CN104268627B - Short-term wind speed forecasting method based on deep neural network transfer model - Google Patents
Short-term wind speed forecasting method based on deep neural network transfer model Download PDFInfo
- Publication number
- CN104268627B CN104268627B CN201410458807.XA CN201410458807A CN104268627B CN 104268627 B CN104268627 B CN 104268627B CN 201410458807 A CN201410458807 A CN 201410458807A CN 104268627 B CN104268627 B CN 104268627B
- Authority
- CN
- China
- Prior art keywords
- data
- neural network
- deep neural
- wind
- wind speed
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Landscapes
- Wind Motors (AREA)
Abstract
The invention discloses a short-term wind speed forecasting method based on a deep neural network transfer model. The method comprises the following steps that (1) normalization preprocessing and division of sample sets are carried out on data of two or more wind power plants, (2) the deep neural network transfer model is established, (3) layered training is started from bottom to top in an unsupervised learning mode, (4) supervised learning is carried out from top to bottom on the basis of the third step, (5) weight parameters of connection of a top layer and hidden layers are finely adjusted so as to obtain an output layer, corresponding to the wind power plants, in a deep neural network, and (6) inverse normalization is carried out on the result output by a deep neural network so as to obtain the predicted value of wind speed. Transfer learning is introduced to the wind speed forecasting field, knowledge of other wind power plants rich in data is transferred to target wind power plants, and the problem that the newly built wind power plants have few data is solved effectively. By means of the effective transfer scheme based on the deep neural network, the wind speed prediction accuracy of the target wind power plants is greatly improved.
Description
Technical field
The present invention is based on machine Learning Theory and Statistical Learning Theory, by building based on deep neural network
Regression model, builds the depth migration models of the multi output of shared hidden layer on this basis, this model is used for minority evidence finally
The short-term wind speed prediction of wind energy turbine set.
Background technology
In terms of wind speed forecasting, many methods have been suggested.These methods can be divided into four classes：1) physical model；2)
Statistical model；3) spatial correlation model；4) artificial intelligence model and other new models.Physical model uses physical factor, meteorology
Data such as landform, pressure and temperature come estimate future wind speed.It is more advantageous in long-term forecast, in short-term forecast
Accurate result cannot be typically given.Therefore, the first step that their Jing are often simply predicted, is input into as the auxiliary of other models.
Dependency of the statistical model based on wind series, sets up prediction mould by steps such as Model Distinguish and parameter estimation, model testings
Type, first describes the change of historical wind speed sequence, then following change is predicted.Conventional statistics equivalent model is in random time
Series model, conventional time series models mainly have：Autoregression model (Autoregressive, AR), moving average model(MA model)
(Moving Average Model, MA), ARMA model (Autoregressive Moving Average
Model, ARMA) and return accumulative formula moving average model(MA model) (Regressive Integrated Moving Average
Model, ARIMA).Another statistical method is Kalman filtering method.It needs the estimated value and nearest of previous wind speed
The method of individual observation, adoption status equation and recursion estimates current air speed value, and its solution is given in the form of estimated value.
Different with additive method, spatial correlation model needs to consider wind energy turbine set and multigroup air speed data in close several places therewith,
With the spatial coherence between several place wind speed, forecasting wind speed is carried out.The method is very big to source data collection amount, therefore
It is difficult than additive method, but as the factor considered during prediction increases, so prediction effect is preferable.Nowadays, with people
The development of work intelligence and other Forecasting Methodologies, various new models are suggested.Including support vector machine (Support
Vector Machine, SVM), fuzzy logic method, artificial neural network (Artificial Neural Network, ANN)
Deng and some combination forecasting methods.
But, for existing solution, do not relate to the forecasting wind speed problem of the newly-built wind energy turbine set of minority evidence.Gather around
There are little data, it is not enough to study to a good forecast model.At this moment, we introduce transfer learning (Transfer
Learning, TL) concept.Transfer learning be it is a kind of by the knowledge acquired in an environment for helping the study in new environment
The method of task.It can migrate knowledge to target domain from other field, migrate the study of knowledge to future from available data,
Efficiently solve few data, updating decision, learning process it is time-consuming the problems such as.
In addition, by the comparison of many results, the modelling effect in short-term wind speed forecasting single method based on ANN compared with
It is good, but neutral net before is shallow Model mostly.2006, Hinton et al. proposed deep learning (Deep
Learning, DL) concept.Obtained extensive concern after this, and be successfully applied to computer vision, speech recognition,
The fields such as natural language processing.Deep learning forms more abstract high-rise expression by combining low-level feature, to find data
Distributed nature represent.It solve before in deep layer network in non-convex objective cost function generally existing local optimum
Problem.Three technologies create the success of deep learning method, are substantial amounts of hidden node respectively, more preferable learning algorithm and more preferably
Parameter initialization technology.In addition, depth structure is very suitable for transfer learning, it can extract the feature of high-level abstractions, and
Feature as some is all suitable for for many fields.
Transfer learning based on deep neural network has achieved good effect in numerous applications.For example, character
Identification, semantic classification, multilingual study, image classification etc..How this model is applied to the short-term wind speed of new wind farm
Forecasting problem be the invention solves the problems that key issue.
The content of the invention
In view of the problem that the neutral net of forecasting wind speed facing challenges and classics is present, the present invention proposes a kind of base
In the short-term wind speed forecasting method for wind energy turbine set of deep neural network, deep learning is incorporated into forecasting wind speed neck by the present invention
Domain, with very strong feature learning ability, the advanced abstract information hidden during wind speed can be extracted.The present invention will be moved
Move study and be incorporated into forecasting wind speed field, to target wind farm, have by by the knowledge migration of the wind energy turbine set of other data riches
Solve the problems, such as that newly-built wind farm data is few to effect.By migration scheme effectively based on deep neural network, greatly
Target wind farm is improve for the precision of prediction of wind speed.
A kind of technical scheme of the short-term wind speed forecasting method based on deep neural network migration models of the present invention is：This is pre-
Reporting method is comprised the following steps：
What step one, setting data collection were noted down is the air speed data with ten minutes to be spaced, first to two or more wind
The data of electric field carry out maximin normalization pretreatment；The data sample pretreated data being divided in units of day
This collection, and as the input of deep neural network migration models, wherein the characteristic vector of the sample at the i-th moment is the wind of 144 dimensions
Fast vector：xi=(xi-143,xi-142,…,xi-1,xi), the deep neural network is output as 8 hours after the sample
Forecasting wind speed data, i.e., the wind velocity vector of 48 dimensions：yi=(xi+1,xi+2,…,xi+47,xi+48), wherein xiFor the wind at the i-th moment
Speed；
Step 2, structure deep neural network migration models, wherein, input layer is the natural law after above-mentioned wind energy turbine set is processed
According to sequence, hidden layer is the noise reduction automatic coding machine of two accumulations, and output layer is the air speed data sequence of prediction in following 8 hours；
Step 3, the initial weight ginseng for using the bottom-up beginning order training method of unsupervised learning, obtaining each hidden layer connection
Number；
Step 4, the initial weight parameter of each hidden layer connection obtained based on step 3 are further had by top-down
Supervised learning carries out retraining to the deep neural network, so as to the weighting parameter of each hidden layer connection after being finely tuned, the step
Used in be that all data sets have label data；
Step 5, the initialization weighting parameter that is connected with hidden layer of top layer, then have label data using each wind energy turbine set
Respectively the weighting parameter that top layer is connected with hidden layer is finely adjusted, the output of corresponding each wind energy turbine set in deep neural network is obtained
Layer, the i.e. wind series of each wind-powered electricity generation field prediction；
Step 6, the result to deep neural network output carry out the pre- of 8 hour wind speed after renormalization is obtained
Measured value, and forecast error is calculated according to predictive value and actual value.
Compared with prior art, the invention has the beneficial effects as follows：
It is shown experimentally that, the deep neural network (Shared-Hidden-Layer of Application share hidden layer in the present invention
Deep Neural Network, SHL-DNN) model with apply non-migratory support vector regression (Support Vector
Regression, SVR) model, non-migratory deep neural network (Deep Neural Network, DNN) compare, to difference
The data of wind energy turbine set carry out short-term wind speed prediction its effect more preferably, reduce forecast error, especially the fewer wind of data volume
Electric field its effect is more projected, and when target wind farm data is enough (as training data includes the wind speed of six months),
Avoid the need for, from other wind energy turbine set there migrating datas, not so, even to reduce precision of prediction in some cases.
Description of the drawings
Fig. 1 is the deep neural network migration models structure chart in the present invention；
Fig. 2 is different error criterions of SVR, DNN, SHL-DNN model in Ningxia wind energy turbine set two weeks data
As a result, wherein, be (a) result of mean absolute error criterion, be (b) knot of mean absolute percentage error criterion
Really, result (c) for mean square error criterion, is (d) result of root-mean-square error criterion；
Fig. 3 is different error criterions of SVR, DNN, SHL-DNN model in Jilin wind energy turbine set two weeks data
As a result, wherein, be (a) result of mean absolute error criterion, be (b) knot of mean absolute percentage error criterion
Really, result (c) for mean square error criterion, is (d) result of root-mean-square error criterion；
The prediction data of tri- models of Fig. 4 (a) initial datas and SVR, DNN, SHL-DNN is two weeks data in data set
Ningxia wind energy turbine set on regard to results contrast that predicted time is 10 minutes；
The prediction data of tri- models of Fig. 4 (b) initial datas and SVR, DNN, SHL-DNN is two weeks data in data set
Ningxia wind energy turbine set on regard to results contrast that predicted time is 30 minutes；
The prediction data of tri- models of Fig. 4 (c) initial datas and SVR, DNN, SHL-DNN is two weeks data in data set
Ningxia wind energy turbine set on regard to results contrast that predicted time is 1 hour；
The prediction data of tri- models of Fig. 4 (d) initial datas and SVR, DNN, SHL-DNN is two weeks data in data set
Ningxia wind energy turbine set on regard to results contrast that predicted time is 2 hours；
The prediction data of tri- models of Fig. 5 (a) initial datas and SVR, DNN, SHL-DNN is 3 months data in data set
Ningxia wind energy turbine set on regard to results contrast that predicted time is 10 minutes；
The prediction data of tri- models of Fig. 5 (b) initial datas and SVR, DNN, SHL-DNN is 3 months data in data set
Ningxia wind energy turbine set on regard to results contrast that predicted time is 30 minutes；
The prediction data of tri- models of Fig. 5 (c) initial datas and SVR, DNN, SHL-DNN is 3 months data in data set
Ningxia wind energy turbine set on regard to results contrast that predicted time is 1 hour；
The prediction data of tri- models of Fig. 5 (d) initial datas and SVR, DNN, SHL-DNN is 3 months data in data set
Ningxia wind energy turbine set on regard to results contrast that predicted time is 2 hours；
The prediction data of tri- models of Fig. 6 (a) initial datas and SVR, DNN, SHL-DNN is 6 months data in data set
Ningxia wind energy turbine set on regard to results contrast that predicted time is 10 minutes；
The prediction data of tri- models of Fig. 6 (b) initial datas and SVR, DNN, SHL-DNN is 6 months data in data set
Ningxia wind energy turbine set on regard to results contrast that predicted time is 30 minutes；SVR, DNN and SHL-DNN
The prediction data of tri- models of Fig. 6 (c) initial datas and SVR, DNN, SHL-DNN is 6 months data in data set
Ningxia wind energy turbine set on regard to results contrast that predicted time is 1 hour；
The prediction data of tri- models of Fig. 6 (d) initial datas and SVR, DNN, SHL-DNN is 6 months data in data set
Ningxia wind energy turbine set on regard to results contrast that predicted time is 2 hours.
Specific embodiment
Technical solution of the present invention is described in further detail with specific embodiment below in conjunction with the accompanying drawings.
The present invention proposes that a kind of application carries out the model of short-term wind speed prediction based on deep neural network migration models.Utilize
Depth learning technology builds the deep neural network model of the multiple-input and multiple-output of shared hidden layer, as shown in Figure 1.In this structure
In, input layer and hidden layer are that all wind energy turbine sets are shared, and can regard a common eigentransformation as.Output layer is each wind
Electric field is separate, because their data distribution is different.This is a type of knowledge migration, because universal spy
Levy and be migrated to each data set.
As shown in figure 1, the present invention proposes a kind of short-term wind speed forecasting method based on deep neural network migration models, its
It is characterised by, comprises the following steps：
What step one, setting data collection were noted down is the air speed data with ten minutes to be spaced, first to two or more wind
The data of electric field carry out maximin normalization pretreatment；The data sample pretreated data being divided in units of day
This collection, and as the input of deep neural network migration models, the characteristic vector of the sample at the i-th moment be 144 dimensions wind speed to
Amount：xi=(xi-143,xi-142,…,xi-1,xi), the deep neural network is output as the wind speed of 8 hours after the sample
Prediction data, i.e., the wind velocity vector of 48 dimensions：yi=(xi+1,xi+2,…,xi+47,xi+48), wherein xiFor the wind speed at i-th moment；
Step 2, structure deep neural network migration models, wherein, input layer is the natural law after above-mentioned wind energy turbine set is processed
According to sequence (144 nodes), hidden layer is noise reduction automatic coding machine (the Stacked Denoising Auto- of two accumulations
Encoder, SDA) (being all 100 nodes), output layer is the air speed data sequence (48 nodes) of prediction in following 8 hours.
Here, we replace traditional automatic coding machine using noise reduction automatic coding machine (Denoising Autoencoder, DAE)
(Auto-encoder, AE).The idea of DAE is very simple.In order that hidden layer represents more robust feature or which is avoided completely
It is equal to input, input is replaced with the treated version in partThis is by a Random MapsIt is complete
Into.We use the mode that part input node is changed into 0 at random.It is identical with AE,It is encoded as hiding expression：
Wherein s is a nonlinear mapping, such as the sigmoid functions in our experiments.θ={ W, b } represents its
Parameter sets, wherein W are the weight matrix that dimension is d' × d, and b is the bias vector that dimension is d'.Afterwards, hiding expression or volume
CodeIt is decoded as by a similar conversionReconstruct：
WhereinCan regard asOne prediction, parameter sets be θ '={ W', b'}.The weight matrix W' of inverse mapping can be set
It is set to W'=WT.Such each trained valuesBe mapped as one it is correspondingReconstruct with oneThe ginseng of this model
Number, respectively θ and θ ' are optimised so that minimizing average reconstructed error：
Wherein, L is loss function, traditional Squared Error Loss that such as we use
Step 3, the excitation function that every layer of DAE is set, learning rate, noise reduction ratio etc..Frequency of training, every time together is set
The number of samples of training, using the bottom-up beginning order training method of SDA unsupervised learnings, obtains the initial weight of each hidden layer connection
Parameter；
Step 4, the excitation function that deep neural network is set, learning rate, frequency of training and the sample trained together every time
Number etc., the initial weight parameter of each hidden layer connection obtained based on step 3 further pass through top-down supervised learning
Retraining is carried out to the deep neural network, so as to the weighting parameter of each hidden layer connection after being finely tuned, used in the step
Be that all data sets have label data；
Step 5, the initialization weighting parameter that is connected with hidden layer of top layer, then have label data using each wind energy turbine set
Respectively the weighting parameter that top layer is connected with hidden layer is finely adjusted, the output of corresponding each wind energy turbine set in deep neural network is obtained
Layer, the i.e. wind series of each wind-powered electricity generation field prediction；
Step 6, the result to deep neural network output carry out the pre- of 8 hour wind speed after renormalization is obtained
Measured value, and forecast error is calculated according to predictive value and actual value.
Research material of the present invention：
In short-term wind speed prediction, non-migratory SVR, non-migratory is compared based on the migration models of deep neural network
For DNN models, substantially good prediction effect is shown.
The evaluation of wind speed forecasting error, the most frequently used index have four, respectively mean absolute error (Mean
Absolute error, MAE), mean absolute percentage error (Mean absolute percentage error, MAPE),
Mean square error (Mean square error, MSE) and root-mean-square error (Root mean square error, RMSE).They
Computing formula it is as follows：
Wherein, numbers of the n for sample, yiIt is i-th actual value, yi' it is i-th predictive value.
1st, migrate to the experiment effect of the few wind energy turbine set of data
It is now assumed that target wind farm has a small amount of data.We have done two experiments altogether.For experiment 1, select peaceful
Summer wind energy turbine set is aiming field, and other three wind energy turbine sets are source domain.The data in Ningxia are two weeks wind speed, and other three power plant are one
The data in year.For experiment 2, selection Jilin wind energy turbine set is aiming field, and other three wind energy turbine sets are source domain.The data in Jilin are half
Individual month wind speed, the data in Ningxia are a year and a half wind speed, and two other is the data of a year.Nothing is moved with aforementioned four measurement index
The SVR of shifting, non-migratory DNN models and we design commented based on three kinds of models of migration models of deep neural network
Valency.The forecast result of two experiments is as shown in Figure 2,3.
To sum up, the migration models using non-migratory SVR models, non-migratory DNN and based on deep neural network model
Respectively the data of different wind energy turbine sets are carried out with short-term wind speed prediction, experimental result explanation, migration models are more non-migratory than two other
Modelling effect it is well a lot.Therefore, the wind energy turbine set fewer for data volume, transfer learning are a kind of effective methods.
2nd, with the experiment effect for increasing of target wind farm data
Design the migration of another group of experimental research model and increase the affecting of being subject to of predicting the outcome with aiming field training data.
Except the difference of target data amount, ibid group experiment is identical for other settings.Table 1-8 is summarized for different targets
Domain, the result of different error measurement indexs.Table 1-4 is that, for testing 1 Ningxia wind energy turbine set, table 5-8 is for testing 2 Jilin wind
Electric field.In addition, Fig. 4 (a) -6 (d) illustrates the prediction data for 1 Ningxia wind energy turbine set initial data of experiment and three models
Comparison in the training set of different sample sizes with regard to different prediction steps, this visually illustrates the pre- of different aspect
Survey effect.
1 different models of table are tied with regard to the experiment of MAE on 1 Ningxia wind farm data is tested for different prediction steps
Really
2 different models of table are tied with regard to the experiment of MAPE on 1 Ningxia wind farm data is tested for different prediction steps
Really
3 different models of table are tied with regard to the experiment of MSE on 1 Ningxia wind farm data is tested for different prediction steps
Really
4 different models of table are tied with regard to the experiment of RMSE on 1 Ningxia wind farm data is tested for different prediction steps
Really
5 different models of table are tied with regard to the experiment of MAE on 2 Jilin wind farm datas are tested for different prediction steps
Really
6 different models of table are tied with regard to the experiment of MAPE on 2 Jilin wind farm datas are tested for different prediction steps
Really
7 different models of table are tied with regard to the experiment of MSE on 2 Jilin wind farm datas are tested for different prediction steps
Really
8 different models of table are tied with regard to the experiment of RMSE on 2 Jilin wind farm datas are tested for different prediction steps
Really
To sum up, using non-migratory SVR models, non-migratory deep neural network model and based on deep neural network mould
The migration models of type carry out short-term wind speed prediction respectively to different how many data of different wind energy turbine sets, and experimental result is illustrated, greatly
In most cases the migration models modelling effect more non-migratory than two other is good.Especially to 10 minutes afterwards and the wind of 30 minutes
Speed prediction, in the case where data volume is fewer, migration models are significantly better than other two methods.But with the increasing of data volume
Plus, the importance of migration has declined, and especially contrasts the prediction after the long period.Therefore, it can it is concluded that, work as target
When wind farm data is enough, avoid the need for, from other wind energy turbine set there migrating datas, even to reduce pre- in some cases
Survey precision.
Although above in conjunction with accompanying drawing, invention has been described, the invention is not limited in above-mentioned being embodied as
Mode, above-mentioned specific embodiment are only schematic rather than restricted, and one of ordinary skill in the art is at this
Under the enlightenment of invention, without deviating from the spirit of the invention, many variations can also be made, these belong to the present invention's
Within protection.
Claims (1)
1. a kind of short-term wind speed forecasting method based on deep neural network migration models, it is characterised in that comprise the following steps：
What step one, setting data collection were noted down is the air speed data with ten minutes to be spaced, first to two or more wind energy turbine set
Data carry out maximin normalization pretreatment；The data sample pretreated data being divided in units of day
Collection, and as the input of deep neural network migration models, wherein the characteristic vector of the sample at the i-th moment is the wind speed of 144 dimensions
Vector：xi=(xi-143,xi-142,…,xi-1,xi), the deep neural network is output as the wind of 8 hours after the sample
Fast prediction data, i.e., the wind velocity vector of 48 dimensions：yi=(xi+1,xi+2,…,xi+47,xi+48), wherein xiFor the wind speed at the i-th moment；
Step 2, structure deep neural network migration models, wherein, input layer is the day data sequence after above-mentioned wind energy turbine set is processed
Row, hidden layer are the noise reduction automatic coding machine of two accumulations, and output layer is the air speed data sequence of prediction in following 8 hours；
Step 3, the initial weight parameter for using the bottom-up beginning order training method of unsupervised learning, obtaining each hidden layer connection；
Step 4, the initial weight parameter of each hidden layer connection obtained based on step 3 further have supervision by top-down
Study carries out retraining to the deep neural network, so as to the weighting parameter of each hidden layer connection after being finely tuned, institute in the step
Use all data sets has label data；
The weighting parameter that step 5, initialization top layer are connected with hidden layer, is then distinguished using the label data that has of each wind energy turbine set
The weighting parameter that top layer is connected with hidden layer is finely adjusted, the output layer of corresponding each wind energy turbine set in deep neural network is obtained,
It is the wind series of each wind-powered electricity generation field prediction；
Step 6, the result to the deep neural network output carry out the predictive value of 8 hour wind speed after renormalization is obtained,
And forecast error is calculated according to predictive value and actual value.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410458807.XA CN104268627B (en) | 2014-09-10 | 2014-09-10 | Short-term wind speed forecasting method based on deep neural network transfer model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410458807.XA CN104268627B (en) | 2014-09-10 | 2014-09-10 | Short-term wind speed forecasting method based on deep neural network transfer model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104268627A CN104268627A (en) | 2015-01-07 |
| CN104268627B true CN104268627B (en) | 2017-04-19 |
Family
ID=52160147
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410458807.XA Active CN104268627B (en) | 2014-09-10 | 2014-09-10 | Short-term wind speed forecasting method based on deep neural network transfer model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104268627B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2017223522A1 (en) * | 2016-06-23 | 2017-12-28 | Mohammad Shami | Neural network systems and methods for generating distributed representations of electronic transaction information |
| CN108399272A (en) * | 2018-01-15 | 2018-08-14 | 北京航空航天大学 | Artificial intelligence program person writes the Decision of Neural Network method of digital aircraft code |
Families Citing this family (29)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104778506B (en) * | 2015-03-31 | 2019-03-26 | 天津大学 | Short-term wind speed forecasting method based on local integrated study |
| CN104965812B (en) * | 2015-07-13 | 2017-12-01 | 深圳市腾讯计算机系统有限公司 | A kind of Deep model processing method and processing device |
| US9990564B2 (en) | 2016-03-29 | 2018-06-05 | Wipro Limited | System and method for optical character recognition |
| CN106371155B (en) * | 2016-08-25 | 2018-12-21 | 华南师范大学 | Method of meteorological forecast and system based on big data and analysis field |
| CN106650982B (en) * | 2016-08-30 | 2020-07-07 | 华北电力大学 | Deep learning power prediction method based on multipoint NWP (non-uniform Power Point) |
| CN110088776A (en) * | 2016-10-06 | 2019-08-02 | 西门子股份公司 | For training the computer equipment of deep neural network |
| CN107122845A (en) * | 2017-03-20 | 2017-09-01 | 华南理工大学 | A kind of modified neutral net overhead transmission line wind speed forecasting method based on division gentle breeze area |
| CN107169598A (en) * | 2017-05-05 | 2017-09-15 | 上海斐讯数据通信技术有限公司 | A kind of weather prediction method and system based on deep learning |
| CN107766877B (en) * | 2017-09-27 | 2020-05-22 | 华南理工大学 | Method for dynamically identifying overweight vehicle in bridge monitoring system |
| WO2019127945A1 (en) * | 2017-12-26 | 2019-07-04 | 佛山科学技术学院 | Structured neural network-based imaging task schedulability prediction method |
| CN108197736B (en) * | 2017-12-29 | 2021-08-13 | 北京工业大学 | Air quality prediction method based on variational self-encoder and extreme learning machine |
| CN108182452B (en) * | 2017-12-29 | 2018-11-20 | 哈尔滨工业大学（威海） | Aero-engine fault detection method and system based on grouping convolution self-encoding encoder |
| CN110070855B (en) * | 2018-01-23 | 2021-07-23 | 中国科学院声学研究所 | Voice recognition system and method based on migrating neural network acoustic model |
| CN109490814B (en) * | 2018-09-07 | 2021-02-26 | 广西电网有限责任公司电力科学研究院 | Metering automation terminal fault diagnosis method based on deep learning and support vector data description |
| CN109784473A (en) * | 2018-12-13 | 2019-05-21 | 天津大学 | A kind of short-term wind power prediction method based on Dual Clocking feature learning |
| CN110334478B (en) * | 2019-07-22 | 2023-07-25 | 山东浪潮科学研究院有限公司 | Machine equipment abnormality detection model construction method, detection method and model |
| CN110889550A (en) * | 2019-11-22 | 2020-03-17 | 江南大学 | Method for predicting dissolved oxygen in different water areas based on transfer learning |
| CN111144303A (en) * | 2019-12-26 | 2020-05-12 | 华北电力大学（保定） | Power line channel transmission characteristic identification method based on improved denoising autoencoder |
| CN111325338B (en) * | 2020-02-12 | 2023-05-05 | 暗物智能科技(广州)有限公司 | Neural network structure evaluation model construction and neural network structure searching method |
| CN111488712B (en) * | 2020-04-13 | 2022-08-26 | 中国船舶重工集团海装风电股份有限公司 | Wind power generator power curve modeling method based on transfer learning |
| CN111967653A (en) * | 2020-07-22 | 2020-11-20 | 易天气(北京)科技有限公司 | Method for constructing airport runway wind forecasting model, forecasting method and forecasting system |
| CN112149857A (en) * | 2020-10-22 | 2020-12-29 | 无锡九方科技有限公司 | Meteorological intelligent forecast correction method based on transfer learning |
| CN112949904B (en) * | 2021-01-28 | 2022-06-07 | 华中科技大学 | Wind power plant short-term power prediction method based on feature selection and multi-level deep transfer learning |
| CN113191524B (en) * | 2021-01-29 | 2023-07-11 | 河北建投新能源有限公司 | Wind speed prediction method, device, storage medium and processor |
| CN112766596B (en) * | 2021-01-29 | 2024-04-16 | 苏州思萃融合基建技术研究所有限公司 | Construction method of building energy consumption prediction model, energy consumption prediction method and device |
| CN113627074B (en) * | 2021-07-13 | 2024-04-19 | 西安理工大学 | Ground wave propagation delay prediction method based on transfer learning |
| CN113792919B (en) * | 2021-09-02 | 2024-02-27 | 华中科技大学 | Wind power prediction method based on combination of transfer learning and deep learning |
| CN114167076A (en) * | 2021-11-10 | 2022-03-11 | 中国地质大学（武汉） | Method for establishing sea surface wind speed and direction inversion model and wind speed and direction inversion method |
| CN116992222B (en) * | 2023-09-27 | 2024-01-26 | 长江三峡集团实业发展(北京)有限公司 | Method, device, equipment and medium for migration learning of wind element correction model |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102479339A (en) * | 2010-11-24 | 2012-05-30 | 香港理工大学 | Method and system for forecasting short-term wind speed of wind farm based on hybrid neural network |
| CN103413174A (en) * | 2013-08-30 | 2013-11-27 | 哈尔滨工业大学 | Short-term wind speed multi-step prediction method based on deep learning method |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPH0949884A (en) * | 1995-08-04 | 1997-02-18 | Aasu Uezaa:Kk | Predicting method for local weather using neural network |
-
2014
- 2014-09-10 CN CN201410458807.XA patent/CN104268627B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102479339A (en) * | 2010-11-24 | 2012-05-30 | 香港理工大学 | Method and system for forecasting short-term wind speed of wind farm based on hybrid neural network |
| CN103413174A (en) * | 2013-08-30 | 2013-11-27 | 哈尔滨工业大学 | Short-term wind speed multi-step prediction method based on deep learning method |
Non-Patent Citations (2)
| Title |
|---|
| Pattern-based wind speed prediction based on generalized principal component analysis;Hu qinghua et al.;《IEEE transactions on sustainable energy》;20140731;第5卷(第3期);第866-873页 * |
| 基于深度学习理论的短期风速多步预测方法研究;万杰等;《智能化电站技术发展研讨暨电站自动化2013年会论文集》;20131012;第73-78页 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2017223522A1 (en) * | 2016-06-23 | 2017-12-28 | Mohammad Shami | Neural network systems and methods for generating distributed representations of electronic transaction information |
| CN108399272A (en) * | 2018-01-15 | 2018-08-14 | 北京航空航天大学 | Artificial intelligence program person writes the Decision of Neural Network method of digital aircraft code |
| CN108399272B (en) * | 2018-01-15 | 2021-05-07 | 北京航空航天大学 | Neural network decision-making method for artificial intelligence programmer to write digital aircraft code |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104268627A (en) | 2015-01-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104268627B (en) | Short-term wind speed forecasting method based on deep neural network transfer model | |
| Fan et al. | Short term load forecasting based on phase space reconstruction algorithm and bi-square kernel regression model | |
| Mahmoud et al. | An advanced approach for optimal wind power generation prediction intervals by using self-adaptive evolutionary extreme learning machine | |
| CN109580215B (en) | Wind power transmission system fault diagnosis method based on deep generation countermeasure network | |
| CN105631483B (en) | A kind of short-term electro-load forecast method and device | |
| CN104077632B (en) | A kind of wind electric field power prediction method based on deep neural network | |
| CN109146162B (en) | A kind of probability wind speed forecasting method based on integrated Recognition with Recurrent Neural Network | |
| CN106952181A (en) | Electric Load Prediction System based on long Memory Neural Networks in short-term | |
| CN108711847A (en) | A kind of short-term wind power forecast method based on coding and decoding shot and long term memory network | |
| CN109255726A (en) | A kind of ultra-short term wind power prediction method of Hybrid Intelligent Technology | |
| Zhuang et al. | Long-lead prediction of extreme precipitation cluster via a spatiotemporal convolutional neural network | |
| CN110276472A (en) | A kind of offshore wind farm power ultra-short term prediction method based on LSTM deep learning network | |
| Li et al. | Short-term prediction of the power of a new wind turbine based on IAO-LSTM | |
| CN112100911A (en) | Solar radiation prediction method based on deep BISLTM | |
| Aslan et al. | Realization of Turkey’s energy demand forecast with the improved arithmetic optimization algorithm | |
| Aslan | Archimedes optimization algorithm based approaches for solving energy demand estimation problem: a case study of Turkey | |
| Qu et al. | A hybrid model based on ensemble empirical mode decomposition and fruit fly optimization algorithm for wind speed forecasting | |
| Ge et al. | A FA-GWO-GRNN method for short-term photovoltaic output prediction | |
| CN115392387B (en) | Low-voltage distributed photovoltaic power generation output prediction method | |
| Han et al. | Interpretation of Stability Assessment Machine Learning Models Based on Shapley Value | |
| Lv et al. | Research on short-term load forecasting approach for smart grid | |
| Chang et al. | Forecasting wind power generation by a new type of radial basis function-based neural network | |
| Yousefi et al. | Strategic planning for minimizing CO2 emissions using LP model based on forecasted energy demand by PSO Algorithm and ANN | |
| Zhenhao et al. | Prediction of wind power ramp events based on deep neural network | |
| Wan et al. | Turbine location wind speed forecast using convolutional neural network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20180419 Address after: 511400 307, 4 East Ring Road, east ring street, Panyu District, Guangzhou, Guangdong 307, 307 Patentee after: Guangzhou about you fly in the Internet of things Technology Co., Ltd. Address before: 300072 Tianjin City, Nankai District Wei Jin Road No. 92 Patentee before: Tianjin University |