CN114239396A - Fan gear box state prediction method and system - Google Patents
Fan gear box state prediction method and system Download PDFInfo
- Publication number
- CN114239396A CN114239396A CN202111533015.0A CN202111533015A CN114239396A CN 114239396 A CN114239396 A CN 114239396A CN 202111533015 A CN202111533015 A CN 202111533015A CN 114239396 A CN114239396 A CN 114239396A
- Authority
- CN
- China
- Prior art keywords
- neural network
- data set
- prediction
- model
- data
- 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
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- 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
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- 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
- G06N3/045—Combinations of networks
-
- 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
- G06N3/048—Activation functions
-
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2113/00—Details relating to the application field
- G06F2113/06—Wind turbines or wind farms
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
- Y02E40/00—Technologies for an efficient electrical power generation, transmission or distribution
- Y02E40/70—Smart grids as climate change mitigation technology in the energy generation sector
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Computer Hardware Design (AREA)
- Geometry (AREA)
- Wind Motors (AREA)
Abstract
The invention discloses a method and a system for predicting the state of a fan gear box, which relate to the field of wind turbines, wherein the method comprises the following steps: acquiring an original data set of a plurality of groups of operating parameters of the wind turbine generator from a wind power plant monitoring and data acquisition system; preprocessing the original data set by using a preset algorithm to reduce the feature vector dimension of the original data set to obtain dimension-reduced data; carrying out convolution operation on the dimensionality reduction data by utilizing a convolution long and short memory neural network model so as to complete reconstruction on a characteristic vector of a data set; establishing a plurality of target prediction models by carrying out deep neural network training on the reconstructed data; and determining an optimal model from the target prediction models as a prediction network model, and predicting the state of the fan gearbox by using the prediction network model.
Description
Technical Field
The invention relates to the field of wind turbines, in particular to a method and a system for predicting the state of a fan gear box.
Background
Wind power is the most important power supply of a novel power system and becomes the third largest energy source in China. The wind turbine generator is severe in operating environment and is very easy to cause damage to a gear box. Damage to the gearbox to a certain extent can cause malfunction and safety accidents. Therefore, the future health state is predicted according to the current operation data and damage situation of the fan gear box, and the method has important significance for arranging maintenance, preventing accidents and guaranteeing safe operation of the wind turbine generator.
In the prior art, the technical problem that a multi-period continuous time prediction cannot be carried out on a state prediction model of a fan gearbox exists.
Disclosure of Invention
The application provides a fan gear box state prediction method and system, and solves the technical problem that a multi-period continuous time prediction cannot be carried out on a state prediction model of a fan gear box in the prior art.
In view of the above problems, the present application provides a method and a system for predicting a state of a fan gearbox.
In one aspect, the present application provides a wind turbine gearbox state prediction method, where the method is applied to a wind turbine gearbox state prediction system, and the method includes: acquiring an original data set of a plurality of groups of operating parameters of the wind turbine generator from a wind power plant monitoring and data acquisition system; preprocessing the original data set by using a preset algorithm to reduce the feature vector dimension of the original data set to obtain dimension-reduced data; carrying out convolution operation on the dimensionality reduction data by utilizing a convolution long and short memory neural network model so as to complete reconstruction on a characteristic vector of a data set; establishing a plurality of target prediction models by carrying out deep neural network training on the reconstructed data; and determining an optimal model from the target prediction models as a prediction network model, and predicting the state of the fan gearbox by using the prediction network model.
In another aspect, the present application further provides a wind turbine gearbox state prediction system, wherein the system comprises: the system comprises a first obtaining unit, a second obtaining unit and a control unit, wherein the first obtaining unit is used for obtaining an original data set of multiple groups of operating parameters of the wind turbine generator from a wind power plant monitoring and data acquisition system; the first execution unit is used for preprocessing the original data set by using a preset algorithm and reducing the feature vector dimension of the original data set to obtain dimension-reduced data; the second execution unit is used for carrying out convolution operation on the dimensionality reduction data by utilizing a convolution long and short memory neural network model so as to complete reconstruction on a data set feature vector; a third execution unit, configured to perform deep neural network training on the reconstructed data to establish a plurality of target prediction models; and the fourth execution unit is used for determining an optimal model from the target prediction models as a prediction network model and predicting the state of the fan gearbox by using the prediction network model.
In a third aspect, the present application provides a comprehensive query and contract-resolving device based on account and contract relation, including a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor implements the steps of the method according to any one of the first aspect when executing the program.
One or more technical solutions provided in the present application have at least the following technical effects or advantages:
acquiring an original data set of a plurality of groups of operating parameters of the wind turbine generator from a wind power plant monitoring and data acquisition system; preprocessing the original data set by using a preset algorithm to reduce the feature vector dimension of the original data set to obtain dimension-reduced data; carrying out convolution operation on the dimensionality reduction data by utilizing a convolution long and short memory neural network model so as to complete reconstruction on a characteristic vector of a data set; establishing a plurality of target prediction models by carrying out deep neural network training on the reconstructed data; and determining an optimal model from the target prediction models as a prediction network model, and predicting the state of the fan gearbox by using the prediction network model. The method has the advantages of being suitable for the wind turbine generator and having the data prediction task with the time sequence characteristic type, high in accuracy, time-consuming, controllable and efficient, further accurately predicting the future health state of the wind turbine generator, and guaranteeing the safe operation of the wind turbine generator.
The foregoing description is only an overview of the technical solutions of the present application, and the present application can be implemented according to the content of the description in order to make the technical means of the present application more clearly understood, and the following detailed description of the present application is given in order to make the above and other objects, features, and advantages of the present application more clearly understandable.
Drawings
In order to more clearly illustrate the technical solutions in the present application or the prior art, the drawings needed to be used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only exemplary, and for those skilled in the art, other drawings can be obtained according to the provided drawings without inventive effort.
FIG. 1 is a schematic flow chart of a method for predicting a state of a gearbox of a wind turbine according to the present application;
FIG. 2 is a flowchart illustrating an algorithm of a convolutional long-short memory neural network (ConvLSTMNN) in a wind turbine gearbox state prediction method according to the present disclosure;
FIG. 3 is a schematic flow chart illustrating a process of determining an optimal model as a prediction network model from the plurality of target prediction models according to the method for predicting the state of the wind turbine gearbox of the present application;
FIG. 4 is a schematic structural diagram of a wind turbine gearbox state prediction system according to the present application;
fig. 5 is a schematic structural diagram of an exemplary electronic device of the present application.
Detailed Description
The application provides a fan gear box state prediction method and system, and solves the technical problem that a multi-period continuous time prediction cannot be carried out on a state prediction model of a fan gear box in the prior art. The method has the advantages of being suitable for the wind turbine generator and having the data prediction task with the time sequence characteristic type, high in accuracy, time-consuming, controllable and efficient, further accurately predicting the future health state of the wind turbine generator, and guaranteeing the safe operation of the wind turbine generator.
Hereinafter, example embodiments according to the present application will be described in detail with reference to the accompanying drawings. It should be apparent that the described embodiments are merely some embodiments of the present application and not all embodiments of the present application, and it should be understood that the present application is not limited to the example embodiments described herein.
According to the technical scheme, the data acquisition, storage, use, processing and the like meet relevant regulations of national laws and regulations.
Wind power is the most important power supply of a novel power system and becomes the third largest energy source in China. The wind turbine generator is severe in operating environment and is very easy to cause damage to a gear box. Damage to the gearbox to a certain extent can cause malfunction and safety accidents. Therefore, the future health state is predicted according to the current operation data and damage situation of the fan gear box, and the method has important significance for arranging maintenance, preventing accidents and guaranteeing safe operation of the wind turbine generator.
In view of the above technical problems, the technical solution provided by the present application has the following general idea:
the application provides a fan gearbox state prediction method, wherein the method is applied to a fan gearbox state prediction system, and the method comprises the following steps: acquiring an original data set of a plurality of groups of operating parameters of the wind turbine generator from a wind power plant monitoring and data acquisition system; preprocessing the original data set by using a preset algorithm to reduce the feature vector dimension of the original data set to obtain dimension-reduced data; carrying out convolution operation on the dimensionality reduction data by utilizing a convolution long and short memory neural network model so as to complete reconstruction on a characteristic vector of a data set; establishing a plurality of target prediction models by carrying out deep neural network training on the reconstructed data; and determining an optimal model from the target prediction models as a prediction network model, and predicting the state of the fan gearbox by using the prediction network model.
For better understanding of the above technical solutions, the following detailed descriptions will be provided in conjunction with the drawings and the detailed description of the embodiments.
Example one
Referring to fig. 1, the present application provides a method for predicting a state of a fan gearbox, wherein the method is applied to a system for predicting a state of a fan gearbox, and the method specifically includes the following steps:
step S100: acquiring an original data set of a plurality of groups of operating parameters of the wind turbine generator from a wind power plant monitoring and data acquisition system;
specifically, a wind farm SCADA (supervisory control and data acquisition) system is utilized to obtain an original data set of multiple groups of operating parameters of the wind turbine generator. The wind power plant SCADA (supervisory control and data acquisition) system is a computer-based production process control and scheduling automation system, and can monitor and control field operation equipment so as to realize functions of data acquisition, equipment control, measurement, parameter adjustment, accident alarm and the like. The wind power plant SCADA (supervisory control and data acquisition system) is used for acquiring data, so that the completeness of the information can be ensured. The original data set of the multiple groups of operating parameters of the wind turbine generator is a data set formed by factors influencing the healthy operation of the wind turbine generator. For example, the wind turbine generator has active power, reactive power, voltage, current, temperature, wind speed, wind direction, yaw starting state, motor operation and the like; the gearbox oil temperature, the gearbox main bearing temperature, the front bearing temperature, the rear bearing temperature and the gearbox oil level are taken as health parameters of the gearbox. The technical effect of collecting complete original data sets of multiple groups of operation parameters of the wind turbine generator and laying a foundation for preprocessing the original data sets to obtain dimension reduction data is achieved.
Step S200: preprocessing the original data set by using a preset algorithm to reduce the feature vector dimension of the original data set to obtain dimension-reduced data;
step S300: carrying out convolution operation on the dimensionality reduction data by utilizing a convolution long and short memory neural network model so as to complete reconstruction on a characteristic vector of a data set;
specifically, on the basis of acquiring an original data set of multiple groups of operating parameters of a wind turbine generator by using a wind power plant monitoring and data acquisition system, the original data set is preprocessed according to a preset algorithm, so that the characteristic vector dimension of the original data set can be reduced to obtain dimension reduction data. The preset algorithm is a spearman correlation coefficient method. The preprocessing is to screen abnormal data with a measurement value exceeding a normal range, remove the whole group of data at a corresponding sampling moment, and reserve the residual data as an effective data set, wherein the residual data is dimension reduction data. And then, carrying out convolution operation on the dimensionality reduction data according to the long and short convolution memory neural network model so as to complete reconstruction on the characteristic vector of the data set. The data set characteristic vector is obtained by performing correlation analysis on collected data of other characteristic variables except 4 characteristic quantities such as gearbox oil temperature, gearbox main bearing temperature, front bearing temperature, rear bearing temperature and gearbox oil level, and discarding the characteristic vector of which the Spearman correlation coefficient is greater than a threshold value. Abnormal data are eliminated, so that the sample size of a characteristic vector in a data set is reduced, and the loss of information amount is minimum after dimensionality reduction, and the operation speed of the convolution long and short memory neural network model on the data is accelerated; and the reconstruction of the characteristic vector of the data set is completed, and the technical effect of providing data support for the subsequent deep neural network training of the data set is achieved.
Step S400: establishing a plurality of target prediction models by carrying out deep neural network training on the reconstructed data;
step S500: and determining an optimal model from the target prediction models as a prediction network model, and predicting the state of the fan gearbox by using the prediction network model.
Specifically, on the basis of completing reconstruction of the feature vectors of the data set, deep neural network training is performed on the reconstructed data, and a plurality of target prediction models can be established. Among them, the Deep Neural Network (DNN) is a technique in the field of machine learning, which can represent a complex function with fewer parameters. Better generalization and smaller training error can be obtained through deep neural network training. And determining the optimal model as a prediction network model according to the target prediction models, and predicting the state of the fan gearbox by using the optimal model. The method has the advantages of being suitable for the wind turbine generator and having the data prediction task with the time sequence characteristic type, high in accuracy, time-consuming, controllable and efficient, further accurately predicting the future health state of the wind turbine generator, and guaranteeing the safe operation of the wind turbine generator. Optionally, the target prediction model is a single hidden layer feedforward neural network (SLF), a support vector machine regression model (SVR), a Random Forest (RF), a convolutional neural network CNN, a long and short memory neural network (LSTM), and a convolutional long and short memory neural network (ConvLSTMNN).
Further, step S200 of the present application further includes:
step S210: calculating the correlation between two variables by adopting a calculation formula of a spearman correlation coefficient method to obtain a variable correlation coefficient, wherein the calculation formula is as follows:
wherein rho is a Spierman correlation coefficient, a and b are two random variables, and the ith values obtained by the two random variables are respectively expressed as ai,bi,Is the average of all random variables a, b;
step S220: judging whether the variable correlation coefficient exceeds a preset threshold value or not;
step S230: when the variable correlation coefficient exceeds the threshold value, the feature vector corresponding to the variable correlation coefficient is discarded, and the feature vector dimension is reduced;
each input feature vector of the data set subjected to dimensionality reduction through the spearman correlation coefficient method is a continuous 12-hour power distribution unit state quantity, and a target prediction value is a 10-day-later wind power gear box state.
Further, step S230 of the present application further includes:
step S231: and the oil temperature of the gearbox, the temperature of a main bearing of the gearbox, the temperature of a front bearing, the temperature of a rear bearing and the oil level of the gearbox are not subjected to characteristic vector rejection operation.
Specifically, the original data set is preprocessed by using a preset algorithm, when dimension of a feature vector of the original data set is reduced to obtain dimension-reduced data, correlation between two variables is calculated by using a calculation formula of a spearman correlation coefficient method, and a variable correlation coefficient can be obtained. And then, judging whether the variable correlation coefficient exceeds a preset threshold value. And when the variable correlation coefficient exceeds a preset threshold value, discarding the feature vector corresponding to the variable correlation coefficient, and reducing the dimension of the feature vector. In particular, the gearbox oil temperature, gearbox main bearing temperature, front bearing temperature, rear bearing temperature and gearbox oil level are contained within predicted targets and are not subjected to eigenvector rejection operations. The variable correlation coefficient is a spearman correlation coefficient and can be obtained by a calculation formula of a spearman correlation coefficient method. The preset threshold is the maximum value of the corresponding spearman correlation coefficient when the fan gearbox operates healthily. Besides 4 characteristic quantities such as the oil temperature of the gear box, the temperature of a main bearing of the gear box, the temperature of a front bearing, the temperature of a rear bearing and the oil level of the gear box, correlation analysis is carried out on collected data of other characteristic variables, characteristic vectors with the Spireman correlation coefficient larger than a threshold value are abandoned, and characteristic dimensions of the data are reduced.
Further, step S300 of the present application further includes:
step S310: the model algorithm of the convolution long and short memory neural network is as follows:
wherein, i (t) is an input gate, f (t) is a forgetting gate, o (t) is an output gate, c (t) is a memory unit state, h (t) is a memory unit output, sigma is a sigmoid activation function, and a is convolution operation.
Further, step S310 of the present application further includes:
step S311: after the sigmoid activation function is positioned at an output gate of the convolution long and short memory neural network, carrying out segmented mapping on an output value h (t) of the convolution long and short memory neural network model;
the segmentation mapping relation is as follows:
where σ is set to 0.02.
Specifically, as shown in fig. 2, a convolution operation is performed on the data set according to a convolution long-short memory neural network model algorithm, so as to reconstruct a feature vector of the data set. Meanwhile, after the Sigmoid activation function is positioned at an output gate of the convolutional long and short memory neural network, the output value h (t) of the convolutional long and short memory neural network model is subjected to segmented mapping, namely, a LeakyReLU activation function is added after an outlet h (t) of the convolutional long and short memory neural network. Wherein, the Sigmoid activation function is a common Sigmoid function in biology, and is also called a Sigmoid growth curve. Due to its simple increment and simple increment of the inverse function, Sigmoid function is often used as a threshold function of neural network, mapping variables between 0 and 1. The Sigmoid activation function can compress data, ensure the amplitude of the data and be suitable for forward transmission; but the phenomenon of gradient disappearance easily occurs, and when the Sigmoid activation function is close to a saturation region, the change is too slow, and deep neural network training cannot be performed. In this case, an LeakyReLU activation function is added after the exit h (t) of the convolutional long and short memory neural network. The LeakyReLU activation function is to assign a non-zero slope to all negative values. For example, the function shown in the above piecewise mapping relation may perform piecewise mapping on the output value h (t) of ConvLSTMNN, and has the effects of accelerating model training and avoiding gradient vanishing caused by the σ function. The technical effect that the characteristic vector of the data set is reconstructed by utilizing the convolution long and short memory neural network model and a foundation is laid for the subsequent deep neural network training of the data set is achieved.
Further, as shown in fig. 3, step S500 of the present application includes:
step S510: constructing a convolution long and short memory neural network model and respectively carrying out fitting training on the prediction target;
step S520: determining a forward data transfer path, calculating loss function loss values, updating weights and offsets of all network layers, calculating loss function loss values and reverse transfer errors, and updating the weights;
step S530: the convolution long-short memory neural network model is preset in batches through training, and when curve trends corresponding to loss function loss values meet expected requirements, evaluation indexes R are selected for the multiple target prediction models2Calculating a value;
step S540: according to the evaluation index R2Value, determining the prediction network model, which is an evaluation index R2The model whose value differs the least from 1.
Specifically, when the optimal model is determined to be the prediction network model from the multiple target prediction models, fitting training is carried out on the prediction targets through the long and short convolution memory neural network models respectively, loss function loss values and reverse transfer errors are calculated, weights are updated, when the models are trained for a certain batch, loss function loss curves tend to be stable, and an evaluation index R is selected2And the model with the highest value is stored to be used as a final prediction network model. The invention adopts a determination coefficient R2As an evaluation index of the model. The closer the decision coefficient is to 1, the higher the fitting accuracy of the model is indicated. Wherein the fitting training is deep neural network training. The loss function loss is used for describing the difference between the predicted value and the true value of the model and guiding the model to move towards the convergence direction in the training process. The better the loss function, the better the performance of the model in general. The loss function loss value is a parameter which helps the training mechanism to optimize at any time. The technical effects of improving the fitting precision of the model, determining the optimal model as a prediction network model and providing data support for the subsequent prediction of the state of the fan gearbox are achieved.
Further, step S530 of the present application includes:
step S531: by the formula:calculating to obtain the evaluation index R2Values where SST is the total square sum and SSR is the regression square sum.
Specifically, the evaluation index R is selected for the plurality of target prediction models2When calculating the value, the inventionApparent coefficient of determination R2As an evaluation index of the model. The closer the decision coefficient is to 1, the higher the fitting accuracy of the model is indicated.
The predictive power of the model is analyzed in the following with a specific example.
The model and comparison related by the invention are completed on the same computer. The computer configuration and simulation environment is as follows: win64, Intel i7-11700K, DDR4-3200-32GB, RTX-3080Ti (10GB), Python3.6, Keras 2.2.4. In addition to the models related to the method, the invention selects a single hidden layer feedforward neural network (SLF), a support vector machine regression model (SVR), a Random Forest (RF), a Convolutional Neural Network (CNN) and a long-short memory neural network (LSTM) as comparison models.
R for performance of each module in experimental simulation2And (6) carrying out statistics. Table I shows a comparison of the fitting performance of the models (R)2)。
In the prediction score of "cabin gear box oil pressure", R of SLF model2The values were lowest (0.69), RF highest (0.82) and ConvLSTMNN (0.81). The ConvLSTMNN model score was only 1.2% lower than the highest score. R of SVR model in prediction of' gearbox low speed bearing temperature2The lowest value (0.74), the ConvLSTMNN model score was the highest (0.85), 3.7% higher than the CNN model score. R of SLF model in prediction of "gearbox high speed bearing temperature2The value was lowest (0.75), the ConvLSTMNN model scored the highest (0.84), 2.4% higher than the CNN model; the predicted score SLF at "gear box side main bearing temperature" was lowest (0.66), SVR was highest (0.82), and ConvLSTMNN was (0.81). The ConvLSTMNN model score was 1.2% lower than the highest score.
Each model R2In the sum of values, the SLF model value is 2.87 at the lowest, the SVR and RF values are 3.10 at the same time, the CNN model value is 3.11, the LSTM model value is 3.07, and the ConvLSTMNN model value is 3.31. Highest score, above SLF model 153%, over the second (CNN model) 6.4%. Although not every target of the proposed model is the highest score, a higher score is maintained in every predicted target calculation. The comprehensive prediction capacity is strong, and different prediction data distributions can be dealt with.
The training time of each model is shown in table two.
Watch two
As the sample size increases (the sample size of a single wind turbine generator exceeds 28000) and the data characteristic quantity increases, the training time of SVR and RF is exponentially increased, but the accuracy is not obviously improved. The temporal complexity of the SVR itself is proportional to the vector product of the sample size and the feature number. The temporal complexity of SVR is O ((s. k. logs)m). Where s is the number of samples, k is the number of features, and m is the number of predicted targets. The training of the SVR takes geometric multiplication in the conditions of large data quantity, feature number and multiple targets. The random forest RF sets w search trees with complexity similar to SVRO (s k logs)m). The SVR and RF operation work is completed on the CPU in a centralized way, which is lower than the deep neural network algorithm completed by the CPU and the GPU together. Model complexity of SVR and RF is exponential, and training time consumption of complex models grows exponentially. SVR and RF are not the preferred algorithms for handling large data, high latitude features and multi-objective predictive tasks.
The time complexity of the deep neural network model isM is the complexity of the monomer network model, K is the convolution scale, and C is the number of convolution channels. The network depth of the ConvLSTMNN model exceeded that of the CNN and LSTM models, requiring 760.54 minutes for the entire training time. The temporal complexity of the ConvLSTMNN model is an additive relationship in predicting the number of targets. Accumulating increased temporal complexity is lower and easier than exponential in large data volume, multi-feature variable, and multi-objective tasksIs widely applied.
Further, step S400 of the present application includes:
step S410: the framework of the target prediction model is composed of three neural networks, including: the long and short convolutional neural network training system comprises a long and short convolutional memory neural network layer, a convolutional neural network and a full connection layer, wherein k time sequence data are simultaneously sent in each batch of deep neural network training, the data are replaced by a three-dimensional tensor from a one-dimensional tensor through the long and short convolutional memory neural network layer, and k is a positive integer.
Specifically, the frame of the target prediction model is composed of three neural networks, namely a convolution long and short memory neural network layer, a convolution neural network and a full connection layer, k time sequence data are simultaneously sent to each batch of training, and the data are m × 1 dimension tensors. And replacing the data by a three-dimensional tensor from a one-dimensional tensor through the long and short convolutional memory neural network layer. And it is inputting xtTo state ht-1And state ht-1To state ct-1The convolution operation is integrated into the conversion of (1). The operation can add time sequences into the vector features, so that the time relation among the features is enhanced, and the fitting precision of the model is improved.
Furthermore, the state of the fan gearbox is mainly predicted according to the health parameters, typical faults of the gearbox can be directly reflected that certain monitoring values exceed a normal range, and the health state of the fan gearbox can be reflected through analysis of the quantities. For example, Table III shows typical failure types for a gearbox.
Watch III
In summary, the method for predicting the state of the fan gearbox provided by the application has the following technical effects:
1. acquiring an original data set of a plurality of groups of operating parameters of the wind turbine generator from a wind power plant monitoring and data acquisition system; preprocessing the original data set by using a preset algorithm to reduce the feature vector dimension of the original data set to obtain dimension-reduced data; carrying out convolution operation on the dimensionality reduction data by utilizing a convolution long and short memory neural network model so as to complete reconstruction on a characteristic vector of a data set; establishing a plurality of target prediction models by carrying out deep neural network training on the reconstructed data; and determining an optimal model from the target prediction models as a prediction network model, and predicting the state of the fan gearbox by using the prediction network model. The method has the advantages of being suitable for the wind turbine generator and having the data prediction task with the time sequence characteristic type, high in accuracy, time-consuming, controllable and efficient, further accurately predicting the future health state of the wind turbine generator, and guaranteeing the safe operation of the wind turbine generator.
2. Besides 4 characteristic quantities such as the oil temperature of the gearbox, the temperature of a main bearing of the gearbox, the temperature of a front bearing, the temperature of a rear bearing and the oil level of the gearbox, correlation analysis is carried out on collected data of other characteristic variables, characteristic vectors with the Spireman correlation coefficient larger than a threshold value are abandoned, and characteristic dimensions of the data can be reduced.
3. The frame of the target prediction model is composed of three neural networks of a convolution long and short memory neural network layer, a convolution neural network and a full connection layer, k time sequence data are simultaneously sent into each batch of training, and the data are m x 1 dimension tensors. And replacing the data by a three-dimensional tensor from a one-dimensional tensor through the long and short convolutional memory neural network layer. And it is inputting xtTo state ht-1And state ht-1To state ct-1The convolution operation is integrated into the conversion of (1). The operation can add time sequences into the vector features, so that the time relation among the features is enhanced, and the fitting precision of the model is improved.
Example two
Based on the same inventive concept as the method for predicting the state of the fan gearbox in the previous embodiment, the invention further provides a system for predicting the state of the fan gearbox, and please refer to fig. 4, wherein the system comprises:
the system comprises a first obtaining unit 11, wherein the first obtaining unit 11 is used for obtaining an original data set of multiple groups of operating parameters of the wind turbine generator from a wind power plant monitoring and data acquisition system;
the first execution unit 12 is configured to pre-process the original data set by using a preset algorithm, so as to reduce the feature vector dimension of the original data set to obtain dimension reduction data;
the second execution unit 13 is configured to perform convolution operation on the dimension reduction data by using a convolution long and short memory neural network model, so as to complete reconstruction of a feature vector of a data set;
a third execution unit 14, where the third execution unit 14 is configured to perform deep neural network training on the reconstructed data to establish a plurality of target prediction models;
and the fourth execution unit 15 is configured to determine an optimal model from the multiple target prediction models as a prediction network model, and predict the state of the fan gearbox by using the prediction network model.
Further, the system further comprises:
a second obtaining unit, configured to calculate a correlation between two variables using a calculation formula of a spearman correlation coefficient method to obtain a variable correlation coefficient;
the first judging unit is used for judging whether the variable correlation coefficient exceeds a preset threshold value or not;
and the fifth execution unit is used for discarding the feature vector corresponding to the variable correlation coefficient when the variable correlation coefficient exceeds the threshold value, and reducing the dimension of the feature vector.
Further, the system further comprises:
and the sixth execution unit is used for not carrying out feature vector abandoning operation on the gearbox oil temperature, the gearbox main bearing temperature, the front bearing temperature, the rear bearing temperature and the gearbox oil level.
Further, the system further comprises:
a seventh execution unit, configured to execute a model algorithm of the convolutional long-short memory neural network, where the model algorithm includes:
wherein, i (t) is an input gate, f (t) is a forgetting gate, o (t) is an output gate, c (t) is a memory unit state, h (t) is a memory unit output, sigma is a sigmoid activation function, and a is convolution operation.
Further, the system further comprises:
an eighth execution unit, configured to perform segment mapping on an output value h (t) of the convolutional long and short memory neural network model after the sigmoid activation function is located at an output gate of the convolutional long and short memory neural network;
the segmentation mapping relation is as follows:
where σ is set to 0.02.
Further, the system further comprises:
a ninth execution unit, configured to construct a convolutional long-short memory neural network model and perform fitting training on the prediction target respectively;
a tenth execution unit, configured to determine a forward transfer path of the data, calculate a loss function loss value, update weights and offsets of network layers, calculate a loss function loss value and a reverse transfer error, and update the weights;
an eleventh execution unit, configured to, by training a preset batch of the long-short convolutional memory neural network model, select an evaluation index R for the multiple target prediction models when a curve trend corresponding to the loss function loss value meets an expected requirement2Calculating a value;
a twelfth execution unit to execute the evaluation index R2Value, determining the prediction network model, which is an evaluation index R2The model whose value differs the least from 1.
Further, the system further comprises:
a thirteenth execution unit to:calculating to obtain the evaluation index R2Values where SST is the total square sum and SSR is the regression square sum.
Further, the system further comprises:
a fourteenth execution unit, wherein a framework for the target prediction model is formed by three neural networks, including: the long and short convolutional neural network training system comprises a long and short convolutional memory neural network layer, a convolutional neural network and a full connection layer, wherein k time sequence data are simultaneously sent in each batch of deep neural network training, the data are replaced by a three-dimensional tensor from a one-dimensional tensor through the long and short convolutional memory neural network layer, and k is a positive integer.
In this specification, the embodiments are described in a progressive manner, and each embodiment focuses on differences from other embodiments, and the fan gearbox state prediction method and the specific example in the first embodiment of fig. 1 are also applicable to a fan gearbox state prediction system in this embodiment, and through the foregoing detailed description of a fan gearbox state prediction method, a fan gearbox state prediction system in this embodiment is clearly known to those skilled in the art, so for the sake of brevity of the description, detailed description is not repeated here. The device disclosed by the embodiment corresponds to the method disclosed by the embodiment, so that the description is simple, and the relevant points can be referred to the method part for description.
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the application. Thus, the present application is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Exemplary electronic device
The computer apparatus of the present application is described below with reference to fig. 5. The computer device may be an application version management server or a terminal, and its internal structure diagram may be as shown in fig. 5. The computer device includes a processor, a memory, and a network interface connected by a system bus. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device comprises a nonvolatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of an operating system and computer programs in the non-volatile storage medium. The network interface of the computer device is used for communicating with an external terminal through a network connection. The computer program is executed by a processor to implement a method of processing an application package.
When the computer device is a terminal, the computer device may further include a display screen and an input device. The display screen of the computer equipment can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer equipment can be a touch layer covered on the display screen, a key, a track ball or a touch pad arranged on the shell of the computer equipment, an external keyboard, a touch pad or a mouse and the like.
Those skilled in the art will appreciate that the architecture shown in fig. 5 is merely a block diagram of some of the structures associated with the disclosed aspects and is not intended to limit the computing devices to which the disclosed aspects apply, as particular computing devices may include more or less components than those shown, or may combine certain components, or have a different arrangement of components.
In one embodiment, a computer device is further provided, which includes a memory and a processor, the memory stores a computer program, and the processor implements the steps of the above method embodiments when executing the computer program.
In an embodiment, a computer-readable storage medium is provided, in which a computer program is stored which, when being executed by a processor, carries out the steps of the above-mentioned method embodiments.
In one embodiment, a computer program product or computer program is provided that includes computer instructions stored in a computer-readable storage medium. The computer instructions are read by a processor of a computer device from a computer-readable storage medium, and the computer instructions are executed by the processor to cause the computer device to perform the steps in the above-mentioned method embodiments.
The application provides a fan gearbox state prediction method, wherein the method is applied to a fan gearbox state prediction system, and the method comprises the following steps: acquiring an original data set of a plurality of groups of operating parameters of the wind turbine generator from a wind power plant monitoring and data acquisition system; preprocessing the original data set by using a preset algorithm to reduce the feature vector dimension of the original data set to obtain dimension-reduced data; carrying out convolution operation on the dimensionality reduction data by utilizing a convolution long and short memory neural network model so as to complete reconstruction on a characteristic vector of a data set; establishing a plurality of target prediction models by carrying out deep neural network training on the reconstructed data; and determining an optimal model from the target prediction models as a prediction network model, and predicting the state of the fan gearbox by using the prediction network model. The technical problem that a multi-period continuous time prediction cannot be carried out on a state prediction model of the fan gearbox in the prior art is solved. The method has the advantages of being suitable for the wind turbine generator and having the data prediction task with the time sequence characteristic type, high in accuracy, time-consuming, controllable and efficient, further accurately predicting the future health state of the wind turbine generator, and guaranteeing the safe operation of the wind turbine generator.
It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by hardware instructions of a computer program, which can be stored in a non-volatile computer-readable storage medium, and when executed, can include the processes of the embodiments of the methods described above. Any reference to memory, storage, database or other medium used in the embodiments provided herein can include at least one of non-volatile and volatile memory. Non-volatile Memory may include Read-Only Memory (ROM), magnetic tape, floppy disk, flash Memory, optical storage, or the like. Volatile Memory can include Random Access Memory (RAM) or external cache Memory. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM), among others.
The technical features of the above embodiments can be arbitrarily combined, and for the sake of brevity, all possible combinations of the technical features in the above embodiments are not described, but should be considered as the scope of the present specification as long as there is no contradiction between the combinations of the technical features.
The above-mentioned embodiments only express several embodiments of the present application, and the description thereof is more specific and detailed, but not construed as limiting the scope of the invention. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the concept of the present application, which falls within the scope of protection of the present application. Therefore, the protection scope of the present patent application should be subject to the appended claims.
Claims (10)
1. A method for predicting the state of a gearbox of a fan, the method comprising:
acquiring an original data set of a plurality of groups of operating parameters of the wind turbine generator from a wind power plant monitoring and data acquisition system;
preprocessing the original data set by using a preset algorithm to reduce the feature vector dimension of the original data set to obtain dimension-reduced data;
carrying out convolution operation on the dimensionality reduction data by utilizing a convolution long and short memory neural network model so as to complete reconstruction on a characteristic vector of a data set;
establishing a plurality of target prediction models by carrying out deep neural network training on the reconstructed data;
and determining an optimal model from the target prediction models as a prediction network model, and predicting the state of the fan gearbox by using the prediction network model.
2. The method of claim 1, wherein the predetermined algorithm is a spearman correlation coefficient method, and the pre-processing the original data set using the predetermined algorithm to reduce the feature vector dimension of the original data set to obtain the dimension-reduced data comprises:
calculating the correlation between two variables by adopting a calculation formula of a spearman correlation coefficient method to obtain a variable correlation coefficient, wherein the calculation formula is as follows:
wherein rho is a Spierman correlation coefficient, a and b are two random variables, and the ith values obtained by the two random variables are respectively expressed as ai,bi,Is the average of all random variables a, b;
judging whether the variable correlation coefficient exceeds a preset threshold value or not;
when the variable correlation coefficient exceeds the threshold value, the feature vector corresponding to the variable correlation coefficient is discarded, and the feature vector dimension is reduced;
each input feature vector of the data set subjected to dimensionality reduction through the spearman correlation coefficient method is a continuous 12-hour power distribution unit state quantity, and a target prediction value is a 10-day-later wind power gear box state.
3. The method of claim 2, wherein the raw data set comprises a gearbox oil temperature, a gearbox main bearing temperature, a front bearing temperature, a rear bearing temperature, and a gearbox oil level, and wherein the gearbox oil temperature, gearbox main bearing temperature, front bearing temperature, rear bearing temperature, and gearbox oil level are not subjected to the feature vector rejection operation.
4. The method of claim 1, wherein the convolving the reduced dimensional data with the long and short convolutional memory neural network model to perform reconstruction on a feature vector of a data set comprises:
the model algorithm of the convolution long and short memory neural network is as follows:
wherein, i (t) is an input gate, f (t) is a forgetting gate, o (t) is an output gate, c (t) is a memory unit state, h (t) is a memory unit output, sigma is a sigmoid activation function, and a is convolution operation.
5. The method of claim 4, wherein the sigmoid activation function is located after an output gate of the convolutional long and short memory neural network, and performs segment mapping on an output value h (t) of the convolutional long and short memory neural network model;
the segmentation mapping relation is as follows:
where σ is set to 0.02.
6. The method of claim 1, wherein determining the optimal model from the plurality of target predictive models as a predictive network model comprises:
constructing a convolution long and short memory neural network model and respectively carrying out fitting training on the prediction target;
determining a forward data transfer path, calculating loss function loss values, updating weights and offsets of all network layers, calculating loss function loss values and reverse transfer errors, and updating the weights;
the convolution long-short memory neural network model is preset in batches through training, and when curve trends corresponding to loss function loss values meet expected requirements, evaluation indexes R are selected for the multiple target prediction models2Calculating a value;
according to the evaluation index R2Value, determining the prediction network model, which is an evaluation index R2The model whose value differs the least from 1.
8. The method of claim 1, wherein the framework of the object prediction model is formed by three neural networks, including: the long and short convolutional neural network training system comprises a long and short convolutional memory neural network layer, a convolutional neural network and a full connection layer, wherein k time sequence data are simultaneously sent in each batch of deep neural network training, the data are replaced by a three-dimensional tensor from a one-dimensional tensor through the long and short convolutional memory neural network layer, and k is a positive integer.
9. A fan gearbox condition prediction system, the system comprising:
the system comprises a first obtaining unit, a second obtaining unit and a control unit, wherein the first obtaining unit is used for obtaining an original data set of multiple groups of operating parameters of the wind turbine generator from a wind power plant monitoring and data acquisition system;
the first execution unit is used for preprocessing the original data set by using a preset algorithm and reducing the feature vector dimension of the original data set to obtain dimension-reduced data;
the second execution unit is used for carrying out convolution operation on the dimensionality reduction data by utilizing a convolution long and short memory neural network model so as to complete reconstruction on a data set feature vector;
a third execution unit, configured to perform deep neural network training on the reconstructed data to establish a plurality of target prediction models;
and the fourth execution unit is used for determining an optimal model from the target prediction models as a prediction network model and predicting the state of the fan gearbox by using the prediction network model.
10. A wind turbine gearbox condition prediction device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor when executing the program implements the steps of the method of any of claims 1 to 8.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111533015.0A CN114239396A (en) | 2021-12-15 | 2021-12-15 | Fan gear box state prediction method and system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111533015.0A CN114239396A (en) | 2021-12-15 | 2021-12-15 | Fan gear box state prediction method and system |
Publications (1)
Publication Number | Publication Date |
---|---|
CN114239396A true CN114239396A (en) | 2022-03-25 |
Family
ID=80756202
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202111533015.0A Pending CN114239396A (en) | 2021-12-15 | 2021-12-15 | Fan gear box state prediction method and system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114239396A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115270993A (en) * | 2022-08-23 | 2022-11-01 | 南通思诺船舶科技有限公司 | Diesel engine unit state detection method and system |
CN115496264A (en) * | 2022-08-26 | 2022-12-20 | 河北大学 | Method for predicting generated power of wind turbine generator |
-
2021
- 2021-12-15 CN CN202111533015.0A patent/CN114239396A/en active Pending
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115270993A (en) * | 2022-08-23 | 2022-11-01 | 南通思诺船舶科技有限公司 | Diesel engine unit state detection method and system |
CN115496264A (en) * | 2022-08-26 | 2022-12-20 | 河北大学 | Method for predicting generated power of wind turbine generator |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Qin et al. | Macroscopic–microscopic attention in LSTM networks based on fusion features for gear remaining life prediction | |
CN110414788B (en) | Electric energy quality prediction method based on similar days and improved LSTM | |
CN110533112B (en) | Internet of vehicles big data cross-domain analysis and fusion method | |
CN114239396A (en) | Fan gear box state prediction method and system | |
Ayodeji et al. | Causal augmented ConvNet: A temporal memory dilated convolution model for long-sequence time series prediction | |
CN114399032B (en) | Method and system for predicting metering error of electric energy meter | |
Huang et al. | A hybrid model for short-term dissolved oxygen content prediction | |
CN116624343B (en) | Wind turbine generator tower abnormal vibration monitoring and health degree evaluation method and system | |
CN108399434B (en) | Analysis and prediction method of high-dimensional time series data based on feature extraction | |
CN111784061B (en) | Training method, device and equipment for power grid engineering cost prediction model | |
CN111461463A (en) | Short-term load prediction method, system and equipment based on TCN-BP | |
CN112365033B (en) | Wind power interval prediction method, system and storage medium | |
CN111931983A (en) | Precipitation prediction method and system | |
CN111525587A (en) | Reactive load situation-based power grid reactive voltage control method and system | |
CN116384244A (en) | Electromagnetic field prediction method based on physical enhancement neural network | |
CN114676645B (en) | Non-stationary time sequence prediction method and system | |
CN115544890A (en) | Short-term power load prediction method and system | |
Wang et al. | Spatiotemporal non-negative projected convolutional network with bidirectional NMF and 3DCNN for remaining useful life estimation of bearings | |
CN115879369A (en) | Coal mill fault early warning method based on optimized LightGBM algorithm | |
Li et al. | Improved LSTM-based prediction method for highly variable workload and resources in clouds | |
Jiang et al. | A short-term wind power prediction approach based on ensemble empirical mode decomposition and improved long short-term memory | |
Sun et al. | Short-term power load prediction based on VMD-SG-LSTM | |
CN112488399A (en) | Power load prediction method and device | |
Che et al. | A modified support vector regression: Integrated selection of training subset and model | |
CN114254904B (en) | Method and device for evaluating operation health degree of engine room of wind turbine generator |
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 |