CN104036328A - Self-adaptive wind power prediction system and prediction method - Google Patents
Self-adaptive wind power prediction system and prediction method Download PDFInfo
- Publication number
- CN104036328A CN104036328A CN201310067802.XA CN201310067802A CN104036328A CN 104036328 A CN104036328 A CN 104036328A CN 201310067802 A CN201310067802 A CN 201310067802A CN 104036328 A CN104036328 A CN 104036328A
- Authority
- CN
- China
- Prior art keywords
- neural network
- wind
- output power
- blower fan
- pivot
- 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
Landscapes
- Wind Motors (AREA)
Abstract
Disclosed is an ultra-short-term blower fan output power prediction system for predicting the output power of target blower fans in multiple wind fields. The prediction system comprises a wind field information acquisition portion for acquiring wind measurement tower data and blower fan operation data of multiple wind fields and recording the data in an association mode in a database as historical information; a principal component analysis extraction portion for reading the historical information from the wind field information acquisition portion, executing principle component analysis and determining principle components according with predetermined conditions; and a nerve network portion which establishes an RBF nerve network by taking the principle components obtained from the principal component analysis extraction portion as input nodes and predicts the output power of the blower fans by use of the obtained RBF nerve network. The prediction system is characterized by further comprising an updating portion which performs online updating on the weight of the output layer of the nerve network by use of a Kalman filtering algorithm.
Description
Technical field
The invention provides a kind of self-adaptation wind power forecasting system and Forecasting Methodology, specifically, relate to and be applicable to the output power of blower fan to carry out ultra-short term prediction, and can adjust online self-adaptation wind power forecasting system and the Forecasting Methodology of forecast model.
Background technology
Wind power prediction contributes to optimize dispatching of power netwoks, reduces the operating cost of wind generator system, and strengthens security, reliability and the controllability of wind generator system.
Chinese Government has issued < < wind farm power prediction forecast management temporary method > > on January 1st, 2012, and the error of the ultra-short term forecast of regulation wind power prediction can not surpass 15%.Ultra-short term was defined as in 4 hours.See at present and do not importing NWP(numerical weather forecast) in the situation that, the precision of ultra-short term wind power prediction is all difficult to reach requirement.Although air motion has continuation, its variation still has suitable uncertainty.Only according to the anemometer tower data of single wind field, be difficult to following wind speed, wind direction to make accurate forecast.
Although the importing of numerical weather forecast can improve precision to a certain extent, also increased the cost of wind-powered electricity generation prognoses system, and the resolution of NWP also need to improve.Ultra-short term forecast requires the data in every 15 minutes forecast futures 4 hours (take 15 minutes as interval), and the renewal frequency of current numerical weather prediction is also lower, cannot reach in so far.
Wind power forecasting method can be divided into persistence forecasting method, autoregressive moving average (ARMA) modelling, Kalman filtering method and intelligent method etc. according to adopted mathematical model difference.Persistence forecasting method is the simplest forecast model, but the predicated error of this model is large and predict the outcome unstable.Improved method has arma modeling and Vector Autoression Models, Kalman filtering algorithm or time series method and Kalman filtering algorithm to combine.Also have in addition some intelligent methods, as Artificial Neural Network etc.
Kalman filtering algorithm is widely used in the signal of linear stochaastic system to be processed, and it utilizes the recursion of state equation, by linear unbias Minimum Mean Square Error estimation criterion, the dbjective state variable of wave filter is done to optimum estimate.Utilize Kalman Prediction iterative equation can realize the prediction of signal.
In prior art, there is the technical scheme (seeing document 1) of utilizing Kalman Prediction recurrence equation to predict power of fan.Specifically, first the sample time series based on wind power is set up the ARMA forecast model of blower fan output power.After setting up ARMA forecast model, can directly predict following blower fan output power, generally speaking, the deviation of the relative measured value that predicts the outcome based on arma modeling is larger.
Owing to having set up wind power seasonal effect in time series arma modeling by time series analysis, thereby arma modeling can be transformed into state space, set up state equation and the measurement equation of Kalman filtering.After setting initial value with reference to engineering custom, utilize recurrence equation to carry out iteration prediction, can obtain the status information of continuous renewal, obtain and predict the outcome more accurately.
In addition, artificial neural network is also widely used in the prediction of blower fan output power.Document 2 has been described a kind of wind power forecasting method, the meteorologic parameter that the numerical weather forecast (NWP) of wherein usining provides (comprising wind speed, wind direction sine, wind direction cosine, atmospheric density etc.) is as input vector, the real output of usining is set up 3 layers of BP neural network as object vector, and the historical data of above-mentioned meteorologic parameter of usining is trained this BP neural network as training set.Consider in numerical weather forecast data and exist systematic error, the technical scheme of document 2 utilizes Kalman filtering algorithm to carry out Kalman filtering to the training set for training in addition, and by BP Application of Neural Network in when prediction, the numerical weather forecast data execute card Kalman Filtering to real-time input.Experimental result shows that Kalman filtering can effectively eliminate the systematic error in air speed data.
Document 3 has been described a kind of wind power prediction scheme based on neural network of not using NWP, wherein the anemometer tower data based on wind field this locality are as input vector, set up and train BP neural network, and predicting the outcome of system operation 3 months analyze, forecast model correction is improved to its precision of prediction.
In the above-mentioned prior art that is applied to wind power prediction, there are the following problems: after wind power forecast model is set up, always maintain forecast model constant and only state variable (input parameter) is upgraded, or need to and re-start training at reconstruct forecast model under off-line state, and can not under on-line operation state, to forecast model itself (weights of predictive equation coefficient or neural network self), revise.Cannot set up like this and adaptively can to forecast model self, carry out online the prognoses system of continuous updating, lower to the utilization ratio of predicted numerical value, measured value etc.
List of documents
The comparison of 12 kinds of wind power forecast models of document (people such as Shi Qinghua, is published in < < energy technology economy > > the 6th phase of 23 volumes in June, 2011)
The wind energy turbine set short term power forecast model of document 2 Kalman filtering corrections (people such as Zhao Pan, is published in the 45th the 5th phase of volume of XI AN JIAOTONG UNIVERSITY Subject Index in May, 2011)
Document 3 Power Output for Wind Power Field ultra-short terms predict the outcome and analyze and improve people such as (, be published in the 35th the 15th phase of volume of < < Automation of Electric Systems > > in August, 2011) Chen Ying
Summary of the invention
The present invention considers the defect existing in above-mentioned prior art, has proposed a kind of adaptive wind power forecasting system and Forecasting Methodology, can under on-line operation state, to wind power forecast model, be optimized adjustment.
Specifically, the invention provides a kind of ultra-short term blower fan output power prognoses system, for the output power of the target blower fan of a plurality of wind fields is predicted, described prognoses system comprises: wind field information acquisition part, for gathering anemometer tower data and the fan operation data of described a plurality of wind fields, and it is recorded in database explicitly as historical information; Pivot analysis Extraction parts, for partly reading described historical information from described wind field information acquisition, and the execution of the data based on extracting from described historical information pivot analysis, to determine the pivot conforming to a predetermined condition; And part of neural network, it utilizes the pivot of obtaining from described pivot analysis Extraction parts to set up radial basis function neural network (RBF) as input node, and uses the output power of the neural network prediction blower fan of gained; Described prognoses system is characterised in that and also comprises more new portion, and described more new portion utilizes Kalman filtering algorithm to carry out online updating to the structural parameters of neural network.
According to a preferred embodiment of the invention, the state variable of upgrading by described Kalman filtering algorithm in described more new portion is the weights from hidden layer node to output layer node in described neural network.
According to a preferred embodiment of the invention, the predetermined condition of extracting pivot in described pivot analysis Extraction parts is set to extract variable that accumulative total variance contribution ratio is greater than predetermined threshold as pivot.In one embodiment, this threshold value is not less than 85%.
According to a preferred embodiment of the invention, wherein said wind field information acquisition partly gathers the anemometer tower data of a plurality of wind fields in same geographic area, calculate the related coefficient between the anemometer tower data of described a plurality of wind fields and the output power of described target blower fan, and extract the anemometer tower data that related coefficient in described a plurality of wind field and between the output power of described target blower fan is greater than predetermined threshold, be used as the input variable of above-mentioned pivot analysis.
