CN115434875A - Wind turbine generator fault state detection method based on space-time neural network - Google Patents
Wind turbine generator fault state detection method based on space-time neural network Download PDFInfo
- Publication number
- CN115434875A CN115434875A CN202211066553.8A CN202211066553A CN115434875A CN 115434875 A CN115434875 A CN 115434875A CN 202211066553 A CN202211066553 A CN 202211066553A CN 115434875 A CN115434875 A CN 115434875A
- Authority
- CN
- China
- Prior art keywords
- data
- wind turbine
- time
- turbine generator
- space
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 127
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 15
- 238000000605 extraction Methods 0.000 claims abstract description 57
- 238000000034 method Methods 0.000 claims abstract description 23
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 22
- 238000007781 pre-processing Methods 0.000 claims abstract description 15
- 238000011897 real-time detection Methods 0.000 claims abstract description 11
- 238000003860 storage Methods 0.000 claims abstract description 8
- 238000012549 training Methods 0.000 claims description 56
- 230000006870 function Effects 0.000 claims description 31
- 241000282461 Canis lupus Species 0.000 claims description 29
- 238000005457 optimization Methods 0.000 claims description 13
- 238000004590 computer program Methods 0.000 claims description 12
- 238000010606 normalization Methods 0.000 claims description 10
- 238000009499 grossing Methods 0.000 claims description 9
- 238000011049 filling Methods 0.000 claims description 8
- 241000282421 Canidae Species 0.000 claims description 6
- 238000012545 processing Methods 0.000 claims description 6
- 230000004927 fusion Effects 0.000 claims description 4
- 210000002569 neuron Anatomy 0.000 claims description 4
- 238000004364 calculation method Methods 0.000 claims description 3
- 239000000463 material Substances 0.000 abstract description 5
- 239000002699 waste material Substances 0.000 abstract description 3
- 230000009286 beneficial effect Effects 0.000 abstract description 2
- 238000012360 testing method Methods 0.000 description 8
- 238000010586 diagram Methods 0.000 description 7
- 230000008569 process Effects 0.000 description 6
- 238000011161 development Methods 0.000 description 5
- 230000008901 benefit Effects 0.000 description 4
- 238000004140 cleaning Methods 0.000 description 4
- 238000012544 monitoring process Methods 0.000 description 4
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000012423 maintenance Methods 0.000 description 3
- 238000003062 neural network model Methods 0.000 description 3
- 238000010248 power generation Methods 0.000 description 3
- 230000002159 abnormal effect Effects 0.000 description 2
- 230000004075 alteration Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 230000009471 action Effects 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000025518 detection of mechanical stimulus involved in sensory perception of wind Effects 0.000 description 1
- 230000006866 deterioration Effects 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 239000000428 dust Substances 0.000 description 1
- 238000005265 energy consumption Methods 0.000 description 1
- 238000011478 gradient descent method Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000001737 promoting effect Effects 0.000 description 1
- 230000008439 repair process Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 239000004576 sand Substances 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000002123 temporal effect Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Images
Classifications
-
- F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
- F03—MACHINES OR ENGINES FOR LIQUIDS; WIND, SPRING, OR WEIGHT MOTORS; PRODUCING MECHANICAL POWER OR A REACTIVE PROPULSIVE THRUST, NOT OTHERWISE PROVIDED FOR
- F03D—WIND MOTORS
- F03D17/00—Monitoring or testing of wind motors, e.g. diagnostics
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Evolutionary Computation (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Business, Economics & Management (AREA)
- Artificial Intelligence (AREA)
- Computational Mathematics (AREA)
- Economics (AREA)
- Pure & Applied Mathematics (AREA)
- Mathematical Optimization (AREA)
- Mathematical Analysis (AREA)
- Mechanical Engineering (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Combustion & Propulsion (AREA)
- Operations Research (AREA)
- Probability & Statistics with Applications (AREA)
- Chemical & Material Sciences (AREA)
- Algebra (AREA)
- Databases & Information Systems (AREA)
- Sustainable Energy (AREA)
- Sustainable Development (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- Human Resources & Organizations (AREA)
- Marketing (AREA)
Abstract
The invention relates to a wind turbine generator fault state detection method based on a time-space neural network, electronic equipment and a storage medium, wherein the method comprises the steps of acquiring real-time detection data of a sensor of a wind turbine generator to be detected and carrying out data preprocessing; inputting the detection data into a trained space-time characteristic extraction model to obtain the multi-dimensional residual error data; calculating the multi-dimensional residual data by adopting the Mahalanobis distance to obtain a real-time performance index of a space-time feature extraction model; inputting the performance index and the real-time detection data into a trained vector regression algorithm model to obtain a real-time dynamic threshold; and detecting the fault state of the wind turbine generator based on the real-time dynamic threshold and the real-time performance index. The method has the beneficial effects that the technical problems of long time consumption, lagging fault feedback, low efficiency, large error and manpower and material resource waste of fault detection of the wind turbine generator set in the prior art can be solved.
Description
Technical Field
The invention relates to the technical field of fault detection of wind turbine generators, in particular to a wind turbine generator fault state detection method based on a space-time neural network.
Background
The wind energy has the advantages of cleanness, large reserve capacity, convenient utilization and the like, and is a renewable energy source with great development potential. With the development of modern society, the wind power generation industry rises rapidly, and the wind power generation technology is continuously developed and tends to be mature and complete. However, because many wind turbine generators are installed in regions with relatively severe working environments such as islands and mountains, and are exposed to environments such as sand, dust, rainfall, high temperature and snow for a long time, faults of the wind turbine generators frequently occur along with wind action and impact influence with uncertain direction and load, and the reliability and safety of the operation of the wind turbine generators are seriously influenced.
In the prior art, a Supervisory Control and Data Acquisition (SCADA) system is widely adopted in a wind farm, but because Data of fault detection based on SCADA Data is huge, manual analysis consumes long time, and the defects of untimely fault feedback, low efficiency, large error and manpower and material waste exist.
Disclosure of Invention
Technical problem to be solved
In view of the above disadvantages and shortcomings of the prior art, the invention provides a wind turbine generator fault state detection method based on a spatiotemporal neural network, an electronic device and a storage medium, which solve the technical problems of long time consumption, lag in fault feedback, low efficiency, large error and waste of manpower and material resources of the fault detection technology based on SCADA data in the prior art.
(II) technical scheme
In order to achieve the above object, in a first aspect, the present invention provides a wind turbine generator fault state detection method based on a spatio-temporal neural network, which is characterized by comprising:
s1, acquiring real-time detection data of a sensor of a wind turbine generator to be detected and performing data preprocessing; the detection data includes: wind speed, power, rotation speed, pitch angle, temperature and detection time;
s2, inputting the detection data into a trained space-time feature extraction model to obtain the multi-dimensional residual data; the multi-dimensional residual data is a residual between the real-time detection data and the output of the space-time characteristic extraction model;
the space-time feature extraction model comprises a deep self-encoder and a gated recursion unit, wherein the deep self-encoder is used for extracting the spatial features of the detection data; the gated recursion unit is used for extracting the time characteristics of the detection data;
s3, calculating the multi-dimensional residual error data by adopting the Mahalanobis distance to obtain a real-time performance index of a space-time feature extraction model;
s4, inputting the performance index and the real-time detection data into a trained vector regression algorithm model to obtain a real-time dynamic threshold;
and S5, detecting the fault state of the wind turbine generator based on the real-time dynamic threshold and the real-time performance index.
Optionally, the data preprocessing in S1 mainly includes:
analyzing the detection data by adopting a fusion strategy based on ideal wind speed power, setting unnecessary data points as outliers and removing the outliers;
filling the detection data with the outliers removed by adopting a local mean filling strategy;
and carrying out normalization processing on the filled detection data.
Optionally, in S3, the performance index is smoothed by an exponentially weighted moving average value, so as to obtain an optimized performance index;
the formula for calculating and obtaining the performance index by adopting the Mahalanobis distance is as follows:
e is a reconstruction error of a space-time feature extraction model, and mu is a dimension mean value of the detection data;
the calculation formula for smoothing the performance index is as follows:
RE t ＝λE t +(1-λ)RE t-1 ；
and the lambda is a smoothing coefficient.
Optionally, the S5 specifically includes:
comparing the dynamic threshold to the performance index;
if the dynamic threshold value is smaller than the performance index, judging that the wind turbine generator is in a fault state;
and if the dynamic threshold is not less than the performance index, judging that the wind turbine generator is in a normal operation state.
Optionally, before the step S1, the method further includes step S0 of training a spatio-temporal feature extraction model and a support vector regression model based on historical detection data of the wind turbine to be detected;
the S0 comprises:
s01, training a time-space feature extraction model based on historical detection data of the wind turbine generator to be detected; the method specifically comprises the following steps:
s011, acquiring historical detection data of the wind turbine generator to be detected, preprocessing the historical detection data, and taking the historical detection data without fault state as a training data set;
s012, inputting the training data set into a pre-constructed space-time feature extraction model for iteration by means of a pre-constructed regularization method, and outputting space-time feature reconstruction data of each iteration stage;
s013, based on the spatio-temporal feature reconstruction data, calculating a target updating weight of each iteration stage by adopting a pre-constructed gradient descent rule and a loss function until the loss function is converged, and finishing training a spatio-temporal feature extraction model;
the loss function is:
the above-mentioned
In order to train the data, it is,
and extracting spatio-temporal feature reconstruction data output by the model data for spatio-temporal features.
Optionally, the S0 further includes:
s02, training a support vector regression model based on historical detection data of the wind turbine generator to be detected, specifically comprising the following steps:
s021, obtaining training data input into a pre-constructed support autoregressive model;
the training data comprise preprocessed historical detection data of the wind turbine generator to be detected and a performance index of the wind turbine generator to be detected;
the performance index is a space-time feature extraction model performance index obtained by inputting the historical detection data into a trained space-time feature extraction model, and calculating the residual error of the historical detection data and the output value of the space-time feature extraction model by adopting the Mahalanobis distance;
s022, inputting the training data into a pre-constructed support vector regression model, and training the support vector regression model, wherein parameters required to be selected by the support vector regression model are obtained through a wolf optimization algorithm.
Optionally, the parameters to be selected by the vector regression model include: a penalty factor C and a kernel function parameter sigma;
in S022, obtaining the parameters to be selected for the support vector regression model based on the grayish optimization algorithm specifically includes:
s0221, initializing parameters of a gray wolf algorithm and a support vector regression model, wherein the parameters are a gray wolf population size N, a maximum allowable iteration time tmax, a penalty factor C and a value range of a kernel function parameter sigma;
s0222, initializing the population by adopting a pre-constructed optimal point set rule, and determining the initial value of the position (C, sigma) of each wolf;
s0223, inputting the training data into the support vector regression model, and calculating the fitness of each wolf under the initial value;
s0224, selecting the first 3 grey wolves with the best fitness, and updating the positions (C and sigma) of the 3 grey wolves;
s0225, calculating the fitness of all individual gray wolfs, comparing the fitness before and after position updating, if the current value is superior to the fitness obtained by the previous iteration, updating the gray wolf positions of the third fitness, otherwise, not updating the positions;
s0226: and comparing the current iteration times with the maximum allowable iteration times, if the current iteration times do not reach N, continuing optimizing, otherwise finishing the training, taking the position (C, sigma) value of the wolf head as the optimal solution, and finishing the training of the corresponding support vector regression model.
Optionally, the structure of the deep self-encoder is 15-100-50-25-50-100-15, and the number of neurons of the gated recursion unit is 100.
In a second aspect, the present invention provides an electronic device, which includes a memory and a processor, where the memory stores a computer program, and the processor executes the computer program stored in the memory, so as to implement the steps of the wind turbine generator fault state detection method according to any one of the above first aspects.
In a third aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored, where the computer program, when executed by a processor, implements the steps of the wind turbine generator fault status detection method according to any one of the above first aspects.
(III) advantageous effects
The invention provides a wind turbine generator fault state detection method based on a space-time neural network, electronic equipment and a storage medium. And calculating a dynamic threshold value based on the wind turbine generator running state changeability characteristic and a gray Wolf optimization algorithm and a pre-constructed Support Vector Regression (GWO-SVR). Compared with the prior art, the wind turbine generator SCADA fault detection method can simultaneously take the time dependence and the space dependence of the wind turbine generator SCADA data into consideration, and improves the accuracy of wind turbine generator fault detection. The dimensionality of the data is reduced through the Mahalanobis distance, the detection difficulty is reduced, the dynamic threshold is set in a self-adaptive mode, the final detection precision of the wind turbine is improved, and the purposes of wide detection range, timely detection and feedback, high efficiency, small error and high precision are achieved.
Drawings
Fig. 1 is a flow chart of a wind turbine generator fault state detection method based on a spatiotemporal neural network according to an embodiment of the present invention;
FIGS. 2 (a) and 2 (b) are graphs comparing wind speed-power curves before and after data cleaning according to an embodiment of the present invention;
FIG. 3 is a logic flow diagram of training spatio-temporal feature extraction models and support vector regression models provided in accordance with an embodiment of the present invention;
FIG. 4 is a schematic structural diagram of a spatiotemporal feature extraction model according to an embodiment of the present invention;
FIG. 5 is a flowchart illustrating a gray wolf optimization algorithm according to an embodiment of the present invention;
fig. 6 is a schematic diagram of a confusion matrix of a wind turbine generator fault detection result according to an embodiment of the present invention.
Detailed Description
For the purpose of better explaining the present invention and to facilitate understanding, the present invention will be described in detail by way of specific embodiments with reference to the accompanying drawings. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention can be embodied in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art.
Wind energy has the advantages of cleanness, large reserve capacity, convenient utilization and the like, and is a renewable energy source with great development potential. The large-scale development and utilization of wind energy promotes the wind power generation industry to rise rapidly. Due to the fact that the placement environment is severe, the reliability and safety of operation of the wind turbine generator cannot be well guaranteed. Therefore, according to the wind turbine generator fault state detection method based on the time-space neural network provided by the embodiment of the invention, based on a Data Acquisition and Supervisory Control (SCADA) system widely adopted by a wind farm, no additional sensor or other equipment needs to be installed on the wind turbine generator, and low-frequency signals such as various powers, wind speeds, temperatures and the like are acquired for detecting the fault state of the wind turbine generator, so that the fault detection cost is reduced.
As shown in fig. 1, the fig. 1 is a schematic flow chart of a wind turbine generator fault state detection method based on a spatiotemporal neural network according to an embodiment of the present invention, where the method may include:
s1, acquiring real-time detection data of a sensor of a wind turbine generator to be detected and performing data preprocessing; the detection data may include: wind speed, power, rotating speed, pitch angle, temperature, detection time and the like.
The data preprocessing may include data cleansing, data population, and/or normalization processing, among others.
S2, inputting the detection data into a trained space-time feature extraction model to obtain the multi-dimensional residual data; and the multi-dimensional residual data is a residual between the real-time detection data and the output of the space-time characteristic extraction model.
The space-time feature extraction model comprises a deep self-encoder and a gated recursion unit, wherein the deep self-encoder is used for extracting the spatial features of the detection data; the gate control recursion unit is used for extracting the time characteristics of the detection data, and the output of the space-time characteristic extraction model is space-time characteristic reconstruction data of the input detection data.
S3, calculating the multi-dimensional residual error data by adopting the Mahalanobis distance to obtain a real-time performance index of a space-time feature extraction model;
s4, inputting the performance index and the real-time detection data into a trained vector regression algorithm model to obtain a real-time dynamic threshold;
and S5, detecting the fault state of the wind turbine generator based on the real-time dynamic threshold and the real-time performance index.
Specifically, in an embodiment, the S5 specifically is:
comparing the dynamic threshold to the performance index;
if the dynamic threshold value is smaller than the performance index, judging that the wind turbine generator is in a fault state;
and if the dynamic threshold is not less than the performance index, judging that the wind turbine generator is in a normal operation state.
According to the wind turbine generator fault state detection method based on the time-space neural network, SCADA detection data collected in real time are fully utilized and are respectively input into the time-space feature extraction model and the support vector regression model, the real-time performance index and the real-time dynamic threshold are calculated, the fault state of the wind turbine generator can be quickly reflected, the feedback efficiency is improved, the operation and maintenance cost is reduced, and the maintenance and the repair of the wind turbine generator are facilitated.
In a specific embodiment, in the wind turbine generator fault state detection method based on the spatiotemporal neural network, the preprocessing of data in S1 may include:
s11, analyzing the detection data by adopting a fusion strategy based on ideal wind speed power, setting unnecessary data points as outliers and removing the outliers; the data can be cleaned, and abnormal real-time data detected by the SCADA system under the influence of interference factors such as wind abandonment, sensor faults and the like of a normally-operated wind turbine generator are eliminated.
In practical applications, the fusion strategy may be a hybrid model considering the relationship (such as wind speed-power, etc.) between the high-dimensional characteristics and the known low-dimensional variables among the SCADA data of the wind turbine, for example, fig. 2 (a) and fig. 2 (b) are the wind speed-power curve distribution comparison diagrams before and after data cleaning provided by an embodiment of the present invention.
S12, filling the detection data without outliers by adopting a local mean filling strategy; temporal discontinuity of the detected data due to data cleansing can be avoided.
In an embodiment, in order to eliminate the influence of the data metric on the accuracy of the final detection model and accelerate the convergence of the model, S13 may be further performed to perform normalization processing on the padded detection data. And (4) carrying out normalization processing on the detection data by using methods such as Max-min normalization extreme value normalization and the like.
Specifically, in some embodiments, the formula of the local mean filling strategy is:
X m indicating the value of each sensor measured at time m to be filled, X i For normal detection data, K is the number of existing data around the missing data.
The formula of the normalization process may be:
further, in the step S2 in this embodiment, the input detection data is a multidimensional data sequence, and after being input to the trained spatio-temporal feature extraction model, spatio-temporal feature reconstruction data is output; and acquiring a multi-dimensional residual sequence between input and output based on the detection data and the space-time characteristic reconstruction data.
Further, in an embodiment, in consideration of a correlation between each variable, the step S3 may be performed to calculate the multi-dimensional residual sequence through mahalanobis distance, and convert the multi-dimensional residual sequence into a one-dimensional performance index. In other embodiments, due to the influence of SCADA data noise and uncertainty in the modeling process, smoothing may be performed on the performance index by an exponentially weighted moving average EWMA to obtain an optimized performance index.
The formula for calculating and obtaining the performance index by adopting the Mahalanobis distance is as follows:
e is a reconstruction error of a space-time feature extraction model, and mu is a dimension mean value of the detection data;
the calculation formula for smoothing the performance index is as follows:
RE t ＝λE t +(1-λ)RE t-1 ；
and the lambda is a smoothing coefficient.
In this embodiment, λ is preferably 0.01, and in other embodiments, there may be other smoothing coefficients, which is not limited herein.
According to the fan resistance fault detection method provided by the embodiment, abnormal data caused by abandoned wind or sensor faults and the like can be effectively eliminated by carrying out data cleaning, filling and normalization on detection data acquired in real time, so that the rationality of the detection data is increased, and the error rate is reduced; the multi-dimensional residual data are converted into the one-dimensional performance data through the Mahalanobis distance, the complexity of data monitoring is simplified, and the detection efficiency is effectively improved.
As shown in fig. 3, fig. 3 is a logic flow diagram of training a spatio-temporal feature extraction model and a support vector regression model according to an embodiment of the present invention. In this embodiment, the following are specific:
and before the step S1, S0 is carried out, and a time-space feature extraction model DAE-GRU and a support vector regression model are trained on the basis of historical detection data of the wind turbine generator to be detected.
The S0 comprises:
s01, training a time-space feature extraction model based on historical detection data of the wind turbine generator to be detected; the method comprises the following specific steps:
s011, historical detection data of wind turbine generator to be detected are obtained
And performing data preprocessing, and taking historical detection data without fault states as a training data set.
In the present embodiment of the present invention,
the data representing the data collected by each sensor of the wind turbine generator in the preset sampling time period may include: wind speed, power, rotational speed, gearbox oil temperature, etc., l being the length of time. M is the dimension of the SCADA data, namely the number of the acquired variables. The data preprocessing process comprises the following steps: data cleansing, data population and/or normalization processing, and the like.
In practical applications, the data preprocessing process may be the same as the data preprocessing process of the real-time monitoring.
Further, the training data set is input to a pre-constructed spatio-temporal feature extraction model for iteration by means of a pre-constructed regularization method, and spatio-temporal feature reconstruction data of each iteration stage is output.
Specifically, as shown in fig. 4, which is a schematic structural diagram of a spatio-temporal feature extraction model provided in an embodiment of the present application, in an embodiment shown in fig. 4, the spatio-temporal feature extraction model, also referred to as a DAE-GRU-based spatio-temporal model normal behavior model, includes a deep self-encoder and a gated recursion unit.
Inputting the training data into the space-time feature extraction model, and the deep-layer self-encoder performs spatial feature extraction on the training data
Carrying out extraction; f. of E Network architecture representing a deep level of autoencoders, W E ,B E Is a parameter of the encoder, W E Is the weight of the internal network of the deep-level self-encoder, B E Is the deviation of the internal network of the deep self-encoder; the gating recursion unit is used for extracting the time characteristics of the training data; the output of the gated recursion unit GRU at time t is:
the output of the space-time feature extraction model is the space-time feature reconstruction data of the input detection data
The model is trained by continuously reducing the root Mean Square Error (MSE) between the input data and the output data of the model until the model training reaches the set iteration number. The pre-constructed regularization strategy can be a Dropout regularization method, and is used for preventing the model from being over-fitted, so that the accuracy of the model can be improved.
In one embodiment, the loss function of the model may be trained to minimize its loss using a gradient descent algorithm in an Adam optimizer. The Adam optimizer replaces a first-order optimization algorithm of the traditional gradient descent process, and the weights of the network are updated by adopting a gradient descent method iteration based on training data.
In this embodiment, M is preferably 15, the input of the gated recursion unit is preferably a time window with a fixed length l =12, the structure of the deep self-encoder is preferably 15-100-50-25-50-100-15, the number of neurons of the gated recursion unit is preferably 100, the amount of training data input per batch is 100, the learning rate is preferably 0.001, and the neuron discarding rate of the gated recursion unit GRU is preferably 0.2.
Further, S013 is implemented, data are reconstructed based on the spatio-temporal features, a gradient descent rule and a loss function which are constructed in advance are adopted, the target updating weight of each iteration stage is calculated until the convergence of the loss function is completed, and the spatio-temporal feature extraction model is trained.
In an embodiment, the gradient descent rule may be a gradient descent rule of an Adam optimizer, and the activation functions for performing nonlinear transformation on the data are sigmoid, so as to enhance the expression capability of the model.
Specifically, the loss function is a function using a root mean square error:
the described
In order to train the data, it is,
and extracting spatio-temporal feature reconstruction data output by the model data for spatio-temporal features.
Further, the S0 further includes:
s02, training a support vector regression model based on historical detection data of the wind turbine generator to be tested and the trained space-time feature extraction model, and specifically comprising the following steps:
and S021, acquiring training data input into a pre-constructed model supporting autoregressive.
The training data comprise preprocessed historical detection data of the wind turbine generator to be detected and performance indexes of the wind turbine generator to be detected.
And the performance index is a spatio-temporal feature extraction model performance index obtained by inputting the historical detection data into a trained spatio-temporal feature extraction model and calculating the residual error of the historical detection data and the output value of the spatio-temporal feature extraction model by adopting the Mahalanobis distance. Specifically, the training data are normal data acquired by the wind turbine generator to be tested in a normal operation state.
S022, inputting the training data into a pre-constructed support vector regression model, and training the support vector regression model, wherein parameters required to be selected by the support vector regression model are obtained through a wolf optimization algorithm.
In one embodiment, the training of the support vector regression model may include:
given training data
The support vector regression model is input and a hyperplane f (X) = ω (X) + b is fitted such that the difference between f (X) and RE is minimized.
The loss function of the model can be formalized as:
where ω is the weight, b is the corresponding deviation, C is the penalty factor, ε is the maximum error allowed by the regression, l ε Is an insensitive loss function.
The constraint may be expressed as:
Z＝(f(X)-RE)。
by introducing a relaxation variable, the loss function of the model is converted into:
the corresponding constraints translate into:
introducing a Lagrange multiplier converts the loss function of the model into:
the corresponding constraints translate into:
finally, introducing a kernel function to convert the expression of the model into:
wherein, beta i ,
Is the Lagrange multiplier, K (X) i And X) is a kernel function, and is used for converting the inner product operation of the low-dimensional space into the kernel function operation of the high-dimensional space.
The kernel function used in this embodiment is a gaussian kernel function:
where σ is a kernel function parameter.
The accuracy of the support vector regression model can be improved by optimizing the penalty factor C and the kernel function parameter sigma by adopting a wolf algorithm. As shown in fig. 5, fig. 5 is a schematic flow chart of the grayish optimization algorithm provided in an embodiment of the present invention. Specifically, the parameters to be selected by the vector regression model (GWO-SVR regression model) include: penalty factor C and kernel function parameter σ.
In S022, obtaining the parameters to be selected for the support vector regression model based on the grayish optimization algorithm may specifically include:
s0221, initializing parameters of a wolf grey scale algorithm and a support vector regression model, wherein the parameters are a wolf grey scale population size N, a maximum allowable iteration time tmax, a penalty factor C and a value range of a kernel function parameter sigma.
S0222, initializing the population by adopting a pre-constructed optimal point set rule, and determining the initial value of the position (C, sigma) of each wolf.
S0223, inputting the training data into the support vector regression model, and calculating the fitness of each wolf head under the initial value.
S0224, selecting the first 3 grey wolves with the best fitness, and updating the positions (C and sigma) of the 3 grey wolves.
S0225, calculating the fitness of all individual gray wolfs, comparing the fitness before and after position updating, if the current value is superior to the fitness obtained in the previous iteration, updating the gray wolf positions of the third fitness, otherwise, not updating the positions.
S0226: and comparing the current iteration times with the maximum allowable iteration times, if the current iteration times do not reach N, continuing optimizing, otherwise finishing the training, taking the position (C, sigma) value of the wolf head as the optimal solution, and finishing the training of the corresponding support vector regression model.
In some other embodiments, the method further includes obtaining a test data set of the spatio-temporal feature extraction model and the support vector regression model, and calculating the fault detection accuracy of the wind turbine generator, specifically including:
a1, historical detection data of a wind turbine generator to be detected are obtained and data preprocessing is carried out, wherein the historical detection data comprise detection data in a normal operation state and detection data in a fault state.
A2, respectively inputting the historical detection data into a trained space-time feature extraction model and a support vector regression model to obtain a performance index and a dynamic threshold;
a3, comparing the performance index with the dynamic threshold value, judging the type (normal operation or fault) of the input historical detection data,
and A4, calculating the detection accuracy of detecting the faults of the wind turbine generator based on the judgment result.
For example, in one embodiment, the detection accuracy is calculated to be 95% and above. Fig. 6 is a schematic diagram of a confusion matrix of the wind turbine generator fault detection result in an embodiment of the present invention.
As shown in FIG. 6, in one embodiment, the detection accuracy for detecting a wind turbine fault is calculated. When the real label is 0, 1089 of the input 1142 test detection data are correctly identified, and 53 identification errors occur; when the real tag is 1, 498 test detection data are correctly identified and 18 test detection data are incorrectly identified in 516 test detection data.
In the above embodiment, the historical SCADA detection data of the wind turbine generator is collected first, and is divided into a training data set and a test data set. And carrying out data cleaning on the training data set to obtain normal data of the running of the wind turbine generator. And establishing a space-time neural network model based on the DAE-GRU, inputting normal data into the space-time model, and continuously reducing the reconstruction error of the space-time neural network model through training. And then converting the output of the space-time neural network model into a performance index for monitoring, training the GWO-SVR regression model by fitting the input normal data and the corresponding performance index to obtain an ideal regression model, and acquiring a dynamic threshold.
Because the training of the regression model is influenced by the overlarge sample size of the training data set, and the final training effect is also influenced by the selection of the punishment factor and the kernel function of the SVR, in some embodiments, a Hui wolf optimization algorithm is introduced to perform self-adaptive optimization on the punishment factor C and the kernel function sigma in the SVR, so that the influence of manual setting on the model regression result can be avoided. And inputting the test data set into the trained spatio-temporal model to obtain a corresponding performance index, and simultaneously inputting the test data set into the trained regressor to obtain a corresponding dynamic threshold. And finally, comparing the performance index with the dynamic threshold value, and judging the detection precision of detecting the faults of the wind turbine generator by finally obtaining the running state of the wind turbine generator.
The wind turbine is a complex system with strong coupling and nonlinearity. The SCADA data is a multidimensional time sequence, each sensor variable has strong time dependence due to the dependence and interaction between different subsystems in the wind turbine, and different sensor variables have spatial correlation. According to the wind turbine generator fault state detection method based on the space-time neural network, the time dependency and the space dependency of SCADA data of the wind turbine generator are considered, the whole wind turbine generator is detected, and false detection and missing detection which are possibly caused by research of a single part can be effectively avoided; a symmetrical structure of the self-encoder is adopted, the spatial feature extraction capability of the self-encoder and the time feature extraction capability of the gating recursion unit are combined, a space-time feature extraction model based on the deep self-encoder and the gating recursion unit is established, and the correlation of variables in time and space is considered. Comprehensively considering the correlation among the variables; and converting the multi-dimensional residual error obtained by the space-time model into a one-dimensional performance index for monitoring by adopting the Mahalanobis distance. The method has the advantages that the gray wolf optimization algorithm and the support vector regression-based algorithm are adopted, the self-adaptive setting of the dynamic threshold is realized, the condition that the running state of the wind turbine generator is changeable is considered, the final detection precision of the wind turbine generator is improved, the fault can be detected more timely, the further deterioration of the fault is avoided, and the economic benefit is improved.
In addition, the present invention further provides an electronic device, which includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program stored in the memory, so as to implement the steps of the wind turbine generator fault state detection method according to any of the above embodiments.
In addition, the present invention further provides a computer readable storage medium, wherein a computer program is stored on the computer readable storage medium, and when being executed by a processor, the computer program implements the steps of the wind turbine generator fault state detection method according to any of the above embodiments.
The wind turbine generator fault state detection method based on the space-time neural network provided in the above embodiments of the present invention does not need to install additional sensors and other devices on the wind turbine generator, and directly judges based on the detection Data of the existing widely adopted SCADA (Supervisory Control and Data Acquisition) system, so that the implementation difficulty is small, and the available range is wide. The method comprehensively considers the problem that the high-dimensional spatial correlation among different variables of SCADA data, the time dependence of the same variable at different moments and the fixed threshold value neglect the variable running state of the wind turbine generator, and by training a space-time feature extraction model and a support vector regression model, the method is high in detection precision and small in error, can achieve the effect of detecting the fault state of the wind turbine generator in real time, improves the running reliability of the wind turbine generator, reduces the operation and maintenance cost, is beneficial to promoting the development of wind power, and further reduces the pollution in the environment caused by fossil energy consumption.
In the description of the present invention, it is to be understood that the terms "first", "second" and the like are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or implying any number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of the present invention, "a plurality" means two or more unless specifically defined otherwise.
In the description of the present specification, the description of "one embodiment", "some embodiments", "examples", "specific examples" or "some examples", etc., means that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, the schematic representations of the terms used above are not necessarily intended to refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Furthermore, various embodiments or examples and features of different embodiments or examples described in this specification can be combined and combined by one skilled in the art without contradiction.
Although embodiments of the present invention have been shown and described above, it should be understood that the above embodiments are exemplary and not to be construed as limiting the present invention and that those skilled in the art may make modifications, alterations, substitutions and alterations to the above embodiments within the scope of the present invention.
Claims (10)
1. A wind turbine generator fault state detection method based on a space-time neural network is characterized by comprising the following steps:
s1, acquiring real-time detection data of a sensor of a wind turbine generator to be detected and performing data preprocessing; the detection data includes: wind speed, power, rotation speed, pitch angle, temperature and detection time;
s2, inputting the detection data into a trained space-time feature extraction model to obtain the multi-dimensional residual error data; the multi-dimensional residual data is a residual between the real-time detection data and the output of the space-time characteristic extraction model;
the space-time feature extraction model comprises a deep self-encoder and a gated recursion unit, wherein the deep self-encoder is used for extracting the spatial features of the detection data; the gated recursion unit is used for extracting the time characteristics of the detection data;
s3, calculating the multi-dimensional residual error data by adopting the Mahalanobis distance to obtain a real-time performance index of a space-time feature extraction model;
s4, inputting the performance index and the real-time detection data into a trained vector regression algorithm model to obtain a real-time dynamic threshold;
and S5, detecting the fault state of the wind turbine generator based on the real-time dynamic threshold and the real-time performance index.
2. The wind turbine generator system fault status detection method of claim 1,
the data preprocessing in the S1 mainly comprises the following steps:
analyzing the detection data by adopting a fusion strategy based on ideal wind speed power, setting unnecessary data points as outliers and removing the outliers;
filling the detection data with the outliers removed by adopting a local mean filling strategy;
and carrying out normalization processing on the filled detection data.
3. The wind turbine generator system fault condition detection method of claim 1,
the S3, smoothing the performance index through an index weighted moving average value to obtain an optimized performance index;
the formula for calculating and obtaining the performance index by adopting the Mahalanobis distance is as follows:
e is a reconstruction error of a space-time feature extraction model, and mu is a dimension mean value of the detection data;
the calculation formula for smoothing the performance index is as follows:
RE t ＝λE t +(1-λ)RE t-1 ；
and the lambda is a smoothing coefficient.
4. The wind turbine generator system fault condition detection method of claim 1,
the S5 specifically comprises the following steps:
comparing the dynamic threshold to the performance index;
if the dynamic threshold value is smaller than the performance index, judging that the wind turbine generator is in a fault state;
and if the dynamic threshold is not less than the performance index, judging that the wind turbine generator is in a normal operation state.
5. The wind turbine generator system fault condition detection method of claim 1,
before the step S1, the method also comprises the step S0 of training a time-space feature extraction model and a support vector regression model based on historical detection data of the wind turbine generator to be detected;
the S0 comprises:
s01, training a time-space feature extraction model based on historical detection data of the wind turbine generator to be detected; the method comprises the following specific steps:
s011, acquiring historical detection data of the wind turbine generator to be detected, preprocessing the historical detection data, and taking the historical detection data without fault state as a training data set;
s012, inputting the training data set into a pre-constructed space-time feature extraction model for iteration by means of a pre-constructed regularization method, and outputting space-time feature reconstruction data of each iteration stage;
s013, based on the spatio-temporal feature reconstruction data, calculating a target updating weight of each iteration stage by adopting a pre-constructed gradient descent rule and a loss function until the loss function is converged, and finishing training a spatio-temporal feature extraction model;
the loss function is:
the above-mentioned
In order to train the data in the form of,
and extracting spatio-temporal feature reconstruction data output by the model data for spatio-temporal features.
6. The wind turbine generator system fault status detection method of claim 5,
the S0 further comprises:
s02, training a support vector regression model based on historical detection data of the wind turbine generator to be detected, and specifically comprises the following steps:
s021, obtaining training data input into a pre-constructed support autoregressive model;
the training data comprise preprocessed historical detection data of the wind turbine generator to be detected and a performance index of the wind turbine generator to be detected;
the performance index is a space-time feature extraction model performance index obtained by inputting the historical detection data into a trained space-time feature extraction model, and calculating the residual error of the historical detection data and the output value of the space-time feature extraction model by adopting the Mahalanobis distance;
s022, inputting the training data into a pre-constructed support vector regression model, and training the support vector regression model, wherein parameters required to be selected by the support vector regression model are obtained through a wolf optimization algorithm.
7. The wind turbine generator system fault condition detection method of claim 6,
the parameters required to be selected by the vector regression model comprise: a penalty factor C and a kernel function parameter sigma;
in S022, obtaining the parameters to be selected for the support vector regression model based on the grayish optimization algorithm specifically includes:
s0221, initializing parameters of a gray wolf algorithm and a support vector regression model, wherein the parameters are a gray wolf population size N, a maximum allowable iteration time tmax, a penalty factor C and a value range of a kernel function parameter sigma;
s0222, initializing the population by adopting a pre-constructed optimal point set rule, and determining the initial value of the position (C, sigma) of each wolf;
s0223, inputting the training data into the support vector regression model, and calculating the fitness of each wolf head under the initial value;
s0224, selecting the front 3 grey wolves with the best fitness, and updating the positions (C and sigma) of the 3 grey wolves;
s0225, calculating the fitness of all the grey wolf individuals, comparing the fitness before and after position updating, if the current value is superior to the fitness obtained by the previous iteration, updating the grey wolf position three before the fitness, otherwise, not updating the position;
s0226: and comparing the current iteration times with the maximum allowable iteration times, if the current iteration times do not reach N, continuing to optimize, otherwise, finishing the training, and finishing the training of the corresponding support vector regression model, wherein the position (C, sigma) values of the wolf head are the optimal solution.
8. The wind turbine generator system fault condition detection method of claim 1,
the structure of the deep self-encoder is 15-100-50-25-50-100-15, and the number of the neurons of the gated recursion unit is 100.
9. An electronic device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program stored in the memory to implement the steps of the wind turbine generator fault status detection method according to any one of claims 1 to 8.
10. A computer-readable storage medium, characterized in that a computer program is stored on the computer-readable storage medium, which computer program, when being executed by a processor, carries out the steps of the wind turbine generator fault status detection method according to any one of the preceding claims 1 to 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211066553.8A CN115434875A (en) | 2022-08-31 | 2022-08-31 | Wind turbine generator fault state detection method based on space-time neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211066553.8A CN115434875A (en) | 2022-08-31 | 2022-08-31 | Wind turbine generator fault state detection method based on space-time neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115434875A true CN115434875A (en) | 2022-12-06 |
Family
ID=84247164
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211066553.8A Pending CN115434875A (en) | 2022-08-31 | 2022-08-31 | Wind turbine generator fault state detection method based on space-time neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115434875A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117057676A (en) * | 2023-10-11 | 2023-11-14 | 深圳润世华软件和信息技术服务有限公司 | Multi-data fusion fault analysis method, equipment and storage medium |
| CN117131315A (en) * | 2023-10-27 | 2023-11-28 | 北京志翔科技股份有限公司 | Out-of-tolerance electric energy meter determining method and medium based on solving multi-element quadratic function extremum |
-
2022
- 2022-08-31 CN CN202211066553.8A patent/CN115434875A/en active Pending
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117057676A (en) * | 2023-10-11 | 2023-11-14 | 深圳润世华软件和信息技术服务有限公司 | Multi-data fusion fault analysis method, equipment and storage medium |
| CN117057676B (en) * | 2023-10-11 | 2024-02-23 | 深圳润世华软件和信息技术服务有限公司 | Multi-data fusion fault analysis method, equipment and storage medium |
| CN117131315A (en) * | 2023-10-27 | 2023-11-28 | 北京志翔科技股份有限公司 | Out-of-tolerance electric energy meter determining method and medium based on solving multi-element quadratic function extremum |
| CN117131315B (en) * | 2023-10-27 | 2024-01-23 | 北京志翔科技股份有限公司 | Out-of-tolerance electric energy meter determining method and medium based on solving multi-element quadratic function extremum |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN115434875A (en) | Wind turbine generator fault state detection method based on space-time neural network | |
| CN111275288B (en) | XGBoost-based multidimensional data anomaly detection method and device | |
| CN107341349B (en) | Method and system for evaluating health of fan, memory and controller | |
| CN110685868A (en) | Wind turbine generator fault detection method and device based on improved gradient elevator | |
| CN112784965A (en) | Large-scale multi-element time series data abnormity detection method oriented to cloud environment | |
| CN110594107B (en) | Wind turbine generator fault detection method and device based on rapid gradient elevator | |
| CN111160620B (en) | Short-term wind power prediction method based on end-to-end memory network | |
| CN117290800B (en) | Timing sequence anomaly detection method and system based on hypergraph attention network | |
| CN111079343B (en) | Wind turbine generator effective wind speed estimation method based on width learning | |
| CN116937579B (en) | Wind power interval prediction considering space-time correlation and interpretable method thereof | |
| CN112070322A (en) | High-voltage cable line running state prediction method based on long-short term memory network | |
| CN113107785A (en) | Real-time monitoring method and device for power performance abnormity of wind turbine generator | |
| Chen et al. | Effective wind speed estimation study of the wind turbine based on deep learning | |
| CN114519923A (en) | Intelligent diagnosis and early warning method and system for power plant | |
| CN112232591A (en) | Icing thickness intelligent early warning method based on meteorological factors | |
| CN114500004A (en) | Anomaly detection method based on conditional diffusion probability generation model | |
| CN116186633A (en) | Power consumption abnormality diagnosis method and system based on small sample learning | |
| CN113283546B (en) | Furnace condition abnormity alarm method and system of heating furnace integrity management centralized control device | |
| CN114742363A (en) | Energy efficiency state evaluation method, system and medium for wind turbine generator | |
| CN116821610B (en) | Method for optimizing wind power generation efficiency by utilizing big data | |
| Li et al. | A novel adaptive STFT-SFA based fault detection method for nonstationary processes | |
| CN110414734B (en) | Method for forecasting and evaluating wind resource utilization rate | |
| Wang et al. | Research on ultra-short term forecasting technology of wind power output based on various meteorological factors | |
| CN116933643A (en) | Intelligent data monitoring method based on partial robust M regression and multiple interpolation | |
| CN116505556A (en) | Wind farm power control system and method based on primary frequency modulation |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |