improved models can be found in time, and the accuracy and reliability of the model are improved.

In this scenario, the model optimization suggestions may include suggestions of various aspects, such as feature selection, hyper-parameter adjustment, and model structure optimization.

The client can receive the model optimization suggestions transmitted by the central terminal through the wireless communication technology, then analyze the received model optimization suggestions, and understand the concrete contents and the suggestion directions. According to the model optimization suggestions and a preset fault diagnosis local model training formula, the local fault diagnosis model is correspondingly updated, specifically, the method can comprise the steps of adjusting model parameters, modifying model structures and the like, testing and evaluating the updated local model, and ensuring that the model reaches a preset fault diagnosis local model evaluation standard. If the local model passes the evaluation, the updated model is transmitted to the central terminal through a wireless communication technology.

In the scheme, the model optimization suggestion transmitted by the central terminal is received, so that the change requirement of the model performance can be responded in real time. The model can adapt to environmental change or service change in time, and the performance and accuracy of the model are improved.

S105, receiving fault sample data generated by the control center according to the abnormal error information and the troubleshooting information, and updating the fault diagnosis local model according to the fault sample data.

The investigation information may be information manually entered by a worker, and specifically may include further description of abnormal error information, cause of occurrence, specific inspection steps, and the like.

The fault sample data may be data generated from abnormal error information and troubleshooting information for updating the fault diagnosis local model. Such data may include sensor data, operating conditions, failure modes, etc. when an abnormal error occurs.

After the client receives the fault sample data, the fault diagnosis local model can be updated by using the data, and specifically, the process of fine tuning, retraining and the like of the model parameters can be included, so that the recognition capability of the model on a new fault sample is improved.

In the scheme, the fault diagnosis global model can monitor the running condition of each wind driven generator gearbox in real time. Once the abnormal error information is found, the system can timely identify the problem and take corresponding measures, so that the response speed to potential faults can be improved. When the fault diagnosis global model identifies abnormal error information, the system can determine the corresponding wind driven generator gear box standard according to the information, can accurately position a fault source, and reduces time and energy expenditure of maintenance personnel in the fault investigation process. Once the system identifies the abnormality, the identification of the gearbox of the wind driven generator and the abnormal error information are transmitted to the control center, and alarm information can be generated, so that a worker can quickly acquire the fault condition, and take necessary maintenance measures, thereby reducing the downtime caused by the fault.

In this scenario, one core idea of federal learning is to perform model training on distributed devices, rather than transmitting all data to a central server. According to the scheme, each wind driven generator gearbox has own local data, the local model is obtained through training on the local data, and then the global model is built through transmission to a central terminal, so that the method is consistent with the idea of federal learning. Federal learning involves updates to model parameters trained on local devices that are sent to a central server to build a global model. The first local update model of the scheme is transmitted to the central end, is fused with the global model, and then generates a fault diagnosis local model. In federal learning, local devices are typically evaluated on a locally trained model and then transmit their updates to a central server. The fault diagnosis local model of the scheme is tested on local data and is transmitted to the central terminal only when the preset evaluation standard is met. The equipment in federal learning typically monitors its local data in real time and feeds back model updates to the central server. According to the fault diagnosis global model, the running conditions of the gear boxes of the wind driven generators are monitored in real time, and when abnormal error information is identified, the information is transmitted to a central end and a control center. In federal learning, a central server gathers model updates from various local devices, which can be used to update the global model. The control center of the scheme realizes the update of the fault diagnosis local model by generating fault sample data and transmitting the fault sample data to the local model.

According to the technical scheme provided by the embodiment, local data of each wind driven generator gearbox are collected, local model training is carried out according to the local data, a first local update model is obtained, the first local update model is transmitted to a central end, and the central end carries out training according to the first local update model to obtain a global model; receiving a global model transmitted by the central terminal, and generating a fault diagnosis local model according to the first local updating model, the global model and a preset fault diagnosis local model determining formula; testing the fault diagnosis local model according to the local data, and if the fault diagnosis local model reaches a preset fault diagnosis local model evaluation standard, transmitting the fault diagnosis local model to a central terminal for training by the central terminal according to the fault diagnosis local model to obtain a fault diagnosis global model; receiving a fault diagnosis global model transmitted by a central terminal, monitoring the running condition of each wind driven generator gearbox in real time according to the fault diagnosis global model, if abnormal error information is identified, determining a corresponding wind driven generator gearbox identifier according to the abnormal error information, and transmitting the wind driven generator gearbox identifier and the abnormal error information to a control center for the control center to generate alarm information so as to inform a worker of fault investigation; and receiving fault sample data generated by the control center according to the abnormal error information and the troubleshooting information, and updating the fault diagnosis local model according to the fault sample data. By the aid of the wind driven generator gearbox fault diagnosis method based on federal learning, a fault diagnosis local model is generated locally, a new fault model can be adapted more quickly, and calculation pressure and communication cost of a central end are reduced. And the fault diagnosis local model is updated according to the fault sample data, so that the accuracy and the reliability of the model can be improved.

Embodiment two: fig. 2 is a flow chart of a method for diagnosing a fault of a gearbox of a wind turbine based on federal learning according to a second embodiment of the present application, as shown in fig. 2, the specific method includes the following steps:

s201, dividing the local data into health sample data and fault sample data, and performing model training according to the health sample data to obtain a health model.

In this solution, the health sample data may be data of the wind turbine gearbox in a normal operation state, where the data corresponds to observations of the system or the device in a good operation state, and specifically may include various sensor measurements and log records.

The health model may be a machine learning model obtained by training using health sample data. This model can capture the characteristics and behavior of the system when it is operating normally, as a comparison, for subsequent use in detecting anomalies or faults.

The locally acquired data may be preprocessed, and specifically may include data cleaning, feature extraction, etc., and the data may be marked as healthy sample data or fault sample data. The data set is then divided into a training set and a test set, ensuring that there are enough healthy samples for model learning during the training process, while retaining the test set for performance evaluation. The machine learning model is trained using the partitioned health sample data, and in particular, an appropriate algorithm may be selected for training, such as a classification algorithm in supervised learning. When training is complete, the performance of the health model can be evaluated using the test set, checking its generalization ability on unseen data.

On the basis of the above technical solution, optionally, dividing the local data into health sample data and fault sample data includes:

In this scheme, the feature data may be a set of values reflecting key information in the original data. The characteristic data of the wind turbine gearbox may include vibration characteristics, temperature characteristics, oil characteristics, current characteristics, sound characteristics, speed characteristics, and acceleration characteristics. The vibration characteristics may include, among other things, the frequency, amplitude, waveform, etc. of the vibrations for analyzing the operating state of the gear. The temperature characteristic may reflect an operating temperature of the gearbox. Abnormal temperature changes may indicate lubrication problems. The oil characteristics can include particles, temperature, fluidity and the like in the oil, and are used for judging friction, abrasion and pollution conditions in the gear box. The current signature may be representative of a current change of the motor or generator for analyzing the operating state of the motor in the gearbox. The audible features may provide information regarding gear wear, engagement problems, and the like. The speed signature may be a speed signature of the operation of the gearbox for analysing the gear mesh status. The acceleration characteristic may be a linear acceleration of the gearbox for identifying whether an anomaly such as an impact or crash exists.

The predetermined health sample data determination criterion may be a predefined criterion for determining whether the characteristic data corresponds to the characteristics of the health sample. This criterion can be set based on historical data analysis, or by statistical analysis of the data distribution, etc.

For the local data of the wind driven generator gearbox, the characteristics representing the characteristics of the data can be extracted through a proper characteristic extraction method. The extracted feature data is compared with a preset health sample data determination standard, and specifically, the similarity and the distance measurement can be calculated for comparison. And judging whether the characteristic data is a health sample or a fault sample according to the comparison result. If the preset health sample data determining standard is met, judging that the health sample is a health sample; otherwise, judging the sample to be a fault sample.

In the scheme, the health sample data and the fault sample data can be rapidly distinguished through the feature extraction, so that the model reasoning and the calculation cost are reduced.

S202, performing model training according to the fault sample data to obtain a fault model.

The fault sample data may include observations of the system or device at the time of a fault, such as sensor measurements, logging, and the like.

The fault model may be a machine learning model that is trained using fault sample data. This model can capture the characteristics and behavior of the system in a fault or abnormal state.

The client may use the collected fault sample data for model training to construct a fault model using a suitable machine learning algorithm, such as a classification algorithm.

S203, performing local model training according to the health model and the fault model to obtain a first local update model, and transmitting the first local update model to a central terminal for the central terminal to train according to the first local update model to obtain a global model.

The resulting fault model may be combined with a previously trained health model, the outputs of the two models may be combined using an ensemble learning approach, or a comprehensive model may be designed. A new first local update model is formed by integrating the fault model and the health model, the model containing the characteristics of the system in normal and fault conditions, and the first local update model is transmitted to the central terminal through the secure channel.

S204, receiving the global model transmitted by the central terminal, and generating a fault diagnosis local model according to the first local updating model, the global model and a preset fault diagnosis local model determining formula.

S205, testing the fault diagnosis local model according to the local data, and if the fault diagnosis local model reaches a preset fault diagnosis local model evaluation standard, transmitting the fault diagnosis local model to a central terminal for training by the central terminal according to the fault diagnosis local model to obtain a fault diagnosis global model.

S206, receiving a fault diagnosis global model transmitted by a central terminal, monitoring the running condition of each wind driven generator gearbox in real time according to the fault diagnosis global model, if abnormal error information is identified, determining a corresponding wind driven generator gearbox identifier according to the abnormal error information, and transmitting the wind driven generator gearbox identifier and the abnormal error information to a control center for the control center to generate alarm information so as to inform a worker of fault investigation.

S207, receiving fault sample data generated by the control center according to the abnormal error information and the troubleshooting information, and updating the fault diagnosis local model according to the fault sample data.

In the embodiment, the health model and the fault model are combined to obtain the local update model, so that the behavior of the system in normal and fault states can be considered more comprehensively, the accuracy of fault diagnosis is improved, and the risk of overfitting can be reduced.

Embodiment III: fig. 3 is a flow chart of a method for diagnosing a fault of a gearbox of a wind turbine based on federal learning according to a third embodiment of the present application, as shown in fig. 3, the specific method includes the following steps:

s301, collecting local data of each wind driven generator gearbox, performing local model training according to the local data to obtain a first local update model, and transmitting the first local update model to a central terminal for training by the central terminal according to the first local update model to obtain a global model.

S302, if an independent model generation instruction transmitted by a central terminal is received, identification information and first weight of a wind driven generator gearbox are determined according to the independent model generation instruction.

The separate model generation instruction may be an instruction sent by the central side to the client side, and is used to inform the client side to generate an independent local model, and specifically may include relevant parameters and configuration information required for generating the model.

The first weight may be a weight of a second local update model generated subsequently from the wind turbine gearbox when the fault diagnosis local model is generated using the second local update model. The first weight may determine a degree of participation of the second locally updated model when the fault diagnosis local model is subsequently generated.

The client receives the single model generation instruction sent from the center end, and can analyze the identification information and the weight information in the single model generation instruction. According to the information in the instruction, an independent local model, specifically a neural network or other machine learning model, is locally generated at the client, and the weight parameters of the model are initialized by using the first weight provided in the instruction.

S303, acquiring local data corresponding to the wind driven generator gear box according to the identification information of the wind driven generator gear box, and performing local model training to obtain a second local updated model.

The second local update model may be an update model obtained by the client acquiring local data corresponding to the wind turbine gearbox according to the received individual model generation instruction and using the local data to perform local model training.

The client can receive the instruction transmitted by the central terminal, and determines the wind driven generator gear box which needs to be updated by the independent model according to the identification information of the wind driven generator gear box which needs to be updated. With the obtained identification information, the client starts to collect local data of the corresponding wind turbine gearbox, and specifically, the data can include running states, sensor readings, vibration information and the like. Using the collected local data, the client may perform local model training to obtain a second local update model. This model will be more tailored to the characteristics and operating environment of the particular wind turbine gearbox.

S304, receiving the global model transmitted by the central terminal, and determining a second weight of the first local update model according to the total weight and the first weight.

The total weight may be a sum of weights of the first local update model and the second local update model in the fault diagnosis local model, and in this scheme, the total weight may be set to 1.

The second weight may be a weight of the first locally updated model subsequent to generation of the fault diagnosis local model from the first locally updated model. The second weight may determine a degree of participation of the first locally updated model when the fault diagnosis local model is subsequently generated.

The total weight and the first weight can be subtracted to obtain a second weight, then the first local updating model endowed with the second weight, the second local updating model endowed with the first weight and the global model are taken as training parameters, and a fault diagnosis local model is generated by combining a preset fault diagnosis local model determination formula.

S305, generating a fault diagnosis local model according to the first local updating model, the second local updating model, the global model, the first weight, the second weight and a preset fault diagnosis local model determining formula.

S306, testing the fault diagnosis local model according to the local data, and if the fault diagnosis local model reaches a preset fault diagnosis local model evaluation standard, transmitting the fault diagnosis local model to a central terminal for training by the central terminal according to the fault diagnosis local model to obtain a fault diagnosis global model.

S307, receiving a fault diagnosis global model transmitted by a central terminal, monitoring the running condition of each wind driven generator gearbox in real time according to the fault diagnosis global model, if abnormal error information is identified, determining a corresponding wind driven generator gearbox identifier according to the abnormal error information, and transmitting the wind driven generator gearbox identifier and the abnormal error information to a control center for the control center to generate alarm information so as to inform a worker of fault investigation.

S308, receiving fault sample data generated by the control center according to the abnormal error information and the troubleshooting information, and updating the fault diagnosis local model according to the fault sample data.

In the embodiment, by setting the independent model for the specific wind driven generator gearbox, personalized adjustment can be performed according to the characteristics of different equipment, and a more detailed and accurate fault diagnosis model is generated, so that the model has customized characteristics, and the flexibility of model generation can be improved. Meanwhile, the models corresponding to the independent equipment are combined with the models corresponding to most of the equipment, and an appropriate scheme can be found on the premise of considering the overall performance of the system, so that the corresponding generated models can perform well under various conditions.

On the basis of the above technical solution, optionally, after generating the fault diagnosis local model according to the first local update model, the second local update model, the global model, the first weight, the second weight, and a preset fault diagnosis local model determination formula, the method further includes:

In this solution, the preset weight adjustment policy may include learning rate adjustment, regularization, weighting update, priority update, and dynamic adjustment. The learning rate is an important parameter for controlling the weight update stride in the training process. The weight updating can be more stable by dynamically adjusting the learning rate. A smaller learning rate may improve the stability of the model, but may result in too slow training, while a larger learning rate may result in unstable weight updates; regularization terms, such as L1 regularization or L2 regularization, are introduced to control the size of the weights. Regularization can prevent the model from being excessively fitted, so that the model is simplified and generalized; different updates are given different weights according to the model's behavior on the validation set. A well behaved model may acquire more weight, while a poorly behaved model may acquire less weight; some weights are given more frequent or greater updates based on the update history of the model. This can be used to preferentially update those weights that were less effective in the previous training step to speed up model convergence; the weights are dynamically adjusted according to the performance of the model in a real-time environment. For example, if the model degrades over a continuous period of time, the update amplitude of the weights may be increased, and vice versa.

The fault diagnosis local model can be tested using local data to evaluate its performance. And checking whether the test result meets the preset fault diagnosis local model evaluation standard. And if the first weight and the second weight do not reach the standard, updating the first weight and the second weight according to a preset weight adjustment strategy. And then regenerating the fault diagnosis local model according to a preset fault diagnosis local model determination formula by using the adjusted weight, the first local update model, the second local update model and the global model. The regenerated fault diagnosis local model is tested and the performance is evaluated. If the evaluation criteria are not met, continuing to dynamically adjust the weights and regenerating the model until the preset fault diagnosis local model evaluation criteria are met.

In the scheme, the continuously updated weights can dynamically adapt to changes in a real-time environment, and as the data distribution can change with time, the continuously updated weights can ensure that the model keeps high efficiency in the changes. And the continuous updating enables the model to respond to new data and environmental changes in time, so that faults can be diagnosed and predicted in time, loss is reduced, and maintenance efficiency is improved.

Embodiment four: fig. 4 is a flow chart of a method for diagnosing a fault of a gearbox of a wind turbine based on federal learning according to a fourth embodiment of the present application, as shown in fig. 4, and the specific method includes the following steps:

s401, collecting local data of each wind driven generator gear box, performing local model training according to the local data to obtain a first local update model, and transmitting the first local update model to a central terminal for training by the central terminal according to the first local update model to obtain a global model.

S402, receiving the global model transmitted by the central terminal, and generating a fault diagnosis local model according to the first local update model, the global model and a preset fault diagnosis local model determination formula.

S403, testing the fault diagnosis local model according to the local data, and if the fault diagnosis local model reaches a preset fault diagnosis local model evaluation standard, transmitting the fault diagnosis local model to a central terminal for training by the central terminal according to the fault diagnosis local model to obtain a fault diagnosis global model.

S404, receiving a fault diagnosis global model transmitted by a central terminal, monitoring the operation condition of each wind driven generator gearbox in real time according to the fault diagnosis global model, if service change information transmitted by the central terminal is received, updating the fault diagnosis local model according to the service change information and a preset fault diagnosis local model training formula until the fault diagnosis local model reaches a preset fault diagnosis local model evaluation standard, transmitting the fault diagnosis local model to the central terminal, and retraining by the central terminal according to the fault diagnosis local model to obtain the fault diagnosis global model; the fault diagnosis local model training formula comprises edge parameter entropy of a fault diagnosis local model before updating.

In this embodiment, the service change information may be some change in the operation of the system, such as a change in the workload, operating conditions, environmental conditions, etc. of the wind turbine gearbox. The purpose of updating the fault diagnosis local model is to enable the model to adapt to new service scenes and improve the generalization performance of the model.

The client can receive the service change information transmitted by the central terminal through a wireless communication technology, analyze the service change information, and extract key information which can influence the performance of the fault diagnosis model, such as the change of workload, the change of temperature, the change of humidity and the like. And then updating the fault diagnosis local model by using the service change information obtained by analysis according to a preset fault diagnosis local model training formula. The training formula contains the edge parameter entropy of the fault diagnosis local model before updating, so that the edge parameter entropy is ensured to be taken into consideration. And evaluating the updated fault diagnosis local model, and checking whether the fault diagnosis local model meets the preset fault diagnosis local model evaluation standard. If the standard is not met, continuing to adjust and update until the standard is met. Once the fault diagnosis local model meets the evaluation standard, the updated model is transmitted to the central terminal through a wireless communication technology, so that the central terminal can train the global model again.

In this embodiment, based on the continuously collected service change information, iterative optimization of the fault diagnosis model can be achieved, and performance and accuracy of the model and adaptability of the model to the working environment are continuously improved.

