CN115130743B - Wind turbine generator set regulation rate prediction method and system based on variation inference - Google Patents
Wind turbine generator set regulation rate prediction method and system based on variation inference Download PDFInfo
- Publication number
- CN115130743B CN115130743B CN202210710993.6A CN202210710993A CN115130743B CN 115130743 B CN115130743 B CN 115130743B CN 202210710993 A CN202210710993 A CN 202210710993A CN 115130743 B CN115130743 B CN 115130743B
- Authority
- CN
- China
- Prior art keywords
- wind turbine
- rate prediction
- turbine generator
- adjustment rate
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 62
- 238000000605 extraction Methods 0.000 claims abstract description 24
- 238000005070 sampling Methods 0.000 claims abstract description 23
- 238000013528 artificial neural network Methods 0.000 claims abstract description 11
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 10
- 238000009499 grossing Methods 0.000 claims abstract description 10
- 239000013598 vector Substances 0.000 claims abstract description 10
- 230000006870 function Effects 0.000 claims description 18
- 238000005457 optimization Methods 0.000 claims description 13
- 230000008569 process Effects 0.000 claims description 10
- 238000012545 processing Methods 0.000 claims description 7
- 239000011159 matrix material Substances 0.000 claims description 4
- 230000004913 activation Effects 0.000 claims description 3
- 230000006835 compression Effects 0.000 abstract description 4
- 238000007906 compression Methods 0.000 abstract description 4
- 238000012549 training Methods 0.000 description 11
- 238000010586 diagram Methods 0.000 description 10
- 230000000694 effects Effects 0.000 description 7
- 230000008859 change Effects 0.000 description 4
- 238000005259 measurement Methods 0.000 description 4
- 108090000672 Annexin A5 Proteins 0.000 description 3
- 238000004422 calculation algorithm Methods 0.000 description 3
- 230000009194 climbing Effects 0.000 description 3
- 238000004146 energy storage Methods 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 108010076504 Protein Sorting Signals Proteins 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 2
- 238000010835 comparative analysis Methods 0.000 description 2
- 230000003750 conditioning effect Effects 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000013213 extrapolation Methods 0.000 description 2
- 238000010248 power generation Methods 0.000 description 2
- 238000012706 support-vector machine Methods 0.000 description 2
- 238000012795 verification Methods 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000004590 computer program Methods 0.000 description 1
- 238000011217 control strategy Methods 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000005315 distribution function Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000003912 environmental pollution Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000007637 random forest analysis Methods 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
-
- 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)
- Business, Economics & Management (AREA)
- Physics & Mathematics (AREA)
- Economics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Marketing (AREA)
- Tourism & Hospitality (AREA)
- Health & Medical Sciences (AREA)
- General Business, Economics & Management (AREA)
- Algebra (AREA)
- Databases & Information Systems (AREA)
- Operations Research (AREA)
- Computational Mathematics (AREA)
- Entrepreneurship & Innovation (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Game Theory and Decision Science (AREA)
- Pure & Applied Mathematics (AREA)
- Quality & Reliability (AREA)
- Software Systems (AREA)
- General Engineering & Computer Science (AREA)
- Development Economics (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- General Health & Medical Sciences (AREA)
- Primary Health Care (AREA)
- Wind Motors (AREA)
Abstract
The invention discloses a wind turbine generator set regulation rate prediction method and system based on variation inference, comprising the following steps: acquiring active power data of the wind turbine generator, and smoothing the active power data by using empirical mode decomposition; inputting the smoothed data into a trained adjustment rate prediction model, and outputting a wind turbine generator adjustment rate prediction value; the method comprises the steps of firstly carrying out feature extraction on input data by the adjustment rate prediction model to obtain a mean value and a variance vector after feature extraction, then sampling Gaussian distribution meeting the mean value and the variance to obtain corresponding hidden variables, enabling the hidden variables to pass through a regression network consisting of a plurality of layers of neural networks, and finally outputting fan adjustment rate prediction values. The invention utilizes the neural network to construct the hidden variable generation model based on Gaussian distribution, thereby realizing the characteristic compression of the input variable; sampling hidden variables and constructing a regression network to realize the prediction of the fan adjusting speed.
Description
Technical Field
The invention relates to the technical field of wind turbine generator set adjustment rate prediction, in particular to a wind turbine generator set adjustment rate prediction method and system based on variation inference.
Background
The statements in this section merely provide background information related to the present disclosure and may not necessarily constitute prior art.
Due to the increasing environmental pollution and energy crisis, more and more countries have begun to explore renewable energy sources in recent decades. Wind power generation is considered a clean, economical, sustainable energy source. However, the fan output is directly affected by wind speed, has obvious intermittent and strong nonlinear characteristics, and the uncertainty and fluctuation of the fan output bring serious risks to the operation and control of the power system. In addition, wind power output often exhibits significant fluctuations and hill climbing due to wind intermittence. Although, to mitigate wind power fluctuations, various energy storage devices are employed in power systems. The control strategy of the wind farm combined with the energy storage system is also widely studied. But such compensation often fails to work in the face of rapid and large fluctuations. Therefore, the wind power prediction technology is used for providing decision support for power dispatching personnel to adjust the power generation plan and control the operation capacity of the fan in time, and also becomes a main basis for the charge-discharge coordination and the optimal control of the energy storage system.
At present, a method for predicting wind power is divided according to a time period, and mainly comprises the following steps: a long-term prediction method in units of years; mid-term prediction in month and week units; short-term prediction methods in units of days and hours and ultra-short-term prediction methods in units of minutes. Aiming at the difference of the prediction models, the method mainly comprises a physical prediction method and a statistical prediction method 2. The physical prediction method mainly uses meteorological elements such as wind speed, wind direction, air pressure, air temperature and the like provided by a numerical weather forecast model (Numerical Weather Prediction, NWP), and combines the landform and the topographic information around a wind farm to estimate the local wind speed so as to provide wind power prediction. However, the power prediction error is amplified due to the NWP error, and the NWP prediction period is long, so that the physical prediction method cannot be used for ultra-short-term prediction. Statistical prediction methods mainly include Auto-Regression Moving Average (ARMA), exponential smoothing (Exponential Smoothing, EM), kalman-filter (Kalman-filter) and the like extrapolation methods, and support vector machine (Support Vector Machine, SVM) methods, neural network methods, and the like, which are mainly characterized by machine learning. But extrapolation methods have a more stringent assumption on the randomly distributed characteristics of the data; the kernel function selection of the SVM has randomness, and the increase of the sample data volume and the increase of the input data dimension can lead to higher computational complexity; although the shallow neural network can fit sample data better, the shallow neural network has the defect of poor overfitting and generalization capability.
The prediction model based on deep learning has stronger nonlinear fitting capability, and is widely applied to new energy prediction and load prediction. In the field of ultra-short-term wind power prediction, a data driving algorithm is also used for estimating and predicting the regulation rate of the wind turbine, for example: in the prior art, a piecewise linear representation method is utilized to extract trend of the fan actual power time sequence, so that the estimation of the fan adjustment rate is realized. In addition, the wind turbine generator system regulation rate is also commonly used for calculating and predicting an event when wind power climbs, for example, a wind power slope event prediction model based on feature extraction and deep learning is provided in the prior art, a convolutional neural network is mainly adopted to extract features from model input, LSTM is utilized to learn a time sequence relation of data, and a climbing prediction result is provided on the basis.
Although the wind power and wind power climbing prediction effects are improved to different degrees by the aid of the above-mentioned work, extreme prediction deviation caused by weather condition uncertainty is ignored because prediction results are deterministic, and the determination of the safety management boundary of the power grid is not facilitated.
The existing probability models give a distribution interval of wind power or wind power regulation rate prediction results, but the distribution relied by the models is simpler, and once training is completed, the distribution function cannot change along with the change of input data. It is thought that the probability function functional space of the predictive model is not efficiently mined.
Disclosure of Invention
In order to solve the problems, the invention provides a wind turbine generator set regulation rate prediction method and a wind turbine generator set regulation rate prediction system based on variation inference, which are used for constructing a hidden variable generation model based on Gaussian distribution by using a neural network so as to realize characteristic compression of input variables; sampling hidden variables and constructing a regression network to realize the prediction of the fan adjusting speed.
In some embodiments, the following technical scheme is adopted:
a wind turbine generator set regulation rate prediction method based on variation inference comprises the following steps:
acquiring active power data of the wind turbine generator, and smoothing the active power data by using empirical mode decomposition;
Inputting the smoothed data into a trained adjustment rate prediction model, and outputting a wind turbine generator adjustment rate prediction value;
The method comprises the steps of firstly carrying out feature extraction on input data by the adjustment rate prediction model to obtain a mean value and a variance vector after feature extraction, then sampling Gaussian distribution meeting the mean value and the variance to obtain corresponding hidden variables, enabling the hidden variables to pass through a regression network consisting of a plurality of layers of neural networks, and finally outputting fan adjustment rate prediction values.
In other embodiments, the following technical solutions are adopted:
A variation inference based wind turbine tuning rate prediction system comprising:
The data processing module is used for acquiring active power data of the wind turbine generator and smoothing the active power data by using empirical mode decomposition;
The prediction module is used for inputting the smoothed data into a trained adjustment rate prediction model and outputting a wind turbine generator adjustment rate prediction value;
The method comprises the steps of firstly carrying out feature extraction on input data by the adjustment rate prediction model to obtain a mean value and a variance vector after feature extraction, then obtaining a corresponding hidden variable by sampling Gaussian distribution meeting the mean value and the variance, enabling the hidden variable to pass through a regression network consisting of a plurality of layers of neural networks, and finally outputting a fan adjustment rate prediction value.
In other embodiments, the following technical solutions are adopted:
A terminal device comprising a processor and a memory, the processor being configured to implement instructions; the memory is used for storing a plurality of instructions adapted to be loaded by the processor and to perform the above-described variation inference based wind turbine adjustment rate prediction method.
In other embodiments, the following technical solutions are adopted:
A computer readable storage medium having stored therein a plurality of instructions adapted to be loaded by a processor of a terminal device and to perform the above-described variation inference based wind turbine tuning rate prediction method.
Compared with the prior art, the invention has the beneficial effects that:
(1) According to the wind power regulation rate probability prediction method, a wind power regulation rate probability prediction model is constructed, the hidden variable after feature compression is generated through the feature extraction network, regression prediction is carried out on the basis of sampling the hidden variable, and then a fan regulation rate probability prediction result is obtained. Meanwhile, in order to avoid the situation that the traditional error function is not suitable for probability model training, monte Carlo ELBO is used as an objective function to optimize model parameters. And finally, verifying the effectiveness and practicability of the model through comparative analysis and interval calculation.
Additional features and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention.
Drawings
FIG. 1 is a schematic diagram of a comparison of the active EMD smoothing of a fan in an embodiment of the present invention;
FIG. 2 is a schematic diagram of fan conditioning rate distribution in an embodiment of the present invention;
FIGS. 3 (a) - (d) are schematic diagrams of a combined wind speed and fan adjustment rate distribution in an embodiment of the present invention;
FIGS. 4 (a) - (d) are schematic diagrams of the combined distribution of air temperature, air pressure, humidity, power and fan conditioning rate in an embodiment of the present invention;
FIG. 5 is a schematic diagram of a wind turbine generator set adjustment rate prediction process based on variation inference in an embodiment of the invention;
FIG. 6 is a diagram illustrating a hyper-parameter contrast based on grid search in accordance with an embodiment of the invention;
FIG. 7 is a schematic diagram showing the comparison of model prediction effects in an embodiment of the present invention;
Fig. 8 is a schematic diagram of a model probability interval prediction effect in an embodiment of the present invention.
Detailed Description
It should be noted that the following detailed description is illustrative and is intended to provide further explanation of the application. Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.
It is noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of exemplary embodiments according to the present application. As used herein, the singular is also intended to include the plural unless the context clearly indicates otherwise, and furthermore, it is to be understood that the terms "comprises" and/or "comprising" when used in this specification are taken to specify the presence of stated features, steps, operations, devices, components, and/or combinations thereof.
Example 1
With the continuous improvement of the new energy duty ratio, the wind turbine generator system which is required to be connected with the grid and runs safely and stably also participates in the rapid adjustment of the frequency of the grid. In order to meet the adjustment requirement, the adjustment capability of the fan needs to be accurately mastered and predicted, namely, the adjustment rate of the unit is accurately predicted.
The unit regulation rate refers to the change of wind power in a given time period and is marked as y
In the method, in the process of the invention,Active power for adjacent time periodsAndIs a difference in (2); Δt=t 1-T2 is the time interval. Here, 15 minutes was used as the time interval for calculating the adjustment rate.
Because the influence factor of the fan adjusting speed is mainly weather factor and has strong randomness, the traditional mean square error (Mean Squared Error, MSE) measurement index of the measurement model performance is limited due to the characteristics of deviation-variance balance.
Let f (x; θ) denote the fan adjustment rate prediction model to be built, θ being the parameter to be estimated. Given an input x and a true output y, the MSE may be written as:
MSE＝Var(y-f)+[E(y)-E(f)]2 (2)
here, var (y-f) represents the variance of the prediction error, and [ E (y) -E (f) ] 2 represents the measurement deviation. Further deriving the formula (2) can obtain:
By equation (3), it is not difficult to find that minimizing the MSE also means minimizing the variance of the prediction result Var (f), but the decrease in variance means that the model can only adapt to local data distribution characteristics, resulting in an increase in the bias term. When the randomness of the input x is strong, two aspects of variance and deviation need to be considered simultaneously, and a new error measurement index is introduced.
Based on this, in one or more embodiments, a method for predicting a wind turbine adjustment rate based on variance inference is disclosed, as shown in fig. 5, including the following procedures:
acquiring active power data of the wind turbine generator, and smoothing the active power data by using empirical mode decomposition;
And inputting the smoothed data into a trained adjustment rate prediction model, and outputting a wind turbine generator adjustment rate predicted value.
Specifically, the wind power signal is affected by external weather, mechanical vibration of a fan, fluctuation of a transmission signal and other factors, and is shown as superposition of signal sources with different frequencies. In order to accurately calculate the regulation rate of the wind turbine, it is necessary to perform high-frequency data rejection on the power signal so as to fully highlight the influence of weather factors on the fan activity.
The Empirical Mode Decomposition (EMD) algorithm is a novel adaptive signal time-frequency processing method, and is particularly suitable for the analysis and processing of nonlinear non-stationary signals. The method can decompose a complex signal into a series of sub-signals called natural mode functions (INTRINSIC MODE FUNCTION, IMFs), each IMF has different frequency components, and can describe the input signal in different scales. IMF satisfies two conditions: (1) The number of signals in the signal sequence passing through the extreme value and the zero value must be equal or at most differ by one; (2) The average of local maxima and minima at any point on the signal sequence is 0.
Given that the signal to be processed is x (t), where t is the time instant, the steps of EMD processing the signal are:
(1) Determining local maximum and minimum values of a signal x (t), calculating an upper envelope and a lower envelope of the signal by using a cubic spline difference function, and further calculating the average value of the upper envelope and the lower envelope;
(2) Subtracting the mean values of the upper envelope curve and the lower envelope curve in the step (1) by using x (t) to obtain a first component c 1 (t);
(3) Judging whether the component c 1 (t) meets two conditions of IMF, if yes: obtaining a first IMF component c 1 (t); if not, repeating the steps (1) and (2) based on c 1 (t);
(4) After subtracting c 1 (t) from the signal x (t), the remaining component was used as a new time sequence r 1 (t), and the EMD method was repeated n times to obtain n IMF subsequences and 1 residue sequence r n (t), the overall expression being as follows:
in formula (1), r i (t) i=1, 2,., n-1 represents the remaining component after the i-th decomposition by the EMD method; r n (t) denotes the last residual component.
Extracting wind power operation data of 7 th month 5 th day to 28 th day of 2 nd month 2021 of a wind farm in Shandong province, wherein the data sampling period is 15 minutes, performing EMD (empirical mode decomposition) on an original active signal to obtain IMFs under different components, and removing high-frequency components of the IMFs 1 to obtain the smoothed fan active power. The result of intercepting part of the data is shown in fig. 1, and it can be known that small disturbance of the signal is eliminated smoothly, the main change process is reserved, and the fan adjusting speed can be calculated more accurately.
The correlation and probability distribution of the related indexes can be primarily known after the combined distribution and the marginal histogram are drawn by selecting the height wind speed, temperature, air pressure, humidity, active power and adjustment rate of 10 meters/30 meters/50 meters/70 meters.
First, a histogram is drawn on the fan adjustment rate data and fitted using gaussian kernel density estimation with a bandwidth parameter set to 0.5. It can be seen that the distribution of the adjustment rate is mainly concentrated between-0.5 and is almost symmetrical. The probability of data occurrence is smaller as the adjustment rates deviate from-0.5 and 0.5, respectively, as shown in fig. 2.
Secondly, a joint probability distribution scatter diagram and a histogram of marginal distribution are respectively drawn for the wind speed and the fan adjusting speed of 10 meters/30 meters/50 meters/70 meters. It can be seen that: the marginal distribution of wind speeds is characterized by a long tail distribution, wherein a wind speed of 5 meters/s occupies a large number of sample points, while other wind speeds correspond to only a few samples. This class imbalance in the number of training samples makes the prediction result of the constant value model incapable of taking into account extreme cases, as shown in fig. 3 (a) - (d). The trained model is easily biased towards the head class with large training data volume, so that the model does not perform well on the tail class with limited data volume.
Finally, drawing a combined probability distribution scatter diagram and a marginal distribution histogram respectively aiming at the temperature, the air pressure, the humidity, the power and the fan adjusting speed. From the marginal distribution, the temperature and air pressure distribution has a distinct double peak, and the humidity and power exhibit long tail distribution, as shown in fig. 4 (a) - (d). From the aspect of the joint distribution of the temperature, the air pressure and the adjusting speed, the joint distribution of the temperature, the air pressure and the adjusting speed has obvious double-peak characteristics, and the joint distribution of the humidity and the adjusting speed is uniform; the joint profile of power and modulation rate has long tail characteristics.
The uncertainty of the input factors means the uncertainty of the sample set { (y i,xi) }, i=1, 2, 3. Wherein y i represents the rate of adjustment of the ith sample; x i represents a single input variable and,D represents the number of feature dimensions. That is, the form of the distribution p (y|x) is unknown.
From the perspective of variance inference, if one wants to get a distribution and probability characteristics for y at a given x, one needs to introduce another distribution q θ (y) to achieve an approximation of p (y|x). Here, θ is a variation parameter to be optimized. Meanwhile, we hope that q θ (y) is x-dependent. Thus, there may be:
Expressed by the average value of g θ(x)μ, Is a multivariate gaussian distribution of covariance matrix. Wherein: Is a diagonal matrix, the elements on the diagonal are Here, g θ (x) serves as a feature extraction input variable dimension, which can be accomplished by a multi-layer feed forward (Multilayer Perceptron, MLP) network.
The predictive model for the overall fan turndown rate consists of three parts, as shown in fig. 5:
(1) The input x is subjected to a feature extraction network g θ (x) to obtain a mean mu and variance sigma vector after feature extraction;
(2) Sampling the Gaussian distribution meeting the mean and the variance to obtain a corresponding hidden variable z;
(3) And (3) allowing the hidden variable z to pass through a regression network f δ (z) consisting of MLP, and finally outputting the fan adjusting speed predicted value y. In this embodiment, the regression network f δ (z) is a layer 2 feed forward network.
Here, the parameters θ and δ are parameters that need to be optimized for the entire process. In addition, to increase the convergence rate in the model optimization process, a modified linear unit activation function (RECTIFIED LINEAR unit, reLU) is used between each layer of the MLP network, expressed as follows:
Finally, it can also be seen from the structure of the model that since g θ (x) varies depending on x, the completed model is trained And also with x, so that the sampling probability of the hidden variable z is not constant.
When optimizing the predictive probability model, since the integral analysis formula of the p (y|x) likelihood function cannot be found, ELBO is used as an optimization target for the factor, and the form is:
here, the objective function This term represents the likelihood of the data given the hidden variable z; the term D KL (q (z|x) ||p (z)) represents the KL divergence between q (z|x) and the a priori distribution p (z); z is a hidden variable; according to the definition of D KL, we get:
substituting formula (7) can give the target form:
since both terms in equation (9) are desired, a Monte Carlo approximation can be used, namely:
wherein L represents the number of samples, q (z|x) is the posterior distribution of hidden variable z after the input x is given, namely the sampling distribution after the feature extraction network in the present formula (5) and fig. 5, and the function of the posterior distribution is to perform dimensional compression on the input variable and output an expression of probability distribution at the same time; p (z) represents the a priori distribution of the hidden variable z, typically set to a standard normal distribution N (0, 1); z j denotes the j-th sample of the L samples. x is an input variable of the model and may include: wind speed, temperature, air pressure, humidity, active power and fan adjusting speed at the previous moment of 10 m/30 m/50 m/70 m height; the hidden variable z is a sample of the sample after passing through the feature extraction network in fig. 5.
It should be noted that the hidden variable z is generated from a normal distribution containing the parameter θ during model training, and the parameter needs to be optimized by gradient descent during the whole model parameter optimization. However, the hidden variable z is sampled to become a value independent of the parameter θ in each training round, so that the backward propagation process of gradient descent cannot be conducted to θ, and therefore cannot be learned. The hidden variable of the embodiment realizes sampling by a sampling re-parameterization method so as to complete the gradient descent optimization process of the optimization objective function. That is, one ε was obtained by sampling N (0, 1) and then letting z=μ+δε, where: the corresponding element is multiplied.
In this embodiment, the smoothed data is divided into a training set, a verification set, and a test set in proportion to 70%, 20%, and 10% of the data amount. Model training and parameter optimization are performed using the data of the training set and the validation set. The characteristics of the model input data x include: wind speed, temperature, air pressure, humidity, active power and fan adjusting speed at the previous moment of 10 m/30 m/50 m/70 m height; the model outputs the fan adjustment rate y at the next moment. The model structure is shown in fig. 5, and is referred to herein as VI prediction model, and the main parameters of the model are shown in table 1.
TABLE 1 VI predictive model parameters
Wherein, represent the number of each batch of data in training. * Represents the hyperparametric dimension of μ, which is determined by hyperparametric optimization. * The hyper-parameter dimension of σ is represented, and the value is determined by hyper-parameter optimization.
In the training process, the number of neurons related to the learning rate, the mean value and the standard deviation is set as 3 super parameters, and the set intervals are respectively:
TABLE 2 super parameter settings
The hyper-parameters of the model are determined by using a grid search method, and the parameter tuning result is shown in fig. 6. It can be seen that the verification set error is the lowest when the learning rate is 1e-3, mu_dim is 100, and var_dim is 100.
Because the variation inference model is a probability model, the model can not only give a mean value prediction result of the fan adjusting speed, but also give a prediction interval through sampling, so as to obtain the upper limit and the lower limit of the prediction result. Therefore, the prediction effect of the model was evaluated using two indices, root mean square error (RootMean Square Error, RMSE) and prediction interval coverage (Prediction Interval Coverage Probability, PICP), respectively. The expression is:
where N represents the number of predicted samples, y i represents the actual value of the ith adjustment rate, A predicted value representing the ith adjustment rate; c i indicates whether the actual value of the ith adjustment rate falls within the corresponding prediction interval, if so, c i is 1, otherwise it is 0.
On the basis of comparative analysis of the deterministic effect of the model, a random forest algorithm Xgboost model is selected as a control group, and the mean value of the VI model in the embodiment is used as a fixed value prediction result. The behavior of the two models is shown in fig. 7, where: xgboost has a RMSE value of 0.1142; the RMSE value of the VI model of this embodiment is 0.1147, and the two effects are relatively close, which indicates that the constant value prediction capability of the VI model of this embodiment is not greatly different from Xgboost commonly used in actual engineering.
Secondly, since the VI model of the present embodiment belongs to a probabilistic model, it can also perform sampling estimation for the predicted value at the same time. Fig. 8 shows a distribution of 5000 predictors sampled at each prediction time point. For the sampled samples, 10%, 90% quantile and 20%, 80% quantile values were calculated as two intervals of prediction, respectively. For the settings with 10% and 90% quantiles as the upper and lower bounds of prediction, the PICP of the model on the test set was 100%; for the settings with 20% and 80% quantiles as the upper and lower bounds of prediction, the PICP of the model on the test set was 99.3%; this illustrates that the prediction horizon of the VI model can cover almost all extreme weather conditions and thus is more conducive to practical use.
Example two
In one or more embodiments, a variation inference based wind turbine tuning rate prediction system is disclosed, comprising:
The data processing module is used for acquiring active power data of the wind turbine generator and smoothing the active power data by using empirical mode decomposition;
The prediction module is used for inputting the smoothed data into a trained adjustment rate prediction model and outputting a wind turbine generator adjustment rate prediction value;
The method comprises the steps of firstly carrying out feature extraction on input data by the adjustment rate prediction model to obtain a mean value and a variance vector after feature extraction, then obtaining a corresponding hidden variable by sampling Gaussian distribution meeting the mean value and the variance, enabling the hidden variable to pass through a regression network consisting of a plurality of layers of neural networks, and finally outputting a fan adjustment rate prediction value.
It should be noted that, the specific implementation manner of each module has been described in the first embodiment, and will not be described in detail herein.
Example III
In one or more embodiments, a terminal device is disclosed, including a server including a memory, a processor, and a computer program stored on the memory and executable on the processor, the processor implementing the variation inference based wind turbine adjustment rate prediction method in embodiment one when executing the program. For brevity, the description is omitted here.
It should be understood that in this embodiment, the processor may be a central processing unit CPU, and the processor may also be other general purpose processors, digital signal processors DSP, application specific integrated circuits ASIC, off-the-shelf programmable gate array FPGA or other programmable logic device, discrete gate or transistor logic devices, discrete hardware components, or the like. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like.
The memory may include read only memory and random access memory and provide instructions and data to the processor, and a portion of the memory may also include non-volatile random access memory. For example, the memory may also store information of the device type.
In implementation, the steps of the above method may be performed by integrated logic circuits of hardware in a processor or by instructions in the form of software.
Example IV
In one or more embodiments, a computer readable storage medium is disclosed, in which a plurality of instructions are stored, the instructions being adapted to be loaded by a processor of a terminal device and to perform the variation inference based wind turbine tuning rate prediction method described in embodiment one.
While the foregoing description of the embodiments of the present invention has been presented in conjunction with the drawings, it should be understood that it is not intended to limit the scope of the invention, but rather, it is intended to cover all modifications or variations within the scope of the invention as defined by the claims of the present invention.
Claims (10)
1. A wind turbine generator system regulation rate prediction method based on variation inference is characterized by comprising the following steps:
acquiring active power data of the wind turbine generator, and smoothing the active power data by using empirical mode decomposition;
Inputting the smoothed data into a trained adjustment rate prediction model, and outputting a wind turbine generator adjustment rate prediction value;
Firstly, carrying out feature extraction on input data by the adjustment rate prediction model to obtain a mean value and a variance vector after feature extraction, sampling Gaussian distribution meeting the mean value and the variance to obtain corresponding hidden variables, enabling the hidden variables to pass through a regression network consisting of a plurality of layers of neural networks, and finally outputting a fan adjustment rate prediction value;
The uncertainty of the input factors means the uncertainty of the sample set { (y i,xi) }, i=1, 2,3,..; wherein N' represents the number of samples; y i represents the rate of adjustment of the ith sample; x i represents a single input variable and, D represents the number of feature dimensions, the form of the distribution p (y|x) is unknown;
Another distribution q θ (y) needs to be introduced to achieve approximation of p (y|x), and θ is a variation parameter to be optimized;
Expressed by the average value of g θ(x)μ, A multi-element Gaussian distribution of a covariance matrix; wherein the method comprises the steps of the method comprises the following steps: Is a diagonal matrix, the elements on the diagonal are G θ (x) plays a role in feature extraction of the input variable dimension, which can be accomplished by a multi-layer feed forward network;
The measuring model of the whole fan adjusting speed consists of three parts:
(1) The input x is subjected to a feature extraction network g θ (x) to obtain a mean mu and variance sigma vector after feature extraction;
(2) Sampling the Gaussian distribution meeting the mean and the variance to obtain a corresponding hidden variable z;
(3) Allowing the hidden variable z to pass through a regression network f δ (z) consisting of a multi-layer feedforward network, and finally outputting a fan adjusting speed predicted value y; the regression network f δ (z) is a layer 2 feed forward network;
Parameters theta and delta are parameters which need to be optimized in the whole process; a modified linear element activation function RELU is used between each layer of a multi-layer feed forward network, expressed as follows:
Using monte carlo as an optimization object, the form is:
Objective function This term represents the likelihood of the data given the hidden variable z; the term D KL (q (z|x) ||p (z)) represents the KL divergence between q (z|x) and the a priori distribution p (z); z is a hidden variable; according to the definition of D KL, we get:
substituting formula (7) can give the target form:
Using the monte carlo approximation, namely:
where L represents the number of samples, q (z|x) is the posterior distribution of the hidden variable z given the input x, and p (z) represents the prior distribution of the hidden variable z, typically set to a standard normal distribution N (0, 1); z j represents the j-th sample of the L samples; x is an input variable of the model, and the hidden variable z is a sampling sample after the feature extraction network; sampling N (0, 1) to obtain ε, and then letting z=μ+δε, where: the corresponding element is multiplied.
2. The variation inference-based wind turbine tuning rate prediction method as defined in claim 1, comprising: and optimizing parameters of the adjustment rate prediction model by taking the subsurface evidence as an optimization objective function.
3. The variation inference-based wind turbine tuning rate prediction method as defined in claim 2, comprising:
The first term of the optimization objective function represents the likelihood of the data given the hidden variables; the second term represents the KL divergence between the hidden variable z posterior distribution q (z|x) and the hidden variable z prior distribution p (z); approximating the first and second terms using monte carlo; x is the input variable of the model.
4. A method for predicting the adjustment rate of a wind turbine generator set based on variation inference as claimed in claim 3, wherein the hidden variable is sampled by a sampling re-parameterization method to complete the gradient descent optimization process of the optimization objective function.
5. The method for predicting the adjustment rate of a wind turbine generator based on variation inference as set forth in claim 1, wherein the processing the input data to obtain the feature extracted mean and variance vectors includes:
The input x is subjected to g θ (x) to obtain the mean value and variance vector of the hidden variable z after feature extraction; g θ (x) is accomplished by a multi-layer feed forward network.
6. A variation inference based wind turbine tuning rate prediction method as defined in claim 5 wherein a modified linear unit activation function is used between each layer of the multi-layer feed forward network.
7. The method for predicting the adjustment rate of the wind turbine generator set based on variation inference as claimed in claim 1, wherein the hidden variable z obtained by sampling is taken as an input, and an output result of the whole adjustment rate prediction model is obtained after entering a regression network f δ (z); the regression network is a layer 2 feed forward network.
8. A system employing the variation inference based wind turbine tuning rate prediction method of claim 1, comprising:
The data processing module is used for acquiring active power data of the wind turbine generator and smoothing the active power data by using empirical mode decomposition;
The prediction module is used for inputting the smoothed data into a trained adjustment rate prediction model and outputting a wind turbine generator adjustment rate prediction value;
The adjusting speed prediction model firstly performs feature extraction on input data to obtain a mean value and a variance vector after feature extraction, and a corresponding hidden variable is obtained by sampling Gaussian distribution meeting the mean value and the variance, and the hidden variable passes through a regression network consisting of a plurality of layers of neural networks to finally output a fan adjusting speed prediction value.
9. A terminal device comprising a processor and a memory, wherein the processor is configured to implement instructions; the memory is used for storing a plurality of instructions;
the instructions are adapted to be loaded by a processor and to perform the variation inference based wind turbine tuning rate prediction method of any one of claims 1-7.
10. A computer readable storage medium storing a plurality of instructions adapted to be loaded by a processor of a terminal device and to perform the variation inference based wind turbine tuning rate prediction method of any of claims 1-7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210710993.6A CN115130743B (en) | 2022-06-22 | 2022-06-22 | Wind turbine generator set regulation rate prediction method and system based on variation inference |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210710993.6A CN115130743B (en) | 2022-06-22 | 2022-06-22 | Wind turbine generator set regulation rate prediction method and system based on variation inference |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115130743A CN115130743A (en) | 2022-09-30 |
| CN115130743B true CN115130743B (en) | 2024-08-02 |
Family
ID=83379624
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210710993.6A Active CN115130743B (en) | 2022-06-22 | 2022-06-22 | Wind turbine generator set regulation rate prediction method and system based on variation inference |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115130743B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117406844B (en) * | 2023-11-14 | 2024-04-19 | 珠海世宁达科技有限公司 | Display card fan control method and related device based on neural network |
Family Cites Families (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104899656A (en) * | 2015-06-05 | 2015-09-09 | 三峡大学 | Wind power combined predication method based on ensemble average empirical mode decomposition and improved Elman neural network |
| CN109272156B (en) * | 2018-09-12 | 2021-12-07 | 河海大学 | Ultra-short-term wind power probability prediction method |
| CN110070201A (en) * | 2019-02-19 | 2019-07-30 | 国网浙江省电力有限公司 | A method of passing through economic growth rate forecast electricity consumption growth rate |
| CN110365053B (en) * | 2019-08-09 | 2023-04-25 | 南京信息工程大学 | Short-term wind power prediction method based on delay optimization strategy |
| EP4115392A4 (en) * | 2020-03-04 | 2023-08-23 | Magic Leap, Inc. | Systems and methods for efficient floorplan generation from 3d scans of indoor scenes |
| CN111626506B (en) * | 2020-05-27 | 2022-08-26 | 华北电力大学 | Regional photovoltaic power probability prediction method based on federal learning and cooperative regulation and control system thereof |
| CN112801350B (en) * | 2021-01-13 | 2022-09-06 | 中国电建集团华东勘测设计研究院有限公司 | Uncertainty-based deep learning ultra-short-time wind power prediction system and method |
| CN112819238A (en) * | 2021-02-19 | 2021-05-18 | 河海大学 | Short-term wind power prediction method based on chaotic chicken flock optimization algorithm |
| CN113205140B (en) * | 2021-05-06 | 2022-11-15 | 中国人民解放军海军航空大学 | Semi-supervised specific radiation source individual identification method based on generative countermeasure network |
| CN113222289B (en) * | 2021-06-02 | 2024-03-12 | 江南大学 | Prediction method of energy power based on data processing |
| CN114298132A (en) * | 2021-11-11 | 2022-04-08 | 华能新能源股份有限公司 | Wind power prediction method and device and electronic equipment |
-
2022
- 2022-06-22 CN CN202210710993.6A patent/CN115130743B/en active Active
Non-Patent Citations (1)
| Title |
|---|
| 基于变分推断的超短期风电功率预测;路宽等;《山东电力技术》;20230425;第50卷(第04期);13-21 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115130743A (en) | 2022-09-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Ma et al. | A hybrid attention-based deep learning approach for wind power prediction | |
| Shang et al. | Enhanced support vector regression based forecast engine to predict solar power output | |
| CN112434848B (en) | Nonlinear weighted combination wind power prediction method based on deep belief network | |
| Zhu et al. | Wind power forecasting based on new hybrid model with TCN residual modification | |
| CN116861201B (en) | Power supply system based on artificial intelligence | |
| Zhang et al. | Wind speed prediction research considering wind speed ramp and residual distribution | |
| CN114757427B (en) | Autoregressive-corrected LSTM intelligent wind power plant ultra-short-term power prediction method | |
| CN114362175B (en) | Wind power prediction method and system based on depth certainty strategy gradient algorithm | |
| Fan et al. | Fluctuation pattern recognition based ultra-short-term wind power probabilistic forecasting method | |
| CN104036328A (en) | Self-adaptive wind power prediction system and prediction method | |
| CN113642767A (en) | Multi-dimensional feature combination prediction method based on MI-VMD-DA-EDLSTM-VEC | |
| CN113111592A (en) | Short-term wind power prediction method based on EMD-LSTM | |
| CN111697560B (en) | Method and system for predicting load of power system based on LSTM | |
| CN115130743B (en) | Wind turbine generator set regulation rate prediction method and system based on variation inference | |
| CN115618922A (en) | Photovoltaic power prediction method and device, photovoltaic power generation system and storage medium | |
| Shabbir et al. | Short-term wind energy forecasting using deep learning-based predictive analytics | |
| Tajjour et al. | Power generation forecasting of a solar photovoltaic power plant by a novel transfer learning technique with small solar radiation and power generation training data sets | |
| Zaman et al. | Wind speed forecasting using ARMA and neural network models | |
| CN118017482A (en) | Flexible climbing capacity demand analysis method based on prediction error feature extraction | |
| Xu et al. | A novel hybrid wind speed interval prediction model based on mode decomposition and gated recursive neural network | |
| Xiong et al. | An ultra-short-term wind speed correction method based on the fluctuation characteristics of wind speed | |
| Chen et al. | Interval prediction of photovoltaic power using improved NARX network and density peak clustering based on kernel mahalanobis distance | |
| CN116960975A (en) | Photovoltaic power generation amount prediction method and device | |
| CN115018156B (en) | Short-term wind power prediction method | |
| CN116187506A (en) | Short-term wind power combination probability prediction method and system considering meteorological classification |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |