A method for predicting remaining life of wind turbine based on multi-channel separable residual neural network

Through the fan residual life prediction method based on multi-channel separable residual neural network, the problem of difficulty in discovering failed components of wind turbine units and high maintenance costs is solved, and the operation efficiency and cost reduction of wind farms are improved.

CN114547967BActive Publication Date: 2024-11-26WUHAN UNIV OF TECH
View PDF 0 Cites 2 Cited by

Patent Information

Application Number
CN202210031823.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-01-12
Publication Date
2024-11-26
Estimated Expiration
2042-01-12

AI Technical Summary

Technical Problem

The existing technology cannot detect failed components of wind turbines in a timely manner, resulting in high maintenance costs of wind farms and affecting the operating efficiency of wind farms.

Method used

The residual life prediction method of fan based on multi-channel separable residual neural network is adopted. By collecting multi-channel vibration data for the entire life cycle of wind turbines, a residual neural network model with separable convolution module is constructed, and the data is trained and verified. , predict the remaining available life of the wind turbine.

Benefits of technology

This method can promptly detect failed components of wind turbines, reduce the maintenance cost of the wind farm, improve the operating efficiency of the wind farm, and improve the training efficiency by reducing model parameters, accurately predicting the remaining available life of the wind turbine.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114547967B_ABST
    Figure CN114547967B_ABST
Patent Text Reader

Abstract

The present invention discloses a method for predicting the remaining life of a wind turbine based on a multi-channel separable residual neural network, comprising: collecting multi-channel vibration data of the entire life cycle of a wind turbine during operation; dividing the multi-channel vibration data into a training data set, a verification data set and a test data set, and taking the distance between the observation time of each sample and the end time as the remaining usable life of the sample; performing normalization preprocessing on the training data set and the verification data set; constructing a residual neural network model of a separable convolution module, and using the training data set to train the residual neural network model, using the verification data set to verify the residual neural network model, and judging whether the residual neural network model meets the iteration condition; when the residual neural network model meets the iteration condition, performing normalization preprocessing on the test data set and inputting it into the residual neural network model for feature extraction, and outputting the remaining life prediction value corresponding to the observation time.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of neural network intelligent algorithms, and in particular to a method for predicting the remaining life of a fan based on a multi-channel separable residual neural network. Background Art

[0002] With the increasing awareness of environmental protection, people have developed some sustainable and carbon-neutral renewable energy sources, which are gradually replacing some traditional fossil raw materials such as oil and coal. According to a latest report on the statistics of renewable energy installed capacity, the total installed capacity of wind power in the world in 2019 was 622,704MW, accounting for 24.55% of the total installed capacity of renewable energy, second only to hydropower. The average annual growth rate of global wind power generation is 10.44%, and its development speed is second only to solar energy. Due to the low efficiency of wind power generation, it has always been a hot research topic to improve the utilization rate of wind energy. In addition to studying the model selection of wind turbines, it is also useful to reasonably design the structure of wind turbines. At the same time, wind turbines are often in dynamic and severe weather conditions and experience changeable and harsh working environments, which makes them more prone to failure than other ordinary machinery. If a component of a wind turbine fails due to wear and tear without the knowledge of the workers, it is likely to cause damage to other components and even cause the wind turbine to shut down, resulting in huge economic losses. The operation and maintenance costs of onshore wind farms have exceeded 25% of the total cost, and this cost for offshore wind power is even higher. Therefore, timely detection of failed components of wind turbines is of great significance to reducing the maintenance cost of wind farms and improving the operating efficiency of wind farms. Summary of the invention

[0003] The purpose of the present invention is to overcome the above technical deficiencies and propose a method for predicting the remaining life of a wind turbine based on a multi-channel separable residual neural network to solve the technical problems in the prior art of being unable to timely detect failed components of wind turbines and the high maintenance cost of wind farms.

[0004] In order to achieve the above technical objectives, in a first aspect, the technical solution of the present invention provides a method for predicting the remaining life of a wind turbine based on a multi-channel separable residual neural network, comprising the following steps:

[0005] Collect multi-channel vibration data of the wind turbine during its entire life cycle;

[0006] Dividing the multi-channel vibration data into a training data set, a validation data set and a test data set, and taking the time difference between the observation time and the end time of each sample as the remaining useful life of the sample;

[0007] Performing normalization preprocessing on the training data set and the verification data set to eliminate data bias;

[0008] Constructing a residual neural network model of a separable convolution module, and using the training data set to train the residual neural network model, using the verification data set to verify the residual neural network model, and determining whether the residual neural network model meets the iteration condition;

[0009] When the residual neural network model meets the iteration condition, the test data set is normalized and preprocessed and input into the residual neural network model for feature extraction, and the remaining life prediction value corresponding to the observation time is output.

[0010] Compared with the prior art, the method for predicting the remaining life of a fan based on a multi-channel separable residual neural network provided by the present invention has the following beneficial effects:

[0011] The wind turbine remaining life prediction method based on a multi-channel separable residual neural network provided by the present invention maps low-level features to high-level features, which can effectively solve the degradation phenomenon of the model; by establishing a separable convolution module, the model parameters can be effectively reduced, thereby greatly improving the training efficiency of the model, and accurately realizing the remaining useful life prediction of the wind turbine. The present invention can timely discover the failed components of the wind turbine set, reduce the maintenance cost of the wind farm, and improve the operating efficiency of the wind farm.

[0012] According to some embodiments of the present invention, the residual neural network model of constructing a separable convolution module comprises the steps of:

[0013] Determine the number of convolution layers, convolution kernel size and number of convolution kernels of the residual neural network to construct a residual neural network model of the separable convolution module.

[0014] According to some embodiments of the present invention, after determining whether the residual neural network model satisfies the iteration condition, the following steps are included:

[0015] When the residual neural network model does not meet the iteration condition, the network parameters of the residual neural network are adjusted.

[0016] According to some embodiments of the present invention, the step of collecting multi-channel vibration data of a wind turbine during operation over its entire life cycle comprises the following steps:

[0017] Acceleration sensors are respectively installed in the first direction and the second direction of the wind turbine to collect the multi-channel vibration data of the wind turbine.

[0018] According to some embodiments of the present invention, the method for predicting the remaining life of a wind turbine based on a multi-channel separable residual neural network further includes the following steps:

[0019] When the amplitude of the multi-channel vibration data collected during the degradation process reaches more than 10 times the maximum amplitude of the multi-channel vibration data during the normal operation stage, it is determined that the life of the wind turbine generator has ended.

[0020] According to some embodiments of the present invention, dividing the multi-channel vibration data into a training data set, a validation data set and a test data set comprises the steps of:

[0021] A preset number of data points are collected each time, and the multi-channel vibration data is divided into 6 samples according to the sample length, and the first 4 samples are selected as the training data set, the fifth sample is selected as the verification data set, and the sixth sample is selected as the test data set;

[0022] The same remaining available lifespan is used as the sample label for the six samples collected each time to generate a labeled data set.

[0023] According to some embodiments of the present invention, the normalization preprocessing of the training data set and the verification data set comprises the steps of:

[0024] The labeled data set is normalized, and the implementation process is as follows:

[0025]

[0026] In the formula, x i is the i-th data set sample, E(x i ) is the sample mean, σ(x i ) is the sample standard deviation.

[0027] According to some embodiments of the present invention, the method of training the residual neural network model using the training data set and verifying the residual neural network model using the verification data set comprises the steps of:

[0028] Constructing a separable convolution module according to the number of convolution layers, convolution kernel size and number of convolution kernels of the residual neural network;

[0029] A residual neural network model is constructed based on the separable convolution module, and the residual neural network model is used to predict the remaining useful life.

[0030] In a second aspect, the technical solution of the present invention provides a system for predicting the remaining life of a wind turbine based on a multi-channel separable residual neural network, comprising: a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein when the processor executes the computer program, it implements a method for predicting the remaining life of a wind turbine based on a multi-channel separable residual neural network as described in any one of the first aspects.

[0031] In the third aspect, the technical solution of the present invention provides a computer-readable storage medium, characterized in that the computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are used to enable a computer to execute a wind turbine remaining life prediction method based on a multi-channel separable residual neural network as described in any one of the first aspects.

[0032] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, wherein the abstract drawings are identical to one of the drawings in the specification:

[0034] Figure 1 A flowchart of a method for predicting the remaining life of a wind turbine based on a multi-channel separable residual neural network provided by an embodiment of the present invention;

[0035] Figure 2 A flowchart of a method for predicting the remaining life of a wind turbine based on a multi-channel separable residual neural network provided by another embodiment of the present invention;

[0036] Figure 3 A flowchart of a method for predicting the remaining life of a wind turbine based on a multi-channel separable residual neural network is provided as another embodiment of the present invention. DETAILED DESCRIPTION

[0037] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0038] It should be noted that, although the functional modules are divided in the system schematic diagram and the logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the system or the order in the flowchart. The terms "first", "second", etc. in the specification, claims and the above drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0039] The present invention provides a method for predicting the remaining useful life of a wind turbine based on a multi-channel separable residual neural network. By mapping low-level features to high-level features, the degradation of the model can be effectively solved. By establishing a separable convolution module, the model parameters can be effectively reduced, thereby greatly improving the training efficiency of the model and accurately realizing the prediction of the remaining useful life of the wind turbine. The present invention can timely discover the failed components of the wind turbine, reduce the maintenance cost of the wind farm, and improve the operating efficiency of the wind farm.

[0040] The embodiments of the present invention are further described below in conjunction with the accompanying drawings.

[0041] Reference Figure 1 , Figure 1 A flowchart of a method for predicting the remaining life of a wind turbine based on a multi-channel separable residual neural network is provided as an embodiment of the present invention. The method for predicting the remaining life of a wind turbine based on a multi-channel separable residual neural network includes but is not limited to steps S110 to S150.

[0042] Step S110, collecting multi-channel vibration data of the wind turbine during its entire life cycle;

[0043] Step S120, dividing the multi-channel vibration data into a training data set, a verification data set and a test data set, and taking the time difference between the observation time and the end time of each sample as the remaining usable life of the sample;

[0044] Step S130, performing normalization preprocessing on the training data set and the verification data set to eliminate data deviation;

[0045] Step S140, constructing a residual neural network model of a separable convolution module, and using a training data set to train the residual neural network model, using a verification data set to verify the residual neural network model, and determining whether the residual neural network model meets the iteration condition;

[0046] Step S150, when the residual neural network model meets the iteration condition, the test data set is normalized and preprocessed and input into the residual neural network model for feature extraction, and the remaining life prediction value corresponding to the observation time is output.

[0047] In one embodiment, a method for predicting the remaining life of a wind turbine based on a multi-channel separable residual neural network includes the following steps: first, collecting multi-channel vibration data of the wind turbine during its entire life cycle during operation; then, dividing the multi-channel vibration data into a training data set, a verification data set, and a test data set, and taking the time difference between the observation time of each sample and the end time as the remaining useful life of the sample; secondly, performing normalization preprocessing on the training data set and the verification data set to eliminate data bias; thirdly, constructing a residual neural network model of a separable convolution module, and using the training data set to train the residual neural network model, using the verification data set to verify the residual neural network model, and judging whether the residual neural network model meets the iteration condition; finally, when the residual neural network model meets the iteration condition, performing normalization preprocessing on the test data set and inputting it into the residual neural network model for feature extraction, and outputting the remaining life prediction value corresponding to the observation time.

[0048] The wind turbine remaining life prediction method based on a multi-channel separable residual neural network provided in this embodiment maps low-level features to high-level features, which can effectively solve the degradation phenomenon of the model; by establishing a separable convolution module, the model parameters can be effectively reduced, thereby greatly improving the training efficiency of the model, and accurately realizing the remaining useful life prediction of the wind turbine. The present invention can timely discover the failed components of the wind turbine set, can reduce the maintenance cost of the wind farm, and improve the operating efficiency of the wind farm.

[0049] In one embodiment, a method for predicting the remaining life of a wind turbine based on a multi-channel separable residual neural network includes the following steps: first, collecting multi-channel vibration data of the entire life cycle of a wind turbine during operation; then, dividing the multi-channel vibration data into a training data set, a verification data set, and a test data set, and taking the time difference between the observation time of each sample and the end time as the remaining usable life of the sample; secondly, normalizing and preprocessing the training data set and the verification data set to eliminate data deviation; thirdly, constructing a residual neural network model of a separable convolution module, and using the training data set to train the residual neural network model, using the verification data set to verify the residual neural network model, and judging whether the residual neural network model meets the iteration condition; finally, when the residual neural network model meets the iteration condition, normalizing and preprocessing the test data set and inputting it into the residual neural network model for feature extraction, and outputting the remaining life prediction value corresponding to the observation time. Constructing a residual neural network model of a separable convolution module includes the steps of determining the number of convolution layers, convolution kernel size, and number of convolution kernels of the residual neural network to construct a residual neural network model of a separable convolution module.

[0050] In one embodiment, a method for predicting the remaining life of a wind turbine based on a multi-channel separable residual neural network includes the following steps: first, collecting multi-channel vibration data of the wind turbine during the entire life cycle of the working process; then, dividing the multi-channel vibration data into a training data set, a verification data set, and a test data set, and taking the time difference between the observation time of each sample and the end time as the remaining usable life of the sample; secondly, normalizing and preprocessing the training data set and the verification data set to eliminate data deviation; thirdly, constructing a residual neural network model of a separable convolution module, and using the training data set to train the residual neural network model, using the verification data set to verify the residual neural network model, and judging whether the residual neural network model meets the iteration condition; finally, when the residual neural network model meets the iteration condition, normalizing and preprocessing the test data set and inputting it into the residual neural network model for feature extraction, and outputting the remaining life prediction value corresponding to the observation time. After judging whether the residual neural network model meets the iteration condition, the method includes the following steps: when the residual neural network model does not meet the iteration condition, adjusting the network parameters of the residual neural network.

[0051] In one embodiment, a method for predicting the remaining life of a wind turbine based on a multi-channel separable residual neural network includes the following steps: first, collecting multi-channel vibration data of the wind turbine during its entire life cycle during operation; then, dividing the multi-channel vibration data into a training data set, a verification data set, and a test data set, and taking the time difference between the observation time of each sample and the end time as the remaining useful life of the sample; secondly, performing normalization preprocessing on the training data set and the verification data set to eliminate data bias; thirdly, constructing a residual neural network model of a separable convolution module, and using the training data set to train the residual neural network model, using the verification data set to verify the residual neural network model, and judging whether the residual neural network model meets the iteration condition; finally, when the residual neural network model meets the iteration condition, performing normalization preprocessing on the test data set and inputting it into the residual neural network model for feature extraction, and outputting the remaining life prediction value corresponding to the observation time.

[0052] The method collects multi-channel vibration data of a wind turbine during its entire life cycle, including the steps of installing acceleration sensors in a first direction and a second direction of the wind turbine respectively to collect multi-channel vibration data of the wind turbine.

[0053] Two acceleration sensors are installed in the X and Y directions of the device under test to collect vibration data of the device. The sampling interval can be 1 minute, each sampling time can last for 1.28 seconds, and the sampling frequency can be 25.6kHz.

[0054] In one embodiment, a method for predicting the remaining life of a wind turbine based on a multi-channel separable residual neural network includes the following steps: first, collecting multi-channel vibration data of the wind turbine during its entire life cycle during operation; then, dividing the multi-channel vibration data into a training data set, a verification data set, and a test data set, and taking the time difference between the observation time of each sample and the end time as the remaining useful life of the sample; secondly, performing normalization preprocessing on the training data set and the verification data set to eliminate data bias; thirdly, constructing a residual neural network model of a separable convolution module, and using the training data set to train the residual neural network model, using the verification data set to verify the residual neural network model, and judging whether the residual neural network model meets the iteration condition; finally, when the residual neural network model meets the iteration condition, performing normalization preprocessing on the test data set and inputting it into the residual neural network model for feature extraction, and outputting the remaining life prediction value corresponding to the observation time.

[0055] Collecting multi-channel vibration data of the wind turbine during its entire life cycle includes the following steps: Acceleration sensors are respectively installed in a first direction and a second direction of the wind turbine to collect multi-channel vibration data of the wind turbine. When the amplitude of the multi-channel vibration data collected during the degradation process reaches more than 10 times the maximum amplitude of the multi-channel vibration data during the normal operation stage, it is determined that the life of the wind turbine has ended.

[0056] refer to Figure 2 , Figure 2 A flowchart of a method for predicting the remaining life of a wind turbine based on a multi-channel separable residual neural network is provided as another embodiment of the present invention; the method for predicting the remaining life of a wind turbine based on a multi-channel separable residual neural network includes but is not limited to steps S210 to S220.

[0057] Step S210, collecting a preset number of data points each time, dividing the multi-channel vibration data into 6 samples according to the sample length, selecting the first 4 samples as the training data set, the 5th sample as the verification data set, and the 6th sample as the test data set;

[0058] Step S210, using the same remaining usable lifespan as sample labels for the six samples collected each time to generate a labeled data set.

[0059] In one embodiment, a method for predicting the remaining life of a wind turbine based on a multi-channel separable residual neural network includes the following steps: first, collecting multi-channel vibration data of the wind turbine during its entire life cycle during operation; then, dividing the multi-channel vibration data into a training data set, a verification data set, and a test data set, and taking the time difference between the observation time of each sample and the end time as the remaining useful life of the sample; secondly, performing normalization preprocessing on the training data set and the verification data set to eliminate data bias; thirdly, constructing a residual neural network model of a separable convolution module, and using the training data set to train the residual neural network model, using the verification data set to verify the residual neural network model, and judging whether the residual neural network model meets the iteration condition; finally, when the residual neural network model meets the iteration condition, performing normalization preprocessing on the test data set and inputting it into the residual neural network model for feature extraction, and outputting the remaining life prediction value corresponding to the observation time.

[0060] The multi-channel vibration data is divided into a training data set, a verification data set and a test data set, including the following steps: collecting a preset number of data points each time, dividing the multi-channel vibration data into 6 samples according to the sample length, selecting the first 4 samples as the training data set, the fifth sample as the verification data set, and the sixth sample as the test data set; using the same remaining usable life as a sample label for the 6 samples collected each time to generate a labeled data set.

[0061] During the data sampling process, 32768 data points are collected each time. The sampled data is divided into 2560 points according to the sample length. The first 4 samples are selected as training data, the 5th sample is used as verification data, and the 6th sample is used as test data. The 6 sample data obtained each time use the same remaining usable life as the sample label to generate a labeled data set.

[0062] In one embodiment, a method for predicting the remaining life of a wind turbine based on a multi-channel separable residual neural network includes the following steps: first, collecting multi-channel vibration data of the wind turbine during its entire life cycle during operation; then, dividing the multi-channel vibration data into a training data set, a verification data set, and a test data set, and taking the time difference between the observation time of each sample and the end time as the remaining useful life of the sample; secondly, performing normalization preprocessing on the training data set and the verification data set to eliminate data bias; thirdly, constructing a residual neural network model of a separable convolution module, and using the training data set to train the residual neural network model, using the verification data set to verify the residual neural network model, and judging whether the residual neural network model meets the iteration condition; finally, when the residual neural network model meets the iteration condition, performing normalization preprocessing on the test data set and inputting it into the residual neural network model for feature extraction, and outputting the remaining life prediction value corresponding to the observation time.

[0063] The multi-channel vibration data is divided into a training data set, a validation data set and a test data set, including the following steps: each time a preset number of data points are collected, the multi-channel vibration data is divided into 6 samples according to the sample length, the first 4 samples are selected as the training data set, the fifth sample is used as the validation data set, and the sixth sample is used as the test data set; the 6 samples collected each time use the same remaining usable life as the sample label to generate a labeled data set. The training data set and the validation data set are normalized and preprocessed, including the following steps: the labeled data set is normalized, and the implementation process is as follows:

[0064]

[0065] In the formula, x i is the i-th data set sample, E(x i ) is the sample mean, σ(x i ) is the sample standard deviation.

[0066] refer to Figure 3 , Figure 3 A flowchart of a method for predicting the remaining life of a wind turbine based on a multi-channel separable residual neural network is provided as another embodiment of the present invention; the method for predicting the remaining life of a wind turbine based on a multi-channel separable residual neural network includes but is not limited to steps S310 to S320.

[0067] Step S310, constructing a separable convolution module according to the number of convolution layers, convolution kernel size and number of convolution kernels of the residual neural network;

[0068] Step S320, constructing a residual neural network model based on the separable convolution module, and the residual neural network model is used to predict the remaining useful life.

[0069] In one embodiment, a method for predicting the remaining life of a wind turbine based on a multi-channel separable residual neural network includes the following steps: first, collecting multi-channel vibration data of the wind turbine during its entire life cycle during operation; then, dividing the multi-channel vibration data into a training data set, a verification data set, and a test data set, and taking the time difference between the observation time of each sample and the end time as the remaining useful life of the sample; secondly, performing normalization preprocessing on the training data set and the verification data set to eliminate data bias; thirdly, constructing a residual neural network model of a separable convolution module, and using the training data set to train the residual neural network model, using the verification data set to verify the residual neural network model, and judging whether the residual neural network model meets the iteration condition; finally, when the residual neural network model meets the iteration condition, performing normalization preprocessing on the test data set and inputting it into the residual neural network model for feature extraction, and outputting the remaining life prediction value corresponding to the observation time.

[0070] Constructing a residual neural network model of a separable convolution module includes the steps of: determining the number of convolution layers, convolution kernel size, and number of convolution kernels of the residual neural network to construct the residual neural network model of the separable convolution module. The residual neural network model is trained using a training data set, and the residual neural network model is verified using a verification data set, including the steps of: constructing a separable convolution module according to the number of convolution layers, convolution kernel size, and number of convolution kernels of the residual neural network; constructing a residual neural network model based on the separable convolution module, and the residual neural network model is used to predict the remaining usable life.

[0071] First, according to the number of convolutional layers, the number and size of convolutional kernels of the residual network, the corresponding separable convolution module is constructed.

[0072] Since deep network models may experience model degradation, this problem can be effectively solved by directly mapping low-level features to high-level features. Therefore, a residual network is constructed based on the separable convolutional model for remaining useful life prediction.

[0073] The implementation process of the model can be described as:

[0074] f(x)=f n (...f2(f1(x,θ 1 ),θ 2 )...,θ n )

[0075] f(.) is each convolutional module of the residual network, and θ is the learnable parameter of the corresponding module.

[0076] For each f(.), it is composed of a separable convolution module and a direct mapping module, namely:

[0077] f i (x) = f i-identity (x)+f i-factorize (x)

[0078] Where f i-identity (x) is a direct mapping module, which combines low-level features with high-level features by converting the convolution layer (convolution kernel of size 1), f i-factorize (x) is a separable convolution module, which is used to lightweight the network model by decomposing the conventional convolution into deep convolution and point convolution. The deep convolution implements the convolution operation by grouping convolution, while the point convolution is a conventional convolution operation with a size of 1.

[0079] Each layer of the residual model consists of three main layers:

[0080] 1) Convolutional layer:

[0081]

[0082] Where K is the convolution kernel, also known as the weight matrix with a shape of w×h. The convolution kernel and the input data x i Perform convolution operation to obtain the output result of the convolution layer.

[0083] 2) Batch Normalization Layer:

[0084]

[0085] Where E[x] and Var[x] are the mean and variance of the input data x, and ε is an infinitesimal. γ and β are the scale factor and translation factor, respectively. In the model training process, the minimum batch method is often used to improve the model fitting effect, but this will also slow down the training speed. Batch normalization can effectively solve the above problem.

[0086] 3) ReLu activation layer:

[0087]

[0088] The activation layer can improve the expression of output features by using nonlinear functions.

[0089] Since the model is used for remaining life prediction, the last layer of the network is a regression layer, and the number of neurons in this layer is 1. During the network training process, the mean square error is used as the loss function of the network model, and the corresponding formula is:

[0090]

[0091] where y i It is real data. is the predicted data, and n is the total number of samples.

[0092] The present invention also provides a system for predicting the remaining life of a fan based on a multi-channel separable residual neural network, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the method for predicting the remaining life of a fan based on a multi-channel separable residual neural network as described above.

[0093] The processor and the memory may be connected via a bus or other means.

[0094] The memory, as a non-transient computer-readable storage medium, can be used to store non-transient software programs and non-transient computer executable programs. In addition, the memory may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory may optionally include a memory remotely disposed relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0095] It should be noted that the wind turbine remaining life prediction system based on a multi-channel separable residual neural network in this embodiment may include a business processing module, an edge database, a server version information register, and a data synchronization module. When the processor executes the computer program, it implements the wind turbine remaining life prediction method based on a multi-channel separable residual neural network as applied to the wind turbine remaining life prediction system based on a multi-channel separable residual neural network as mentioned above.

[0096] The device embodiments described above are merely illustrative, and the units described as separate components may or may not be physically separated, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0097] In addition, an embodiment of the present invention also provides a computer-readable storage medium, which stores computer-executable instructions, and the computer-executable instructions are executed by a processor or a controller, for example, by a processor in the above-mentioned terminal embodiment, so that the above-mentioned processor can execute the wind turbine remaining life prediction method based on a multi-channel separable residual neural network in the above-mentioned embodiment.

[0098] It will be appreciated by those skilled in the art that all or some of the steps and systems in the disclosed method above may be implemented as software, firmware, hardware and appropriate combinations thereof. Some physical components or all physical components may be implemented as software executed by a processor, such as a central processing unit, a digital signal processor or a microprocessor, or may be implemented as hardware, or may be implemented as an integrated circuit, such as an application specific integrated circuit. Such software may be distributed on a computer-readable medium, which may include a computer storage medium (or a non-transitory medium) and a communication medium (or a temporary medium). As known to those skilled in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tapes, disk storage or other magnetic storage devices, or any other medium that may be used to store desired information and may be accessed by a computer. Furthermore, it is well known to those skilled in the art that communication media typically embodies computer readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism, and may include any information delivery media.

[0099] The above is a specific description of the preferred implementation of the present invention, but the present invention is not limited to the above-mentioned implementation mode. Technical personnel familiar with the field can also make various equivalent deformations or substitutions without violating the spirit of the present invention. These equivalent deformations or substitutions are all included in the scope defined by the claims of the present invention.

[0100] The specific implementation of the present invention described above does not constitute a limitation on the protection scope of the present invention. Any other corresponding changes and modifications made based on the technical concept of the present invention should be included in the protection scope of the claims of the present invention.

Claims

1. A method for predicting the remaining life of a fan based on a multi-channel separable residual neural network, characterized in that: The following steps are involved: Collect multi-channel vibration data of the wind turbine during its entire life cycle; Dividing the multi-channel vibration data into a training data set, a validation data set and a test data set, and taking the time difference between the observation time and the end time of each sample as the remaining useful life of the sample; Performing normalization preprocessing on the training data set and the verification data set to eliminate data bias; Constructing a residual neural network model of a separable convolution module, and using the training data set to train the residual neural network model, using the verification data set to verify the residual neural network model, and determining whether the residual neural network model meets the iteration condition; When the residual neural network model meets the iteration condition, the test data set is normalized and preprocessed and input into the residual neural network model for feature extraction, and the remaining life prediction value corresponding to the observation time is output; The implementation process of the model is described as follows: For each convolutional module of the residual network, is the learnable parameter of the corresponding module; For each , which is specifically composed of a separable convolution module and a direct mapping module, namely: In the formula It is a direct mapping module that combines low-level features with high-level features by converting the convolutional layer. It is a separable convolution module, which is used to lightweight the network model by decomposing the conventional convolution into deep convolution and point convolution. The deep convolution implements the convolution operation by grouping convolution, while the point convolution is a conventional convolution operation with a size of 1. Each layer of the residual model consists of three layers: 1) Convolutional layer: In the formula, is the convolution kernel, also known as the shape The weight matrix, convolution kernel and input data Perform convolution operation to obtain the output result of the convolution layer; 2) Batch Normalization Layer: In the formula and For input data The mean and variance of is an infinite decimal, and are the scale factor and the translation factor respectively; 3) ReLu activation layer: The last layer of the network is the regression layer, and the number of neurons in this layer is 1. During the training process of the network, the mean square error is used as the loss function of the network model. The corresponding formula is: in It is real data. is the forecast data, is the total number of samples.

2. The method for predicting the remaining life of a wind turbine based on a multi-channel separable residual neural network according to claim 1 is characterized in that: The residual neural network model for constructing the separable convolution module comprises the following steps: Determine the number of convolution layers, convolution kernel size and number of convolution kernels of the residual neural network to construct a residual neural network model of the separable convolution module.

3. The method for predicting the remaining life of a wind turbine based on a multi-channel separable residual neural network according to claim 1 is characterized in that: After determining whether the residual neural network model satisfies the iteration condition, the method comprises the following steps: When the residual neural network model does not meet the iteration condition, the network parameters of the residual neural network are adjusted.

4. The method for predicting the remaining life of a wind turbine based on a multi-channel separable residual neural network according to claim 1 is characterized in that: The method of collecting multi-channel vibration data of the wind turbine during its entire life cycle comprises the following steps: Acceleration sensors are respectively installed in the first direction and the second direction of the wind turbine to collect the multi-channel vibration data of the wind turbine.

5. The method for predicting the remaining life of a wind turbine based on a multi-channel separable residual neural network according to claim 4 is characterized in that: Also includes the steps: When the amplitude of the multi-channel vibration data collected during the degradation process reaches more than 10 times the maximum amplitude of the multi-channel vibration data during the normal operation stage, it is determined that the life of the wind turbine generator has ended.

6. The method for predicting the remaining life of a wind turbine based on a multi-channel separable residual neural network according to claim 1 is characterized in that: The step of dividing the multi-channel vibration data into a training data set, a verification data set and a test data set comprises the following steps: A preset number of data points are collected each time, and the multi-channel vibration data is divided into 6 samples according to the sample length, and the first 4 samples are selected as the training data set, the fifth sample is selected as the verification data set, and the sixth sample is selected as the test data set; The same remaining available lifespan is used as the sample label for the six samples collected each time to generate a labeled data set.

7. The method for predicting the remaining life of a wind turbine based on a multi-channel separable residual neural network according to claim 6 is characterized in that: The normalization preprocessing of the training data set and the verification data set comprises the steps of: The labeled data set is normalized, and the implementation process is as follows: In the formula, For the A sample of the data set, is the sample mean, is the sample standard deviation.

8. The method for predicting the remaining life of a wind turbine based on a multi-channel separable residual neural network according to claim 2 is characterized in that: The method of training the residual neural network model using the training data set and verifying the residual neural network model using the verification data set comprises the steps of: Constructing a separable convolution module according to the number of convolution layers, convolution kernel size and number of convolution kernels of the residual neural network; A residual neural network model is constructed based on the separable convolution module, and the residual neural network model is used to predict the remaining useful life.

9. A wind turbine remaining life prediction system based on a multi-channel separable residual neural network, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the remaining life prediction method for a wind turbine based on a multi-channel separable residual neural network as described in any one of claims 1 to 8 when executing the computer program.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are used to enable a computer to execute the method for predicting the remaining life of a wind turbine based on a multi-channel separable residual neural network as described in any one of claims 1 to 8.

Citation Information

Cited By

  • Method for predicting residual life of key component of wind driven generator under multiple working conditions

    CN116306302A

  • A method for predicting the residual life of key components of a wind turbine under multiple working conditions

    CN116306302B