CN115860247A

CN115860247A - Method and device for training fan loss power prediction model in extreme weather

Info

Publication number: CN115860247A
Application number: CN202211638366.2A
Authority: CN
Inventors: 卓毅鑫; 蒙文川; 唐健; 胡甲秋; 杨再敏; 饶志
Original assignee: Guangxi Power Grid Co Ltd; Energy Development Research Institute of China Southern Power Grid Co Ltd
Current assignee: Guangxi Power Grid Co Ltd; Energy Development Research Institute of China Southern Power Grid Co Ltd
Priority date: 2022-12-19
Filing date: 2022-12-19
Publication date: 2023-03-28

Abstract

The application relates to a method and a device for training a prediction model of the power loss of a fan in extreme weather. The method comprises the following steps: acquiring geographic position information and meteorological information corresponding to at least one wind turbine generator in a wind power station; extracting the characteristics of each piece of geographic position information and each piece of meteorological information to obtain a wind turbine generator generation characteristic set corresponding to each wind turbine generator; inputting the generating characteristic set of each wind turbine generator into a loss power prediction model to be trained to obtain a wind turbine generator loss power prediction value corresponding to each wind turbine generator; and training the loss power prediction model to be trained according to the operating characteristics of each wind turbine and the loss power prediction value of each wind turbine to obtain the trained loss power prediction model. By adopting the method, the prediction accuracy of the fan loss power prediction model of the wind power generation unit on the power loss value under the extreme weather condition can be improved.

Description

Method and device for training fan loss power prediction model in extreme weather

Technical Field

The present application relates to the field of computer technologies, and in particular, to a method and an apparatus for training a model for predicting a power loss of a wind turbine in extreme weather, a computer device, a storage medium, and a computer program product.

Background

With the development of computer technology, a wind turbine power prediction technology appears, and with the accelerated construction of a novel power system with new energy as a main body, a wind turbine is very susceptible to the influence of extreme climates such as systematic precipitation, cold and tide ice disasters, typhoons and the like, particularly under the extreme climatic conditions such as ice coating, typhoons and the like, the wind turbine photovoltaic volatility is further amplified, the output power of the wind turbine is usually seriously deviated from a predicted value, and the wind turbine power prediction technology, power supply guarantee and safe and stable operation of a power grid are more challenged.

In the traditional technology, the influence of the wind turbine generator under extreme weather is mainly considered to include that typhoon causes overspeed offline and ice coating causes ice coating shutdown, and the fault model of the wind turbine generator under extreme weather is obtained by establishing a typhoon path and influence model and finite element analysis models of extreme cold weather and ice coating influence. The traditional modeling method mainly considers modeling of the whole extreme weather, but because the extreme weather has certain contingency and paroxysmal property, the model established by the traditional method is difficult to adapt to the extreme weather of the long-time scale wind turbine generator, and the accuracy rate of predicting the fan loss power of the extreme weather is low.

Disclosure of Invention

In view of the foregoing, it is desirable to provide a method, an apparatus, a computer device, a computer readable storage medium, and a computer program product for training a wind turbine loss power prediction model in extreme weather, which can improve the wind turbine loss power prediction accuracy in extreme weather.

In a first aspect, the application provides a method for training a wind turbine loss power prediction model in extreme weather. The method comprises the following steps: acquiring geographic position information and meteorological information corresponding to at least one wind turbine generator in a wind power station; performing feature extraction on each geographic position information and each meteorological information to obtain a wind turbine generator set power generation feature set corresponding to each wind turbine generator set; inputting the power generation characteristic set of each wind turbine generator into a loss power prediction model to be trained to obtain a wind turbine generator loss power prediction value corresponding to each wind turbine generator; the loss power prediction model to be trained sequentially comprises a wind turbine generator fault type judging sub-model, a wind turbine generator fault clustering sub-model and a wind turbine generator power loss value calculating sub-model; and training the loss power prediction model to be trained according to the operating characteristics of each wind turbine generator and the loss power prediction value of each wind turbine generator to obtain the trained loss power prediction model.

In a second aspect, the application further provides a device for training the fan loss power prediction model in extreme weather. The device comprises: the information acquisition module is used for acquiring the geographic position information and the meteorological information corresponding to at least one wind turbine generator in the wind power station; the characteristic extraction module is used for carrying out characteristic extraction on the geographic position information and the meteorological information to obtain a wind turbine generator set power generation characteristic set corresponding to each wind turbine generator set; the loss power prediction module is used for inputting the power generation feature set of each wind turbine generator into a loss power prediction model to be trained to obtain a wind turbine generator loss power prediction value corresponding to each wind turbine generator; the loss power prediction model to be trained sequentially comprises a wind turbine generator fault type judging sub-model, a wind turbine generator fault clustering sub-model and a wind turbine generator power loss value calculating sub-model; and the model training module is used for training the loss power prediction model to be trained according to the operating characteristics of each wind turbine generator and the loss power prediction value of each wind turbine generator to obtain the trained loss power prediction model.

In a third aspect, the present application also provides a computer device. The computer device comprises a memory storing a computer program and a processor implementing the following steps when executing the computer program: acquiring geographic position information and meteorological information corresponding to at least one wind turbine generator in a wind power station; performing feature extraction on each geographic position information and each meteorological information to obtain a wind turbine generator set power generation feature set corresponding to each wind turbine generator set; inputting the power generation feature set of each wind turbine generator into a loss power prediction model to be trained to obtain a wind turbine generator loss power prediction value corresponding to each wind turbine generator; the loss power prediction model to be trained sequentially comprises a wind turbine generator fault type judging sub-model, a wind turbine generator fault clustering sub-model and a wind turbine generator power loss value calculating sub-model; and training the loss power prediction model to be trained according to the operating characteristics of each wind turbine generator and the loss power prediction value of each wind turbine generator to obtain the trained loss power prediction model.

In a fourth aspect, the present application further provides a computer-readable storage medium. The computer-readable storage medium having stored thereon a computer program which, when executed by a processor, performs the steps of: acquiring geographic position information and meteorological information corresponding to at least one wind turbine generator in a wind power station; performing feature extraction on each geographic position information and each meteorological information to obtain a wind turbine generator set power generation feature set corresponding to each wind turbine generator set; inputting the power generation characteristic set of each wind turbine generator into a loss power prediction model to be trained to obtain a wind turbine generator loss power prediction value corresponding to each wind turbine generator; the loss power prediction model to be trained sequentially comprises a wind turbine generator fault type judging sub-model, a wind turbine generator fault clustering sub-model and a wind turbine generator power loss value calculating sub-model; and training the loss power prediction model to be trained according to the operating characteristics of each wind turbine generator and the loss power prediction value of each wind turbine generator to obtain the trained loss power prediction model.

In a fifth aspect, the present application further provides a computer program product. The computer program product comprising a computer program which when executed by a processor performs the steps of: acquiring geographic position information and meteorological information corresponding to at least one wind turbine generator in a wind power station; performing feature extraction on each geographical position information and each meteorological information to obtain a wind turbine generator set power generation feature set corresponding to each wind turbine generator set; inputting the power generation feature set of each wind turbine generator into a loss power prediction model to be trained to obtain a wind turbine generator loss power prediction value corresponding to each wind turbine generator; the loss power prediction model to be trained sequentially comprises a wind turbine generator fault type judging sub-model, a wind turbine generator fault clustering sub-model and a wind turbine generator power loss value calculating sub-model; and training the loss power prediction model to be trained according to the operating characteristics of each wind turbine generator and the loss power prediction value of each wind turbine generator to obtain the trained loss power prediction model.

According to the method, the device, the computer equipment, the storage medium and the computer program product for training the wind turbine loss power prediction model under extreme weather, the geographical position information and the meteorological information corresponding to at least one wind turbine generator in the wind power plant station are obtained; extracting the characteristics of each piece of geographic position information and each piece of meteorological information to obtain a wind turbine generator generation characteristic set corresponding to each wind turbine generator; inputting the generating characteristic set of each wind turbine generator into a loss power prediction model to be trained to obtain a wind turbine generator loss power prediction value corresponding to each wind turbine generator; the loss power prediction model to be trained sequentially comprises a wind turbine generator fault type judging sub-model, a wind turbine generator fault clustering sub-model and a wind turbine generator power loss value calculating sub-model; and training the loss power prediction model to be trained according to the operating characteristics of each wind turbine and the loss power prediction value of each wind turbine to obtain the trained loss power prediction model.

The method has the advantages that the power loss value of the wind turbine generator under extreme weather is predicted through the wind turbine generator fault type judging submodel, the wind turbine generator fault clustering submodel and the wind turbine generator power loss value calculating operator model in the fan loss power prediction model, the time sequence prediction can be carried out on the output of the wind turbine generator under a long time scale, the mass data and the wind turbine generator operation mechanism are used for training the neural network, and the parameter correction is carried out in combination with the operation rule, so that the convergence speed and the prediction precision of the neural network are improved, the output influence of the extreme weather on the wind turbine generator under the long time scale is quantized more accurately, and the prediction precision of the fan loss power prediction model of the wind turbine generator on the power loss value under the extreme weather condition is improved.

Drawings

FIG. 1 is an application environment diagram of a model training method for predicting power loss of a wind turbine in extreme weather according to an embodiment;

FIG. 2 is a schematic flow chart illustrating a method for training a model for predicting power loss of a wind turbine in extreme weather according to an embodiment;

FIG. 3 is a schematic flow chart of a method for obtaining a predicted value of power loss of a wind turbine generator in one embodiment;

FIG. 4 is a schematic flow chart of a method for obtaining a predicted value of loss power of a wind turbine generator in another embodiment;

FIG. 5 is a schematic flow chart of a method for obtaining wind turbine generator system fault type discrimination information in one embodiment;

FIG. 6 is a schematic flow chart illustrating a method for obtaining a set of power generation characteristics of a wind turbine generator in one embodiment;

FIG. 7 is a flow diagram illustrating a method for obtaining a trained lost power prediction model in one embodiment;

FIG. 8 is a schematic flow chart illustrating a method for predicting power loss of a wind turbine in extreme weather according to an embodiment;

FIG. 9 is a schematic diagram illustrating logic implemented in the method for training a model for predicting power loss of a wind turbine in extreme weather according to an embodiment;

FIG. 10 is a schematic flow chart diagram illustrating a lost power prediction model training method in accordance with one embodiment;

FIG. 11 is a block diagram illustrating an exemplary embodiment of a model training apparatus for predicting a power loss of a wind turbine in extreme weather;

FIG. 12 is a diagram of an internal structure of a computer device in one embodiment.

Detailed Description

In order to make the objects, technical solutions and advantages of the present application more clearly understood, the present application is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the present application.

The method for training the fan loss power prediction model under extreme weather can be applied to the application environment shown in fig. 1. Wherein the terminal 102 communicates with the server 104 via a network. The data storage system may store data that the server 104 needs to process. The data storage system may be integrated on the server 104, or may be located on the cloud or other network server. The server 104 acquires geographic position information and meteorological information corresponding to at least one wind turbine generator in the wind power plant from the terminal 102; extracting the characteristics of each piece of geographic position information and each piece of meteorological information to obtain a wind turbine generator generation characteristic set corresponding to each wind turbine generator; inputting the generating characteristic set of each wind turbine generator into a loss power prediction model to be trained to obtain a wind turbine generator loss power prediction value corresponding to each wind turbine generator; the method comprises the steps that a loss power prediction model to be trained sequentially comprises a wind turbine generator fault type distinguishing sub-model, a wind turbine generator fault clustering sub-model and a wind turbine generator power loss value calculating sub-model; and training the loss power prediction model to be trained according to the operating characteristics of each wind turbine and the loss power prediction value of each wind turbine to obtain the trained loss power prediction model. The terminal 102 may be, but not limited to, various personal computers, notebook computers, smart phones, tablet computers, internet of things devices and portable wearable devices, and the internet of things devices may be smart speakers, smart televisions, smart air conditioners, smart car-mounted devices, and the like. The portable wearable device can be a smart watch, a smart bracelet, a head-mounted device, and the like. The server 104 may be implemented as a stand-alone server or as a server cluster comprised of multiple servers.

In one embodiment, as shown in fig. 2, a method for training a prediction model of power loss of a wind turbine in extreme weather is provided, which is described by taking the method as an example applied to the server in fig. 1, and includes the following steps:

step 202, obtaining geographic position information and meteorological information corresponding to at least one wind turbine generator in the wind power station.

The geographic information position may be formed by latitude and longitude coordinates of the earth, that is, longitude and latitude, which represent the position of the target object on the earth.

Specifically, the server responds to an instruction of the terminal, acquires geographic position information and meteorological information corresponding to at least one wind turbine from the terminal, stores the acquired geographic position information and meteorological information in the storage unit, and calls volatile storage resources from the storage unit for the central processing unit to calculate when the server needs to process any data record in the geographic position information and the meteorological information. Any data record can be a single data input to the central processing unit, or a plurality of data can be simultaneously input to the central processing unit.

Acquiring geographical position information of the wind turbine generator set by using a station wind measuring tower and a GIS system, wherein the geographical position information comprises geographical coordinates and an altitude; and meteorological information including wind direction, wind speed, temperature, humidity, precipitation and air pressure within a height of 10-150 meters. Then, a Monte Carlo statistical simulation method is utilized, random variables are set on the basis of a large amount of original meteorological data, similarity expansion is carried out on partial missing meteorological data in extreme weather, and a random disturbance noise generation characteristic data set is set.

And 204, extracting the characteristics of each piece of geographic position information and each piece of meteorological information to obtain a wind turbine generation characteristic set corresponding to each wind turbine.

The wind turbine generator generation feature set can be a data set which embodies the geographic position information and meteorological information features of the wind turbine generator, wherein the feature set can be a feature vector set or a matrix.

Specifically, after extracting the features of the geographic position information and the meteorological information, calculating a Pearson correlation coefficient of the feature information of different geographic positions of the wind turbine generator and the meteorological feature information of different heights by using a Pearson correlation analysis method, defining that the Pearson correlation coefficient between data is greater than 0.8 as strong correlation data, and extracting the strong correlation data as a power generation feature set of the wind turbine generator;

the pearson correlation coefficient is equal to the product of the covariance divided by the respective standard deviation, and the specific calculation formula is as follows:

wherein x is _i A specific value of a certain data in the meteorological characteristic data,

is the average value of the characteristic data; y is _i For a particular value of another data in the meteorological feature data, is/are>

Is the average of the characteristic data.

And step 206, inputting the generating characteristic set of each wind turbine generator into a loss power prediction model to be trained to obtain a wind turbine generator loss power prediction value corresponding to each wind turbine generator.

The loss power prediction model to be trained may be an artificial intelligence model for calculating the feature data to obtain a prediction result.

The predicted value of the loss power of the wind turbine generator can be a power loss value of the wind turbine generator caused by extreme weather.

Specifically, the first step: the method comprises the steps of taking a power generation characteristic set of each wind turbine generator with a time sequence as an input quantity of a wind turbine generator fault type distinguishing sub-model, selecting typical fault types of the wind turbine generator, such as normal operation, overspeed offline, icing shutdown and the like, as an output quantity of the wind turbine generator fault type distinguishing sub-model, training the RNN-based wind turbine generator fault type distinguishing sub-model, respectively obtaining probability predicted values corresponding to fault types of the wind turbine generators in extreme weather, and finally outputting wind turbine generator fault type distinguishing information corresponding to the wind turbine generators. The specific calculation process of the wind turbine generator fault type discrimination submodel can be expressed as follows:

n _t ＝φ(UI _t +Wn _t-1 )

O _t ＝f(Vn _t )

wherein, U, V and W are network weight matrixes, and training adjustment is carried out through back propagation; I.C. A _t For input at time t, n _t-1 、n _t Hidden layer states at time t-1 and t, respectively, O _t For the output at time t, φ and f are activation functions.

The specific implementation process of the wind turbine generator fault type distinguishing submodel is that the generation characteristic set of each wind turbine generator with time sequence is used as the wind turbine generator fault type distinguishing submodel to be input into I _t ＝{I ₁ ,I ₂ ,...I _k And outputting O = { O } by using the fault state of the wind turbine as a fault type judgment sub-model of the wind turbine ₁ ,O ₂ ,...O _k Denotes hidden layer as n = { n = } ₁ ,n ₂ ,...n _k }. At time i, n _k-1 And current I _k As input, a calculation result O is obtained _k And the output is taken as the output and is transmitted to the time k +1, so that the next layer can always obtain the output weight of the previous layer, and the sub-model for judging the fault type of the wind turbine generator has the memory capacity for time sequence data.

The second step is that: and inputting the wind turbine generator fault type distinguishing information corresponding to each wind turbine generator of the wind turbine generator fault type distinguishing submodel into a wind turbine generator fault clustering submodel consisting of a plurality of layers of SVM (support vector machines) for clustering. Firstly, normalization processing is carried out on the wind turbine generator fault type judging information, for example: 7000 groups of wind turbine generator fault types are selected as a training set, and 3000 groups of wind turbine generator fault types are selected as a testing set; then, selecting RBF as a kernel function; searching for optimal parameters C and gamma by using a cross verification method, and training to obtain a wind turbine generator fault clustering sub-model consisting of at least 3-layer SVM classifiers; finally, a hyperplane obtained by using the wind turbine generator fault clustering submodel is substituted into a sample to be identified, and feature clustering is carried out on the wind turbine generator fault type distinguishing information.

The third step: using the wind turbine generator generation characteristic set obtained after characteristic extraction and the wind turbine generator fault type clustering information obtained by characteristic clustering as the operator model input quantity of the wind turbine generator power loss value, constructing a time sequence sample and a time sequence mark, wherein the time sequence sample and the time sequence mark are respectively (R) ₁ ,R ₂ ,...R _m ) And (y) ₁ ,y ₂ ,...y _m ) Wherein R is ₁ Is (r) ₁ ,r ₂ ,...r _k-1 ) Vector of composition, y ₁ Is r of _k ；R ₂ Is (r) ₂ ,r ₃ ,...r _k ) Vector of composition, y ₂ Is r _k+1 (ii) a By analogy, R _m Is (r) _m ,r _m+1 ,...r _m+k-1 )，y _m Is r of _m+k . Then, normalization processing is performed on the time series by using a normalization method of R' = (R-R) _min )/(r _max -r _min ). And finally, constructing an operator model of the power loss value of the motor generator based on the LSTM network model, training by using a time sequence back propagation algorithm (BPTT) according to the normalized time sequence and the time sequence marks, setting a root mean square error as a loss function, setting Sigmoid as an activation function, storing the operator model of the power loss value of the motor generator when the model loss value tends to be stable, and calculating the operator model of the power loss value of the wind generator to obtain the predicted value of the power loss of the wind generator corresponding to each wind generator.

The specific calculation steps of the electric machine set power loss value operator model are as follows:

when the wind turbine generator generation feature set and the wind turbine generator fault type clustering information are input into the electric machine power loss value calculation operator model, firstly, the cell state is filtered through a forgetting gate in the electric machine power loss value calculation operator model. The forgetting door is oneA gating mechanism activated by Sigmoid function, the gate being input X at the current time step t _t Hidden layer output h of sum time step t-1 _t-1 As a common input, the forgetting gate function is expressed as follows:

f _t ＝σ(U _f X _t +W _f h _t-1 +b _f )

wherein f is _t The current time step t of the forgetting gate is output, sigma is a Sigmoid function, X _t For the input of the current time step t, h _t-1 Hidden layer output, U, for time step t-1 _f 、W _f A weight matrix for passing the forgetting gate at the present moment, b _f A biased term for the current time passing the forgetting gate.

The input gate function in the electric machine set power loss value operator model is to update the cell state according to the current input, and the function expression is as follows:

i _t ＝σ(U _i X _t +W _i h _t-1 +b _i )

g _t ＝tanh(U _g X _t +W _g h _t-1 +b _g )

wherein the input gate is also denoted by X _t And h _t-1 As input, mapping the combination value between 0 and 1 by a Sigmoid activation function to obtain an input gate update weight i _t (ii) a At the same time, the input gate couples the input value X through the Tanh activation function _t And hidden state h of time step t-1 _t-1 Processing to obtain a cell state candidate vector g at the current time step _t Wherein, U _i 、W _i 、U _g 、W _g Respectively, the weight matrix of the current time passing through the input gate, b _i 、b _g Respectively, the offset terms that pass through the input gate at the current time.

The outputs of the forgetting gate and the input gate together constitute the cell state c of the current time step _t The specific calculation formula is as follows:

c _t ＝f _t c _t-1 +i _t g _t

the output gate output coefficient and the output signal calculation formula are as follows:

o _t ＝σ(U _o X _t +W _o h _t-1 +b _o )

h _t ＝o _t ×tanh(c _t )

wherein o is _t For the current time step output coefficient, U _o 、W _o Respectively, a weight matrix passing through the output gate at the current moment, b _o Offset term for the current time passing through the output gate, h _t And outputting a signal for the current time step.

And 208, training the loss power prediction model to be trained according to the operating characteristics of each wind turbine and the loss power prediction value of each wind turbine to obtain the trained loss power prediction model.

The trained loss power prediction model can be an artificial intelligence model which is trained and can meet business requirements.

Specifically, a wind turbine generator loss power prediction value of a loss power prediction model to be trained is subjected to a post-correction link by utilizing a wind turbine generator energy conversion characteristic, a wind turbine generator flicker characteristic and a wind turbine generator power control characteristic, a prediction result which does not accord with an operation rule is limited, a constraint result is fed back to a weight parameter of the loss power prediction model to be trained, and the step of returning to execute' obtaining geographic position information of the wind turbine generator by utilizing a station wind measuring tower and a GIS system, wherein the geographic position information comprises a geographic coordinate and an altitude; and meteorological information including wind direction, wind speed, temperature, humidity, precipitation and air pressure within a height of 10-150 meters. And then, setting random variables on the basis of a large amount of original meteorological data by using a Monte Carlo statistical simulation method, performing similarity expansion on partial missing meteorological data in extreme weather, and setting random disturbance noise to generate a characteristic data set until the output quantity of the loss power prediction model to be trained meets the service requirement to obtain the trained loss power prediction model. A schematic diagram of implementation logic of the model training method for predicting the power loss of the wind turbine in extreme weather is shown in fig. 9; a flow diagram of the method of training the predictive model for power loss is shown in fig. 10.

In the method for training the model for predicting the power loss of the wind turbine in extreme weather, the geographical position information and the meteorological information corresponding to at least one wind turbine in the wind power station are obtained; extracting the characteristics of each piece of geographic position information and each piece of meteorological information to obtain a wind turbine generator generation characteristic set corresponding to each wind turbine generator; inputting the generating characteristic set of each wind turbine generator into a loss power prediction model to be trained to obtain a wind turbine generator loss power prediction value corresponding to each wind turbine generator; the loss power prediction model to be trained sequentially comprises a wind turbine generator fault type judging sub-model, a wind turbine generator fault clustering sub-model and a wind turbine generator power loss value calculating sub-model; and training the loss power prediction model to be trained according to the operating characteristics of each wind turbine and the loss power prediction value of each wind turbine to obtain the trained loss power prediction model.

The method has the advantages that the power loss value of the wind turbine generator under extreme weather is predicted through the wind turbine generator fault type judging submodel, the wind turbine generator fault clustering submodel and the wind turbine generator power loss value calculating operator model in the fan loss power prediction model, time sequence prediction can be conducted on the output of the wind turbine generator under a long time scale, the neural network is trained through mass data and the wind turbine generator operation mechanism, parameter correction is conducted through combination of operation rules, and therefore the convergence speed and the prediction precision of the neural network are improved, the output influence quantification of the wind turbine generator under the long time scale is more accurate, and the prediction precision of the power loss value of the fan loss power prediction model of the wind turbine generator under the extreme weather condition is improved.

In one embodiment, as shown in fig. 3, inputting the generation feature set of each wind turbine into the lost power prediction model to be trained to obtain a wind turbine lost power prediction value corresponding to each wind turbine, including:

step 302, inputting the generating feature set of each wind turbine into a wind turbine fault type distinguishing submodel to obtain wind turbine fault type distinguishing information corresponding to each wind turbine.

The wind turbine generator fault type judging submodel can be one of submodels in a loss power prediction model and is composed of a circulating neural network.

The wind turbine generator fault type judging information can be a classification result classified by a wind turbine generator fault type judging sub-model.

Specifically, the generation characteristic set of each wind turbine generator with time sequence is used as the input quantity of a wind turbine generator fault type distinguishing submodel, typical fault types of the wind turbine generators such as normal operation, overspeed offline and icing shutdown are selected as the output quantity of the wind turbine generator fault type distinguishing submodel, the RNN-based wind turbine generator fault type distinguishing submodel is trained, probability predicted values corresponding to fault types of the wind turbine generators in extreme weather are respectively obtained, and finally wind turbine generator fault type distinguishing information corresponding to the wind turbine generators is output. The specific calculation process of the wind turbine generator fault type discrimination submodel can be expressed as follows:

n _t ＝φ(UI _t +Wn _t-1 )

O _t ＝f(Vn _t )

wherein, U, V and W are network weight matrixes, and training adjustment is carried out through back propagation; i is _t For input at time t, n _t-1 、n _t Hidden layer states at time t-1 and t, respectively, O _t For the output at time t, φ and f are activation functions.

The specific implementation process of the wind turbine generator fault type distinguishing submodel is that the generation characteristic set of each wind turbine generator with time sequence is used as the wind turbine generator fault type distinguishing submodel to be input into I _t ＝{I ₁ ,I ₂ ,...I _k And outputting O = { O } by using the fault state of the wind turbine as a fault type judgment sub-model of the wind turbine ₁ ,O ₂ ,...O _k And the hidden layer is marked as n = { n = } ₁ ,n ₂ ,...n _k }. At time i, n _k-1 And current I _k As input, a calculation result O is obtained _k And the output is transmitted to the moment k +1, so that the next layer can always obtain the output weight of the previous layer, and the wind turbine generator fault type judging sub-model has the memory capacity of time sequence data.

And step 304, inputting the wind turbine fault type distinguishing information into the wind turbine fault clustering sub-model to obtain wind turbine fault type clustering information corresponding to each wind turbine.

The wind turbine generator fault clustering submodel can be one of submodels in a loss power prediction model and is composed of a plurality of layers of vector machines.

The wind turbine fault type clustering information can be a result obtained by carrying out feature clustering on the wind turbine fault clustering sub-model.

Specifically, the wind turbine generator fault type distinguishing information corresponding to each wind turbine generator of the wind turbine generator fault type distinguishing submodel is input into a wind turbine generator fault clustering submodel composed of multiple layers of SVM for clustering. Firstly, normalization processing is carried out on the wind turbine generator fault type judging information, for example: selecting 7000 groups of wind turbine generator fault types as a training set, and 3000 groups of wind turbine generator fault types as a test set; then, selecting RBF as a kernel function; searching for optimal parameters C and gamma by using a cross verification method, and training to obtain a wind turbine generator fault clustering sub-model consisting of at least 3-layer SVM classifiers; finally, a hyperplane obtained by the wind turbine fault clustering submodel is substituted into a sample to be identified, and feature clustering is carried out on the wind turbine fault type distinguishing information.

And step 306, inputting the wind turbine generation feature set and the wind turbine fault type clustering information into an operator model of the wind turbine power loss value, and obtaining wind turbine loss power predicted values corresponding to the wind turbines.

The operator model of the power loss value of the wind turbine generator can be one of submodels in the loss power prediction model.

Specifically, the wind turbine generator generation feature set obtained after feature extraction and the wind turbine generator fault type clustering information obtained after feature clustering are used as the operator model input quantity of the wind turbine generator power loss value, time sequence samples and time sequence marks are constructed, and the time sequence samples and the time sequence marks are (P) ₁ ,R ₂ ,...R _m ) And (y) ₁ ,y ₂ ,...y _m ) Wherein R is ₁ Is (r) ₁ ,r ₂ ,...r _k-1 ) Vector of composition, y ₁ Is r _k ；R ₂ Is (r) ₂ ,r ₃ ,...r _k ) Vector of composition, y ₂ Is r _k+1 (ii) a By analogy, R _m Is (r) _m ,r _m+1 ,...r _m+k-1 )，y _m Is r of _m+k . Then, normalization processing is performed on the time series by using a normalization method of R' = (R-R) _min )/(r _max -r _min ). And finally, constructing an operator model of the power loss value of the motor generator based on the LSTM network model, training by using a time sequence back propagation algorithm (BPTT) according to the normalized time sequence and the time sequence marks, setting a root mean square error as a loss function, setting Sigmoid as an activation function, storing the operator model of the power loss value of the motor generator when the model loss value tends to be stable, and calculating the operator model of the power loss value of the wind generator to obtain the predicted value of the power loss of the wind generator corresponding to each wind generator.

when the wind turbine generator generation feature set and the wind turbine generator fault type clustering information are input into the electric machine power loss value calculation operator model, firstly, the cell state is filtered through a forgetting gate in the electric machine power loss value calculation operator model. The forgetting gate is a gate control mechanism activated by a Sigmoid function and inputs X at the current time step t _t Hidden layer output h of sum time step t-1 _t-1 As a common input, the forgetting gate function is expressed as follows:

f _t ＝σ(U _f X _t +W _f h _t-1 +b _f )

wherein f is _t For the current time step t output of the forgetting gate, sigma is Sigmoid function, X _t For the input of the current time step t, h _t-1 Hidden layer output, U, for time step t-1 _f 、W _f A weight matrix for passing the forgetting gate at the present moment, b _f A biased term for the current time passing the forgetting gate.

i _t ＝σ(U _i X _t +W _i h _t-1 +b _i )

g _t ＝tanh(U _g X _t +W _g h _t-1 +b _g )

c _t ＝f _t c _t-1 +i _t g _t

o _t ＝σ(U _o X _t +W _o h _t-1 +b _o )

h _t ＝o _t ×tanh(c _t )

wherein o is _t For the current time step output coefficient, U _o 、W _o Respectively, a weight matrix passing through the output gate at the current moment, b _o Offset term for the current time passing through the output gate, h _t Outputting a signal for the current time step.

In the embodiment, different neural networks are set to adapt to the power generation characteristics and the fault set data types of the wind turbine, so that the scale required by model training is shortened, the power prediction calculation application of the wind turbine is effectively promoted, and the prediction precision of the power loss value of the wind turbine under extreme weather conditions is improved.

In one embodiment, as shown in fig. 4, inputting the wind turbine generation feature set and the wind turbine fault type clustering information to a wind turbine power loss value calculation operator model to obtain a wind turbine loss power prediction value corresponding to each wind turbine, including:

step 402, constructing a time sequence sample and a time sequence mark according to the wind turbine generator generation feature set and the wind turbine generator fault type clustering information.

Where the time series of samples may be samples of data collected at different times for the case where the described phenomenon changes over time, such data reflects the state or extent of change over time of an object, phenomenon, etc.

The time-series marker may be information for marking a time-series sample.

Specifically, the wind turbine generator generation characteristic set obtained after characteristic extraction and wind turbine generator fault type clustering information obtained after characteristic clustering are used as the operator model input quantity of the wind turbine generator power loss value, a time sequence sample and a time sequence mark are constructed, wherein the operator model input quantity is (R) respectively ₁ ,R ₂ ,..R _m ) And (y) ₁ ,y ₂ ,...y _m ) Wherein R is ₁ Is (r) ₁ ,r ₂ ,...r _k-1 ) Vector of composition, y ₁ Is r of _k ；R ₂ Is (r) ₂ ,r ₃ ,...r _k ) Vector of composition, y ₂ Is r _k+1 (ii) a By analogy, R _m Is (r) _m ,r _m+1 ,...r _m+k-1 )，y _m Is r _m+k 。

Step 404, performing normalization processing on the time sequence samples to obtain a normalized time sequence.

Specifically, normalization processing is performed on the time series to obtain a normalized time series, wherein the normalization method is R' = (R-R) _min )/(r _max -r _min )。

And 406, obtaining a wind turbine generator loss power predicted value corresponding to each wind turbine generator according to the normalized time sequence and the time sequence marks.

Specifically, an operator model of the power loss value of the motor generator based on the LSTM network model is built, a time sequence back propagation algorithm (BPTT) is used for training according to a normalized time sequence and a time sequence mark, a root mean square error is set as a loss function, sigmoid is set as an activation function, the operator model of the power loss value of the motor generator is stored when the model loss value tends to be stable, and the predicted value of the loss power of the wind generator corresponding to each wind generator is obtained through calculation of the operator model of the power loss value of the wind generator.

when the wind turbine generator generation feature set and the wind turbine generator fault type clustering information are input into the electric machine power loss value calculation operator model, firstly, the cell state is filtered through a forgetting gate in the electric machine power loss value calculation operator model. Forgetting gate is a gating mechanism activated by Sigmoid function, which is input X at current time step t _t Hidden layer output h of sum time step t-1 _t-1 As a common input, the forgetting gate function is expressed as follows:

f _t ＝(U _f X _t + _f h _t-1 + _f )

wherein f is _t For the current time step t output of the forgetting gate, sigma is Sigmoid function, X _t For the input of the current time step t, h _t-1 Hidden layer output, U, for time step t-1 _f 、W _f A weight matrix for passing the forgetting gate at the current moment, b _f The offset item of passing the forgetting gate for the current time.

i _t ＝(U _i X _t + _i h _t-1 + _i )

g _t ＝anh(U _g X _t + _g h _t-1 + _g )

wherein the input gate is also denoted by X _t And h _t-1 As input, it is activated by Sigmoid activation functionThe combined value is mapped between 0 and 1 to obtain the input gate update weight i _t (ii) a At the same time, the input gate couples the input value X through the Tanh activation function _t And hidden state h of time step t-1 _t-1 Processing to obtain a cell state candidate vector g at the current time step _t Wherein, U _i 、W _i 、U _g 、W _g Respectively, the weight matrix of the current time passing through the input gate, b _i 、b _g Respectively, the offset terms that pass through the input gate at the current time.

c _t ＝ _t c _t-1 + _t g _t

o _t ＝(U _o X _t + _o h _t-1 + _o )

h _t ＝ _t ×tanh(c _t )

In the embodiment, the extraction and marking of the time sequence samples are set through a sub-model sampling link of the power loss value of the wind turbine generator, and the model training learning rate is improved while the wind power output time sequence change characteristics are kept.

In an embodiment, as shown in fig. 5, the step of inputting the power generation feature set of each wind turbine into the wind turbine fault type determination submodel to obtain the wind turbine fault type determination information corresponding to each wind turbine includes:

and 502, determining at least one model prediction fault type corresponding to the wind turbine fault type discrimination sub-model according to the power generation characteristic set of each wind turbine.

The model prediction fault type can be a fault type of the wind turbine generator which can be predicted by the wind turbine generator fault type judging sub-model.

Specifically, the generating characteristic set of each wind turbine generator with time sequence is used as the input quantity of the wind turbine generator fault type judging submodel, and typical fault types of normal operation, overspeed offline, icing shutdown and the like of the wind turbine generator are selected as the output quantity of the wind turbine generator fault type judging submodel.

And step 504, calculating a prediction value set of the power generation characteristic set of each wind turbine generator, wherein the prediction value set belongs to the prediction fault type of each model.

Specifically, according to the power generation characteristic set of each wind turbine generator, probability predicted values corresponding to fault types of each wind turbine generator in extreme weather are respectively calculated, and a predicted value set is obtained.

And 506, obtaining wind turbine generator fault type judging information corresponding to each wind turbine generator according to each prediction value set.

Specifically, for any wind turbine generator, a mapping relation is established with each fault type from high to low according to each probability predicted value, and the mapping relation is used as each fault type of the wind turbine generator to obtain wind turbine generator fault type judging information corresponding to each wind turbine generator. The specific calculation process of the wind turbine generator fault type discrimination submodel can be expressed as follows:

n _t ＝φ(UI _t +Wn _t-1 )

O _t ＝f(Vn _t )

wherein, U, V, W are network weight matrix, and training adjustment is carried out through back propagation; i is _t For input at time t, n _t-1 、n _t Hidden layer states at time t-1 and t, respectively, O _t For the output at time t, φ and f are activation functions.

The wind turbine generator fault type distinguishing submodel is specifically realized in the process that the power generation characteristic set of each wind turbine generator with time sequence is used as the wind turbine generator fault type distinguishing submodel to be input into I _t ＝{I ₁ ,I ₂ ,...I _k And outputting O = { O } by using the fault state of the wind turbine as a fault type judgment sub-model of the wind turbine ₁ ,O ₂ ,...O _k The hidden layer is marked asn＝{n ₁ ,n ₂ ,...n _k }. At time i, n _k-1 And current I _k As input, a calculation result O is obtained _k And the output is transmitted to the moment k +1, so that the next layer can always obtain the output weight of the previous layer, and the wind turbine generator fault type judging sub-model has the memory capacity of time sequence data.

In the embodiment, the hidden layer local link and the weight network are set in the training process through the established wind turbine generator fault type discrimination sub-model, and the predicted value set of the model prediction fault type is calculated, so that the convergence search speed of the wind turbine generator fault type identification training process can be increased, and the training is prevented from falling into local optimization.

In an embodiment, as shown in fig. 6, performing feature extraction on each piece of geographic position information and each piece of meteorological information to obtain a wind turbine generation feature set corresponding to each wind turbine, includes:

step 602, performing feature extraction on the geographic position information and the meteorological information to obtain an extracted wind turbine generator generation feature set.

Specifically, the neural network is used for extracting the features of the geographic position information and the meteorological information to obtain an extracted wind turbine generator generation feature set, wherein the extracted wind turbine generator generation feature set can be a feature vector or a matrix.

Step 604, calculating correlation coefficients corresponding to the geographic position characteristic information and the meteorological characteristic information according to a correlation algorithm.

Wherein the correlation algorithm may be a pearson correlation analysis method.

Wherein the correlation coefficient may be a pearson correlation coefficient.

Specifically, according to each piece of geographic position feature information and each piece of feature meteorological information, a corresponding correlation coefficient (pearson correlation coefficient) is calculated, wherein the pearson correlation coefficient is equal to a product of covariance divided by each standard deviation, and a specific calculation formula is as follows:

Is the average of the characteristic data.

And 606, eliminating the geographic position characteristic information and the meteorological characteristic information of which the correlation coefficients do not meet the preset threshold value to obtain the wind turbine generator generation characteristic set corresponding to each wind turbine generator.

Specifically, a preset threshold value of a correlation coefficient between data is defined to be 0.8, and for each piece of geographic position characteristic information and each piece of meteorological characteristic information of which the correlation coefficient does not meet the preset threshold value, the wind turbine generator generation characteristic set corresponding to each wind turbine generator is obtained by removing the geographic position characteristic information and each piece of meteorological characteristic information from the original data set.

In the embodiment, the characteristic data set with the strong correlation of meteorological data larger than the threshold value is selected, so that the influence of high-volume high-dimensional data on the training efficiency of the wind turbine generator fault type distinguishing submodel is eliminated, the input dimension of a neural network is reduced, and the accuracy and the efficiency of characteristic identification in network training are improved.

In an embodiment, as shown in fig. 7, training the loss power prediction model to be trained according to the operation characteristics of each wind turbine and the loss power prediction value of each wind turbine to obtain a trained loss power prediction model, includes:

and 702, limiting the loss power predicted value of the wind generation set which does not accord with the preset operation rule corresponding to each wind generation set according to the operation characteristics of each wind generation set to obtain the loss power predicted value after limitation.

Specifically, a wind turbine generator loss power predicted value of a loss power prediction model to be trained is subjected to a post-correction link by utilizing the wind turbine generator energy conversion characteristic, the wind turbine generator flicker characteristic and the wind turbine generator power control characteristic, and a prediction result which is not in accordance with an operation rule is limited to obtain a loss power predicted value after limitation.

And 704, training the loss power prediction model to be trained according to the limited loss power prediction value to obtain the trained loss power prediction model.

Specifically, according to the limited loss power predicted value, a constraint result is fed back to a weight parameter of a loss power prediction model to be trained, and the constraint result is returned to execute' obtaining the geographical position information of the wind turbine generator by using a station anemometer tower and a GIS system, wherein the geographical position information comprises geographical coordinates and altitude; and meteorological information including wind direction, wind speed, temperature, humidity, precipitation and air pressure within a height of 10-150 meters. And then, setting random variables on the basis of a large amount of original meteorological data by using a Monte Carlo statistical simulation method, performing similarity expansion on partial missing meteorological data in extreme weather, and setting random disturbance noise to generate a characteristic data set until the output quantity of the loss power prediction model to be trained meets the service requirement to obtain the trained loss power prediction model.

In the embodiment, the reliability and effectiveness of the output result are ensured by setting the prediction correction link based on the wind turbine generator operation rule, and the precision of the loss power prediction model for predicting the power loss of the wind turbine generator can be improved by assisting the training of the guided loss power prediction model.

In one embodiment, as shown in fig. 8, a method for predicting a power loss of a fan in extreme weather includes:

step 802, a trained loss power prediction model is obtained.

Specifically, the server obtains a loss power prediction model obtained by training through a fan loss power prediction model training method under extreme weather from the terminal.

Step 804, obtaining geographic position information to be predicted and meteorological information to be predicted corresponding to at least one wind turbine generator in the wind power station.

Specifically, the server obtains geographical position information to be predicted and meteorological information to be predicted corresponding to at least one wind turbine generator in a wind power plant which needs to perform fan loss power prediction under the condition of extreme weather from the terminal.

Step 806, inputting the geographic position information to be predicted and the meteorological information to be predicted into the trained loss power prediction model to obtain wind turbine loss power prediction values corresponding to the wind turbines in the wind power station.

Specifically, inputting geographic position information to be predicted and meteorological information to be predicted into a trained loss power prediction model, and judging through a wind turbine generator fault type judging sub-model in the trained loss power prediction model to obtain wind turbine generator fault type judging information corresponding to each wind turbine generator; secondly, inputting the wind turbine fault type distinguishing information into a wind turbine fault clustering sub-model for feature clustering to obtain wind turbine fault type clustering information corresponding to each wind turbine; and finally, inputting the wind turbine fault type clustering information into an operator model of the power loss value of the generator to obtain the wind turbine loss power predicted value corresponding to each wind turbine.

In the embodiment, the trained loss power prediction model is used for predicting the loss power of the wind turbine generator, so that the accuracy of prediction of the loss power of the fan in extreme weather can be ensured, and the capability of a power grid system for dealing with emergency situations can be improved.

It should be understood that, although the steps in the flowcharts related to the embodiments are shown in sequence as indicated by the arrows, the steps are not necessarily executed in sequence as indicated by the arrows. The steps are not performed in the exact order shown and described, and may be performed in other orders, unless explicitly stated otherwise. Moreover, at least a part of the steps in the flowcharts related to the above embodiments may include multiple steps or multiple stages, which are not necessarily performed at the same time, but may be performed at different times, and the order of performing the steps or stages is not necessarily sequential, but may be performed alternately or alternately with other steps or at least a part of the steps or stages in other steps.

Based on the same inventive concept, the embodiment of the application also provides a device for training the wind turbine loss power prediction model under extreme weather, which is used for realizing the method for training the wind turbine loss power prediction model under extreme weather. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the method, so that the specific limitations in the following embodiment of the one or more extreme weather fan loss power prediction model training devices can be referred to the limitations of the above fan loss power prediction model training method in extreme weather, and are not described herein again.

In one embodiment, as shown in fig. 11, there is provided a training apparatus for a prediction model of power loss of a wind turbine in extreme weather, including: an information acquisition module 1102, a feature extraction module 1104, a loss power prediction module 1106, and a model training module 1108, wherein:

the information acquisition module 1102 is configured to acquire geographic position information and meteorological information corresponding to at least one wind turbine generator in a wind farm;

the feature extraction module 1104 is configured to perform feature extraction on each piece of geographic position information and each piece of meteorological information to obtain a wind turbine generator generation feature set corresponding to each wind turbine generator;

the loss power prediction module 1106 is used for inputting the power generation feature set of each wind turbine generator into a loss power prediction model to be trained to obtain a wind turbine generator loss power prediction value corresponding to each wind turbine generator; the loss power prediction model to be trained sequentially comprises a wind turbine generator fault type judging sub-model, a wind turbine generator fault clustering sub-model and a wind turbine generator power loss value calculating sub-model;

and the model training module 1108 is configured to train the loss power prediction model to be trained according to the operating characteristics of each wind turbine and the loss power prediction value of each wind turbine, so as to obtain a trained loss power prediction model.

In one embodiment, the loss power prediction module 1106 is further configured to input the power generation feature set of each wind turbine into the wind turbine fault type determination sub-model to obtain wind turbine fault type determination information corresponding to each wind turbine; inputting the wind turbine fault type distinguishing information into a wind turbine fault clustering sub-model to obtain wind turbine fault type clustering information corresponding to each wind turbine; and inputting the generating characteristic set of the wind turbine generator and the fault type clustering information of the wind turbine generator to an operator model of a power loss value of the wind turbine generator to obtain a wind turbine generator loss power predicted value corresponding to each wind turbine generator.

In one embodiment, the lost power prediction module 1106 is further configured to construct a time series sample and a time series flag according to the wind turbine generator generation feature set and the wind turbine generator fault type clustering information; carrying out normalization processing on the time sequence samples to obtain a normalized time sequence; and obtaining the wind turbine generator loss power predicted value corresponding to each wind turbine generator according to the normalized time sequence and the time sequence mark.

In one embodiment, the loss power prediction module 1106 is further configured to determine at least one model prediction fault type corresponding to the wind turbine fault type discrimination sub-model according to the power generation feature set of each wind turbine; calculating a prediction value set belonging to each model prediction fault type in the power generation characteristic set of each wind turbine; and obtaining wind turbine generator fault type judging information corresponding to each wind turbine generator according to each prediction value set.

In one embodiment, the feature extraction module 1104 is further configured to perform feature extraction on the geographic position information and the meteorological information to obtain an extracted wind turbine generator generation feature set; the extracted wind turbine generator set power generation feature set comprises each geographic position feature information and each meteorological feature information; calculating the characteristic information of each geographical position and a correlation coefficient corresponding to the characteristic information of each weather according to a correlation algorithm; and eliminating each piece of geographic position characteristic information and each piece of meteorological characteristic information of which the correlation coefficient does not meet the preset threshold value to obtain a wind turbine generator generation characteristic set corresponding to each wind turbine generator.

In one embodiment, the model training module 1108 is further configured to limit the wind turbine loss power prediction value that does not meet the preset operation rule corresponding to each wind turbine according to the operation characteristics of each wind turbine, so as to obtain a loss power prediction value after limitation; and training the loss power prediction model to be trained according to the loss power prediction value after limitation to obtain the trained loss power prediction model.

All modules in the fan loss power prediction model training device under extreme weather can be wholly or partially realized through software, hardware and a combination of the software and the hardware. The modules can be embedded in a hardware form or independent from a processor in the computer device, and can also be stored in a memory in the computer device in a software form, so that the processor can call and execute operations corresponding to the modules.

In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as shown in fig. 12. 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 includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of an operating system and computer programs in the non-volatile storage medium. The database of the computer device is used for storing server data. 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 for training a model for predicting the power loss of a wind turbine in extreme weather.

Those skilled in the art will appreciate that the architecture shown in fig. 12 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.

It should be noted that, the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, presented data, etc.) referred to in the present application are information and data authorized by the user or sufficiently authorized by each party.

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, database, or other medium used in the embodiments provided herein may include at least one of non-volatile and volatile memory. The nonvolatile Memory may include Read-Only Memory (ROM), magnetic tape, floppy disk, flash Memory, optical Memory, high-density embedded nonvolatile Memory, resistive Random Access Memory (ReRAM), magnetic Random Access Memory (MRAM), ferroelectric Random Access Memory (FRAM), phase Change Memory (PCM), graphene Memory, and the like. Volatile Memory can include Random Access Memory (RAM), external cache Memory, and the like. 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 databases referred to in various embodiments provided herein may include at least one of relational and non-relational databases. The non-relational database may include, but is not limited to, a block chain based distributed database, and the like. The processors referred to in the various embodiments provided herein may be, without limitation, general purpose processors, central processing units, graphics processors, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, or the like.

All possible combinations of the technical features in the above embodiments may not be described for the sake of brevity, but should be considered as being within the scope of the present disclosure 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 present application. 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 application shall be subject to the appended claims.

Claims

1. A method for training a fan loss power prediction model in extreme weather is characterized by comprising the following steps:

acquiring geographic position information and meteorological information corresponding to at least one wind turbine generator in a wind power station;

performing feature extraction on each geographic position information and each meteorological information to obtain a wind turbine generator set power generation feature set corresponding to each wind turbine generator set;

inputting the power generation feature set of each wind turbine generator into a loss power prediction model to be trained to obtain a wind turbine generator loss power prediction value corresponding to each wind turbine generator; the loss power prediction model to be trained sequentially comprises a wind turbine generator fault type judging sub-model, a wind turbine generator fault clustering sub-model and a wind turbine generator power loss value calculating sub-model;

and training the loss power prediction model to be trained according to the operating characteristics of each wind turbine generator and the loss power prediction value of each wind turbine generator to obtain the trained loss power prediction model.

2. The method according to claim 1, wherein the step of inputting the generating feature set of each wind turbine generator into a lost power prediction model to be trained to obtain a wind turbine generator lost power prediction value corresponding to each wind turbine generator comprises:

inputting the generating characteristic set of each wind turbine generator into the wind turbine generator fault type judging submodel to obtain wind turbine generator fault type judging information corresponding to each wind turbine generator;

inputting the wind turbine generator fault type distinguishing information into the wind turbine generator fault clustering sub-model to obtain wind turbine generator fault type clustering information corresponding to each wind turbine generator;

and inputting the wind turbine generation feature set and the wind turbine fault type clustering information into an operator model of the wind turbine power loss value, and obtaining wind turbine loss power predicted values corresponding to the wind turbines.

3. The method according to claim 2, wherein the step of inputting the wind turbine generation feature set and the wind turbine fault type clustering information into the wind turbine power loss value calculation operator model to obtain the wind turbine loss power prediction value corresponding to each wind turbine comprises:

constructing a time sequence sample and a time sequence mark according to the wind turbine generator generation feature set and the wind turbine generator fault type clustering information;

normalizing the time sequence samples to obtain a normalized time sequence;

and obtaining a wind turbine generator loss power predicted value corresponding to each wind turbine generator according to the normalized time sequence and the time sequence marks.

4. The method according to claim 2, wherein the inputting each wind turbine generator generation feature set to the wind turbine generator fault type determination submodel to obtain wind turbine generator fault type determination information corresponding to each wind turbine generator comprises:

determining at least one model prediction fault type corresponding to the wind turbine fault type discrimination submodel according to the power generation characteristic set of each wind turbine;

calculating a prediction value set which belongs to each model prediction fault type in each wind turbine generator set power generation characteristic set;

and obtaining wind turbine generator set fault type judging information corresponding to each wind turbine generator set according to each prediction value set.

5. The method of claim 1, wherein the performing feature extraction on each of the geographic location information and the meteorological information to obtain a wind turbine generator set generation feature set corresponding to each of the wind turbine generators comprises:

extracting the geographic position information and the meteorological information to obtain an extracted wind turbine generator set power generation characteristic set; the extracted wind turbine generator generation feature set comprises feature information of each geographic position and feature information of each weather;

calculating the correlation coefficient corresponding to each geographical position characteristic information and each meteorological characteristic information according to a correlation algorithm;

and removing the geographic position characteristic information and the meteorological characteristic information of which the correlation coefficients do not meet a preset threshold value to obtain a wind turbine generator set power generation characteristic set corresponding to each wind turbine generator set.

6. The method of claim 1, wherein the training the loss power prediction model to be trained according to the operating characteristics of each wind turbine generator and the loss power prediction value of each wind turbine generator to obtain a trained loss power prediction model comprises:

according to the operating characteristics of each wind turbine generator, limiting a wind turbine generator loss power predicted value which does not accord with a preset operating rule corresponding to each wind turbine generator to obtain a limited loss power predicted value;

and training the loss power prediction model to be trained according to the limited loss power prediction value to obtain a trained loss power prediction model.

7. A method for predicting the power loss of a fan in extreme weather is characterized by comprising the following steps:

obtaining a trained lost power prediction model, wherein the trained lost power prediction model is obtained by training according to the method for training the wind turbine lost power prediction model in extreme weather according to any one of claims 1 to 6;

acquiring geographic position information to be predicted and meteorological information to be predicted, which correspond to at least one wind turbine generator in a wind power station;

and inputting the geographical position information to be predicted and the meteorological information to be predicted into the trained loss power prediction model to obtain a wind turbine generator loss power prediction value corresponding to each wind turbine generator in the wind power station.

8. A device for training a fan loss power prediction model under extreme weather is characterized by comprising:

the information acquisition module is used for acquiring geographic position information and meteorological information corresponding to at least one wind turbine generator in the wind power station;

the characteristic extraction module is used for carrying out characteristic extraction on the geographic position information and the meteorological information to obtain a wind turbine generator set power generation characteristic set corresponding to each wind turbine generator set;

the loss power prediction module is used for inputting the power generation characteristic set of each wind turbine generator into a loss power prediction model to be trained to obtain a wind turbine generator loss power prediction value corresponding to each wind turbine generator; the loss power prediction model to be trained sequentially comprises a wind turbine generator fault type judging sub-model, a wind turbine generator fault clustering sub-model and a wind turbine generator power loss value calculating sub-model;

and the model training module is used for training the loss power prediction model to be trained according to the operating characteristics of each wind turbine generator and the loss power prediction value of each wind turbine generator to obtain the trained loss power prediction model.

9. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor, when executing the computer program, implements the steps of the method of any of claims 1 to 6.

10. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the steps of the method of any one of claims 1 to 6.