CN117150705A - Marine wind farm short-term wind speed prediction method based on characteristic deconstruction - Google Patents
Marine wind farm short-term wind speed prediction method based on characteristic deconstruction Download PDFInfo
- Publication number
- CN117150705A CN117150705A CN202310046981.2A CN202310046981A CN117150705A CN 117150705 A CN117150705 A CN 117150705A CN 202310046981 A CN202310046981 A CN 202310046981A CN 117150705 A CN117150705 A CN 117150705A
- Authority
- CN
- China
- Prior art keywords
- model
- wind speed
- gru
- data
- sample set
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 47
- 238000012549 training Methods 0.000 claims abstract description 50
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 20
- 238000012360 testing method Methods 0.000 claims abstract description 12
- 238000010606 normalization Methods 0.000 claims abstract description 10
- 238000012545 processing Methods 0.000 claims abstract description 4
- YHXISWVBGDMDLQ-UHFFFAOYSA-N moclobemide Chemical compound C1=CC(Cl)=CC=C1C(=O)NCCN1CCOCC1 YHXISWVBGDMDLQ-UHFFFAOYSA-N 0.000 claims description 37
- 238000004364 calculation method Methods 0.000 claims description 6
- 230000008569 process Effects 0.000 claims description 6
- 230000004913 activation Effects 0.000 claims description 3
- 230000000694 effects Effects 0.000 description 17
- 238000011156 evaluation Methods 0.000 description 6
- 238000010586 diagram Methods 0.000 description 5
- 230000006870 function Effects 0.000 description 5
- 238000013528 artificial neural network Methods 0.000 description 3
- 238000000780 augmented Dickey–Fuller test Methods 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000010248 power generation Methods 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 238000002790 cross-validation Methods 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000000053 physical method Methods 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000012797 qualification Methods 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000012876 topography Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- General Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Molecular Biology (AREA)
- Computational Mathematics (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Artificial Intelligence (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Pure & Applied Mathematics (AREA)
- Mathematical Optimization (AREA)
- Mathematical Analysis (AREA)
- Biomedical Technology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computer Hardware Design (AREA)
- Operations Research (AREA)
- Probability & Statistics with Applications (AREA)
- Geometry (AREA)
- Algebra (AREA)
- Databases & Information Systems (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a method for predicting the short-term wind speed of an offshore wind farm based on characteristic deconstruction, which comprises the following steps: obtaining an actually measured wind speed signal through a wind measuring tower of the offshore wind farm, and taking the actually measured wind speed signal as sample data; deconstructing the sample data into an IMF component and a residual sequence by a variational modal decomposition algorithm; converting all IMF components and residual sequences into a target data format through normalization and sliding window slicing processing, and obtaining a training sample set and a test sample set; building a GRU-ARIMA model, inputting a training sample set into the GRU-ARIMA model for training until the GRU-ARIMA model converges, and storing model parameters of the GRU-ARIMA model; predicting the values of the IMF component and the residual sequence through a trained GRU-ARIMA model; and integrating all the predicted results and obtaining a wind speed predicted value. According to the method and the device, more time sequence information can be extracted from the original wind speed signal, so that the accuracy of wind speed prediction is improved.
Description
Technical Field
The invention relates to the technical field of wind speed prediction, in particular to a method for predicting the short-term wind speed of an offshore wind farm based on characteristic deconstruction.
Background
Wind speed is one of the main factors influencing the power generation of a fan, and as the offshore environment is quite different from the onshore environment, the offshore environment is not influenced by landforms, plants, buildings and other landform features, the turbulence intensity is small, and the wake flow influence distance of a wind turbine generator is long and the range is wide; at the same time there is also a regional numerical mode on the sea, based on ocean-ocean wave-atmosphere mode coupling. These characteristics make the relevant research of onshore wind speed prediction unsuitable for offshore wind power. At present, wind speed prediction is mainly divided into a physical modeling method based on atmosphere, environmental topography and the like and a statistical learning method based on data. The method utilizes physical methods such as hydrodynamics, thermodynamics and the like, comprehensively considers information such as meteorological conditions, ocean and atmospheric thermodynamic processes and the like, has the defects of high implementation difficulty, complex calculation and the like, and is generally used for medium-long term prediction.
Statistical learning methods typically make predictions of wind speed over a short time period in the future based on numerical weather forecast data (e.g., wind speed, wind direction, temperature, etc.) and historical wind speed data. Currently, short-term wind speed predictions have been studied in some cases, but most of them are directed to wind speeds in terrestrial environments and are not suitable for use offshore. A few researches are performed on predicting the offshore wind speed, but the researches or the data sets of the overseas wind farms are used, so that the prediction accuracy is not high, and the actual engineering application requirements of the overseas wind farms in China are difficult to meet. Therefore, how to accurately predict the wind speed of an offshore wind farm is a problem that needs to be solved by the person skilled in the art.
Disclosure of Invention
The invention aims to provide a method for predicting the short-term wind speed of an offshore wind farm based on characteristic deconstruction. According to the method and the device, more time sequence information can be extracted from the original wind speed signal, so that the accuracy of wind speed prediction is improved.
The technical scheme of the invention is as follows: a method for predicting the short-term wind speed of an offshore wind farm based on characteristic deconstruction comprises the following steps:
s1: obtaining an actually measured wind speed signal through a wind measuring tower of the offshore wind farm, and taking the actually measured wind speed signal as sample data;
s2: deconstructing the sample data into an IMF component and a residual sequence by a variational modal decomposition algorithm;
s3: converting all IMF components and residual sequences into a target data format through normalization and sliding window slicing processing, and obtaining a training sample set and a test sample set;
s4: building a GRU-ARIMA model, inputting a training sample set into the GRU-ARIMA model for training until the GRU-ARIMA model converges, and storing model parameters of the GRU-ARIMA model;
s5: predicting the values of the IMF component and the residual sequence through a trained GRU-ARIMA model;
s6: and integrating all the predicted results and obtaining a wind speed predicted value.
In the above method for predicting the short-term wind speed of the offshore wind farm based on the feature deconstruction, the step S2 includes the following steps:
s2.1: decomposing the actually measured wind speed signal into an intrinsic mode IMF component, and extracting signal frequency domain characteristics, wherein the constraint variation expression is as follows:
the expression of the intrinsic mode IMF component is as follows:
where k is the number of decomposed intrinsic mode IMF components, { u k }＝{u 1 ,...,u k The k intrinsic mode IMF components are represented, { w k }＝{w 1 ,...,w k The central frequency of the IMF component of the natural mode is shown, delta (t) is a dirichlet function, the convolution operation is shown, t is a time sequence, and a is shown in the specification k (t) is a non-negative envelope,for the phase +.>Representing partial derivatives of time t, K representing the total modal number, j representing the imaginary number in the Fourier transform process;
s2.2: introducing a quadratic term punishment factor alpha and Lagrange multiplication operator to convert the constraint variation problem into an unconstrained variation problem, wherein the Lagrange expression is as follows:
where f (t) represents the original signal and λ (t) represents the lagrangian multiplier.
In the above method for predicting the short-term wind speed of the offshore wind farm based on the feature deconstruction, in the step S3, the normalization is maximum and minimum normalization, and the normalized expression is:
wherein X is max X is the maximum value of the sample data min X is the minimum value of the sample data norm To normalize the result, X norm The numerical interval of (2) is [0,1 ]]。
In the above method for predicting the short-term wind speed of the offshore wind farm based on the feature deconstruction, in S3, the sliding window slice includes: the sliding window divides the data of the IMF component into input window data and output window data, wherein the number of steps of the sliding window is the number of data needing to be predicted at different time points.
In the above-mentioned method for predicting the short-term wind speed of the offshore wind farm based on feature deconstruction, in the step S4, the training sample set includes an IMF component training sample set and a residual sequence training sample set;
the step S4 comprises the following steps:
s4.1: building a GRU model, inputting an IMF component training sample set into the GRU model for training until the GRU model converges, and storing parameters of the GRU model;
s4.2: and constructing an ARIMA model, inputting a residual sequence training sample set into the ARIMA model for training until the ARIMA model converges, and storing parameters of the ARIMA model.
In the above method for predicting the short-term wind speed of the offshore wind farm based on characteristic deconstruction, the expression of the GRU model is as follows:
wherein z is t 、r t 、h t Respectively corresponding to an update gate, a reset gate, a candidate state at the current moment and a hidden state at the current moment in the GRU network unit, and x t H is the input variable at the current moment t-1 Is in a history state, r t To reset the current state of the gate, h t-1 Is the output of the last moment. W (W) z 、W r W is a training parameter inside the GRU model, and sigma is a nonlinear activation function.
In the above method for predicting the short-term wind speed of the offshore wind farm based on feature deconstruction, the types of the ARIMA model include an autoregressive model, a moving average model and an autoregressive moving average model, wherein the expression of the autoregressive model is as follows:
the expression of the moving average model is:
the expression of the autoregressive moving average model is as follows:
wherein i represents the ith data before the predicted data, j represents the jth residual quantity, t represents the predicted time, p and q are model training parameters, p measured data and q residual quantities before the predicted data point are taken are represented, { epsilon }, and t and is the term of deviation(s),as a parameter of an autoregressive model, θ 1 、θ 2 ···θ q Is a parameter of the moving average model.
In the above method for predicting the short-term wind speed of the offshore wind farm based on the feature deconstruction, the step S4.1 includes: respectively constructing a GRU model for each IMF component, respectively inputting each IMF component training sample set which is sliced by a sliding window into a corresponding GRU model, and storing parameters of each GRU model after model convergence;
the S4.2 includes: and (3) independently constructing an ARIMA model for the residual sequence components, inputting a residual sequence training sample set which is sliced by a sliding window into the ARIMA model, and storing parameters of the ARIMA model after model convergence.
In the above method for predicting the short-term wind speed of the offshore wind farm based on the feature deconstruction, the step S5 includes the following steps:
s5.1: inputting test sample set data of each IMF component into a corresponding trained GRU model, wherein the output result of each GRU model is the predicted value of each IMF component;
s5.2: and inputting test sample set data of the residual sequence into a trained ARIMA model, wherein an output result of the ARIMA model is a predicted value of the residual sequence.
In the above-mentioned method for predicting the short-term wind speed of the offshore wind farm based on the feature deconstruction, the calculation formula of S6 is as follows:
V p ＝V 1 +V 2 +…+V m +V m+1 ；
wherein V is p For final wind speed prediction value, { V m Is the predicted value of each IMF component, V m+1 Is a predicted value of the residual sequence.
Compared with the prior art, the invention has the following beneficial effects:
1. according to the invention, the sample data is deconstructed into the IMF component and the residual sequence through a variation modal decomposition algorithm, so that the time sequence prediction model is beneficial to extracting more time sequence information from the time sequence prediction model, and the prediction accuracy is improved; moreover, the invention makes accurate prediction on each IMF component with strong time sequence characteristic based on the GRU cyclic neural network, and can realize better prediction effect by constructing an ARIMA model for residual sequences with weak time sequence but certain stability, thereby further improving the accuracy of wind speed information prediction.
2. The method utilizes different points of the variation modal decomposition algorithm, the GRU and the ARIMA algorithm, and combines the algorithms to obtain a better short-term wind speed prediction effect, improves the prediction accuracy and provides accurate and reliable basis for power grid scheduling.
Drawings
FIG. 1 is a flow chart of a predictive method of the present invention;
FIG. 2 is a VMD3 layer exploded effects diagram of the present invention;
FIG. 3 is a VMD4 layer exploded effects diagram of the present invention;
FIG. 4 is a schematic view of a data sliding window slice of the present invention;
FIG. 5 is a schematic diagram of the GRU network element structure of the invention;
FIG. 6 is a schematic diagram of the predictive effect of components in the present invention;
FIG. 7 is a schematic diagram showing the effect of residual sequence prediction in the present invention;
FIG. 8 is a graph showing the predicted results of low wind speed value data between cut-in and cut-out wind speeds in accordance with the present invention;
FIG. 9 is a schematic representation of the result of predicting low wind speed value data in accordance with the present invention;
FIG. 10 is a graph showing the prediction results of high wind speed data between cut-in and cut-out wind speeds in the present invention.
Detailed Description
The invention is further illustrated by the following figures and examples, which are not intended to be limiting.
Examples: a method for predicting the short-term wind speed of an offshore wind farm based on characteristic deconstruction is shown in the attached figure 1, and comprises the following steps:
s1: obtaining an actually measured wind speed signal through a wind measuring tower of the offshore wind farm, and taking the actually measured wind speed signal as sample data; specifically, the wind speed signal collected in the embodiment is measured wind speed data of a domestic offshore wind farm.
S2: deconstructing the sample data into an IMF component and a residual sequence by a variational modal decomposition algorithm; the variational modal decomposition is to adopt a VMD method in a vmdpy library in python to carry out modal decomposition on the acquired data. The VMD is a new self-adaptive, completely non-recursive mode variation and signal processing method, can effectively avoid the influence of signal length selection on a decomposition result, and the decomposition process is essentially a process for solving the constraint variation problem to obtain an optimal solution. The step S2 comprises the following steps:
s2.1: decomposing the actually measured wind speed signal into IMF components of inherent modes, and extracting signal frequency domain characteristics, wherein the actually measured wind speed signal is a one-dimensional signal, the constraint condition is that the sum of estimated bandwidths of all modes is minimum, the sum of all modes is equal to the original signal, and the constraint variation expression is:
the expression of the intrinsic mode IMF component is as follows:
where k is the number of decomposed intrinsic mode IMF components, { u k }＝{u 1 ,…,u k The k intrinsic mode IMF components are represented, { w k }＝{w 1 ,…,w k The central frequency of the IMF component of the natural mode is shown, delta (t) is a dirichlet function, the convolution operation is shown, t is a time sequence, and a is shown in the specification k (t) is a non-negative envelope,for the phase +.>Representing partial derivatives of time t, K representing the total modal number, j representing the imaginary number in the Fourier transform process;
s2.2: introducing a quadratic term punishment factor alpha and Lagrange multiplication operator to convert the constraint variation problem into an unconstrained variation problem, wherein the Lagrange expression is as follows:
where f (t) represents the original signal and λ (t) represents the lagrangian multiplier. When the decomposition operation of the variation modes is carried out, the number k of the decomposition modes and the bandwidth limit a are required to be defined, the k takes a value according to the actual situation, and the empirical value of a is 1.5-2.0 times of the length of the sample data. Decomposing the actually measured wind speed signal by using a variation modal decomposition algorithm, and decomposing one original signal into a plurality of signals with different center frequencies under the condition that the original signal is not lost. The effect of the original signal decomposed into 3 layers and 4 layers by VMD is shown in fig. 2 and 3, respectively. When the number of decomposition layers is 3, the frequency of the component 3 is high, and the amplitude is small; when the number of decomposition layers is 4, both the component 3 and the component 4 have the characteristics of high frequency and small amplitude. Components with high frequency and small amplitude have weak effect on improving the overall prediction accuracy, and more components mean more operation resources are required to be consumed. Therefore, the number of VMD optimal decomposition modes selected in this embodiment is 3.
S3: converting all IMF components and residual sequences into a target data format through normalization and sliding window slicing processing, and obtaining a training sample set and a test sample set; in the step S3, the normalization is maximum and minimum normalization, and the normalized expression is:
wherein X is max X is the maximum value of the sample data min X is the minimum value of the sample data norm To normalize the result, X norm The numerical interval of (2) is [0,1 ]]. The sliding window slice includes: the sliding window divides the IMF component data into input window data and output window data, wherein the step number of the sliding window is the number of data needing to be predicted at different time points。
Further, regarding the specific operation of the data sliding window slice: each component obtained through VMD decomposition needs to be respectively input into an independent time sequence prediction model for training. As shown in fig. 4, assuming that the values of n time points (including t) after t time points need to be predicted, the length of the history data of the input model is k, the data is divided into input x and output y by using a sliding window, and in order to avoid overlapping of the predicted values, the number of sliding steps is set to n.
In general, the larger the k value, the better the prediction effect, but the larger the calculation amount is required. In order to reduce the calculation amount as much as possible while ensuring the prediction accuracy, a loss curve (control n is the same) varying with the k value may be established with the k value as the x-axis and the prediction result mean square value as the y-axis. When the curvature of the decreasing curve of the loss function reaches the inflection point, the k value is determined. And finally, dividing the data set, taking 70% of data as a training sample set and 30% of data as a test sample set. The training sample set is used to train the time series prediction model, and the test sample set is used to evaluate the prediction accuracy of the model built. Furthermore, in order to improve the training effect of the model, the training sample set needs to be shuffled again and randomly disturbed.
S4: building a GRU-ARIMA model, inputting a training sample set into the GRU-ARIMA model for training until the GRU-ARIMA model converges, and storing model parameters of the GRU-ARIMA model; in the step S4, the training sample set includes an IMF component training sample set and a residual sequence training sample set; the step S4 comprises the following steps:
s4.1: according to the GRU circulating neural network principle, a GRU model is built, an IMF component training sample set is input into the GRU model for training until the GRU model converges, and parameters of the GRU model are stored; the specific unit structure of the GRU recurrent neural network is shown in fig. 5, and the expression of the GRU model is as follows:
wherein z is t 、r t 、h t Respectively corresponding to an update gate, a reset gate, a candidate state at the current moment and a hidden state at the current moment in the GRU network unit, and x t H is the input variable at the current moment t-1 Is in a history state, r t To reset the current state of the gate, h t-1 Is the output of the last moment. W (W) z 、W r W is a training parameter in the GRU model, and sigma is a nonlinear activation function; the S4.1 includes: respectively constructing a GRU model for each IMF component, respectively inputting each IMF component training sample set which is sliced by a sliding window into a corresponding GRU model, and storing parameters of each GRU model after model convergence;
s4.2: and constructing an ARIMA model, inputting a residual sequence training sample set into the ARIMA model for training until the ARIMA model converges, and storing parameters of the ARIMA model. Types of the ARIMA model comprise an autoregressive model, a moving average model and an autoregressive moving average model, wherein the autoregressive model has the following expression:
the expression of the moving average model is:
the expression of the autoregressive moving average model is as follows:
wherein i represents the ith data before the predicted data, j represents the jth residual quantity, t represents the predicted time, p and q are model training parameters, p measured data and q residual quantities before the predicted data point are taken are represented, { epsilon }, and t is the deviation } ofThe term "is used to refer to,as a parameter of an autoregressive model, θ 1 、θ 2 ···θ q Is a parameter of the moving average model.
Three parameters (p, d, q) are determined before the ARIMA model is established, and two methods exist for determining the parameters (p, q): firstly, the tail-end and tail-order of the graph by means of the Autocorrelation Coefficients (ACF) and the Partial Autocorrelation Coefficients (PACF), and the red pool information criterion (Akaike Information Criterion, AIC), the Bayesian information criterion (Bayesian Information Criterion, BIC); and secondly, carrying out parameter iteration by combining with an information criterion, and searching an optimal value. The differential order d is determined by ADF (Augmented Dickey-Fuller test), typically d.ltoreq.3.
Firstly, the stability of the residual sequence is checked by using an ADF stability detection method, and the differential order when the residual sequence is found to be stable is verified to be 0, namely, the determination parameter d=0. Parameters (p, q) are then determined using a grid search cross-validation method, a decision is made using information criteria BIC, and parameter adjustment ranges are empirically set. The final parameters are shown in table 1.
TABLE 1 ARIMA parameters
The S4.2 includes: and (3) independently constructing an ARIMA model for the residual sequence components, inputting a residual sequence training sample set which is sliced by a sliding window into the ARIMA model, and storing parameters of the ARIMA model after model convergence. It is noted that in S4.2, the residual sequence component for which the ARIMA model is built is a residual sequence component that is poorly time-sequential but has some stationarity.
S5: predicting the values of the IMF component and the residual sequence through a trained GRU-ARIMA model; the step S5 comprises the following steps:
s5.1: inputting test sample set data of each IMF component into a corresponding trained GRU model, wherein the output result of each GRU model is the predicted value of each IMF component; as shown in fig. 6, the result of the GRU model prediction for each IMF component for one hour in the future is shown. Because the IMF3 component has higher frequency, only 100 predicted points are taken in the figure to intuitively reflect the predicted effect.
S5.2: and inputting test sample set data of the residual sequence into a trained ARIMA model, wherein an output result of the ARIMA model is a predicted value of the residual sequence. As shown in fig. 7, the prediction result of the residual sequence is one hour in the future. As can be seen from the figure, compared with the GRU model, the ARIMA model has significantly improved prediction accuracy for the residual sequence.
Furthermore, the evaluation index of the GRU-ARIMA model refers to the file of China energy industry Standard NB/T10205-2019 wind Power prediction technology Specification implemented in 2019, and root mean square error E is adopted in the file RMSE Correlation coefficient R and accuracy R 1 And percent of pass r 2 As an evaluation index of the model performance. Wherein:
wherein V is M,t For the actual wind speed at time t, V P,t For the predicted wind speed at time t,for predicting the average value of the actual wind speed corresponding to the result sample in the period of time,/for the prediction>Is the average value of the prediction results in the prediction time period. Root mean square error E RMSE And accuracy r 1 The correlation coefficient R is used for reflecting the degree of close correlation between the predicted wind speed and the actual wind speed, and the closer R is to 1, the closer the correlation is. Yield r 2 For determining the proportion of the prediction result meeting the precision requirement, and the accuracy r 1 And more than or equal to 85 percent of the products are qualified.
S6: integrating all the prediction results and obtaining a wind speed prediction value, wherein the calculation formula of the S6 is as follows:
V p ＝V 1 +V 2 +…+V m +V m+1 ；
wherein V is p For final wind speed prediction value, { V m Is the predicted value of each IMF component, V m+1 Is a predicted value of the residual sequence.
And evaluating the predicted result by using the performance evaluation index. In this embodiment, the fan parameters of the offshore wind farm are: the cut-in wind speed is 3m/s, and the cut-out wind speed is 25m/s. When the wind speed is lower than the cut-in wind speed, the fan is in a state to be started; when the wind speed is higher than the cut-out wind speed, the blade angle of the fan is changed into 90 degrees, at the moment, the fan blades are parallel to the wind direction, the fan cannot capture wind energy, and the fan is in a stop state; when the actual wind speed value is between the cut-in wind speed and the cut-out wind speed, the fan is in a normal power generation state. And when the wind speed value is smaller, an infinite value is easy to appear because the actual value is positioned at the denominator position, so that the evaluation index is abnormal. Taking the above conditions into consideration comprehensively, the prediction performance evaluation of the prediction result of the actual wind speed value between the cut-in wind speed and the cut-out wind speed is more in line with the actual conditions. The evaluation results of the different methods are shown in table 2:
table 2 comparison of experimental results of different models
As can be seen from Table 2, compared with the method of directly predicting wind speed by using the GRU model, when the VMD is decomposed and then GRU is used for prediction, the effects of different time periods are obviously improved, and the prediction effect of the VMD-GRU-ARIMA model is obviously improved on the basis of the VMD-GRU model. With the increase of the prediction duration, the prediction effect of the GRU model is obviously reduced, and after VMD decomposition, the prediction effect of the model is slowly reduced. When the prediction time is 60 minutes, the root mean square error E of the VMD-GRU-ARIMA model prediction result RMSE Is 0.101, the correlation coefficient R is 0.977, and the accuracy R 1 Is 89.9 percent of pass rate r 2 90.7% root mean square error E compared to GRU model RMSE The correlation coefficient R and the accuracy R are reduced by 0.095 1 And percent of pass r 2 The lifting rates are respectively 0.049%, 10.5% and 18.7%. Root mean square error E of GRU model prediction result when prediction time length increases from 30 minutes to 150 minutes RMSE Improves the correlation coefficient R and the accuracy R by 0.086 1 And percent of pass r 2 Reduced by 0.071, 8.6% and 17.4%, respectively, and the root mean square error E of the VMD-GRU-ARIMA model prediction result RMSE Only improves the correlation coefficient R and the accuracy R by 0.027 1 And percent of pass r 2 The reduction of 0.012, 2.7% and 6.6% is achieved respectively.
As shown in FIGS. 8-10, the data in the figures do not cull the data below the cut-in wind speed and above the cut-out wind speed for comparison of the results of the different models to predict the future one hour offshore wind speed. Fig. 8, 9 and 10 are three pieces of data randomly taken from the prediction results, in fig. 8, low wind speed data between cut-in and cut-out wind speeds is mainly used, in fig. 9, high wind speed data between cut-in and cut-out wind speeds is mainly used, and in fig. 10. As can be seen from the figure, the VMD-GRU-ARIMA prediction model can obtain good prediction effect on the data in each wind speed section, and is obviously superior to the GRU model and the VMD-GRU model.
The method for predicting the short-term wind speed of the offshore wind farm based on the variation modal decomposition and the GRU-ARIMA has practical application value. The prediction accuracy of wind speed can be effectively improved by deconstructing the original signal characteristics by using a VMD (virtual machine decomposition) algorithm, and the prediction effect of the model can be further improved by using an ARIMA algorithm prediction residual. For short-term wind speed prediction within one hour, the accuracy and qualification rate of the VMD-GRU-ARIMA model prediction result can reach more than 90%, and the requirements in the file of China energy industry standard NB/T10205-2019 wind power prediction technology regulation are met.
The above description is only of the preferred embodiments of the present invention and is not intended to limit the present invention, but various modifications and variations can be made to the present invention by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the scope of the claims of the present invention should be included in the scope of the present invention.
Claims (10)
1. A method for predicting the short-term wind speed of an offshore wind farm based on characteristic deconstruction is characterized by comprising the following steps: the method comprises the following steps:
s1: obtaining an actually measured wind speed signal through a wind measuring tower of the offshore wind farm, and taking the actually measured wind speed signal as sample data;
s2: deconstructing the sample data into an IMF component and a residual sequence by a variational modal decomposition algorithm;
s3: converting all IMF components and residual sequences into a target data format through normalization and sliding window slicing processing, and obtaining a training sample set and a test sample set;
s4: building a GRU-ARIMA model, inputting a training sample set into the GRU-ARIMA model for training until the GRU-ARIMA model converges, and storing model parameters of the GRU-ARIMA model;
s5: predicting the values of the IMF component and the residual sequence through a trained GRU-ARIMA model;
s6: and integrating all the predicted results and obtaining a wind speed predicted value.
2. The method for predicting the short-term wind speed of an offshore wind farm based on characteristic deconstruction according to claim 1, wherein: the step S2 comprises the following steps:
s2.1: decomposing the actually measured wind speed signal into an intrinsic mode IMF component, and extracting signal frequency domain characteristics, wherein the constraint variation expression is as follows:
the expression of the intrinsic mode IMF component is as follows:
where k is the number of decomposed intrinsic mode IMF components, { u k }＝{u 1 ,...,u k The k intrinsic mode IMF components are represented, { w k }＝{w 1 ,…,w k The central frequency of the IMF component of the natural mode is shown, delta (t) is a dirichlet function, the convolution operation is shown, t is a time sequence, and a is shown in the specification k (t) is a non-negative envelope,for the phase +.>Representing partial derivatives of time t, K representing the total modal number, j representing the imaginary number in the Fourier transform process;
s2.2: introducing a quadratic term punishment factor alpha and Lagrange multiplication operator to convert the constraint variation problem into an unconstrained variation problem, wherein the Lagrange expression is as follows:
where f (t) represents the original signal and λ (t) represents the lagrangian multiplier.
3. The method for predicting the short-term wind speed of an offshore wind farm based on characteristic deconstruction according to claim 1, wherein: in the step S3, the normalization is maximum and minimum normalization, and the normalized expression is:
wherein X is max X is the maximum value of the sample data min X is the minimum value of the sample data norm To normalize the result, X norm The numerical interval of (2) is [0,1 ]]。
4. The method for predicting the short-term wind speed of an offshore wind farm based on characteristic deconstruction according to claim 1, wherein: in the S3, the sliding window slice includes: the sliding window divides the data of the IMF component into input window data and output window data, wherein the number of steps of the sliding window is the number of data needing to be predicted at different time points.
5. The method for predicting the short-term wind speed of an offshore wind farm based on characteristic deconstruction according to claim 1, wherein: in the step S4, the training sample set includes an IMF component training sample set and a residual sequence training sample set;
the step S4 comprises the following steps:
s4.1: building a GRU model, inputting an IMF component training sample set into the GRU model for training until the GRU model converges, and storing parameters of the GRU model;
s4.2: and constructing an ARIMA model, inputting a residual sequence training sample set into the ARIMA model for training until the ARIMA model converges, and storing parameters of the ARIMA model.
6. The method for predicting the short-term wind speed of an offshore wind farm based on characteristic deconstruction according to claim 5, wherein: the expression of the GRU model is as follows:
wherein z is t 、r t 、h t Respectively corresponding to an update gate, a reset gate, a candidate state at the current moment and a hidden state at the current moment in the GRU network unit, and x t H is the input variable at the current moment t-1 Is in a history state, r t To reset the current state of the gate, h t-1 Is the output of the last moment. W (W) z 、W r W is a training parameter inside the GRU model, and sigma is a nonlinear activation function.
7. The method for predicting the short-term wind speed of an offshore wind farm based on characteristic deconstruction according to claim 5, wherein: types of the ARIMA model comprise an autoregressive model, a moving average model and an autoregressive moving average model, wherein the autoregressive model has the following expression:
the expression of the moving average model is:
the expression of the autoregressive moving average model is as follows:
wherein i represents the ith data before the predicted data, j represents the jth residual quantity, t represents the predicted time, p and q are model training parameters, p measured data and q residual quantities before the predicted data point are taken are represented, { epsilon }, and t and is the term of deviation(s),as a parameter of an autoregressive model, θ 1 、θ 2 ···θ q Is a parameter of the moving average model.
8. The method for predicting the short-term wind speed of an offshore wind farm based on characteristic deconstruction according to claim 5, wherein: the S4.1 includes: respectively constructing a GRU model for each IMF component, respectively inputting each IMF component training sample set which is sliced by a sliding window into a corresponding GRU model, and storing parameters of each GRU model after model convergence;
the S4.2 includes: and (3) independently constructing an ARIMA model for the residual sequence components, inputting a residual sequence training sample set which is sliced by a sliding window into the ARIMA model, and storing parameters of the ARIMA model after model convergence.
9. The method for predicting the short-term wind speed of an offshore wind farm based on characteristic deconstruction according to claim 5, wherein: the step S5 comprises the following steps:
s5.1: inputting test sample set data of each IMF component into a corresponding trained GRU model, wherein the output result of each GRU model is the predicted value of each IMF component;
s5.2: and inputting test sample set data of the residual sequence into a trained ARIMA model, wherein an output result of the ARIMA model is a predicted value of the residual sequence.
10. The method for predicting the short-term wind speed of an offshore wind farm based on characteristic deconstruction according to claim 1, wherein: the calculation formula of the S6 is as follows:
V p ＝V 1 +V 2 +…+V m +V m+1 ；
wherein V is p For final wind speed prediction value, { V m Is the predicted value of each IMF component, V m+1 Is a predicted value of the residual sequence.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310046981.2A CN117150705A (en) | 2023-01-31 | 2023-01-31 | Marine wind farm short-term wind speed prediction method based on characteristic deconstruction |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310046981.2A CN117150705A (en) | 2023-01-31 | 2023-01-31 | Marine wind farm short-term wind speed prediction method based on characteristic deconstruction |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117150705A true CN117150705A (en) | 2023-12-01 |
Family
ID=88885545
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310046981.2A Pending CN117150705A (en) | 2023-01-31 | 2023-01-31 | Marine wind farm short-term wind speed prediction method based on characteristic deconstruction |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117150705A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117748501A (en) * | 2024-02-19 | 2024-03-22 | 西安热工研究院有限公司 | Wind power prediction method and system for energy storage auxiliary black start |
-
2023
- 2023-01-31 CN CN202310046981.2A patent/CN117150705A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117748501A (en) * | 2024-02-19 | 2024-03-22 | 西安热工研究院有限公司 | Wind power prediction method and system for energy storage auxiliary black start |
| CN117748501B (en) * | 2024-02-19 | 2024-05-07 | 西安热工研究院有限公司 | Wind power prediction method and system for energy storage auxiliary black start |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109299430A (en) | The short-term wind speed forecasting method with extreme learning machine is decomposed based on two stages | |
| CN111695290A (en) | Short-term runoff intelligent forecasting hybrid model method suitable for variable environment | |
| CN104573876A (en) | Wind power plant short-period wind speed prediction method based on time sequence long memory model | |
| CN110168220A (en) | A kind of method and system for assessing wind turbine generator performance | |
| CN113822418A (en) | Wind power plant power prediction method, system, device and storage medium | |
| CN115013258B (en) | Intelligent soft measurement method for wind speed in front of wind turbine generator | |
| CN117150705A (en) | Marine wind farm short-term wind speed prediction method based on characteristic deconstruction | |
| CN111931983A (en) | Precipitation prediction method and system | |
| CN111159900B (en) | Method and device for predicting wind speed of fan | |
| CN115511657A (en) | Wind power output and photovoltaic output evaluation method based on combined prediction model | |
| CN109919382A (en) | Consider the non-precision probability forecasting method of photovoltaic power climbing event and system that date periodicity influences | |
| CN116029465A (en) | Method, device, medium and equipment for predicting wind power before day in turning weather | |
| Wang et al. | A novel wind power forecasting system integrating time series refining, nonlinear multi-objective optimized deep learning and linear error correction | |
| CN109886488B (en) | Distributed wind power plant layered hybrid short-term prediction method considering wind speed time lag | |
| Reddy et al. | Hybrid approach for short term wind power forecasting | |
| CN114897204A (en) | Method and device for predicting short-term wind speed of offshore wind farm | |
| Lai et al. | Multiscale wavelet-driven graph convolutional network for blade icing detection of wind turbines | |
| Silva et al. | Air pressure forecasting for the Mutriku oscillating-water-column wave power plant: Review and case study | |
| CN116885711A (en) | Wind power prediction method, device, equipment and readable storage medium | |
| CN115977874B (en) | Wind turbine generator yaw self-adaptive calibration method and system based on laser wind-finding radar | |
| CN117252288A (en) | Regional resource active support capacity prediction method and system | |
| CN114298300A (en) | Uncertainty prediction method, uncertainty prediction device, electronic device, and storage medium | |
| CN116384586A (en) | Wind speed prediction method based on VMD-SVR | |
| CN115712860A (en) | Fan blade vibration alarm method and device | |
| CN114320770A (en) | Effective wind speed measuring method and device for wind turbine generator |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |