CN116544922A - Wind power probability prediction method based on VMDS-CNN-BGRU model - Google Patents
Wind power probability prediction method based on VMDS-CNN-BGRU model Download PDFInfo
- Publication number
- CN116544922A CN116544922A CN202310533079.3A CN202310533079A CN116544922A CN 116544922 A CN116544922 A CN 116544922A CN 202310533079 A CN202310533079 A CN 202310533079A CN 116544922 A CN116544922 A CN 116544922A
- Authority
- CN
- China
- Prior art keywords
- wind power
- bgru
- model
- cnn
- vmds
- 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 44
- 238000000605 extraction Methods 0.000 claims abstract description 28
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 27
- 238000012549 training Methods 0.000 claims description 9
- 230000008569 process Effects 0.000 claims description 7
- 230000004927 fusion Effects 0.000 claims description 6
- 238000004364 calculation method Methods 0.000 claims description 5
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 4
- 238000012545 processing Methods 0.000 claims description 4
- 230000001419 dependent effect Effects 0.000 claims description 3
- 230000001154 acute effect Effects 0.000 abstract description 2
- 230000006870 function Effects 0.000 description 18
- 238000013527 convolutional neural network Methods 0.000 description 12
- 238000010586 diagram Methods 0.000 description 6
- 238000004458 analytical method Methods 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 5
- 238000013136 deep learning model Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000010801 machine learning Methods 0.000 description 3
- 238000009827 uniform distribution Methods 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 2
- 230000000052 comparative effect Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 239000002803 fossil fuel Substances 0.000 description 2
- 238000010248 power generation Methods 0.000 description 2
- 238000013179 statistical model Methods 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 238000012795 verification Methods 0.000 description 2
- 230000006978 adaptation Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000015556 catabolic process Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000006731 degradation reaction Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 238000005192 partition Methods 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Classifications
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J3/00—Circuit arrangements for ac mains or ac distribution networks
- H02J3/004—Generation forecast, e.g. methods or systems for forecasting future energy generation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/047—Probabilistic or stochastic networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J2300/00—Systems for supplying or distributing electric power characterised by decentralized, dispersed, or local generation
- H02J2300/20—The dispersed energy generation being of renewable origin
- H02J2300/28—The renewable source being wind energy
-
- 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)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Evolutionary Computation (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Power Engineering (AREA)
- Probability & Statistics with Applications (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a wind power probability prediction method based on a VMDS-CNN-BGRU model, which comprises the following steps: constructing a VMDS-CNN-BGRU model formed by sequentially connecting a data decomposition module, a multi-source feature extraction module, a probability prediction module and a prediction result output module; inputting the collected wind power sequence data into a data decomposition module to decompose the wind power sequence data into a plurality of components; inputting the decomposed components and NWP data into a multi-source feature extraction module for feature extraction; inputting the extracted characteristics into a probability prediction module to output wind power predicted values under different quantiles; and inputting the wind power predicted value into a predicted result output module, and obtaining a probability density function of the wind power predicted value to realize wind power probability prediction. By adopting the method, accurate deterministic prediction of wind power, reliable and acute interval prediction and reliable and effective probability prediction can be simultaneously realized.
Description
Technical Field
The invention belongs to the technical field of new energy power generation and intelligent power grids, and particularly relates to a wind power probability prediction method based on a VMDS-CNN-BGRU model.
Background
The limitations of fossil fuels and environmental degradation caused by fossil fuels limit the worldwide growing demand for electricity, and the world is pushing the rapid development of clean energy sources, typified by wind energy. Because of the randomness, volatility and intermittence of wind power, large scale grid connection of wind power presents great uncertainty and risk to the supply side of the power system. Therefore, accurate and reliable wind power prediction has important significance for safe operation of a power system and utilization of wind energy.
The wind power prediction method is mainly divided into deterministic prediction and probabilistic prediction, wherein the deterministic prediction is realized by a point prediction model, and the model has three types: statistical models, machine learning models, and deep learning models. Many statistical models have been applied to wind power predictions, such as kalman filters, autoregressive moving averages, autoregressive integrated averages, and the like. Common machine learning models include artificial neural networks, support vector machines, and the like. The deep learning model has a stronger nonlinear mapping capability than the statistical and machine learning models. With the development of deep learning techniques, some deep neural network models have been applied to renewable energy prediction, including Deep Belief Networks (DBNs), convolutional neural networks, recurrent neural networks, and the like. Due to problems of gradient dissipation and gradient explosion, the general RNN model is rarely used for predicting wind power generation, and its variant models LSTM and GRU solve the above problems and are widely used.
The above models are all point prediction models, and the obtained future wind power certainty value cannot describe the wind power uncertainty reliably. When the wind power fluctuation is large, the reliability of the point prediction result may be low, and the actual scheduling requirement cannot be satisfied.
Disclosure of Invention
Aiming at the problems existing in the prior art, the invention provides a wind power probability prediction method based on a VMDS-CNN-BGRU model, which ensures the accuracy of deterministic prediction and simultaneously realizes interval prediction with high reliability and acuity and reliable and effective probability prediction.
In order to achieve the technical purpose, the invention adopts the following technical scheme: a wind power probability prediction method based on a VMDS-CNN-BGRU model specifically comprises the following steps:
step 1, constructing a VMDS-CNN-BGRU model, wherein the VMDS-CNN-BGRU model is formed by sequentially connecting a data decomposition module, a multi-source feature extraction module, a probability prediction module and a prediction result output module;
step 2, collecting wind power sequence data and NWP data of a wind power plant;
step 3, inputting the collected wind power sequence data into a data decomposition module to decompose the wind power sequence data into a plurality of components;
step 4, inputting the decomposed components and NWP data into a multi-source feature extraction module for feature extraction;
step 5, inputting the extracted characteristics into a probability prediction module to output wind power predicted values under different quantiles;
and step 6, inputting the wind power predicted value into a predicted result output module, and obtaining a probability density function of the wind power predicted value to realize wind power probability prediction.
Further, step 3 comprises the following sub-steps:
step 3.1, inputting the collected wind power sequence data X into a data decomposition module, decomposing the wind power sequence data X into subsequences X through a variation mode decomposition VMD IMF1 ～X IMF5 Wherein X is IMF1 And X IMF2 As low frequency component, X IMF3 ～X IMF5 Is a high frequency component;
step 3.2, extracting high frequency components X using SSA, respectively IMF3 ～X IMF5 Trend component X with time sequence characteristic IMFS1 、X IMFS2 And X IMFS3 。
Further, parameters of the variant modal decomposition VMD are set as follows: penalty parameter α=1000, initial center frequency ω=0, convergence criterion r=10 -6 The method comprises the steps of carrying out a first treatment on the surface of the The parameters of the SSA are set as follows: embedded window length ρ=10.
Further, the multi-source feature extraction module is formed by sequentially connecting two layers of CNN, one layer of convolution layer and two layers of BGRU.
Further, step 4 comprises the following sub-steps:
step 4.1, combining the low frequency component X IMF1 And X IMF2 Inputting into two BGRUs, extracting X IMF Features;
step 4.2, trend component X IMFS1 、X IMFS2 And X IMFS3 X is extracted by the first layer CNN input of the multi-source feature extraction module IMFS Complex dynamic characteristics;
and 4.3, inputting the NWP data by a first layer CNN of the multi-source feature extraction module, and extracting space-time features affecting wind power.
Further, the probability prediction module is formed by sequentially connecting a BGRU layer, a full connection layer and a quantile regression module.
Further, the specific process of step 5 is as follows: x extracted by the feature extraction module IMF Features, X IMFS The complex dynamic characteristics and the space-time characteristics are fused, the complex dynamic characteristics and the space-time characteristics are input into a trained probability prediction model, the BGRU is used for extracting deeper time sequence characteristics, the result is sent to a full-connection layer for processing, and the complex dynamic characteristics and the space-time characteristics are input into a quantile regression module to output predicted values of wind power under different quantiles:
wherein,,predicted value of wind power in quantile τ, +.>For optimal network weights of the probabilistic predictive model at quantiles τ +.>For the optimal deviation of the probabilistic predictive model at quantiles tau,Afor fused features, f () is a nonlinear function that reflects the relationship between the independent and dependent variables in the quantile regression calculation.
Further, the training process of the probability prediction model is as follows: inputting the extracted fusion features into a probability prediction model, and optimizing a quantile loss function by using an Adam gradient descent algorithm when the quantile tau is continuously taken within the range of (0, 1), so as to obtain the optimal network weight under different quantile conditionsAnd optimum deviation->
The fractional loss function is constructed as follows:
wherein m represents the number of samples of each batch of training set, y t Representing the actual value of the wind power,is the wind power predicted value at tau quantiles corresponding to the fusion features, W is the network weight of the probability prediction model, b is the deviation of the probability prediction model, ρ τ To represent a function +.>u is a variable that requires quantile calculation.
Further, the specific process of step 6 is as follows: inputting the predicted values of the wind power under different quantiles into a predicted result output module, and obtaining a probability density function of the wind power predicted value y by adopting KDE kernel density estimationAnd realizing wind power probability prediction, wherein B is bandwidth, N is a plurality of quantiles and K () is a kernel function.
Further, the kernel function is:
compared with the prior art, the invention has the beneficial effects that:
1. the invention designs a combined decomposition technology combining the VMD and the SSA, can effectively reduce the complexity of wind power data through the proposed decomposition technology, has excellent effect and accurate result, and can effectively improve the probability prediction precision of the subsequent wind power;
2. according to the invention, the multisource feature extraction module based on CNN and BGRU is adopted, CNN has excellent feature extraction function, and the complex features of meteorological data and wind power high-frequency components can be fully and effectively extracted by combining the predication performance of BGRU on time sequence sequences, so that the predication performance of wind power is effectively improved;
3. the invention realizes the probability interval prediction of wind power, not only can provide a deterministic prediction value of wind power, but also can provide an interval prediction result under certain confidence level, and can contain more information;
4. the invention realizes the probability prediction of the predicted observation points, combines quantile regression with kernel density estimation, can obtain the probability density curve of each observation point, realizes the probability prediction, and ensures the reliability of the probability prediction.
Drawings
FIG. 1 is a diagram of a VMDS-CNN-BGRU model of the present invention;
FIG. 2 is a diagram showing the original wind power sequence and VMDS decomposition results according to the present invention;
FIG. 3 is a PDF curve of the VMDS-CNN-BGRU model of the present invention at different times;
FIG. 4 is a graph showing the probability prediction result of the VMDS-CNN-BGRU model of the present invention;
FIG. 5 is a graph showing the prediction result of the VMDS-CNN-BGRU model of the present invention;
FIG. 6 is a schematic diagram of the interval prediction result of the comparative model VMDS-QR-GRU of the present invention;
FIG. 7 is a schematic diagram of the interval prediction result of the comparative model QRGRU of the present invention;
FIG. 8 is a schematic diagram of the point prediction results of the prediction model of the present invention compared with different models.
Detailed Description
The technical scheme of the invention is further described below with reference to the accompanying drawings. The present invention is applicable to other fields and fields such as load and photovoltaic output, and the like.
The invention provides a wind power probability prediction method based on a VMDS-CNN-BGRU model, which specifically comprises the following steps:
step 1, constructing a VMDS-CNN-BGRU model, wherein as shown in fig. 1, the VMDS-CNN-BGRU model is formed by sequentially connecting a data decomposition module, a multi-source feature extraction module, a probability prediction module and a prediction result output module;
step 2, collecting wind power sequence data and NWP data of a wind power plant;
step 3, inputting the collected wind power sequence data into a data decomposition module to decompose the wind power sequence data into a plurality of components; the method specifically comprises the following substeps:
step 3.1, inputting the collected wind power sequence data X into a data decomposition module, decomposing the wind power sequence data X into subsequences X through a variation mode decomposition VMD IMF1 ～X IMF5 Wherein X is IMF1 And X IMF2 As low frequency component, X IMF3 ～X IMF5 For high frequency divisionAn amount of;
step 3.2, extracting high frequency components X using SSA, respectively IMF3 ～X IMF5 Trend component X with time sequence characteristic IMFS1 、X IMFS2 And X IMFS3 The complexity of the wind power sequence data is reduced, and the time sequence characteristics in the wind power sequence data are highlighted.
The parameters of the variant mode decomposition VMD in the invention are set as follows: penalty parameter α=1000, initial center frequency ω=0, convergence criterion r=10 -6 The method comprises the steps of carrying out a first treatment on the surface of the The parameters of SSA are set as: embedded window length ρ=10. As shown in fig. 2, the wind power sequence is VMDS processed to obtain two low frequency components and three high frequency components, where IMF1 through IMF5 are the results of VMD and IMFs1 through IMFs3 are the results of SSA further processing.
Step 4, inputting the decomposed components and NWP data into a multi-source feature extraction module for feature extraction; the multisource feature extraction module is formed by sequentially connecting two layers of CNN, one layer of convolution layer and two layers of BGRU, wherein the CNN layer is used for extracting high-order features in data; the BGRU is combined with two GRU networks with opposite time sequences to fully acquire hidden information before and after the current unit, and further excavates time sequence characteristics. The method specifically comprises the following substeps:
step 4.1, combining the low frequency component X IMF1 And X IMF2 Inputting into two BGRUs, extracting X IMF Characterised by X IMF The characteristic fluctuation frequency is low, and the trend is obvious, so that two layers of BGRUs are used for extracting X IMF Is characterized by (2);
step 4.2, trend component X IMFS1 、X IMFS2 And X IMFS3 X is extracted by the first layer CNN input of the multi-source feature extraction module IMFS Complex dynamic characteristics due to X IMFS The fluctuation frequency is high, and the trend is not obvious, so the feature extraction module combined with CNN and BGRU is used for extracting X IMFS Complex dynamic characteristics of (a);
step 4.3, X NWP The NWP data is input by the first layer CNN of the multi-source feature extraction module to extract spatio-temporal features affecting wind power, including many non-linear and low correlation factors.
Step 5,The point prediction model generally only obtains a prediction expected value, namely, realizes deterministic prediction, and cannot realize interval prediction and probability prediction. Based on the quantile regression (quantile regression, QR) principle, a mixed model combining the point prediction model and the QR can be realized to obtain the predicted value of the point prediction model under different quantile conditions, so that interval prediction and probability prediction are realized; inputting the extracted characteristics into a probability prediction module to output wind power predicted values under different quantiles; the probability prediction module is formed by sequentially connecting a BGRU layer, a full connection layer and a quantile regression module. Specifically, X extracted by the feature extraction module IMF Features, X IMFS The complex dynamic characteristics and the space-time characteristics are fused, the complex dynamic characteristics and the space-time characteristics are input into a trained probability prediction model, the BGRU is used for extracting deeper time sequence characteristics, the result is sent to a full-connection layer for processing, and the complex dynamic characteristics and the space-time characteristics are input into a quantile regression module to output predicted values of wind power under different quantiles:
wherein,,predicted value of wind power in quantile τ, +.>For optimal network weights of the probabilistic predictive model at quantiles τ +.>For the optimal deviation of the probabilistic predictive model at quantiles tau,Afor fused features, f () is a nonlinear function that reflects the relationship between the independent and dependent variables in the quantile regression calculation.
The training process of the probability prediction model in the invention is as follows: inputting the extracted fusion features into a probability prediction model, and optimizing a quantile loss function by using an Adam gradient descent algorithm when the quantile tau is continuously taken within the range of (0, 1) to obtainOptimal network weight under different quantile conditionsAnd optimum deviation->
The fractional loss function of the invention is constructed as follows:
wherein m represents the number of samples of each batch of training set, y t Representing the actual value of the wind power,is the wind power predicted value at tau quantiles corresponding to the fusion features, W is the network weight of the probability prediction model, b is the deviation of the probability prediction model, ρ τ To represent a function +.>u is a variable that requires quantile calculation.
Step 6, inputting the wind power predicted value into a predicted result output module, and obtaining a probability density function of the wind power predicted value to realize wind power probability prediction; specifically, the prediction model of QR is combined to only obtain the wind power predicted value under each quantile, but not directly obtain the probability density function (probability density function, PDF) of the predicted value. Inputting the predicted values of the wind power under different quantiles into a predicted result output module, and obtaining a probability density function of the wind power predicted value y by adopting KDE kernel density estimationImplementing wind power probability predictionWherein B is bandwidth, N is quantile number, K () is kernel function,
the wind power probability prediction method based on the VMDS-CNN-BGRU model can simultaneously realize accurate deterministic prediction, reliable and sharp interval prediction and reliable and effective probability prediction of wind power.
Examples
The raw data of this embodiment is from a certain wind farm in Jiangsu of China, including 2017 annual wind power sequence data and NWP data. Wind power data is collected every 15 minutes, so there are 96 data points per day. NWP data includes wind speed at different heights, wind direction at different heights, temperature, air pressure, humidity, etc. In this embodiment, the input dimensions of the wind power data and NWP data are set to 10, and the first 80% of each data set is used as a training set, and the second 20% is used as a test set.
In order to verify the effectiveness of the model provided by the invention, several prediction models are provided as comparison models to verify the superiority of the comprehensive prediction performance of the VMDS-CNN-BGRU model. The comparison models are VMD-CNN-BGRU, VMDS-QRGRU and QRCNN-GRU, QRLSTM, QRGRU, wherein the VMD-CNN-BGRU is used for explaining the superiority of VMDS; the VMDS-QRGRU is used for reflecting the effect of the proposed multi-source feature extraction module; QRCNN-GRU, QRLSTM, QRGRU is a common deep learning model that does not use data decomposition techniques to reflect the overall performance of wind power probability prediction methods. The model set up and training configuration of all the comparison models presented in this embodiment are the same: the training round number is 200, the optimizer is Adam, the advanced stop-wait round number is 10, and the verification set ratio is 0.1. 199 quantiles were set, quantile τ= [0.005,0.01,.. 0.99,0.995]. The super parameter settings of the VMDS-CNN-BGRU model are shown in Table 1:
TABLE 1 VMDS-CNN-BGRU Supermarameter setting
In order to verify the prediction effect of the method, the method performs comparison analysis on three aspects of probability prediction, interval prediction and point prediction and a comparison model.
(1) Probability prediction results
The model uses the predicted values of different quantiles obtained by the QR, and the PDF of each observation point can be estimated through the KDE. Fig. 3 shows PDF of six randomly selected observation points of the data set, and fig. 3 shows that most of the actual values are close to the peak of PDF, close to the predicted median, which indicates that the wind power probability prediction method based on VMDS-CNN-BGRU model of the present invention is effective.
Table 2 shows CRPS values of probability prediction results for different models, where CRPS is the smallest in the VMDS-CNN-BGRU model, indicating that the probability prediction overall performance is the highest in the VMDS-CNN-BGRU model.
TABLE 2 CRPS values of probability predictions for different models
| Model | CRPS value |
| VMDS-CNN-BGRU | 0.621 |
| VMD-CNN-BGRU | 0.713 |
| VMDS-BGRU | 0.686 |
| QRCNN-BGRU | 0.696 |
| QRGRU | 0.716 |
| QRCNN | 0.722 |
By calculating a probability integration transformation (probability integral transformation, PIT) of the predicted values, analyzing whether they obey a uniform distribution, the reliability of the probability prediction model can be verified. The partition map (QQ) is used to intuitively analyze whether the PIT values of the prediction model result follow a uniform distribution, fig. 4 is a QQ map of the PIT values of the VMDS-CNN-BGRU probability prediction result, the straight line is a uniform distribution of theoretical cases, and "·" represents the actual probability integral change value, and it can be seen from fig. 4 that the PIT value distribution is in Kolmogorov 5% significant band, indicating that the probability prediction result of VMDS-CNN-BGRU is reliable.
(2) Analysis of interval prediction results
The excellent interval prediction performance requirement improves the predicted acuity as much as possible while guaranteeing the prediction reliability, and the interval prediction error statistics of each model are given in table 3, including evaluation indexes: PINAW, ACE, IS. The following can be concluded:
TABLE 3 comparison of section predictions for different models
| Model | PINAW 95％ | ACE 95％ | IS 95％ |
| VMDS-CNN-BGRU | 0.137 | 1.197 | -0.249 |
| VMD-CNN-BGRU | 0.161 | 1.283 | -0.307 |
| VMDS-BGRU | 0.172 | 2.972 | -0.291 |
| QRCNN-BGRU | 0.189 | 2.297 | -0.321 |
| QRGRU | 0.236 | 2.253 | -0.432 |
| QRCNN | 0.219 | 1.621 | -0.401 |
The QRCNN-GRU, QRLSTM, QRGRU model has a much poorer evaluation index than the model using the data decomposition technique. The ACE value of a model without using a decomposition technology is generally higher than that of the model with using the decomposition technology, but the PINAW value is far lower than that of the model with using the decomposition technology, which indicates that the prediction interval of the model such as QRCNN-GRU is wider, so that the reliability of prediction is improved, but the sensitivity of prediction is reduced, so that the overall performance of interval prediction is reduced; the IS value of the model using the VMD or VMDS IS 0.1 to 0.5 lower than the IS value of the model of QRCNN-GRU or the like. In summary, the interval prediction of VMDS-CNN-BGRU has better performance than the conventional model without data decomposition technique.
The interval prediction performance comparison of the VMDS-CNN-BGRU and the VMD-CNN-BGRU is to verify the effectiveness of the proposed VMDS, and the ACE value of the VMDS-CNN-BGRU is lower than that of the VMD-CNN-BGRU, which indicates that the reliability of the VMD-CNN-BGRU is slightly higher than that of the VMDS-CNN-BGRU. However, the PINAW values of VMDS-CNN-BGRU were 14.9% lower than that of VMD-CNN-BGRU, respectively. The IS values of VMDS-CNN-BGRU are 18.8% higher than VMD-CNN-BGRU, respectively. The above analysis shows that the proposed VMDS method is helpful for improving the prediction performance of the model interval.
The validity of the proposed combination model is verified by comparing the interval prediction performance of the VMDS-CNN-BGRU and the VMDS-QR-GRU: PINAW values of VMDS-CNN-BGRU are 20.3% lower than VMDS-QR-GRU, respectively; in addition, the IS values of VMDS-CNN-BGRU were 14.4% higher than VMDS-QR-GRU, respectively. The analysis shows that the combination method of the dynamic characteristics of the NWP data and the high-frequency trend component extracted by using the CBG is effective.
Fig. 5, fig. 6 and fig. 7 are schematic diagrams of the interval prediction results of VMDS-CNN-BGRU, VMDS-QR-CBG and QRGRU, respectively, wherein the interval width of VMDS-CNN-BGRU is significantly smaller than QRGRU and VMD-QR-GRU.
(3) Deterministic predictive result analysis
Selecting the median of the probability prediction results of each model as the deterministic prediction result of wind power, wherein table 4 shows that the RMSE and NMAPE of VMDS-CNN-BGRU are the lowest, and compared with other models, the RMSE is respectively reduced by 13.6%, 8.5%, 16.1%, 31.4% and 32.2%; NMAPE was reduced by 13.3%, 8.3%, 14.1%, 26.4% and 28.4% compared to the other models, respectively.
Table 4 comparison of point predictions for different models
| Model | RMSE/MW | NMAPE/％ |
| VMDS-CNN-BGRU | 0.492 | 1.940 |
| VMD-CNN-BGRU | 0.570 | 2.240 |
| VMDS-BGRU | 0.538 | 2.116 |
| QRCNN-BGRU | 0.587 | 2.261 |
| QRGRU | 0.718 | 2.639 |
| QRCNN | 0.726 | 2.713 |
FIG. 8 is a graph comparing predicted values of each model with actual wind power, and it can be seen from FIG. 8 that each model can accurately predict the trend of wind power, and the predicted value of the VMDS-CNN-BGRU model is closest to the actual wind power. In conclusion, the VMDS-CNN-BGRU can better ensure accurate deterministic prediction of wind power.
According to the wind power probability prediction method based on the VMDS-CNN-BGRU model, an original wind power sequence is decomposed by using a combined decomposition method VMDS combined with VMD and SSA, so that the data complexity is reduced; extracting complex dynamic characteristics of NWP data and high-frequency components by using a characteristic extractor based on CNN and BGRU; the QR modeling is then performed to obtain predictions for different quantiles. Finally, estimating a probability density curve of the future wind power predicted value by using the KDE. Through verification, the model provided by the invention can simultaneously realize accurate deterministic prediction of wind power, reliable and acute interval prediction and reliable and effective probability prediction.
The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above examples, and all technical solutions belonging to the concept of the present invention belong to the protection scope of the present invention. It should be noted that modifications and adaptations to the invention without departing from the principles thereof are intended to be within the scope of the invention as set forth in the following claims.
Claims (10)
1. A wind power probability prediction method based on a VMDS-CNN-BGRU model is characterized by comprising the following steps:
step 1, constructing a VMDS-CNN-BGRU model, wherein the VMDS-CNN-BGRU model is formed by sequentially connecting a data decomposition module, a multi-source feature extraction module, a probability prediction module and a prediction result output module;
step 2, collecting wind power sequence data and NWP data of a wind power plant;
step 3, inputting the collected wind power sequence data into a data decomposition module to decompose the wind power sequence data into a plurality of components;
step 4, inputting the decomposed components and NWP data into a multi-source feature extraction module for feature extraction;
step 5, inputting the extracted characteristics into a probability prediction module to output wind power predicted values under different quantiles;
and step 6, inputting the wind power predicted value into a predicted result output module, and obtaining a probability density function of the wind power predicted value to realize wind power probability prediction.
2. The method for predicting wind power probability based on VMDS-CNN-BGRU model as claimed in claim 1, wherein the step 3 comprises the following sub-steps:
step 3.1, inputting the collected wind power sequence data X into a data decomposition module, decomposing the wind power sequence data X into subsequences X through a variation mode decomposition VMD IMF1 ～X IMF5 Wherein X is IMF1 And X IMF2 As low frequency component, X IMF3 ～X IMF5 Is a high frequency component;
step 3.2, extracting high frequency components X using SSA, respectively IMF3 ～X IMF5 Trend component X with time sequence characteristic IMFS1 、X IMFS2 And X IMFS3 。
3. The wind power probability prediction method based on the VMDS-CNN-BGRU model according to claim 2, wherein parameters of the variant mode decomposition VMD are set as follows: penalty parameter α=1000, initial center frequency ω=0, convergence criterion r=10 -6 The method comprises the steps of carrying out a first treatment on the surface of the The parameters of the SSA are set as follows: embedded window length ρ=10.
4. The wind power probability prediction method based on the VMDS-CNN-BGRU model of claim 1, wherein the multi-source feature extraction module is composed of two layers of CNN, one layer of convolution layer and two layers of BGRU which are sequentially connected.
5. The method for predicting wind power probability based on VMDS-CNN-BGRU model as claimed in claim 4, wherein step 4 comprises the sub-steps of:
step 4.1, combining the low frequency component X IMF1 And X IMF2 Inputting into two BGRUs, extracting X IMF Features;
step 4.2, trend component X IMFS1 、X IMFS2 And X IMFS3 X is extracted by the first layer CNN input of the multi-source feature extraction module IMFS Complex dynamic characteristics;
and 4.3, inputting the NWP data by a first layer CNN of the multi-source feature extraction module, and extracting space-time features affecting wind power.
6. The wind power probability prediction method based on the VMDS-CNN-BGRU model as claimed in claim 1, wherein the probability prediction module is formed by sequentially connecting a BGRU layer, a full connection layer and a quantile regression module.
7. The method for predicting wind power probability based on VMDS-CNN-BGRU model as claimed in claim 6, wherein the specific process of step 5 is as follows: x extracted by the feature extraction module IMF Features, X IMFS The complex dynamic characteristics and the space-time characteristics are fused, the complex dynamic characteristics and the space-time characteristics are input into a trained probability prediction model, the BGRU is used for extracting deeper time sequence characteristics, the result is sent to a full-connection layer for processing, and the complex dynamic characteristics and the space-time characteristics are input into a quantile regression module to output predicted values of wind power under different quantiles:
wherein,,predicted value of wind power in quantile τ, +.>For optimal network weights of the probabilistic predictive model at quantiles τ +.>For the optimal deviation of the probabilistic predictive model at quantiles tau,Afor fused features, f () is the auto-change reflecting quantile regression calculationA nonlinear function of the relationship between the quantity and the dependent variable.
8. The method for predicting wind power probability based on VMDS-CNN-BGRU model as set forth in claim 7, wherein the training process of the probability prediction model is as follows: inputting the extracted fusion features into a probability prediction model, and optimizing a quantile loss function by using an Adam gradient descent algorithm when the quantile tau is continuously taken within the range of (0, 1), so as to obtain the optimal network weight under different quantile conditionsAnd optimum deviation->
The fractional loss function is constructed as follows:
wherein m represents the number of samples of each batch of training set, y t Representing the actual value of the wind power,is the wind power predicted value at tau quantiles corresponding to the fusion features, W is the network weight of the probability prediction model, b is the deviation of the probability prediction model, ρ τ To represent a function +.>u is a variable that requires quantile calculation.
9. The wind power probability prediction method based on the VMDS-CNN-BGRU model as claimed in claim 1, wherein the specific process of step 6 is as follows: inputting the predicted values of wind power under different quantiles into a predicted result output module, and collectingObtaining a probability density function of the wind power predicted value y by KDE core density estimationAnd realizing wind power probability prediction, wherein B is bandwidth, N is a plurality of quantiles and K () is a kernel function.
10. The method for predicting wind power probability based on VMDS-CNN-BGRU model of claim 9, wherein the kernel function is:
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310533079.3A CN116544922A (en) | 2023-05-11 | 2023-05-11 | Wind power probability prediction method based on VMDS-CNN-BGRU model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310533079.3A CN116544922A (en) | 2023-05-11 | 2023-05-11 | Wind power probability prediction method based on VMDS-CNN-BGRU model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116544922A true CN116544922A (en) | 2023-08-04 |
Family
ID=87446703
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310533079.3A Pending CN116544922A (en) | 2023-05-11 | 2023-05-11 | Wind power probability prediction method based on VMDS-CNN-BGRU model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116544922A (en) |
-
2023
- 2023-05-11 CN CN202310533079.3A patent/CN116544922A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106529814B (en) | Distributed photovoltaic ultra-short term prediction method based on Adaboost clustering and Markov chain | |
| CN112632794B (en) | Power grid reliability evaluation method based on cross entropy parameter subset simulation optimization | |
| Chen et al. | Short-term wind speed forecasting with principle-subordinate predictor based on Conv-LSTM and improved BPNN | |
| CN109635928A (en) | A kind of voltage sag reason recognition methods based on deep learning Model Fusion | |
| CN110321934B (en) | Method and system for detecting abnormal data of user electricity consumption | |
| CN111709454B (en) | Multi-wind-field output clustering evaluation method based on optimal copula model | |
| CN114861533A (en) | Wind power ultra-short-term prediction method based on time convolution network | |
| CN114792158B (en) | Multi-wind power plant short-term power prediction method based on space-time fusion graph neural network | |
| CN112819238A (en) | Short-term wind power prediction method based on chaotic chicken flock optimization algorithm | |
| CN117078048B (en) | Digital twinning-based intelligent city resource management method and system | |
| CN115034485A (en) | Wind power interval prediction method and device based on data space | |
| CN113673768A (en) | Wind power probability prediction model based on VMD and CNN-LSTM | |
| CN116842337A (en) | Transformer fault diagnosis method based on LightGBM (gallium nitride based) optimal characteristics and COA-CNN (chip on board) model | |
| CN112633565A (en) | Photovoltaic power aggregation interval prediction method | |
| CN118095891A (en) | Active power distribution network payload prediction method and system considering source load meteorological characteristic decoupling | |
| CN114298136A (en) | Wind speed prediction method based on local mean decomposition and deep learning neural network | |
| CN117689082A (en) | Short-term wind power probability prediction method, system and storage medium | |
| CN112132344A (en) | Short-term wind power prediction method based on similar day and FRS-SVM | |
| CN116544922A (en) | Wind power probability prediction method based on VMDS-CNN-BGRU model | |
| CN117134315A (en) | Distribution transformer load prediction method and device based on BERT algorithm | |
| CN116596129A (en) | Electric vehicle charging station short-term load prediction model construction method | |
| CN111126645A (en) | Wind power prediction algorithm based on data mining technology and improved support vector machine | |
| CN116628488A (en) | Training method of wind power prediction model, wind power prediction method and device | |
| CN114036845B (en) | Wind farm group model construction 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 |