Fan life prediction method, device, equipment and medium based on separable convolution

Through the fan life prediction method based on separable convolution, the Bayesian optimization algorithm is used to optimize the network hyperparameters, which solves the problems of complex fan life prediction model and slow training speed in the existing technology, and improves the efficiency of wind turbine life prediction and accuracy.

CN114997052BActive Publication Date: 2024-11-01WUHAN UNIV OF TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202210605172.6
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-05-30
Publication Date
2024-11-01
Estimated Expiration
2042-05-30

AI Technical Summary

Technical Problem

The existing technology of stroke fans' life prediction model is complex, with many hyperparameters and requires artificial settings, resulting in slow model training speed and low prediction efficiency.

Method used

Using the fan life prediction method based on separable convolution, the initial separable convolution network model is constructed by obtaining the life sample set of wind turbines, and the hyperparameters are optimized by using Bayesian optimization algorithm to optimize the network model, and quickly Obtain a target separable convolutional network model to improve prediction efficiency.

Benefits of technology

It effectively reduces the model parameters, improves the efficiency and accuracy of wind turbine life prediction, simplifies the manual feature extraction process, and improves the model training speed.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114997052B_ABST
    Figure CN114997052B_ABST
Patent Text Reader

Abstract

The present invention provides a method, device, equipment and medium for predicting the life of a wind turbine based on separable convolution, and the method comprises: obtaining a life sample set of a wind turbine; constructing an initial separable convolutional network model, the initial separable convolutional network model including initial network hyperparameters; optimizing the initial network hyperparameters based on the life sample set and the Bayesian optimization algorithm to obtain optimized network hyperparameters and an optimized separable convolutional network model; training the optimized separable convolutional network model based on the life sample set to obtain a target separable convolutional network model; obtaining vibration data of a wind turbine to be predicted, and predicting the usable life of the wind turbine to be predicted based on the target separable convolutional network model. The present invention can improve the life prediction efficiency and life prediction accuracy of wind turbines.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of fan life prediction, and in particular to a fan life prediction method, device, equipment and medium based on separable convolution. Background Art

[0002] The development of efficient wind turbine life prediction technology has become an urgent need for major wind power manufacturers. At present, the most widely used life prediction technologies are mainly simulation analysis technology based on finite element simulation and signal analysis technology based on time-frequency analysis. Specifically: simulation analysis technology mostly uses finite element analysis method to simulate and simulate from the perspective of load analysis to obtain component fatigue life, or proposes different modified rated life calculation methods based on the life calculation formula of the national standard. However, the influence of the actual operating environment is not taken into account, and the above analysis methods are mostly used to obtain the theoretical life results of the components. It is difficult to effectively apply it to the operation, maintenance and inspection of wind turbine tower connectors in actual sites. The technology based on time-frequency analysis is highly dependent on expert knowledge and engineer experience, and the effect in practical applications is not satisfactory.

[0003] As a typical representative of data-driven methods, machine learning has been widely used in the field of fault diagnosis. As a branch of machine learning, deep learning can adaptively extract features from a large amount of data to meet the requirements of adaptive feature extraction, effectively overcome the shortcomings of poor generalization ability and poor robustness of traditional manually extracted features, and reduce the uncertainty of traditional fault diagnosis methods in the process of manual design and extraction. Therefore, an end-to-end diagnostic model can be established using deep learning technology, that is, a life prediction model is directly established on the basis of collecting raw data, omitting the complicated manual feature extraction process. However, due to the complex structure of the fan and the changeable environment in which it is located, a deep-structured convolutional network model is required to extract features from the signal, resulting in a slow model training speed and low efficiency in predicting the life of the fan. In addition, the deep-structured convolutional network model includes multiple hyperparameters, and the hyperparameters in the prior art need to be manually configured, which further slows down the model training speed and further reduces the efficiency of fan life prediction.

[0004] Therefore, it is urgent to propose a fan life prediction method, device, equipment and medium based on separable convolution to solve the technical problems in the prior art that the network model for predicting the fan life is complex, has many hyperparameters, and requires manual setting of hyperparameters, resulting in a slow model training process and low efficiency in predicting the fan life. Summary of the invention

[0005] In view of this, it is necessary to provide a fan life prediction method, device, equipment and medium based on separable convolution to solve the technical problems in the prior art that the network model for predicting the fan life is complex, has many hyperparameters, and requires manual setting of hyperparameters, resulting in a slow model training process and low efficiency in predicting the fan life.

[0006] On the one hand, the present invention provides a wind turbine life prediction method based on separable convolution, comprising:

[0007] Obtain a life sample set of wind turbines;

[0008] Constructing an initial separable convolutional network model, wherein the initial separable convolutional network model includes initial network hyperparameters;

[0009] Optimizing the initial network hyperparameters based on the life sample set and the Bayesian optimization algorithm to obtain optimized network hyperparameters and an optimized separable convolutional network model;

[0010] Training the optimized separable convolutional network model based on the life sample set to obtain a target separable convolutional network model;

[0011] Vibration data of a wind turbine to be predicted is acquired, and the usable life of the wind turbine to be predicted is predicted based on the target separable convolutional network model.

[0012] In some possible implementations, optimizing the initial network hyperparameters based on the life sample set and the Bayesian optimization algorithm to obtain optimized network hyperparameters and optimize the separable convolutional network model includes:

[0013] Step 1: determining a first loss value of the initial separable convolutional network model based on a preset loss function and the life sample set;

[0014] Step 2: obtaining the candidate network hyperparameters based on the acquisition function, and obtaining the candidate separable convolutional network model according to the candidate network hyperparameters;

[0015] Step 3: determining a second loss value of the separable convolutional network model to be selected based on the loss function and the life sample set;

[0016] Step 4: evaluating the initial network hyperparameters and the selected network hyperparameters based on the Gaussian process model, the first loss value, and the second loss value, and obtaining the optimal potential separable convolutional network model according to the optimal potential hyperparameters;

[0017] Step 5: determining a third loss value of the optimal potential separable convolutional network model based on the loss function and the life sample set, and judging whether the third loss value is less than a loss threshold and whether the number of times the acquisition function obtains the candidate network hyperparameters is greater than a threshold number;

[0018] Step six: when the third loss value is less than the loss threshold, or the number of times the acquisition function obtains the network hyperparameters to be selected is greater than the threshold number of times, the optimal potential hyperparameters are the optimized network hyperparameters, and the optimal potential separable convolutional network model is the optimized separable convolutional network model; when the third loss value is greater than or equal to the loss threshold and the number of times the acquisition function obtains the network hyperparameters to be selected is less than or equal to the threshold number of times, repeat steps two to six.

[0019] In some possible implementations, obtaining a life sample set of a wind turbine generator includes:

[0020] Collecting vibration data of the wind turbine over its entire life cycle, wherein the vibration data over its entire life cycle includes a plurality of observation moments, a plurality of vibration sub-data corresponding to the plurality of observation moments, and a life end moment;

[0021] using the time difference between each of the plurality of observation times and the end-of-life time as the remaining usable life of each of the vibrator data;

[0022] The life sample set is constructed according to the plurality of vibration sub-data and the remaining available life of each of the vibration sub-data.

[0023] In some possible implementations, after collecting the vibration data of the wind turbine over its entire life cycle, the method further includes:

[0024] The whole life cycle vibration data of the wind turbine is normalized.

[0025] In some possible implementations, the initial network hyperparameters include the initial number of backbone modules, the initial embedding form between different backbone modules, the initial number of convolution layers in the backbone modules, the initial number of convolution kernels, and the initial convolution kernel size; the optimized network hyperparameters include the optimized number of backbone modules, the optimized embedding form between different backbone modules, the optimized number of convolution layers in the backbone modules, the optimized number of convolution kernels, and the optimized convolution kernel size.

[0026] In some possible implementations, the training the optimized separable convolutional network model based on the life sample set to obtain a target separable convolutional network model includes:

[0027] Determine a life training set and a life test set from the life sample set;

[0028] Training the optimized separable convolutional network model based on the lifespan training set to obtain a separable convolutional network model to be tested;

[0029] Testing the separable convolutional network model to be tested based on the life test set to obtain a test accuracy, and determining whether the test accuracy is greater than a threshold accuracy;

[0030] If the test accuracy is greater than the threshold accuracy, the separable convolutional network model to be tested is the target separable convolutional network model;

[0031] If the test accuracy is less than or equal to the threshold accuracy, the optimized separable convolutional network model is retrained based on the life sample set.

[0032] In some possible implementations, the loss function is:

[0033]

[0034] Wherein, MSE is the loss function; n is the total number of samples in the life sample set; is the actual usable life; To predict the useful life.

[0035] On the other hand, the present invention also provides a wind turbine life prediction device based on separable convolution, comprising:

[0036] A sample set acquisition module is used to acquire a life sample set of a wind turbine;

[0037] An initial model building module, used to build an initial separable convolutional network model, wherein the initial separable convolutional network model includes initial network hyperparameters;

[0038] A model optimization module, used to optimize the initial network hyperparameters based on the life sample set and the Bayesian optimization algorithm, obtain optimized network hyperparameters and optimize the separable convolutional network model;

[0039] A model training module, used for training the optimized separable convolutional network model based on the life sample set to obtain a target separable convolutional network model;

[0040] The life prediction module is used to obtain vibration data of the wind turbine to be predicted and predict the usable life of the wind turbine to be predicted based on the target separable convolutional network model.

[0041] On the other hand, the present invention also provides an electronic device, including a memory and a processor, wherein:

[0042] The memory is used to store programs;

[0043] The processor is coupled to the memory and is used to execute the program stored in the memory to implement the steps in the wind turbine life prediction method based on separable convolution described in any one of the above implementations.

[0044] On the other hand, the present invention also provides a computer-readable storage medium for storing computer-readable programs or instructions, which, when executed by a processor, can implement the steps of the wind turbine life prediction method based on separable convolution described in any of the above-mentioned implementation methods.

[0045] The beneficial effect of adopting the above embodiment is that the wind turbine life prediction method based on separable convolution provided by the present invention can effectively reduce the model parameters by constructing an initial separable convolutional network model, thereby improving the efficiency of obtaining the target separable convolutional network model, so as to improve the efficiency of life prediction of wind turbines. In addition, the present invention optimizes the initial network hyperparameters through the Bayesian optimization algorithm, and can quickly obtain the optimized separable convolutional network model, thereby further improving the efficiency of obtaining the target separable convolutional network model, and further improving the efficiency of life prediction of wind turbines.

[0046] Furthermore, the present invention is arranged that after obtaining the optimized separable convolutional network model through the Bayesian optimization algorithm, the optimized separable convolutional network model is trained based on the life sample set to obtain the target separable convolutional network model, which can improve the accuracy of the target separable convolutional network model in predicting the useful life of the wind turbine. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0048] Figure 1 A schematic flow chart of an embodiment of a wind turbine life prediction method based on separable convolution provided by the present invention;

[0049] Figure 2 For the present invention Figure 1 A schematic flow chart of an embodiment of S103;

[0050] Figure 3 For the present invention Figure 1 A schematic flow chart of an embodiment of S101;

[0051] Figure 4 For the present invention Figure 1A schematic flow chart of an embodiment of S104;

[0052] Figure 5 A schematic diagram of the structure of an embodiment of the target separable convolutional network model provided by the present invention;

[0053] Figure 6 A schematic diagram of the structure of an embodiment of a wind turbine life prediction device based on separable convolution provided by the present invention;

[0054] Figure 7 A schematic structural diagram of an embodiment of an electronic device provided by the present invention. DETAILED DESCRIPTION

[0055] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0056] It should be understood that the schematic drawings are not drawn to scale. The flow chart used in the present invention shows the operations implemented according to some embodiments of the present invention. It should be understood that the operations of the flow chart can be implemented out of order, and the steps without logical context can be reversed in order or implemented simultaneously. In addition, those skilled in the art can add one or more other operations to the flow chart under the guidance of the content of the present invention, and can also remove one or more operations from the flow chart.

[0057] Some of the block diagrams shown in the accompanying drawings are functional entities that do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software form, or in one or more hardware modules or integrated circuits, or in different networks and / or processor systems and / or microcontroller systems.

[0058] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present invention. The appearance of the phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0059] The embodiments of the present invention provide a method, device, equipment and medium for predicting the life of a wind turbine based on separable convolution, which are described below respectively.

[0060] Figure 1A flow chart of an embodiment of a wind turbine life prediction method based on separable convolution provided by the present invention is shown in FIG. Figure 1 As shown in the figure, the wind turbine life prediction method based on separable convolution includes:

[0061] S101, obtaining a life sample set of a wind turbine;

[0062] S102, constructing an initial separable convolutional network model, where the initial separable convolutional network model includes initial network hyperparameters;

[0063] S103, optimizing the initial network hyperparameters based on the life sample set and the Bayesian optimization algorithm to obtain optimized network hyperparameters and an optimized separable convolutional network model;

[0064] S104, training the optimized separable convolutional network model based on the life sample set to obtain a target separable convolutional network model;

[0065] S105. Obtain vibration data of the wind turbine to be predicted, and predict the usable life of the wind turbine to be predicted based on the target separable convolutional network model.

[0066] Compared with the prior art, the wind turbine life prediction method based on separable convolution provided by the embodiment of the present invention can effectively reduce the model parameters by constructing an initial separable convolutional network model, thereby improving the efficiency of obtaining the target separable convolutional network model, so as to improve the efficiency of life prediction of wind turbines. In addition, the embodiment of the present invention optimizes the initial network hyperparameters through the Bayesian optimization algorithm, and can quickly obtain the optimized separable convolutional network model, thereby further improving the efficiency of obtaining the target separable convolutional network model, and further improving the efficiency of life prediction of wind turbines.

[0067] Furthermore, an embodiment of the present invention is arranged that after obtaining the optimized separable convolutional network model through the Bayesian optimization algorithm, the optimized separable convolutional network model is trained based on the life sample set to obtain the target separable convolutional network model, which can improve the accuracy of the target separable convolutional network model in predicting the useful life of the wind turbine.

[0068] In some embodiments of the present invention, the initial network hyperparameters include the initial number of backbone modules, the initial embedding form between different backbone modules, the initial number of convolution layers in the backbone modules, the initial number of convolution kernels, and the initial convolution kernel size; the optimized network hyperparameters include the optimized number of backbone modules, the optimized embedding form between different backbone modules, the optimized number of convolution layers in the backbone modules, the optimized number of convolution kernels, and the optimized convolution kernel size.

[0069] The initial separable convolutional network model in the embodiment of the present invention can be expressed as:

[0070]

[0071]

[0072] In the formula, is the kth convolutional layer in the i-th backbone module; are the initial network hyperparameters.

[0073] Each convolution layer includes channel convolution and point-by-point convolution. Compared with the conventional convolution layer in the prior art, the number of parameters can be reduced, and the time-phase initial separable convolution network model can be lightweight.

[0074] Specifically: Channel convolution applies a single convolution kernel to each input channel to map the temporal correlation of each sensor sequence separately. Then, point-wise convolution performs a convolution of size 1 to create a linear combination of the channel convolution outputs, that is, mapping the cross-channel correlation of different sensors.

[0075] In some embodiments of the present invention, Figure 2 As shown, step S103 includes:

[0076] S201, determining a first loss value of an initial separable convolutional network model based on a preset loss function and a life sample set;

[0077] S202, obtaining a candidate network hyperparameter based on the acquisition function, and obtaining a candidate separable convolutional network model according to the candidate network hyperparameter;

[0078] S203, determining a second loss value of the separable convolutional network model to be selected based on the loss function and the life sample set;

[0079] S204, evaluating the initial network hyperparameters and the candidate network hyperparameters based on the Gaussian process model, the first loss value, and the second loss value, and obtaining the optimal potential separable convolutional network model according to the optimal potential hyperparameters;

[0080] S205, determining a third loss value of the optimal potential separable convolutional network model based on the loss function and the life sample set, and judging whether the third loss value is less than the loss threshold and whether the number of times the acquisition function obtains the candidate network hyperparameters is greater than the threshold number;

[0081] S206. When the third loss value is less than the loss threshold, or the number of times the acquisition function obtains the hyperparameters of the network to be selected is greater than the threshold number of times, the optimal potential hyperparameters are the optimized network hyperparameters, and the optimal potential separable convolutional network model is the optimized separable convolutional network model; when the third loss value is greater than or equal to the loss threshold and the number of times the acquisition function obtains the hyperparameters of the network to be selected is less than or equal to the threshold number of times, repeat steps S202 to S206.

[0082] It should be understood that the loss threshold and the threshold number can be adjusted according to the actual application scenario and are not specifically limited here.

[0083] It should be noted that the acquisition function in step S202 can be any one of Upper confidence bound (UCB), Probability of Improvement (PI), and Expected Improvement (EI).

[0084] In a specific embodiment of the present invention, the loss function is:

[0085]

[0086] Where, MSE is the loss function; n is the total number of samples in the life sample set; is the actual usable life; To predict the useful life.

[0087] In some embodiments of the present invention, Figure 3 As shown, step S101 includes:

[0088] S301, collecting vibration data of the wind turbine throughout its life cycle, where the vibration data of the wind turbine throughout its life cycle includes a plurality of observation moments, a plurality of vibration sub-data corresponding to the plurality of observation moments one by one, and a life end moment;

[0089] S302, taking the time difference between each observation time among the multiple observation times and the end time of life as the remaining usable life of each vibrator data;

[0090] S303 : Construct a life sample set according to a plurality of vibrator sub-data and the remaining available life of each vibrator sub-data.

[0091] In some real-time examples of the present invention, step S301 is specifically as follows: using a vibration acceleration sensor to collect vibration data of the wind turbine during operation throughout its life cycle, each sampling lasts 0.1s, the sampling time interval is 10s, and the sampling frequency is 25.6kHz. During the sampling process, the maximum amplitude of the vibration acceleration is recorded. When the maximum amplitude of the vibration acceleration continuously exceeds the threshold acceleration amplitude, the moment when the maximum amplitude exceeds the threshold acceleration amplitude is considered to be the end of life. With the above sampling parameters, 2560 observation moments can be obtained for each sampling.

[0092] In a specific embodiment of the present invention, the threshold acceleration amplitude is 20g, where g is the acceleration due to gravity.

[0093] In order to eliminate data deviation and improve the reliability and accuracy of the obtained life sample set, in some embodiments of the present invention, after step S301, the following is further included:

[0094] Normalize the vibration data of wind turbines throughout their life cycle.

[0095] The embodiment of the present invention can eliminate data bias and improve the reliability and accuracy of the obtained life sample set by normalizing the vibration data of the entire life cycle, thereby improving the accuracy of the available life prediction of the wind turbine by the target separable convolutional network model obtained by training the life sample set.

[0096] In a specific embodiment of the present invention, the normalization process is a z-score process, and the implementation process is as follows:

[0097]

[0098] In the formula, is the normalized oscillator data; is the oscillator data at the i-th observation moment; is the mean of the oscillator data at all observation times; is the standard deviation of the oscillator data at all observation times; n is the total number of observation times.

[0099] In some embodiments of the present invention, Figure 4 As shown, step S104 includes:

[0100] S401, determining a life training set and a life test set from a life sample set;

[0101] S402, training the optimized separable convolutional network model based on the life training set to obtain a separable convolutional network model to be tested;

[0102] S403, testing the separable convolutional network model to be tested based on the life test set, obtaining the test accuracy, and determining whether the test accuracy is greater than a threshold accuracy;

[0103] S404: If the test accuracy is greater than the threshold accuracy, the separable convolutional network model to be tested is the target separable convolutional network model;

[0104] S405: If the test accuracy is less than or equal to the threshold accuracy, the optimized separable convolutional network model is retrained based on the life sample set.

[0105] The embodiment of the present invention sets a life test set for the separable convolutional network model to be tested obtained by training with a life training set, and sets that when the test accuracy is greater than a threshold accuracy, the separable convolutional network model to be tested is the target separable convolutional network model, which can further improve the prediction accuracy of the target separable convolutional network model for the useful life of the wind turbine.

[0106] It should be understood that the threshold accuracy can be adjusted or set according to actual application scenarios or empirical values, and is not specifically limited here.

[0107] In some embodiments of the present invention, a vibration acceleration sensor may be used to collect full life cycle data of a wind turbine three times, with the first two full life cycle data being used as life training sets, and the latter full life cycle data being used as a life test set.

[0108] It should be noted that when the optimized separable convolutional network model is trained in step S402, the optimized separable convolutional network model can also be trained according to the loss function in step S203, and the specific training process is not described here.

[0109] In a specific application example of the present invention, Figure 5 As shown, the target separable convolutional network model includes three nested connected backbone modules, each backbone module includes three convolutional layers. Figure 5 The arrow in the figure indicates the embedded form.

[0110] In order to better implement the fan life prediction method based on separable convolution in the embodiment of the present invention, on the basis of the fan life prediction method based on separable convolution, the embodiment of the present invention also provides a fan life prediction device based on separable convolution, such as Figure 6 As shown, the wind turbine life prediction device 600 based on separable convolution includes:

[0111] The sample set acquisition module 601 is used to acquire the life sample set of the wind turbine;

[0112] An initial model building module 602 is used to build an initial separable convolutional network model, where the initial separable convolutional network model includes initial network hyperparameters;

[0113] A model optimization module 603 is used to optimize the initial network hyperparameters based on the life sample set and the Bayesian optimization algorithm to obtain the optimized network hyperparameters and optimize the separable convolutional network model;

[0114] A model training module 604 is used to train the optimized separable convolutional network model based on the life sample set to obtain a target separable convolutional network model;

[0115] The life prediction module 605 is used to obtain the vibration data of the wind turbine to be predicted, and predict the usable life of the wind turbine to be predicted based on the target separable convolutional network model.

[0116] The separable convolution-based wind turbine life prediction device 600 provided in the above embodiment can implement the technical solution described in the above embodiment of the separable convolution-based wind turbine life prediction method. The specific implementation principles of the above modules or units can refer to the corresponding contents in the above embodiment of the separable convolution-based wind turbine life prediction method, which will not be repeated here.

[0117] like Figure 7 As shown, the present invention also provides an electronic device 700. The electronic device 700 includes a processor 701, a memory 702 and a display 703. Figure 7 Only some components of the electronic device 700 are shown, but it should be understood that it is not required to implement all of the components shown, and more or fewer components may be implemented instead.

[0118] In some embodiments, the processor 701 may be a central processing unit (CPU), a microprocessor or other data processing chip, used to run program codes or process data stored in the memory 702, such as the wind turbine life prediction method based on separable convolution in the present invention.

[0119] In some embodiments, the processor 701 may be a single server or a server group. The server group may be centralized or distributed. In some embodiments, the processor 701 may be local or remote. In some embodiments, the processor 701 may be implemented in a cloud platform. In one embodiment, the cloud platform may include a private cloud, a public cloud, a hybrid cloud, a community cloud, a distributed cloud, an internal cloud, a multi-cloud, etc., or any combination thereof.

[0120] In some embodiments, the memory 702 may be an internal storage unit of the electronic device 700, such as a hard disk or memory of the electronic device 700. In other embodiments, the memory 702 may also be an external storage device of the electronic device 700, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc., equipped on the electronic device 700.

[0121] Furthermore, the memory 702 may include both an internal storage unit of the electronic device 700 and an external storage device. The memory 702 is used to store application software installed in the electronic device 700 and various data.

[0122] In some embodiments, the display 703 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, an OLED (Organic Light-Emitting Diode) touch device, etc. The display 703 is used to display information on the electronic device 700 and to display a visual user interface. The components 701-703 of the electronic device 700 communicate with each other via a system bus.

[0123] In some embodiments of the present invention, when the processor 701 executes the wind turbine life prediction program based on separable convolution in the memory 702, the following steps may be implemented:

[0124] Obtain a life sample set of wind turbines;

[0125] Construct an initial separable convolutional network model, which includes initial network hyperparameters;

[0126] Based on the life sample set and Bayesian optimization algorithm, the initial network hyperparameters are optimized to obtain the optimized network hyperparameters and the optimized separable convolutional network model;

[0127] The optimized separable convolutional network model is trained based on the life sample set to obtain the target separable convolutional network model;

[0128] The vibration data of the wind turbine to be predicted is obtained, and the usable life of the wind turbine to be predicted is predicted based on the target separable convolutional network model.

[0129] It should be understood that: when the processor 701 executes the separable convolution-based wind turbine life prediction program in the memory 702, in addition to the above functions, other functions can also be implemented. For details, please refer to the description of the corresponding method embodiment above.

[0130] Furthermore, the embodiment of the present invention does not specifically limit the type of the electronic device 700 mentioned, and the electronic device 700 may be a portable electronic device such as a mobile phone, a tablet computer, a personal digital assistant (PDA), a wearable device, a laptop computer, etc. Exemplary embodiments of portable electronic devices include but are not limited to portable electronic devices equipped with IOS, Android, Microsoft or other operating systems. The above-mentioned portable electronic devices may also be other portable electronic devices, such as a laptop computer with a touch-sensitive surface (e.g., a touch panel). It should also be understood that in some other embodiments of the present invention, the electronic device 700 may not be a portable electronic device, but a desktop computer with a touch-sensitive surface (e.g., a touch panel).

[0131] Accordingly, an embodiment of the present application also provides a computer-readable storage medium, which is used to store computer-readable programs or instructions. When the program or instructions are executed by a processor, it can implement the steps or functions of the wind turbine life prediction method based on separable convolution provided in the above-mentioned method embodiments.

[0132] Those skilled in the art will appreciate that all or part of the processes of the above-mentioned embodiments can be implemented by instructing related hardware (such as a processor, a controller, etc.) through a computer program, and the computer program can be stored in a computer-readable storage medium, wherein the computer-readable storage medium is a disk, an optical disk, a read-only storage memory, or a random access memory, etc.

[0133] The above is a detailed introduction to the fan life prediction method, device, equipment and medium based on separable convolution provided by the present invention. Specific examples are used in this article to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea; at the same time, for technical personnel in this field, according to the idea of ​​the present invention, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A wind turbine life prediction method based on separable convolution, characterized in that: include: Obtain a life sample set of wind turbines; Constructing an initial separable convolutional network model, wherein the initial separable convolutional network model includes initial network hyperparameters; Optimizing the initial network hyperparameters based on the life sample set and the Bayesian optimization algorithm to obtain optimized network hyperparameters and an optimized separable convolutional network model; Training the optimized separable convolutional network model based on the life sample set to obtain a target separable convolutional network model; Acquire vibration data of the wind turbine to be predicted, and predict the usable life of the wind turbine to be predicted based on the target separable convolutional network model; The optimizing the initial network hyperparameters based on the life sample set and the Bayesian optimization algorithm to obtain the optimized network hyperparameters and the optimized separable convolutional network model includes: Step 1: determining a first loss value of the initial separable convolutional network model based on a preset loss function and the life sample set; Step 2: obtaining the candidate network hyperparameters based on the acquisition function, and obtaining the candidate separable convolutional network model according to the candidate network hyperparameters; Step 3: determining a second loss value of the separable convolutional network model to be selected based on the loss function and the life sample set; Step 4: evaluating the initial network hyperparameters and the selected network hyperparameters based on the Gaussian process model, the first loss value, and the second loss value, and obtaining the optimal potential separable convolutional network model according to the optimal potential hyperparameters; Step 5: determining a third loss value of the optimal potential separable convolutional network model based on the loss function and the life sample set, and judging whether the third loss value is less than a loss threshold and whether the number of times the acquisition function obtains the candidate network hyperparameters is greater than a threshold number; Step six: when the third loss value is less than the loss threshold, or the number of times the acquisition function obtains the network hyperparameters to be selected is greater than the threshold number of times, the optimal potential hyperparameters are the optimized network hyperparameters, and the optimal potential separable convolutional network model is the optimized separable convolutional network model; when the third loss value is greater than or equal to the loss threshold and the number of times the acquisition function obtains the network hyperparameters to be selected is less than or equal to the threshold number of times, repeat steps two to six.

2. The wind turbine life prediction method based on separable convolution according to claim 1 is characterized in that: The step of obtaining a life sample set of a wind turbine generator comprises: Collecting vibration data of the wind turbine over its entire life cycle, wherein the vibration data over its entire life cycle includes a plurality of observation moments, a plurality of vibration sub-data corresponding to the plurality of observation moments, and a life end moment; using the time difference between each of the plurality of observation times and the end-of-life time as the remaining usable life of each of the vibrator data; The life sample set is constructed according to the plurality of vibration sub-data and the remaining available life of each of the vibration sub-data.

3. The wind turbine life prediction method based on separable convolution according to claim 2 is characterized in that: After collecting the vibration data of the wind turbine generator throughout its life cycle, the method further includes: The whole life cycle vibration data of the wind turbine is normalized.

4. The wind turbine life prediction method based on separable convolution according to claim 1 is characterized in that: The initial network hyperparameters include the initial number of backbone modules, the initial embedding form between different backbone modules, the initial number of convolution layers in the backbone modules, the initial number of convolution kernels, and the initial convolution kernel size; the optimized network hyperparameters include the optimized number of backbone modules, the optimized embedding form between different backbone modules, the optimized number of convolution layers in the backbone modules, the optimized number of convolution kernels, and the optimized convolution kernel size.

5. The wind turbine life prediction method based on separable convolution according to claim 1 is characterized in that: The step of training the optimized separable convolutional network model based on the life sample set to obtain a target separable convolutional network model includes: Determine a life training set and a life test set from the life sample set; Training the optimized separable convolutional network model based on the lifespan training set to obtain a separable convolutional network model to be tested; Testing the separable convolutional network model to be tested based on the life test set to obtain a test accuracy, and determining whether the test accuracy is greater than a threshold accuracy; If the test accuracy is greater than the threshold accuracy, the separable convolutional network model to be tested is the target separable convolutional network model; If the test accuracy is less than or equal to the threshold accuracy, the optimized separable convolutional network model is retrained based on the life sample set.

6. The wind turbine life prediction method based on separable convolution according to claim 1 is characterized in that: The loss function is: Wherein, MSE is the loss function; n is the total number of samples in the life sample set; is the actual usable life; To predict the useful life.

7. A wind turbine life prediction device based on separable convolution, characterized in that: The wind turbine life prediction method based on separable convolution according to any one of claims 1 to 6, the device comprising: A sample set acquisition module is used to acquire a life sample set of a wind turbine; An initial model building module, used to build an initial separable convolutional network model, wherein the initial separable convolutional network model includes initial network hyperparameters; A model optimization module, used to optimize the initial network hyperparameters based on the life sample set and the Bayesian optimization algorithm, obtain optimized network hyperparameters and optimize the separable convolutional network model; A model training module, used for training the optimized separable convolutional network model based on the life sample set to obtain a target separable convolutional network model; The life prediction module is used to obtain vibration data of the wind turbine to be predicted and predict the usable life of the wind turbine to be predicted based on the target separable convolutional network model.

8. An electronic device, characterized in that: comprising a memory and a processor, wherein: The memory is used to store programs; The processor is coupled to the memory and is used to execute the program stored in the memory to implement the steps in the wind turbine life prediction method based on separable convolution as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that: Used to store computer-readable programs or instructions, which, when executed by a processor, can implement the steps in the wind turbine life prediction method based on separable convolution as described in any one of claims 1 to 6.