The foregoing description is only of the preferred embodiments of the present application and the technical principles employed. The present application is not limited to the specific embodiments described herein, but is capable of numerous obvious changes, rearrangements and substitutions as will now become apparent to those skilled in the art without departing from the scope of the present application. Therefore, while the present application has been described in connection with the above embodiments, the present application is not limited to the above embodiments, but may include many other equivalent embodiments without departing from the spirit of the present application, and the scope of the present application is determined by the scope of the claims.

Claims

1. A method for diagnosing a failure of a gearbox of a wind turbine based on federal learning, the method being performed by a client, the method comprising:

s101: collecting local data of each wind driven generator gearbox, performing local model training according to the local data to obtain a first local update model, and transmitting the first local update model to a central end for the central end to train according to the first local update model to obtain a global model; collecting local data of each wind driven generator gearbox, performing local model training according to the local data to obtain a first local update model, wherein the local data are divided into health sample data and fault sample data, and performing model training according to the health sample data to obtain a health model; model training is carried out according to the fault sample data to obtain a fault model; performing local model training according to the health model and the fault model to obtain a first local update model;

S102: receiving a global model transmitted by the central terminal, and generating a fault diagnosis local model according to the first local updating model, the global model and a preset fault diagnosis local model determining formula; the preset fault diagnosis local model determining formula is as follows:

；

wherein,representing the value of the fault diagnosis local model after the t+1st iteration; n represents the number of first locally updated models; i is the index of the first local update model; />Representing an initial value of a fault diagnosis local model of the t+1st iteration;

s103: testing the fault diagnosis local model according to the local data, and if the fault diagnosis local model reaches a preset fault diagnosis local model evaluation standard, transmitting the fault diagnosis local model to a central terminal for training by the central terminal according to the fault diagnosis local model to obtain a fault diagnosis global model;

s104: receiving a fault diagnosis global model transmitted by a central terminal, monitoring the running condition of each wind driven generator gearbox in real time according to the fault diagnosis global model, if abnormal error information is identified, determining a corresponding wind driven generator gearbox identifier according to the abnormal error information, and transmitting the wind driven generator gearbox identifier and the abnormal error information to a control center for the control center to generate alarm information so as to inform a worker of fault investigation;

S105: and receiving fault sample data generated by the control center according to the abnormal error information and the troubleshooting information, and updating the fault diagnosis local model according to the fault sample data.

2. The federally learned based wind turbine gearbox fault diagnosis method according to claim 1, wherein after testing the fault diagnosis local model according to the local data, the method further comprises:

3. The federally learned wind turbine gearbox fault diagnosis method according to claim 1, wherein dividing the local data into health sample data and fault sample data comprises:

4. The federally learned wind turbine gearbox fault diagnosis method according to claim 1, wherein after receiving a fault diagnosis global model transmitted from a central terminal and monitoring the operation condition of each wind turbine gearbox in real time according to the fault diagnosis global model, the method further comprises:

5. The federally learned wind turbine gearbox fault diagnosis method according to claim 1, wherein prior to receiving the central-transmitted global model and generating a fault diagnosis local model from the first locally updated model, the global model, and a pre-set fault diagnosis local model determination formula, the method further comprises:

6. The federally learning-based wind turbine gearbox fault diagnosis method according to claim 5, wherein after generating a fault diagnosis local model according to the first local update model, the second local update model, the global model, the first weight, the second weight, and a preset fault diagnosis local model determination formula, the method further comprises:

7. The federally learned based wind turbine gearbox fault diagnosis method according to claim 1, wherein after monitoring the operating condition of each wind turbine gearbox in real time according to the fault diagnosis global model, the method further comprises:

8. The federally learned based wind turbine gearbox fault diagnosis method according to claim 1, wherein after monitoring the operating condition of each wind turbine gearbox in real time according to the fault diagnosis global model, the method further comprises:

9. The federally learned based wind turbine gearbox fault diagnosis method according to claim 1, wherein after monitoring the operating condition of each wind turbine gearbox in real time according to the fault diagnosis global model, the method further comprises: