Wind turbine generator fault prediction model and early warning system based on big data analysis

Abstract

The invention discloses a wind turbine generator fault prediction model and an early warning system based on big data analysis, and relates to the technical field of wind turbine generator fault prediction. The system comprises a wind turbine generator set fault prediction model and an early warning system. The wind turbine generator set fault prediction model comprises a data acquisition module used for acquiring fault early warning and fault generation signals of a wind turbine generator set; the model training module is used for establishing a wind turbine generator fault prediction model through deep learning in combination with an extreme learning machine; and a fault diagnosis module of the early warning system is used for sorting the result output by the wind turbine generator fault prediction model and the samples in the fault knowledge base according to the matching success rate, and determining the type of the fault. According to the method, the fault early warning and fault generation signals of the wind turbine generator are collected, data processing and feature extraction are carried out to train the wind turbine generator fault prediction model, risk early warning is carried out on the wind turbine generator, and the fault diagnosis efficiency and accuracy are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of wind turbine fault prediction, and in particular relates to a wind turbine fault prediction model and an early warning system based on big data analysis. Background Art

[0002] Wind turbines are the main equipment of wind farms, and their price accounts for 74-82% of the total investment of wind farms. Due to the high equipment failure rate caused by the harsh operating environment and the high maintenance cost, the maintenance cost of wind turbines has become the main operating cost of wind farms. Reducing the maintenance cost of wind turbines is an important way to improve the economic benefits of wind farm operations. In order to effectively reduce the maintenance cost of wind turbines, wind power companies have introduced technologies such as condition monitoring, fault diagnosis and condition maintenance. However, most of the condition monitoring and fault diagnosis systems on the market at this stage can only realize data acquisition and signal processing functions, while fault diagnosis, positioning and prediction functions often require experts with rich fault diagnosis experience to make judgments based on the frequency domain analysis of the collected signal data.

[0003] From the above, it can be seen that the intelligent diagnosis method of wind turbine faults in the existing technology has the following drawbacks: since on-site maintenance personnel often do not have rich expert experience in fault diagnosis, the existing software based on expert experience is difficult to fully play its role in actual work, resulting in the difficulty in realizing functions such as intelligent fault diagnosis and intelligent early warning. Summary of the invention

[0004] The purpose of the present invention is to provide a wind turbine fault prediction model and an early warning system based on big data analysis, which collects fault warning and fault occurrence signals of wind turbines, performs data processing and feature extraction to train the wind turbine fault prediction model, and provides risk warning for wind turbines, thereby solving the problem that the existing wind turbine fault warning needs to be judged based on expert experience, resulting in high warning cost, untimely warning and low warning efficiency.

[0005] In order to solve the above technical problems, the present invention is achieved through the following technical solutions:

[0006] The present invention is a wind turbine fault prediction model and an early warning system based on big data analysis, including a wind turbine fault prediction model and an early warning system;

[0007] The wind turbine fault prediction model comprises a data acquisition module, a data preprocessing module, a feature extraction module and a model training module connected in sequence; the data acquisition module is used to collect fault warning and fault occurrence signals of the wind turbine; the data preprocessing module is used to process the data collected by the data acquisition module and clean up abnormal data; the feature extraction module is used to extract feature parameters from the preprocessed data to obtain the state feature parameters of the wind turbine; the model training module is used to establish a wind turbine fault prediction model;

[0008] The early warning system includes a fault diagnosis module, an abnormal alarm module and a report printing module; the fault diagnosis module is used to sort the results output by the wind turbine fault prediction model and the samples in the fault knowledge base according to the matching success rate, and determine the type of fault.

[0009] As a preferred technical solution, the early warning system has a built-in fault knowledge base; the fault knowledge base stores known wind turbine fault types and characteristic parameter samples corresponding to the fault types; the fault occurrence signal includes working data and monitoring control data of the wind turbine.

[0010] As a preferred technical solution, the fault knowledge base establishment process is as follows:

[0011] Step S1: Acquire fault log data during the operation of the wind turbine generator set;

[0012] Step S2: Count the frequently occurring fault information in the fault log and establish a fault information tree;

[0013] Step S3: constructing a Bayesian network model using the associated fault chain information of the fault information tree;

[0014] Step S4: Send the Bayesian network model to the inference engine, and store the inference results of the inference engine to form a fault knowledge base.

[0015] As a preferred technical solution, after the data acquisition module acquires the fault warning and fault occurrence signals of the wind turbine generator set, it performs fault marking according to the fault warning and fault occurrence time of the wind turbine generator set.

[0016] As a preferred technical solution, the fault warning and fault signal of the wind turbine generator set are used to perform warning levels and fault markings at the time of occurrence, including:

[0017] Whenever the early warning system generates fault warning information, the corresponding time is recorded and the warning level is marked in the data;

[0018] Whenever a fault occurrence signal appears in the early warning system, the corresponding time is recorded and the fault is marked in the data.

[0019] As a preferred technical solution, when the fault diagnosis module diagnoses that a new fault has occurred, the samples in the fault knowledge base are arranged to be sorted according to the matching success rate, and the sample with the highest matching success rate and the fault type corresponding to the sample are selected to form the training sample for this diagnosis. The intelligent diagnosis model trained with the training samples will be applied to this fault diagnosis. The diagnosis result shows which fault type in the fault knowledge base the new fault belongs to, and the maintenance personnel determine whether the diagnosis result is correct. If correct, the matching success rate of the samples corresponding to the fault type shown in the diagnosis result increases. If wrong, the matching success rate of the samples corresponding to the fault type shown in the diagnosis result decreases.

[0020] As a preferred technical solution, the feature extraction module extracts characteristic parameters of wind turbine fault warning and fault occurrence signals including: absolute mean and range; the features extracted by the feature extraction module are divided into a training data set and a test data set, and the signals in the data set are normalized.

[0021] As a preferred technical solution, the parameters involved in the model training module include: the overall observation matrix , memory matrix , observation matrix and the estimation matrix ;in;

[0022] Overall observation matrix P nb = [ X ( 1 ), X ( 2 ),..., X ( B ) ] = [ x 11 x 12 ... x 1 b x 21 x 22 ... x 2 b ⋮ ⋮ ⋮ ⋮ x n 1 x n 2 ... x nb ] ;

[0023] Where n is the number of observed variables at each time, b is the time label, X(1) represents the first signal vector to be estimated, and X(B) represents the Bth signal vector to be estimated;

[0024] From the overall observation matrix The parameters of any period of time are selected as the training matrix K. There are k observation vectors in the training matrix K. The calculation formula of the training matrix K is as follows:

[0025] K = [ X ( 1 ), X ( 2 ),..., X ( K ) ] = [ x 11 x 12 ... x 1 k x 21 x 22 ... x 2 k ⋮ ⋮ ⋮ ⋮ x n 1 x n 2 ... x n ] ;

[0026] Where X(K) represents the Kth signal vector to be estimated;

[0027] Extract part of the data from the training matrix K to form a memory matrix , then the memory matrix The calculation formula is:

[0028] D = [ X ( 1 ), X ( 2 ),..., X ( D ) ] = [ x 11 x 12 ... x 1 d x 21 x 22 ... x 2 d ⋮ ⋮ ⋮ ⋮ x n 1 x n 2 ... x nd ] ;

[0029] The observation matrix Input the wind turbine fault prediction model and get the estimation matrix , then the wind turbine fault prediction model will generate a matrix The weight vector W of the same dimension is calculated as follows:

[0030] ,in W = [ w 1 , w 2 ,..., w m ] .

[0031] As a preferred technical solution, the framework of the wind turbine fault prediction model includes four layers: the first layer is a data preprocessing layer, the second layer is a multidimensional spatiotemporal sequence fusion layer, the third layer is a long short-term memory neural network layer LSTM, and the fourth layer is a metric learning fault classification layer based on Mahalanobis distance.

[0032] LSTM is a type of recurrent neural network. LSTM can be seen as a group of identical networks, each of which passes information to the next one. The long short-term artificial memory network uses a special "gate" connection structure to memorize valid information within a certain time range and forget useless "memories" in time to achieve a deep representation of time series data. The gates of LSTM include forget gate, input gate and output gate.

[0033] The LSTM network is structured by the grid parameter adjustment method to achieve the optimal prediction of the algorithm. The input and output features of this network are both 20 spatiotemporal fusion features, the number of LSTM hidden layers is 256, the training time step is 64 time steps, and the prediction time step is 32 time steps. The time series sampling method of this training adopts adjacent sampling. In order to prevent gradient explosion, it is necessary to design a clipping gradient structure. The clipping threshold is determined to be 0.01 through grid parameter adjustment. Finally, we will use the error back propagation algorithm to perform 5000 residual training on the network, and the training update weight is 0.1. The idea of ​​the back propagation algorithm is to first randomly generate the weight value of the neural node, then input the training data into the network to obtain an output result and compare it with the label value to calculate the loss function value and its gradient. According to the back propagation of the gradient, the residual is fed back to each weight in the network, and then the weight is adjusted to obtain a model with a low loss function.

[0034] The present invention has the following beneficial effects:

[0035] The present invention collects fault warning and fault occurrence signals of wind turbines, performs data processing and feature extraction to train a wind turbine fault prediction model, thereby providing risk warning for wind turbines and improving fault diagnosis efficiency and accuracy.

[0036] The present invention establishes an NSET model through real wind turbine data and data preprocessing, and verifies the feasibility of the model and the feasibility of the early warning strategy through a fixed early warning threshold, thereby improving the efficiency of fault diagnosis.

[0037] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for describing the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0039] Figure 1 A schematic diagram of the structure of a wind turbine fault prediction model and an early warning system based on big data analysis according to the present invention;

[0040] Figure 2 Flowchart created for the fault knowledge base. DETAILED DESCRIPTION

[0041] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions 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 ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0042] In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0043] To make the purpose, technical solutions and advantages of the present invention more clear, the following will be combined with the attached Figure 1 The embodiments of the present invention are described in further detail.

[0044] Before introducing the embodiments of the present invention, firstly, the wind turbine fault warning is described.

[0045] There are abundant sources of wind turbine operation data, among which the data acquisition and monitoring control system equipped by the unit itself generates a large amount of operation data every day. Reasonable use of this information-rich SCADA data can realize early warning of wind turbine failures and improve the reliability and power quality of the unit. The wind turbine early warning system based on SCADA big data is easy to implement and has high reliability. It has the following advantages: no additional investment is required. The wind turbine itself is equipped with a real-time monitoring system, which can obtain accurate and reliable SCADA data, and is convenient for taking a large amount of experimental data. The richer the experimental data, the more scientific the experiment is, avoiding contingency, and more conducive to the analysis and prediction of the operation status of the wind turbine; the amount of historical stored data is large, and a large amount of health data can be used for experiments to establish a health model. Since the data before the failure is relatively complete, the health model and various parameters before abnormal operation can be used for analysis and comparison to determine the relationship between normal working state and abnormal working state; it is convenient to implement on site, without the need for special operation and maintenance personnel, and there is no subsequent maintenance cost, which greatly reduces the workload and repair and maintenance costs, which is of great benefit to the operation of wind farms, reducing investment costs and improving safety assurance, so the wind turbine fault early warning strategy has a large application space.

[0046] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the following Figure 1 It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0047] See also Figure 1 As shown, the present invention is a wind turbine fault prediction model and an early warning system based on big data analysis, including a wind turbine fault prediction model and an early warning system;

[0048] The wind turbine fault prediction model includes a data acquisition module, a data preprocessing module, a feature extraction module and a model training module connected in sequence; the data acquisition module is used to collect fault warning and fault occurrence signals of the wind turbine; the data preprocessing module is used to process the data collected by the data acquisition module and clean up abnormal data; the feature extraction module is used to extract feature parameters from the preprocessed data to obtain the state feature parameters of the wind turbine; the model training module is used to establish a wind turbine fault prediction model;

[0049] The early warning system includes a fault diagnosis module, an abnormal alarm module and a report printing module; the fault diagnosis module is used to sort the results output by the wind turbine fault prediction model and the samples in the fault knowledge base according to the matching success rate, and determine the type of fault.

[0050] The early warning system has a built-in fault knowledge base; the fault knowledge base stores known wind turbine fault types and characteristic parameter samples corresponding to the fault types; the fault occurrence signal includes the working data and monitoring control data of the wind turbine.

[0051] The wind turbine equipment status monitoring and fault warning system can effectively understand the equipment operating status and determine equipment operating failures. Equipment status monitoring and fault diagnosis are two important processes in equipment maintenance. Among them, equipment diagnosis requires judgment on the cause, location and severity of equipment failure. Any deviation during equipment operation will affect the normal operation of the power generation system. Traditional fault inspection and repair methods are mainly manual inspection and repair. During on-site use, the monitoring instrument cannot grasp the equipment operating status and fault development trend, and cannot accurately and effectively monitor hidden faults. The lack of prediction and diagnosis of equipment failures often requires emergency repairs when equipment fails, which poses a risk of large-scale power outages.

[0052] The fault diagnosis of wind turbines includes rotating machine failure, mechanical failure, parts failure, engineering structure failure, and electrical equipment failure. The causes of the failure include design defects, installation and maintenance, and damage during operation. In the design process, the wind turbine has defects in dynamics or unreasonable structure, which is easy to produce forced vibration and self-excited vibration during operation. The poor processing and manufacturing process of parts and components and serious quality problems of materials lead to insufficient strength and also cause fan failure. Misalignment of parts during installation, improper adjustment of bearing parameters, and excessive pipeline stress will destroy the dynamic balance during the operation of the fan and cause equipment damage. If the operation is improper, the equipment will not be fully preheated or cooled, which is easy to cause local damage. In addition, the fan is running under overload, which is easy to change the working characteristics of the equipment and cause stress damage. Common diagnostic methods include vibration diagnosis, temperature diagnosis, acoustic diagnosis, strength diagnosis, pressure diagnosis, electrical parameter diagnosis, etc.

[0053] See also Figure 2 As shown in the figure, the process of establishing the fault knowledge base is as follows:

[0054] Step S1: Acquire fault log data during the operation of the wind turbine generator set;

[0055] Step S2: Count the frequently occurring fault information in the fault log and establish a fault information tree;

[0056] Step S3: constructing a Bayesian network model using the associated fault chain information of the fault information tree;

[0057] Step S4: Send the Bayesian network model to the inference engine, and store the inference results of the inference engine to form a fault knowledge base.

[0058] To realize intelligent diagnosis of wind turbine faults, the types of currently known faults and the corresponding characteristic parameter samples must be stored in the fault knowledge base; when a new fault occurs, an intelligent diagnosis algorithm is used to compare the characteristic parameter samples of the new fault with the known faults in the fault knowledge base, and the type of the new fault is determined based on the matching success rate.

[0059] As time goes by, the number of training samples of the same fault from different units in the fault knowledge base will gradually increase, so there is a problem of how to select appropriate training samples. When the number of training samples is large, wind turbine fault data with a high matching success rate should be selected as training samples for intelligent diagnosis. The fault knowledge base structure and matching degree update algorithm process are as follows: Figure 2 shown.

[0060] For each component, each type of fault that has ever occurred will be recorded; for each component and each type of fault, there is historical status monitoring data of the fault that occurred at different times and on different wind turbines. These data are called training samples, and each training sample has a matching success rate attribute. The matching degree update algorithm process is as follows:

[0061] When a new fault occurs, the training samples are sorted according to the matching success rate;

[0062] All samples with the highest historical matching success rate will be selected and used together with the corresponding fault types to form the training samples for this diagnosis. For example, sample A1 is used for fault type A, and sample B4 is used for fault type B.

[0063] The intelligent diagnosis model trained with the training samples will be applied to this fault diagnosis;

[0064] The diagnosis result will show which fault type in the fault knowledge base the new fault belongs to, for example, fault type A, and the maintenance personnel will judge whether the diagnosis result is correct; update the matching success rate. If the judgment is correct, the matching success rate of sample A1 corresponding to fault type A will increase, otherwise the matching success rate of sample A1 will decrease.

[0065] After the data acquisition module collects the fault warning and fault occurrence signals of the wind turbine generator set, the fault is marked according to the fault warning and fault occurrence time of the wind turbine generator set.

[0066] The fault warning and fault signal of the wind turbine are marked with warning level and fault mark at the time of occurrence, including:

[0067] Whenever the early warning system generates a fault warning message, the corresponding time is recorded and the warning level is marked in the data;

[0068] Whenever a fault occurs in the early warning system, the corresponding time is recorded and the fault is marked in the data.

[0069] When the fault diagnosis module diagnoses that a new fault has occurred, the samples in the fault knowledge base are set to be sorted according to the matching success rate, and the sample with the highest matching success rate and the fault type corresponding to the sample are selected to form the training sample for this diagnosis. The intelligent diagnosis model trained with the training samples will be applied to this fault diagnosis. The diagnosis result shows which fault type in the fault knowledge base the new fault belongs to, and the maintenance personnel determine whether the diagnosis result is correct. If correct, the matching success rate of the corresponding sample of the fault type shown in the diagnosis result increases. If wrong, the matching success rate of the corresponding sample of the fault type shown in the diagnosis result decreases.

[0070] The feature extraction module extracts characteristic parameters of wind turbine fault warning and fault occurrence signals, including absolute mean and range. The features extracted by the feature extraction module are divided into training data sets and test data sets, and the signals in the data sets are normalized.

[0071] The parameters involved in the model training module include: overall observation matrix , memory matrix , observation matrix and the estimation matrix ;in;

[0072] Overall observation matrix P nb = [ X ( 1 ), X ( 2 ),..., X ( B ) ] = [ x 11 x 12 ... x 1 b x 21 x 22 ... x 2 b ⋮ ⋮ ⋮ ⋮ x n 1 x n 2 ... x nb ] ;

[0073] Where n is the number of observed variables at each time, b is the time label, X(1) represents the first signal vector to be estimated, and X(B) represents the Bth signal vector to be estimated;

[0074] From the overall observation matrix The parameters of any period of time are selected as the training matrix K. There are k observation vectors in the training matrix K. The calculation formula of the training matrix K is as follows:

[0075] K = [ X ( 1 ), X ( 2 ),..., X ( K ) ] = [ x 11 x 12 ... x 1 k x 21 x 22 ... x 2 k ⋮ ⋮ ⋮ ⋮ x n 1 x n 2 ... x n ] ;

[0076] Where X(K) represents the Kth signal vector to be estimated;

[0077] Extract some data from the training matrix K to form a memory matrix , then the memory matrix The calculation formula is:

[0078] D = [ X ( 1 ), X ( 2 ),..., X ( D ) ] = [ x 11 x 12 ... x 1 d x 21 x 22 ... x 2 d ⋮ ⋮ ⋮ ⋮ x n 1 x n 2 ... x nd ] ;

[0079] The observation matrix Input the wind turbine fault prediction model and get the estimation matrix , then the wind turbine fault prediction model will generate a matrix The weight vector W of the same dimension is calculated as follows:

[0080] ,in W = [ w 1 , w 2 ,..., w m ] ;

[0081] Let the residual be , then the residual minimization process is:

[0082] ;

[0083] If Minimum, find The minimum value of , then:

[0084] ;

[0085] but and The Euclidean distance between is:

[0086] ;

[0087] The weight vector W is changed to:

[0088] ;

[0089] Then the current state estimation matrix of the early warning system is:

[0090] .

[0091] The framework of the wind turbine fault prediction model consists of four layers: the first layer is the data preprocessing layer, the second layer is the multi-dimensional spatiotemporal sequence fusion layer, the third layer is the long short-term memory neural network layer LSTM, and the fourth layer is the metric learning fault classification layer based on Mahalanobis distance.

[0092] LSTM is a type of recurrent neural network. LSTM can be seen as a group of identical networks, each of which passes information to the next one. The long short-term artificial memory network uses a special "gate" connection structure to memorize valid information within a certain time range and forget useless "memories" in time to achieve a deep representation of time series data. The gates of LSTM include forget gate, input gate and output gate.

[0093] The LSTM network is structured by the grid parameter adjustment method to achieve the optimal prediction of the algorithm. The input and output features of this network are both 20 spatiotemporal fusion features, the number of LSTM hidden layers is 256, the training time step is 64 time steps, and the prediction time step is 32 time steps. The time series sampling method of this training adopts adjacent sampling. In order to prevent gradient explosion, it is necessary to design a clipping gradient structure. The clipping threshold is determined to be 0.01 through grid parameter adjustment. Finally, we will use the error back propagation algorithm to perform 5000 residual training on the network, and the training update weight is 0.1. The idea of ​​the back propagation algorithm is to first randomly generate the weight value of the neural node, then input the training data into the network to obtain an output result and compare it with the label value to calculate the loss function value and its gradient. According to the back propagation of the gradient, the residual is fed back to each weight in the network, and then the weight is adjusted to obtain a model with a low loss function.

[0094] It is worth noting that in the above system embodiment, the various units included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of the present invention.

[0095] In addition, those skilled in the art can understand that all or part of the steps in the above-mentioned embodiments can be completed by instructing related hardware through a program, and the corresponding program can be stored in a computer-readable storage medium.

[0096] The preferred embodiments of the present invention disclosed above are only used to help illustrate the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to the specific implementation methods described. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can understand and use the present invention well. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. A wind turbine fault prediction model and an early warning system based on big data analysis, comprising a wind turbine fault prediction model and an early warning system, characterized in that: The wind turbine fault prediction model comprises a data acquisition module, a data preprocessing module, a feature extraction module and a model training module connected in sequence; the data acquisition module is used to collect fault warning and fault occurrence signals of the wind turbine; the data preprocessing module is used to process the data collected by the data acquisition module and clean up abnormal data; the feature extraction module is used to extract feature parameters from the preprocessed data to obtain the state feature parameters of the wind turbine; the model training module is used to establish a wind turbine fault prediction model; The early warning system includes a fault diagnosis module, an abnormality alarm module and a report printing module; The fault diagnosis module is used to sort the results output by the wind turbine fault prediction model and the samples in the fault knowledge base according to the matching success rate, and determine the type of fault.

2. A wind turbine fault prediction model and early warning system based on big data analysis according to claim 1, characterized in that: The early warning system has a built-in fault knowledge base; the fault knowledge base stores known wind turbine fault types and characteristic parameter samples corresponding to the fault types; the fault occurrence signal includes working data and monitoring control data of the wind turbine.

3. A wind turbine fault prediction model and an early warning system based on big data analysis according to claim 2, characterized in that: The fault knowledge base establishment process is as follows: Step S1: Acquire fault log data during the operation of the wind turbine generator set; Step S2: Count the frequently occurring fault information in the fault log and establish a fault information tree; Step S3: constructing a Bayesian network model using the associated fault chain information of the fault information tree; Step S4: Send the Bayesian network model to the inference engine, and store the inference results of the inference engine to form a fault knowledge base.

4. The wind turbine fault prediction model and early warning system based on big data analysis according to claim 1, characterized in that: After the data acquisition module acquires the fault warning and fault occurrence signals of the wind turbine generator set, the data acquisition module performs fault marking according to the fault warning and fault occurrence time of the wind turbine generator set.

5. A wind turbine fault prediction model and early warning system based on big data analysis according to claim 4, characterized in that: The fault warning and fault signal of the wind turbine generator set are used to mark the warning level and fault at the time of occurrence, including: Whenever the early warning system generates fault warning information, the corresponding time is recorded and the warning level is marked in the data; Whenever a fault occurrence signal appears in the early warning system, the corresponding time is recorded and the fault is marked in the data.

6. The wind turbine fault prediction model and early warning system based on big data analysis according to claim 1, characterized in that: When the fault diagnosis module diagnoses that a new fault has occurred, the samples in the fault knowledge base are arranged to be sorted according to the matching success rate, and the sample with the highest matching success rate and the fault type corresponding to the sample are selected to form the training sample for this diagnosis. The intelligent diagnosis model trained with the training samples will be applied to this fault diagnosis. The diagnosis result shows which fault type in the fault knowledge base the new fault belongs to, and the maintenance personnel determine whether the diagnosis result is correct. If correct, the matching success rate of the samples corresponding to the fault type displayed in the diagnosis result increases. If wrong, the matching success rate of the samples corresponding to the fault type displayed in the diagnosis result decreases.

7. The wind turbine fault prediction model and early warning system based on big data analysis according to claim 1, characterized in that: The feature extraction module extracts characteristic parameters of wind turbine fault warning and fault occurrence signals including absolute mean and range. The features extracted by the feature extraction module are divided into a training data set and a test data set, and the signals in the data set are normalized.

8. The wind turbine fault prediction model and early warning system based on big data analysis according to claim 1, characterized in that: The parameters involved in the model training module include: the overall observation matrix , memory matrix , observation matrix and the estimation matrix ;in; Overall observation matrix ; Where n is the number of observed variables at each time, b is the time label, X(1) represents the first signal vector to be estimated, and X(B) represents the Bth signal vector to be estimated; From the overall observation matrix The parameters of any period of time are selected as the training matrix K. There are k observation vectors in the training matrix K. The calculation formula of the training matrix K is as follows: ； Where X(K) represents the Kth signal vector to be estimated; Extract part of the data from the training matrix K to form a memory matrix , then the memory matrix The calculation formula is: ； The observation matrix Input the wind turbine fault prediction model and get the estimation matrix , then the wind turbine fault prediction model will generate a matrix The weight vector W of the same dimension is calculated as follows: ,in .

9. The wind turbine fault prediction model and early warning system based on big data analysis according to claim 1, characterized in that: The framework of the wind turbine fault prediction model includes four layers: the first layer is a data preprocessing layer, the second layer is a multi-dimensional spatiotemporal sequence fusion layer, the third layer is a long short-term memory neural network layer LSTM, and the fourth layer is a metric learning fault classification layer based on Mahalanobis distance.

Images