The present invention also provides a kind of ultra-short term blower fan output power predicting method, comprises the steps: to gather anemometer tower data and the fan operation data of a plurality of wind fields, and it is recorded in database explicitly as historical information; The anemometer tower data that comprise based on described historical information and fan operation data are carried out pivot analysis, extract the pivot conforming to a predetermined condition; The pivot of being extracted of usining is set up RBF neural network as input node, and uses the output power of the RBF neural network prediction blower fan of gained; With utilize Kalman filtering algorithm to carry out online updating to neural network.
Accompanying drawing explanation
Below in connection with accompanying drawing, the structure of wind power forecasting system of the present invention and preferred embodiment are described.
Fig. 1 is the schematic diagram of the total structure of wind power forecasting system of the present invention;
Fig. 2 is the organigram of wind power forecasting system of the present invention in the off-line modeling stage;
Fig. 3 is the configuration diagram of the RBF neural network used in wind power forecasting system of the present invention;
Organigram when Fig. 4 is wind power forecasting system on-line operation of the present invention;
Fig. 5 is the process flow diagram of wind power forecasting system on-line operation of the present invention;
Fig. 6 is that the more new portion in wind power forecasting system of the present invention is carried out the process flow diagram that online Kalman upgrades.
Fig. 7 is the example of predetermined period for wind power forecasting system of the present invention.
Embodiment
1, the collection of wind field information
The present invention is not importing NWP, does not increase under the condition of system cost, and the anemometer tower data of other wind fields of comprehensive utilization the same area, realize the high-precision ultra-short term prediction of wind power.
The process of wind power forecast model being carried out to off-line modeling is under the state of off-line, collects the process that historical data arranges modeling, for each blower fan modeling process, is identical.Total prediction output power of wind field is each blower fan prediction output power sum.When gathering wind field information, need to consider following aspect.
1) first the data of collecting pass through pre-service, comprise and get rid of exceptional value, filtering etc.
2) the fan operation data that use comprise impeller radius, tip speed ratio, slurry elongation, blower fan output power etc.
3) the anemometer tower data of the wind field this locality using comprise wind direction, wind speed, temperature, air pressure.
4) wind field terrain information can be in addition as with reference to information.
5) other anemometer tower data of this region comprise wind direction, wind speed etc.
6) when determining other anemometer tower inputs, the correlativity between each anemometer tower and target blower fan output power is judged, select the data of relevant stronger anemometer tower as input parameter.
Related coefficient between two variablees can be calculated by following formula.
Wherein r is the related coefficient between variable x and y, and n is the sample size of variable x and y.According to one embodiment of present invention, can select wind speed in anemometer tower data as a variable, and carry out correlation calculations with the output power of target blower fan; For each the anemometer tower data in a plurality of wind fields, calculate, and extract the anemometer tower data that the related coefficient of calculating is greater than predetermined threshold.The threshold value here can suitably arrange according to wind field scope and quantity; In one embodiment of the invention, extract related coefficient | the anemometer tower data of r|>0.8.
According to the present invention, for the prediction of the output power of target blower fan anemometer tower data based on local wind field not only, can also effectively utilize the anemometer tower data of other wind field of the same area.
2, pivot analysis (PCA)
Carry out preliminary screening input data by the related coefficient between calculating anemometer tower data and target power of fan after, further apply pivot analysis (PCA) and choose suitable pivot for modeling.
Pivot analysis is for find the method for appropriate character representation in data.PCA is generally used for input sample set to carry out pre-service, a plurality of characteristic parameters of complex nonlinear relation that exist can be converted into less incoherent generalized variable each other, then with these less vectors, form new network input variable, thereby reduce the variable number of neural network input, optimized network structure.PCA to as if the sample data matrix of process variable.Line display sampled value or the observed value of data matrix, variable is shown in list.PCA produces statistical model---the principal component model of a compression, model the linear combination of variable, the main trend of data variation has been described.Principal component model makes the poor square redistribution of primary standard, and poor square of most standard can be distributed in the first pivot, and next is distributed in the second pivot, and the rest may be inferred.By certain criterion, last several pivots are considered as decomposing residual error and are ignored, utilize minimum pivot to represent maximum information.Principal component model has been given up part residual error and has been retained the main direction that embodies data variation, thereby reaches extraction system information, the object that scavenge system disturbs.
The main calculation procedure of pivot analysis is as follows:
(1) original sample standardization
In order to eliminate the impact that dimension is different with the order of magnitude, adopt average value standard deviation standardized method to process raw sample data.
(2) covariance matrix of Criterion variable, the eigenwert of solution matrix and proper vector
Utilize standardized value to calculate the related coefficient between variable, have k characteristic parameter can set up k rank correlation matrix.Matrix can obtain the eigenwert Ai (i=1,2, L, k) of descending arrangement thus, k corresponding k the proper vector of eigenwert, and each proper vector comprises k component.
(3) contribution rate of accumulative total as requested, chooses pivot
Calculate the contribution rate of i pivot to population variance, i.e. variance contribution ratio.Conventionally select accumulative total variance contribution ratio to be greater than 85% required pivot, can represent most information that k original variable can provide.
The computing formula of variance contribution ratio is as follows:
(4) set up pivot equation, calculate each pivot value
Each pivot value equation is shown in following formula, and wherein aj is the component corresponding to the proper vector of j, the standardized value that xj is each variable.
In wind power forecasting system of the present invention, be provided with pivot analysis Extraction parts, in the manner described above the data that partly gather by wind field information acquisition being carried out to pivot analysis, extract the pivot of right quantity.
3, the modeling of neural network prediction model and training
The present invention adopts radial basis function (RBF) neural network wind power forecast model, and the pivot that the aforementioned pivot analysis Extraction parts of usining provides is as input node, sets up to take 3 layers of neural network that blower fan output power is output node.Selecting RBF neural network is mainly to consider the following aspects.
In prior art, utilize the wind power forecasting system of artificial neural network conventionally to use BP neural network, have no and use RBF neural network to be used as wind power forecast model.These two kinds of neural networks are all applicable to approach any non-linear process by the calculating of training set, but have different separately features.
BP neural network is all carried out power and is connected, and RBF neural network input layer is to be directly to connect between hidden layer unit, and hidden layer is to weigh connection to outgoing side.BP neural network is three layers or above static feedforward neural network, and its hidden layer node number is not easy to determine there is no blanket rule; RBF neural network is three layers of static feedforward neural network, and hidden layer unit number (being network structure) can be adjusted in training stage self-adaptation.
From training algorithm, the learning process convergence of BP neural network is slower, easily be confined to local minimum, hidden layer node number is difficult to determine, and whether the final convergence of BP neural network depends on the factors such as the capacity of training sample, the initial network structure of setting up, train epochs.Therefore comparatively redundant and complicated is calculated conventionally in the training of BP neural network, and pace of learning is not ideal enough.
In RBF network structure, input layer is comprised of signal source node, and directly transmission of signal is to hidden layer; Output layer is generally simple linear function, and input pattern is responded.The transforming function transformation function of hidden layer node (basis function) is the non-negative nonlinear function to central point radial symmetry and decay, to input signal, will produce in part response, that is to say that this network has partial approximation ability, so radial primary function network is also referred to as local sensing field network.As the form of basis function, the most frequently used is generally Gaussian function.
Compare with BP neural network, online and the off-line training of the training algorithm support of RBF neural network, can dynamically determine numerical value center and the expansion constant of network structure, hidden layer unit, pace of learning is very fast, non-linear continuous function is had to Uniform Approximation, embody than the better performance of BP algorithm.
In view of network structure is dynamically determined in the support of RBF neural network, from previously described the object of the invention, in order under on-line operation state, wind power forecast model to be adjusted, embodiments of the invention are selected RBF neural network forecast model.Its ordinary construction as shown in Figure 3, wherein input node x for the anemometer tower that extracts by PCA pivot analysis and the variable in fan operation data, z, for the basis function (being hidden layer node) that rule of thumb numerical value is selected, obtains single output: the output power of blower fan after the weight w i being multiplied by separately.
The learning process of RBF neural network is mainly divided into two stages.First, according to input sample, obtain the center ci of each hidden layer node gaussian kernel function and the radius R i at each center.After definite hidden layer parameter, obtain the weight w i between hidden layer node and output layer node.
4, the online updating of RBF neural network prediction model
The present invention selects RBF neural network forecast model not only to consider the feasibility of its hidden layer node quantity of online updating, the most important thing is to utilize the linear relationship of RBF neural network from hidden layer node to output layer node.
As mentioned before, Kalman filtering algorithm has considered the state-noise of system and has measured noise, to linear system, is a kind of very effectively update method.In prior art, conventionally adopt Kalman filtering algorithm direct prediction wind power on the basis of arma modeling, or for the training dataset application card Kalman Filtering of the neural network model of having set up, to obtain the training set of systematic error reduction.
Embodiments of the invention are under the framework of wind power forecast model that uses RBF neural network support online updating (adjustment neural network structure), further utilize Kalman filtering algorithm to upgrade the structural parameters of RBF network, so that the RBF neural network of setting up more high efficient and reliable is approached real process under on-line operation state.
The system of a discrete control procedure is described with any linear random differential equation below.
X(k)=A X(k-1)+B U(k)+W(k)………(3)
The measured value of system is:
Z(k)=H X(k)+V(k)………(4)
X in above-mentioned expression formula (k) is k system state constantly, and U (k) is the k controlled quentity controlled variable to system constantly.A and B are systematic parameter (being matrix for Multi-model System).Z (k) is k measured value constantly, and H is the parameter (for many measuring systems, H is matrix) of measuring system.W (k) and V (k) represent respectively the noise of process and measurement, can assume white Gaussian noise here, and its covariance (covariance) is respectively Q, R.
First utilize the process model of system, carry out the NextState of prognoses system.Suppose that present system state is k, according to the model of system, laststate that can be based on system and predict current state:
X(k|k-1)=A X(k-1|k-1)+B U(k)………(5)
In formula (5), X (k|k-1) is the result of utilizing laststate prediction, and X (k-1|k-1) is the result of laststate optimum, the controlled quentity controlled variable that U (k) is present status.
Corresponding to the covariance P of X (k|k-1) as shown in the formula upgrading.
P(k|k-1)=A P(k-1|k-1)A’+Q………(6)
In formula (6), P (k|k-1) is the covariance that X (k|k-1) is corresponding, and P (k-1|k-1) is the covariance that X (k-1|k-1) is corresponding, and A ' represents the transposed matrix of A, and Q is the covariance of systematic procedure.
In conjunction with predicted value and measured value, can obtain the optimization estimated value X (k|k) of present status (k):
X(k|k)=X(k|k-1)+Kg(k)(Z(k)-H X(k|k-1))………(7)
Wherein Kg is kalman gain (Kalman Gain):
Kg(k)=P(k|k-1)H’/(H P(k|k-1)H’+R)………(8)
Obtained now estimated value X (k|k) optimum under k state.In order to make the continuous iteration operation of Kalman filter, upgrade the covariance of X (k|k) under k state:
P(k|k)=（I-Kg(k)H）P(k|k-1)………(9)
Above-mentioned formula (5) to (9) is 5 fundamental formulars of Kalman filtering algorithm.According to one embodiment of present invention, using the output layer weight w of the RBF neural network after aforementioned modeling training as dbjective state variable, utilize above-mentioned formula (5) to (9) to set up state equation, measurement equation and the optimal estimation equation for output layer weights, and when the deviation between blower fan output power predicted value and the actual value of blower fan output power exceeds predetermined threshold value, start renewal and optimal estimation for output layer weights.
Formula (5) is the general formula of Kalman filtering to (9).In a specific embodiment of the present invention, RBF network can be predicted once for every 15 minutes, each predict future power stage of 4 hours.Lap is as the criterion with new predicted value, as shown in Figure 7.On-line prediction surpassed after 4 hours, often obtained a measured value, in conjunction with 15 points of the point in 15 minutes intervals before it, had just had 16 measured values.For example the k+16 in upper figure, during the moment, can utilize its front 16 measured values to upgrade k model constantly.When if the predicted value of 4 hours that k starts constantly and the deviation between measured value surpass predetermined threshold, can utilize following formula (10)-(14) to carry out an iteration to model and upgrade.
For RBF network, input x obtains z after nonlinear transformation, output layer be linear
y(k)=a(k)z(k)+v(k) (10)
What formula (10) was corresponding is formula (4).Y (k) is that model prediction output is the column vector of 16*1, and it has represented the power prediction (15 minutes predicted values, 4 hours totally 16) of following 4 hours.A (k) is matrix of coefficients, the parameter that namely will upgrade during model modification.V (k) measures noise.
In order to apply kalman filtering, also needing structural regime equation, be model coefficient a (k), so state equation is chosen as follows due to what will upgrade here
a(k)=a(k-1)+w(k) (11)
What above formula was corresponding is formula (3), coefficient has been regarded as to a stochastic variable here.Contrast (3) can obtain A is unit matrix, and system is not inputted U.The meaning of Q and R is the same; H is z.Now
P(k|k-1)=P(k-1|k-1)+Q………(12)
Kg(k)=P(k|k-1)z’/(z P(k|k-1)z’+R)………(13)
a(k|k)=a(k|k-1)+Kg(k)(y(k)-a(k|k-1)z(k))………(14)
Formula (14) is the right value update to the linear output layer of RBF network; Its physical significance is very clear and definite, and the coefficient after renewal is that last coefficient adds that actual measurement and the deviation of prediction are multiplied by kalman gain.
According to the present invention, in not importing digital weather forecast (NWP) data, do not increase under the condition of system cost, can integrate the windy field information in the same area, consider the many factors such as wind speed, wind direction, weather conditions, running of wind generating set state, set up the forecast model of Power Output for Wind Power Field, can make high-precision forecast to the blower fan output power of following 4 hours.During on-line operation, by utilizing Kalman filtering algorithm to upgrade the output layer weights of radial basis function (RBF) neural network, can upgrade efficiently forecast model according to the deviation of prediction output power and measured value, guarantee prediction effect.
Claims (11)
1. a ultra-short term blower fan output power prognoses system, predicts for the output power of the target blower fan to a plurality of wind fields, and described prognoses system comprises:
Wind field information acquisition part, for gathering anemometer tower data and the fan operation data of described a plurality of wind fields, and is recorded in database it as historical information explicitly;
Pivot analysis Extraction parts, for partly reading described historical information from described wind field information acquisition, and the execution of the data based on extracting from described historical information pivot analysis, to determine the pivot conforming to a predetermined condition; With
Part of neural network, it utilizes the pivot of obtaining from described pivot analysis Extraction parts to set up radial basis function neural network (RBF) as input node, and uses the output power of the neural network prediction blower fan of gained;
Described prognoses system is characterised in that and also comprises more new portion, and described more new portion utilizes Kalman filtering algorithm to carry out online updating to the structural parameters of neural network.
2. prognoses system as claimed in claim 1, wherein said more new portion is using the weights from hidden layer node to output layer node in described neural network as carrying out by described Kalman filtering algorithm the state variable of online updating.
3. prognoses system as claimed in claim 2, wherein said pivot analysis Extraction parts is before carrying out pivot analysis, also calculate the related coefficient between the anemometer tower data of described a plurality of wind fields and the output power of described target blower fan, and extract the anemometer tower data that related coefficient in described a plurality of wind field and between the output power of described target blower fan is greater than predetermined threshold.
4. prognoses system as claimed in claim 3, the anemometer tower data of wherein said a plurality of wind fields comprise at least one in wind speed and direction.
5. prognoses system as claimed any one in claims 1 to 3, wherein said part of neural network before carrying out prediction under off-line state the historical information by described extraction radial basis function neural network is trained.
6. prognoses system as claimed in claim 1, determines pivot according to contribution rate of accumulative total over 85% criterion in wherein said pivot analysis Extraction parts.
7. a ultra-short term blower fan output power predicting method, comprises the steps:
Gather anemometer tower data and the fan operation data of a plurality of wind fields, and it is recorded in database explicitly as historical information;
Data based on extracting from described historical information are carried out pivot analysis, to determine the pivot conforming to a predetermined condition;
The determined pivot of usining is set up radial basis function neural network (RBF) as input node, and uses the output power of the radial basis function neural network prediction blower fan obtaining; With
Utilize Kalman filtering algorithm to carry out online updating to neural network.
8. Forecasting Methodology as claimed in claim 7, wherein usings in described neural network the weights from hidden layer node to output layer node as carry out the state variable of online updating by described Kalman filtering algorithm.
9. Forecasting Methodology as claimed in claim 7 or 8, before carrying out pivot analysis, also calculate the related coefficient between the anemometer tower data of described a plurality of wind fields and the output power of described target blower fan, and extract the anemometer tower data that related coefficient in described a plurality of wind field and between the output power of described target blower fan is greater than predetermined threshold.
10. Forecasting Methodology as claimed in claim 9, the anemometer tower data of wherein said a plurality of wind fields comprise at least one in wind speed and direction.
11. Forecasting Methodologies as claimed in claim 7 or 8, wherein before the described radial basis function neural network of application is carried out prediction, under off-line state, the historical information by described extraction is trained described radial basis function neural network.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310067802.XA CN104036328A (en) | 2013-03-04 | 2013-03-04 | Self-adaptive wind power prediction system and prediction method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310067802.XA CN104036328A (en) | 2013-03-04 | 2013-03-04 | Self-adaptive wind power prediction system and prediction method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN104036328A true CN104036328A (en) | 2014-09-10 |
Family
ID=51467092
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201310067802.XA Pending CN104036328A (en) | 2013-03-04 | 2013-03-04 | Self-adaptive wind power prediction system and prediction method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104036328A (en) |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104200290A (en) * | 2014-09-26 | 2014-12-10 | 国家电网公司 | Wind power forecast method |
| CN107133702A (en) * | 2017-05-18 | 2017-09-05 | 北京唐浩电力工程技术研究有限公司 | A kind of wind power plant whole audience power forecasting method |
| CN107179692A (en) * | 2017-06-27 | 2017-09-19 | 星际（重庆）智能装备技术研究院有限公司 | A kind of wind field adaptive power tracing control method based on neutral net |
| CN108376298A (en) * | 2018-02-12 | 2018-08-07 | 湘潭大学 | A kind of Wind turbines generator-temperature detection fault pre-alarming diagnostic method |
| CN108960309A (en) * | 2018-06-11 | 2018-12-07 | 宁波大学 | A kind of dynamic process monitoring method rejected based on RBF neural autocorrelation |
| CN109886126A (en) * | 2019-01-23 | 2019-06-14 | 长安大学 | A kind of region traffic density estimation method based on dynamic sampling mechanism and RBF neural |
| CN110175639A (en) * | 2019-05-17 | 2019-08-27 | 华北电力大学 | A kind of short-term wind power forecast method based on Feature Selection |
| CN111091236A (en) * | 2019-11-27 | 2020-05-01 | 长春吉电能源科技有限公司 | Multi-classification deep learning short-term wind power prediction method classified according to pitch angles |
| CN113107785A (en) * | 2021-05-12 | 2021-07-13 | 浙江浙能技术研究院有限公司 | Real-time monitoring method and device for power performance abnormity of wind turbine generator |
| CN114744623A (en) * | 2022-06-09 | 2022-07-12 | 深圳万甲荣实业有限公司 | New energy power generation prediction method and system based on deep learning |
-
2013
- 2013-03-04 CN CN201310067802.XA patent/CN104036328A/en active Pending
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104200290B (en) * | 2014-09-26 | 2017-05-10 | 国家电网公司 | Wind power forecast method |
| CN104200290A (en) * | 2014-09-26 | 2014-12-10 | 国家电网公司 | Wind power forecast method |
| CN107133702B (en) * | 2017-05-18 | 2020-11-13 | 北京唐浩电力工程技术研究有限公司 | Full-field power prediction method for wind power plant |
| CN107133702A (en) * | 2017-05-18 | 2017-09-05 | 北京唐浩电力工程技术研究有限公司 | A kind of wind power plant whole audience power forecasting method |
| CN107179692A (en) * | 2017-06-27 | 2017-09-19 | 星际（重庆）智能装备技术研究院有限公司 | A kind of wind field adaptive power tracing control method based on neutral net |
| CN108376298A (en) * | 2018-02-12 | 2018-08-07 | 湘潭大学 | A kind of Wind turbines generator-temperature detection fault pre-alarming diagnostic method |
| CN108376298B (en) * | 2018-02-12 | 2023-11-10 | 湘潭大学 | Early warning and diagnosing method for temperature faults of engine of wind turbine generator |
| CN108960309A (en) * | 2018-06-11 | 2018-12-07 | 宁波大学 | A kind of dynamic process monitoring method rejected based on RBF neural autocorrelation |
| CN109886126A (en) * | 2019-01-23 | 2019-06-14 | 长安大学 | A kind of region traffic density estimation method based on dynamic sampling mechanism and RBF neural |
| CN110175639A (en) * | 2019-05-17 | 2019-08-27 | 华北电力大学 | A kind of short-term wind power forecast method based on Feature Selection |
| CN110175639B (en) * | 2019-05-17 | 2021-06-11 | 华北电力大学 | Short-term wind power prediction method based on feature selection |
| CN111091236A (en) * | 2019-11-27 | 2020-05-01 | 长春吉电能源科技有限公司 | Multi-classification deep learning short-term wind power prediction method classified according to pitch angles |
| CN113107785A (en) * | 2021-05-12 | 2021-07-13 | 浙江浙能技术研究院有限公司 | Real-time monitoring method and device for power performance abnormity of wind turbine generator |
| CN114744623A (en) * | 2022-06-09 | 2022-07-12 | 深圳万甲荣实业有限公司 | New energy power generation prediction method and system based on deep learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Wang et al. | Research and application of a combined model based on multi-objective optimization for multi-step ahead wind speed forecasting | |
| CN104036328A (en) | Self-adaptive wind power prediction system and prediction method | |
| Di Piazza et al. | An artificial neural network-based forecasting model of energy-related time series for electrical grid management | |
| Song et al. | Hourly heat load prediction model based on temporal convolutional neural network | |
| Wang et al. | Deep belief network based deterministic and probabilistic wind speed forecasting approach | |
| CN107591844B (en) | Active power distribution network robust reconstruction method considering node injection power uncertainty | |
| Korprasertsak et al. | Robust short-term prediction of wind power generation under uncertainty via statistical interpretation of multiple forecasting models | |
| Alencar et al. | Hybrid approach combining SARIMA and neural networks for multi-step ahead wind speed forecasting in Brazil | |
| CN103117546B (en) | A kind of Ultrashort-term slide prediction method for wind power | |
| CN110766200A (en) | Method for predicting generating power of wind turbine generator based on K-means mean clustering | |
| CN112101669B (en) | Photovoltaic power interval prediction method based on improved extreme learning machine and quantile regression | |
| Sengar et al. | Ensemble approach for short term load forecasting in wind energy system using hybrid algorithm | |
| Wu et al. | Stratification-based wind power forecasting in a high-penetration wind power system using a hybrid model | |
| CN110837915B (en) | Low-voltage load point prediction and probability prediction method for power system based on hybrid integrated deep learning | |
| Kim et al. | Short-term electric load prediction using multiple linear regression method | |
| CN109636066A (en) | A kind of wind power output power prediction technique based on fuzzy time series data mining | |
| CN113988481B (en) | Wind power prediction method based on dynamic matrix prediction control | |
| Mahmoud et al. | Advanced method for short‐term wind power prediction with multiple observation points using extreme learning machines | |
| CN108694475B (en) | Short-time-scale photovoltaic cell power generation capacity prediction method based on hybrid model | |
| CN114429248A (en) | Transformer apparent power prediction method | |
| Yang et al. | Short‐term photovoltaic power dynamic weighted combination forecasting based on least squares method | |
| CN113112085A (en) | New energy station power generation load prediction method based on BP neural network | |
| CN107016470A (en) | Wind power plant wind energy Forecasting Methodology and device | |
| WO2020161624A1 (en) | Method and device for controlling a wastewatertank pumping system | |
| Yuan et al. | Short-term wind speed forecasting using STLSSVM hybrid model |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C02 | Deemed withdrawal of patent application after publication (patent law 2001) | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20140910 |