CN104636823A - Wind power prediction method - Google Patents
Wind power prediction method Download PDFInfo
- Publication number
- CN104636823A CN104636823A CN201510037230.XA CN201510037230A CN104636823A CN 104636823 A CN104636823 A CN 104636823A CN 201510037230 A CN201510037230 A CN 201510037230A CN 104636823 A CN104636823 A CN 104636823A
- Authority
- CN
- China
- Prior art keywords
- relative error
- initial
- error state
- training
- threshold
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 38
- 238000012549 training Methods 0.000 claims abstract description 70
- 238000003062 neural network model Methods 0.000 claims abstract description 45
- 238000012937 correction Methods 0.000 claims abstract description 11
- 238000004364 calculation method Methods 0.000 claims abstract description 8
- 239000011159 matrix material Substances 0.000 claims description 19
- 238000012546 transfer Methods 0.000 claims description 17
- 238000005259 measurement Methods 0.000 claims description 16
- 230000008569 process Effects 0.000 claims description 15
- 238000013277 forecasting method Methods 0.000 claims description 11
- 230000006870 function Effects 0.000 claims description 11
- 230000007704 transition Effects 0.000 claims description 10
- 230000005284 excitation Effects 0.000 claims description 6
- 238000012545 processing Methods 0.000 abstract description 2
- 238000010248 power generation Methods 0.000 abstract 2
- 239000010410 layer Substances 0.000 description 52
- 238000013528 artificial neural network Methods 0.000 description 12
- 230000005611 electricity Effects 0.000 description 5
- 238000010586 diagram Methods 0.000 description 4
- 238000011161 development Methods 0.000 description 3
- 238000013507 mapping Methods 0.000 description 3
- 238000005096 rolling process Methods 0.000 description 3
- 238000004422 calculation algorithm Methods 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000009472 formulation Methods 0.000 description 2
- 230000001394 metastastic effect Effects 0.000 description 2
- 206010061289 metastatic neoplasm Diseases 0.000 description 2
- 239000000203 mixture Substances 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 241001269238 Data Species 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 230000003749 cleanliness Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000003912 environmental pollution Methods 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 239000011229 interlayer Substances 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 230000008520 organization Effects 0.000 description 1
- 238000005309 stochastic process Methods 0.000 description 1
- 238000000714 time series forecasting Methods 0.000 description 1
Classifications
-
- 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
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a wind power prediction method. The method comprises the following steps: collecting sample data and processing; establishing a BP (Back Propagation) neural network model by using the processed sample data, and training the BP neural network model to obtain a final weight, a threshold value and a relative error sequence of a prediction value relative to a sample value; solving an initial prediction value of output power according to the weight and the threshold value after the training is completed; calculating a calculation relative error state corresponding to the initial prediction value of the output power by using a markov chain error correction model; combining the initial prediction value of the output power with the corresponding calculation relative error state to calculate corrected power. According to the method, the accuracy of wind power prediction is further improved, and an accurate wind power prediction value is provided for daily power generation scheduling and secure economic dispatch of a wind power generation power grid.
Description
Technical field
The present invention relates to power prediction field, more specifically relate to a kind of wind power forecasting method.
Background technology
Because environmental pollution and energy shortage problem are on the rise, wind-powered electricity generation with its aboundresources, cleanliness without any pollution, actual to take up an area less, the advantage such as recyclability is subject to extensive concern.But wind energy, as a kind of energy of instability, has randomness, intermittence and uncontrollability.Along with the development of wind-powered electricity generation, wind energy turbine set penetrate continuing to increase of power, grid connected wind power adds the difficulty of electric power system dispatching plan.
Unsound due to current China wind power forecasting system, lack some basic datas, degree of accuracy is not enough in addition, very difficult to the accurately predicting of wind power, therefore accurately cannot formulate the operation plan of system after wind-electricity integration, also cannot arrange the rational method of operation.
Summary of the invention
(1) technical matters that will solve
The technical problem to be solved in the present invention is the electromotive power output of how accurately predicting wind-powered electricity generation network.
(2) technical scheme
In order to solve the problems of the technologies described above, the invention provides a kind of wind power forecasting method, said method comprising the steps of:
S1, collection sample data, and described sample data is revised and normalized;
S2, utilization set up BP neural network model through the described sample data of correction and normalized, and described BP neural network model is trained, obtain the relative error sequence of predicted value relative to described sample data of BP neural network model described in the first training weights, the first training threshold value, the second training weights, the second training threshold value and training process;
S3, the described relative error sequence obtained according to described step S2, set up Markov chain VEC;
S4, input input parameter data needed for value to be predicted, according to described BP neural network model, utilize the initial prediction of described first training weights, the first training threshold value, the second training weights, the second training threshold calculations output power; Described Markov chain VEC is utilized to calculate calculating relative error state corresponding to described output power initial prediction;
S5, by described calculating relative error combinations of states corresponding with it for the initial prediction of described output power, calculate corrected output.
Preferably, described BP neural network model comprises input layer, hidden layer and output layer;
Described hidden layer exports:
Wherein, f is the excitation function of described hidden layer, L
jfor the output of a described hidden layer jth node, W
ijfor the first initial weight between described input layer i-th node and a described hidden layer jth node, b
jfor the first initial threshold between described input layer and a described hidden layer jth node, x
ifor the input of input layer i-th node, n is the nodes of described input layer;
Described output layer exports:
Wherein, f is the excitation function of described output layer, O
kfor the output of described output layer K node, W
jkfor the second initial weight between a described hidden layer jth node and a described output layer kth node, b
kfor the second initial threshold between described hidden layer and a described output layer kth node, for l is the nodes of described hidden layer;
The learning error of described BP neural network model is:
Wherein, y
kfor revising and the described sample data of normalized, m is the nodes of described output layer, and E is the learning error of described BP neural network model.
Preferably, in described step S2, train described BP neural network model, the predicted value obtaining the first training weights, the first training threshold value, the second training weights, the second training threshold value and described BP neural network model relative to the detailed process of the relative error sequence of described sample data is:
S21, described second initial weight, the second initial threshold obtain more new increment respectively according to the following formula:
Wherein, η is initial learn speed;
S22, described first initial weight, the first initial threshold obtain more new increment respectively according to the following formula:
S24, described first initial weight, the first initial threshold, the second initial weight, the second initial threshold upgrade respectively according to the following formula:
w
ij(t+1)=w
ij(t)+Δw
ij(t) (8)
w
jk(t+1)=w
jk(t)+Δw
jk(t) (9)
b
j(t+1)=b
j(t)+Δb
j(t) (10)
b
k(t+1)=b
k(t)+Δb
k(t) (11)
;
S25, utilize described step S24 to obtain output valve that described first initial weight, the first initial threshold, the second initial weight, the second initial threshold and formula (1), (2) after upgrading calculate described output layer, calculate described learning error in conjunction with formula (3);
If the described learning error of S6 is less than or equal to specification error, or described first initial weight, the first initial threshold, the second initial weight, the second initial threshold update times exceed setting iterations, then train end, described first initial weight utilizing formula (8), (9), (10), (11) to obtain for the last time, the first initial threshold, the second initial weight, the second initial threshold are the first training weights, the first training threshold value, the second training weights, the second training threshold value, and utilize relative error APE described in formulae discovery below:
Wherein, y (t) is the described sample data of correction and normalized, and O (t) is the training predicted value of described BP neural network model; The training predicted value of described BP neural network model trains weights, second to train the output valve of the described output layer of threshold calculations for utilizing formula (1), (2) and the first training weights, the first training threshold value, second;
Otherwise, get back to step S21.
Preferably, described initial learn speed η adjusts in mode below:
If the output valve of described output layer calculated according to described first initial weight after renewal, the first initial threshold, the second initial weight, the second initial threshold and the error of described sample data reduce, then increase the value of described initial learn speed; Otherwise reduce the value of described initial learn speed, and give up described first initial weight, the first initial threshold, the second initial weight, the second initial threshold after renewal, take described first initial weight, the first initial threshold, the second initial weight, the second initial threshold before upgrading.
Preferably, in described step S3, set up described Markov chain VEC and be specially:
S41, divide multiple predetermined relative error state according to the distribution density of described relative error;
S42, set up state transition probability matrix, described state transition probability matrix is:
Wherein, P
ijk () represents that relative error is from the first predetermined relative error state S
iwhen k, step transfers to the second predetermined relative error state S
jprobability.
Preferably, described relative error is from the first predetermined relative error state S
iwhen k, step transfers to the second predetermined relative error state S
jprobability P
ijk () is according to formulae discovery below:
Wherein, M
ijk () is the first predetermined relative error state S
idescribed second predetermined relative error state S is transferred to through k step
jnumber of times; M
ifor relative error is in described first-phase to error state S
inumber.
Preferably, in described step S5, calculating relative error state corresponding to the initial prediction of described Markov chain VEC calculating output power is utilized to be specially:
S43, choose the actual measurement relative error state group of described calculating relative error state; Described actual measurement relative error state group by before described calculating relative error state m calculate relative error state or predetermined relative error state, form according to time order and function order;
S44, in described actual measurement relative error state group, each calculates relative error state or predetermined relative error state, according to described state-transition matrix, calculates its time step through correspondence and transfers to predetermined relative error shape probability of state described in each;
S45, the probability utilizing described step S44 to obtain, each calculating respectively in described actual measurement relative error state group calculate relative error state or predetermined relative error state shift each predetermined relative error shape probability of state and;
S46, select in described step S45, probability and predetermined relative error state corresponding to maximal value are as described calculating relative error state.
Preferably, in described step S5, according to corrected output described in formulae discovery below:
Wherein, F (x) is corrected output; F (x) is the initial prediction of BP neural network model gained output power; Δ
dand Δ
ube respectively lower limit and the higher limit of relative error corresponding to described calculating relative error state.
Preferably, in described step S1, described sample data is normalized and is specially: adopt history maximal value to be normalized described sample data for wind speed and power; Temperature is adopted to the value of the maximal value of absolute value in historical high temperature and minimum temperature, be normalized; Wind angle sine value and cosine value are got respectively for wind direction.
(3) beneficial effect
The invention provides a kind of wind power forecasting method, the present invention is based on BP neural network and Markov chain VEC two kinds of Forecasting Methodologies, establish new wind power forecasting method, wherein adopt the rule of development of BP neural network prediction wind power, residual GM is carried out again with Markov VEC, compare single BP neural network prediction method, its result is closer to measured value; The method improves the accuracy of wind power prediction further, for the electrical network formulation daily trading planning containing wind-power electricity generation and safety economy scheduling provide wind power prediction value accurately.
Accompanying drawing explanation
In order to be illustrated more clearly in the embodiment of the present invention or technical scheme of the prior art, be briefly described to the accompanying drawing used required in embodiment or description of the prior art below, apparently, accompanying drawing in the following describes is only some embodiments of the present invention, for those of ordinary skill in the art, under the prerequisite not paying creative work, other accompanying drawing can also be obtained according to these accompanying drawings.
Fig. 1 is a kind of wind power forecasting method process flow diagram of the present invention;
Fig. 2 is the structural drawing of BP neural network in the present invention;
Fig. 3 is BP neural metwork training process flow diagram in the present invention;
Fig. 4 is a kind of wind power forecasting method of a preferred embodiment of the present invention.
Embodiment
Below in conjunction with drawings and Examples, the present invention is described in further detail.Following examples for illustration of the present invention, but can not be used for limiting the scope of the invention.
Fig. 1 is a kind of wind power forecasting method process flow diagram of the present invention, said method comprising the steps of:
S1, collection sample data, and described sample data is revised and normalized;
S2, utilization set up BP neural network model through the described sample data of correction and normalized, and described BP neural network model is trained, obtain the relative error sequence of predicted value relative to described sample data of BP neural network model described in the first training weights, the first training threshold value, the second training weights, the second training Threshold-training process;
S3, the described relative error sequence obtained according to described step S2, set up Markov chain VEC;
S4, the input parameter data needed for value to be predicted are inputted described BP neural network model, utilize the initial prediction of described first training weights, the first training threshold value, the second training weights, the second training threshold calculations output power; Described Markov chain VEC is utilized to calculate calculating relative error state corresponding to described output power initial prediction;
S5, by described calculating relative error combinations of states corresponding with it for the initial prediction of described output power, calculate corrected output.
Described BP neural network model comprises input layer, hidden layer and output layer, and as shown in Figure 2, x1, x2, x3 are input, and Y is for exporting;
Described hidden layer exports:
Wherein, f is the excitation function of described hidden layer, general Sigmoid function, and Lj is the output of a described hidden layer jth node, W
ijfor the first initial weight between described input layer i-th node and a described hidden layer jth node, b
jfor the first initial threshold between described input layer and a described hidden layer jth node, x
ifor the input of input layer i-th node, n is the nodes of described input layer;
Described output layer exports:
Wherein, f is the excitation function of described output layer, is generally Purelin function, and Ok is the output of described output layer K node, W
jkfor the second initial weight between a described hidden layer jth node and a described output layer kth node, bk is the second initial threshold between described hidden layer and a described output layer kth node, for l is the nodes of described hidden layer;
The learning error of described BP neural network model is:
Wherein, y
kfor revising and the described sample data of normalized, m is the nodes of described output layer, and E is the learning error of described BP neural network model.
In described step S2, train described BP neural network model, the predicted value obtaining the first training weights, the first training threshold value, the second training weights, the second training threshold value and described BP neural network model relative to the detailed process of the relative error sequence of described sample data is:
S21, described second initial weight, the second initial threshold obtain more new increment respectively according to the following formula:
Wherein, η is initial learn speed;
S22, described first initial weight, the first initial threshold obtain more new increment respectively according to the following formula:
S24, described first initial weight, the first initial threshold, the second initial weight, the second initial threshold upgrade respectively according to the following formula:
w
ij(t+1)=w
ij(t)+Δw
ij(t) (8)
w
jk(t+1)=w
jk(t)+Δw
jk(t) (9)
b
j(t+1)=b
j(t)+Δb
j(t) (10)
b
k(t+1)=b
k(t)+Δb
k(t) (11)
;
S25, utilize described step S24 to obtain output valve that described first initial weight, the first initial threshold, the second initial weight, the second initial threshold and formula (1), (2) after upgrading calculate described output layer, calculate described learning error in conjunction with formula (3);
If the described learning error of S6 is less than or equal to specification error, or described first initial weight, the first initial threshold, the second initial weight, the second initial threshold update times exceed setting iterations, then train end, described first initial weight utilizing formula (8), (9), (10), (11) to obtain for the last time, the first initial threshold, the second initial weight, the second initial threshold are the first training weights, the first training threshold value, the second training weights, the second training threshold value, and utilize relative error APE described in formulae discovery below:
Wherein, y (t) is the described sample data of correction and normalized, and O (t) is the training predicted value of described BP neural network model; The training predicted value of described BP neural network model trains weights, second to train the output valve of the described output layer of threshold calculations for utilizing formula (1), (2) and the first training weights, the first training threshold value, second;
Otherwise, get back to step S21.
Described initial learn speed η adjusts in mode below:
If the output valve of described output layer calculated according to described first initial weight after renewal, the first initial threshold, the second initial weight, the second initial threshold and the error of described sample data reduce, then increase the value of described initial learn speed; Otherwise reduce the value of described initial learn speed, and give up described first initial weight, the first initial threshold, the second initial weight, the second initial threshold after renewal, take described first initial weight, the first initial threshold, the second initial weight, the second initial threshold before upgrading.
Fig. 3 is BP neural network model training process flow diagram, as shown in the figure, first initialization BP neural network model, given input vector and object vector, described input vector is input to BP neural network model, hidden layer and output layer export, calculate the value of BP neural network model output and the difference of object vector, calculate learning error, weights and threshold is trained, then judge whether study terminates, if terminate, export threshold value and the weights of training, otherwise change learning rate, new weights and threshold is utilized to try to achieve the output valve of output layer, calculate the difference of itself and object vector.
Because BP algorithm learns based on error correction, the size of correction is subject to the control of learning rate by checking whether the weights after upgrading reduce the size of error to learning rate and regulate, namely when error is tending towards target in a decreasing manner, illustrate that correction is in the right direction, so step-length increases, therefore learning rate is multiplied by increment factor, and learning rate is increased; And when error increases above the value of setting, illustrate that correction is excessive, should reduce step-length, therefore learning rate is multiplied by decrement factor, and learning rate is reduced, and casts out the back makeover process that error increases simultaneously.Embodying formula is:
In described step S3, set up described Markov chain VEC and be specially:
S41, divide multiple predetermined relative error state according to the distribution density of described relative error;
S42, set up state transition probability matrix, described state transition probability matrix is:
Wherein, P
ijk () represents that relative error is from the first predetermined relative error state S
iwhen k, step transfers to the second predetermined relative error state S
jprobability.
Described relative error is from the first predetermined relative error state S
iwhen k, step transfers to the second predetermined relative error state S
jprobability P
ijk () is according to formulae discovery below:
Wherein, M
ijk () is the first predetermined relative error state S
idescribed second predetermined relative error state S is transferred to through k step
jnumber of times; M
ifor relative error is in described first-phase to error state S
inumber.
In described step S5, calculating relative error state corresponding to the initial prediction of described Markov chain VEC calculating output power is utilized to be specially:
S43, choose the actual measurement relative error state group of described calculating relative error state; Described actual measurement relative error state group by before described calculating relative error state m calculate relative error state or predetermined relative error state, form according to time order and function order;
S44, in described actual measurement relative error state group, each calculates relative error state or predetermined relative error state, according to described state-transition matrix, calculates its time step through correspondence and transfers to predetermined relative error shape probability of state described in each;
During calculating, in described actual measurement relative error state group, each calculating relative error state or predetermined relative error state are deformed into a matrix of a line multiple row, its columns is the number of described predetermined relative error state, each element of matrix is from left to right corresponding in turn to each predetermined relative error state, 1 is set at corresponding relative error state place, all the other are 0, again with being multiplied to state transition probability matrix during correspondence with it, obtain the matrix of a line multiple row, each element correspondence calculating relative error state of this matrix or predetermined relative error state are through transferring to each predetermined relative error shape probability of state to correspondence, the error state adjacent with calculating relative error state corresponding time walk be 1, other error states time walk correspondence and add 1 successively,
S45, the probability utilizing described step S44 to obtain, each calculating respectively in described actual measurement relative error state group calculate relative error state or predetermined relative error state shift each predetermined relative error shape probability of state and;
Be specially and the element of the matrix same column of the multiple 1 row multiple rows obtained be added, obtain arriving each predetermined relative error shape probability of state and;
S46, select in described step S45, probability and predetermined relative error state corresponding to maximal value are as described calculating relative error state.
In described step S5, according to corrected output described in formulae discovery below:
Wherein, F (x) is corrected output; F (x) is the initial prediction of BP neural network model gained output power; Δ
dand Δ
ube respectively lower limit and the higher limit of relative error corresponding to described calculating relative error state.
In described step S1, pre-service is carried out to data, namely pick out misdata and normalized; Described rejecting misdata mainly comprises rejects Wind turbines instrument for wind measurement fault data, Wind turbines normal or disorderly closedown data, communication failure data etc.; Described normalized comprises and adopts history maximal value to be normalized sample data to wind speed and power, temperature adopts the maximal value of historical high temperature and minimum temperature absolute value to be normalized, wind direction gets sine value and cosine value respectively, be converted to (l, 1) interval to be normalized.
BP (Back Propagation) neural network belongs to multilayer forward direction type network, can store information by multiple step format, have good fault-tolerance and robustness; Because it adopts error backpropagation algorithm to learn, therefore there is the self study to unknown message and self organization ability; Due to the mapping relations between its energy accurate description input value and output target, so any Nonlinear Mapping can be approached with arbitrary accuracy, be applicable to process challenge.
Markov chain (Markov Chain, MC) be the theory of Study system state metastatic rule, since 20 beginning of the century Russia mathematicians Markov (Markov) propose, be applied in various fields such as traffic, ecology, the hydrology, geology, achieve great successes.But in wind power prediction field, relate to markovian research application basic for blank.Markov process is the stochastic process of metastatic rule between the state of research event and state.Its state transition probability is only relevant with the initial state of transfer, the step-length of transfer, state after transfer, and state before transfer is irrelevant, Markov process markov property.Markov chain is state and time all discrete Markov process.Because MC is not by the impact of past state, for the time series forecasting by various factors, there is certain superiority.
The present invention is based on BP neural network and Markov VEC two kinds of Forecasting Methodologies, establish new wind power forecasting method, wherein adopt the rule of development of BP neural network prediction wind power, residual GM is carried out again with Markov VEC, compare single BP neural network prediction method, its result is closer to measured value; The method improves the accuracy of wind power prediction further, for the electrical network formulation daily trading planning containing wind-power electricity generation and safety economy scheduling provide wind power prediction value accurately.
Embodiment: Fig. 4 is a kind of wind power forecasting method of a preferred embodiment of the present invention;
Wind power influence factor mainly comprises wind speed, wind direction, air pressure, humidity, temperature etc., and for the newly-built wind energy turbine set of master data scarcity, also available historical wind speed and history wind power are studied.Collect the historical data of actual wind energy turbine set, data are revised and normalized.After data are disposed, BP neural network model can be set up, and to neural network model initialization.Neural network has very strong non-linear mapping capability, is particularly suitable for processing the wind power data with randomness, non-linear behavior.According to environmental impact factor and the data of the basis itself of actual wind energy turbine set, each input basic parameter of Confirming model, comprises the selection of input quantity, the nodes of each layer, maximum frequency of training, convergence error etc.According to historical sample, neural network model is trained; Judge whether study terminates, if terminate, determine the weights of each interlayer and the relative error sequence of threshold value and sample value and predicted value, and export; Otherwise change learning rate, continue studying; According to the weights exported and threshold value, the input quantity inputting data to be predicted can obtain prediction curve tentatively.
Relative error sequence according to training sample gained sets up Markov VEC.The distribution density of relative error and size are divided into n predetermined relative error state, be designated as S=[S
1, S
1, S
1... S
n], relative error is from a relative error state S
iwhen k, step transfers to another relative error state S
jprobability be
In formula: M
ijk () is for relative error is from a relative error state S
iwhen k, step transfers to another relative error state S
jnumber of times; M
ifor relative error is in relative error state S
inumber; P
ijk () is for relative error is from a relative error state S
iwhen k, step transfers to another relative error state S
jprobability.
Then walk time sequence status transition matrix P (k) during kth can be expressed as
After setting up state-transition matrix P (k), error prediction can be carried out according to P (k), revise the predicted value obtained based on BP neural network and reduce error.First choose from calculating m nearest actual measurement relative error state of relative error state, according to obtain in state-transition matrix i-th survey relative error state through k (k=m, k=m-1 ... k=1; K+i=m+1) walk shape probability of state when step transfers to prediction time, and obtained m probability summation, the predetermined relative error state residing for maximal value can think required calculating relative error state.
According to classical prediction theory, during prediction, step is more, and the error between predicted value and actual value is larger, so rolling forecast method can be adopted to reduce error for single time series.Namely the known the 1st is utilized, 2, m predetermined relative error state or calculating relative error state (the 1st relative error state can be surveyed the last relative error state rolling forecast of relative error state by the history of proxima luce (prox. luc)) predict the calculating relative error state time sequence status of m+1 thereafter by state-transition matrix P (k), then up-to-date Actual measurement relative error state is utilized, m of carrying out next time calculates relative error status predication, namely the 2nd is used, 3, the calculating relative error state of m+1 or the predetermined calculating relative error state relatively being predicted m+2 by P (k), by that analogy, rolling forecast utilizes up-to-date measured data to predict sequential below, greatly enhance the precision of prediction.
Determine and calculate relative error state, namely determine the variation range of BP neural network prediction value relative error, then the modified value of output power is
Wherein, F (x) is modified value; F (x) is the initial prediction of the output power of BP neural network; Δ
dand Δ
ube respectively the lower limit and higher limit that calculate relative error corresponding to relative error state.
Above embodiment is only for illustration of the present invention, but not limitation of the present invention.Although with reference to embodiment to invention has been detailed description, those of ordinary skill in the art is to be understood that, various combination, amendment or equivalent replacement are carried out to technical scheme of the present invention, do not depart from the spirit and scope of technical solution of the present invention, all should be encompassed in the middle of right of the present invention.
Claims (9)
1. a wind power forecasting method, is characterized in that, said method comprising the steps of:
S1, collection sample data, and described sample data is revised and normalized;
S2, utilization set up BP neural network model through the described sample data of correction and normalized, and described BP neural network model is trained, obtain the relative error sequence of predicted value relative to described sample data of BP neural network model described in the first training weights, the first training threshold value, the second training weights, the second training threshold value and training process;
S3, the described relative error sequence obtained according to described step S2, set up Markov chain VEC;
S4, the input parameter data needed for value to be predicted are inputted described BP neural network model, utilize the initial prediction of described first training weights, the first training threshold value, the second training weights, the second training threshold calculations output power; Described Markov chain VEC is utilized to calculate calculating relative error state corresponding to described output power initial prediction;
S5, by described calculating relative error combinations of states corresponding with it for the initial prediction of described output power, calculate corrected output.
2. method according to claim 1, is characterized in that, described BP neural network model comprises input layer, hidden layer and output layer;
Described hidden layer exports:
Wherein, f is the excitation function of described hidden layer, L
jfor the output of a described hidden layer jth node, W
ijfor the first initial weight between described input layer i-th node and a described hidden layer jth node, b
jfor the first initial threshold between described input layer and a described hidden layer jth node, x
ifor the input of input layer i-th node, n is the nodes of described input layer;
Described output layer exports:
Wherein, f is the excitation function of described output layer, O
kfor the output of described output layer K node, W
jkfor the second initial weight between a described hidden layer jth node and a described output layer kth node, b
kfor the second initial threshold between described hidden layer and a described output layer kth node, l is the nodes of described hidden layer;
The learning error of described BP neural network model is:
Wherein, y
kfor revising and the described sample output parameter data of normalized, m is the nodes of described output layer, and E is the learning error of described BP neural network model.
3. method according to claim 2, it is characterized in that, in described step S2, train described BP neural network model, the predicted value obtaining the first training weights, the first training threshold value, the second training weights, the second training threshold value and described BP neural network model relative to the detailed process of the relative error sequence of described sample data is:
S21, described second initial weight, the second initial threshold obtain more new increment respectively according to the following formula:
Wherein, η is initial learn speed;
S22, described first initial weight, the first initial threshold obtain more new increment respectively according to the following formula:
S24, described first initial weight, the first initial threshold, the second initial weight, the second initial threshold upgrade respectively according to the following formula:
w
ij(t+1)=w
ij(t)+Δw
ij(t) (8)
w
jk(t+1)=w
jk(t)+Δw
jk(t) (9)
b
j(t+1)=b
j(t)+Δb
j(t) (10)
b
k(t+1)=b
k(t)+Δb
k(t) (11)
;
S25, utilize described step S24 to obtain output valve that described first initial weight, the first initial threshold, the second initial weight, the second initial threshold and formula (1), (2) after upgrading calculate described output layer, calculate described learning error in conjunction with formula (3);
If the described learning error of S6 is less than or equal to specification error, or described first initial weight, the first initial threshold, the second initial weight, the second initial threshold update times exceed setting iterations, then train end, described first initial weight utilizing formula (8), (9), (10), (11) to obtain for the last time, the first initial threshold, the second initial weight, the second initial threshold are the first training weights, the first training threshold value, the second training weights, the second training threshold value, and utilize relative error APE described in formulae discovery below:
Wherein, y (t) is the described sample data of correction and normalized, and O (t) is the training predicted value of described BP neural network model; The training predicted value of described BP neural network model trains weights, second to train the output valve of the described output layer of threshold calculations for utilizing formula (1), (2) and the first training weights, the first training threshold value, second;
Otherwise, get back to step S21.
4. method according to claim 3, is characterized in that, described initial learn speed η adjusts in mode below:
If the output valve of described output layer calculated according to described first initial weight after renewal, the first initial threshold, the second initial weight, the second initial threshold and the error of described sample data reduce, then increase the value of described initial learn speed; Otherwise reduce the value of described initial learn speed, and give up described first initial weight, the first initial threshold, the second initial weight, the second initial threshold after renewal, take described first initial weight, the first initial threshold, the second initial weight, the second initial threshold before upgrading.
5. the method according to any one of Claims 1-4, is characterized in that, in described step S3, sets up described Markov chain VEC and is specially:
S41, divide multiple predetermined relative error state according to the distribution density of described relative error;
S42, set up state transition probability matrix, described state transition probability matrix is:
Wherein, P
ijk () represents that relative error is from the first predetermined relative error state S
iwhen k, step transfers to the second predetermined relative error state S
jprobability.
6. method according to claim 5, is characterized in that, described relative error is from the first predetermined relative error state S
iwhen k, step transfers to the second predetermined relative error state S
jprobability P
ijk () is according to formulae discovery below:
Wherein, M
ijk () is the first predetermined relative error state S
idescribed second predetermined relative error state S is transferred to through k step
jnumber of times; M
ifor relative error is in described first-phase to error state S
inumber.
7. method according to claim 6, is characterized in that, in described step S5, utilizes calculating relative error state corresponding to the initial prediction of described Markov chain VEC calculating output power to be specially:
S43, choose the actual measurement relative error state group of described calculating relative error state; Described actual measurement relative error state group by before described calculating relative error state m calculate relative error state or predetermined relative error state, form according to time order and function order;
S44, in described actual measurement relative error state group, each calculates relative error state or predetermined relative error state, according to described state transition probability matrix, calculates its time step through correspondence and transfers to predetermined relative error shape probability of state described in each;
S45, the probability utilizing described step S44 to obtain, each calculating respectively in described actual measurement relative error state group calculate relative error state or predetermined relative error state shift each predetermined relative error shape probability of state and;
S46, select in described step S45, probability and predetermined relative error state corresponding to maximal value are as described calculating relative error state.
8. method according to claim 7, is characterized in that, in described step S5, according to corrected output described in formulae discovery below:
Wherein, F (x) is corrected output; F (x) is the initial prediction of BP neural network model gained output power; Δ
dand Δ
ube respectively lower limit and the higher limit of relative error corresponding to described calculating relative error state.
9. method according to claim 8, is characterized in that, in described step S1, is normalized is specially described sample data: adopt history maximal value to be normalized described sample data for wind speed and power; For temperature, the value that in historical high temperature and minimum temperature, absolute value is large is adopted to be normalized; Wind angle sine value and cosine value are got respectively for wind direction.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510037230.XA CN104636823B (en) | 2015-01-23 | 2015-01-23 | A kind of wind power forecasting method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510037230.XA CN104636823B (en) | 2015-01-23 | 2015-01-23 | A kind of wind power forecasting method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN104636823A true CN104636823A (en) | 2015-05-20 |
CN104636823B CN104636823B (en) | 2018-02-16 |
Family
ID=53215547
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201510037230.XA Expired - Fee Related CN104636823B (en) | 2015-01-23 | 2015-01-23 | A kind of wind power forecasting method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN104636823B (en) |
Cited By (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107947738A (en) * | 2017-12-11 | 2018-04-20 | 南京航空航天大学 | A kind of Forecasting Methodology of solar energy unmanned plane cell plate voltage |
CN107958297A (en) * | 2016-10-17 | 2018-04-24 | 华为技术有限公司 | A kind of product demand forecasting method and product demand prediction meanss |
CN108564206A (en) * | 2018-03-27 | 2018-09-21 | 中国农业大学 | A kind of wind power forecasting method based on distributed optimization and spatial coherence |
CN109165797A (en) * | 2018-10-12 | 2019-01-08 | 国网宁夏电力有限公司 | A kind of wind power forecasting method and device |
CN110059392A (en) * | 2019-04-11 | 2019-07-26 | 桂林电子科技大学 | A kind of landslide deformation prediction method |
CN110165693A (en) * | 2019-06-27 | 2019-08-23 | 上海电机学院 | A kind of wind power plant optimization energy storage configuration method |
CN110717610A (en) * | 2018-07-11 | 2020-01-21 | 华北电力大学(保定) | Wind power prediction method based on data mining |
CN110874616A (en) * | 2019-11-18 | 2020-03-10 | 苏文电能科技股份有限公司 | Transformer operation prediction method based on LSTM network and Markov chain correction error |
CN111765075A (en) * | 2020-05-20 | 2020-10-13 | 天津市天锻压力机有限公司 | Hydraulic forging press pump source fault prediction method and system |
CN112749792A (en) * | 2021-02-02 | 2021-05-04 | 南京信息工程大学 | Wind speed prediction method based on BP algorithm |
CN112948462A (en) * | 2021-02-26 | 2021-06-11 | 武汉大学 | Ultra-short-term wind speed prediction method based on improved singular spectrum analysis and Bp neural network |
CN113591957A (en) * | 2021-07-21 | 2021-11-02 | 国网上海市电力公司 | Wind power output short-term rolling prediction and correction method based on LSTM and Markov chain |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101706335A (en) * | 2009-11-11 | 2010-05-12 | 华南理工大学 | Wind power forecasting method based on genetic algorithm optimization BP neural network |
CN102102626A (en) * | 2011-01-30 | 2011-06-22 | 华北电力大学 | Method for forecasting short-term power in wind power station |
CN103400204A (en) * | 2013-07-26 | 2013-11-20 | 华南理工大学 | Forecasting method for solar photovoltaic electricity generation amount based on SVM (support vector machine) - Markov combination method |
CN103473618A (en) * | 2013-09-18 | 2013-12-25 | 上海电机学院 | System and method for predicting short-term wind power of wind power plant |
CN104299031A (en) * | 2014-06-13 | 2015-01-21 | 国家电网公司 | Ultra-short-term load prediction method of BP neural network |
-
2015
- 2015-01-23 CN CN201510037230.XA patent/CN104636823B/en not_active Expired - Fee Related
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101706335A (en) * | 2009-11-11 | 2010-05-12 | 华南理工大学 | Wind power forecasting method based on genetic algorithm optimization BP neural network |
CN102102626A (en) * | 2011-01-30 | 2011-06-22 | 华北电力大学 | Method for forecasting short-term power in wind power station |
CN103400204A (en) * | 2013-07-26 | 2013-11-20 | 华南理工大学 | Forecasting method for solar photovoltaic electricity generation amount based on SVM (support vector machine) - Markov combination method |
CN103473618A (en) * | 2013-09-18 | 2013-12-25 | 上海电机学院 | System and method for predicting short-term wind power of wind power plant |
CN104299031A (en) * | 2014-06-13 | 2015-01-21 | 国家电网公司 | Ultra-short-term load prediction method of BP neural network |
Cited By (18)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107958297A (en) * | 2016-10-17 | 2018-04-24 | 华为技术有限公司 | A kind of product demand forecasting method and product demand prediction meanss |
CN107958297B (en) * | 2016-10-17 | 2021-10-26 | 华为技术有限公司 | Product demand prediction method and product demand prediction device |
CN107947738A (en) * | 2017-12-11 | 2018-04-20 | 南京航空航天大学 | A kind of Forecasting Methodology of solar energy unmanned plane cell plate voltage |
CN108564206B (en) * | 2018-03-27 | 2021-08-17 | 中国农业大学 | Wind power prediction method based on distributed optimization and spatial correlation |
CN108564206A (en) * | 2018-03-27 | 2018-09-21 | 中国农业大学 | A kind of wind power forecasting method based on distributed optimization and spatial coherence |
CN110717610B (en) * | 2018-07-11 | 2023-10-31 | 华北电力大学(保定) | Wind power prediction method based on data mining |
CN110717610A (en) * | 2018-07-11 | 2020-01-21 | 华北电力大学(保定) | Wind power prediction method based on data mining |
CN109165797A (en) * | 2018-10-12 | 2019-01-08 | 国网宁夏电力有限公司 | A kind of wind power forecasting method and device |
CN110059392A (en) * | 2019-04-11 | 2019-07-26 | 桂林电子科技大学 | A kind of landslide deformation prediction method |
CN110165693B (en) * | 2019-06-27 | 2023-05-02 | 上海电机学院 | Wind farm optimized energy storage configuration method |
CN110165693A (en) * | 2019-06-27 | 2019-08-23 | 上海电机学院 | A kind of wind power plant optimization energy storage configuration method |
CN110874616A (en) * | 2019-11-18 | 2020-03-10 | 苏文电能科技股份有限公司 | Transformer operation prediction method based on LSTM network and Markov chain correction error |
CN111765075A (en) * | 2020-05-20 | 2020-10-13 | 天津市天锻压力机有限公司 | Hydraulic forging press pump source fault prediction method and system |
CN112749792A (en) * | 2021-02-02 | 2021-05-04 | 南京信息工程大学 | Wind speed prediction method based on BP algorithm |
CN112749792B (en) * | 2021-02-02 | 2023-07-07 | 南京信息工程大学 | BP algorithm-based wind speed prediction method |
CN112948462A (en) * | 2021-02-26 | 2021-06-11 | 武汉大学 | Ultra-short-term wind speed prediction method based on improved singular spectrum analysis and Bp neural network |
CN113591957A (en) * | 2021-07-21 | 2021-11-02 | 国网上海市电力公司 | Wind power output short-term rolling prediction and correction method based on LSTM and Markov chain |
CN113591957B (en) * | 2021-07-21 | 2023-10-27 | 国网上海市电力公司 | Wind power output short-term rolling prediction and correction method based on LSTM and Markov chain |
Also Published As
Publication number | Publication date |
---|---|
CN104636823B (en) | 2018-02-16 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN104636823A (en) | Wind power prediction method | |
CN104779611B (en) | Micro-capacitance sensor economic load dispatching method based on centralized and distributed dual-layer optimization strategy | |
Rajkumar et al. | Techno-economical optimization of hybrid pv/wind/battery system using Neuro-Fuzzy | |
CN107425520A (en) | A kind of probabilistic active distribution network three-phase section method for estimating state of injecting power containing node | |
Capizzi et al. | Recurrent neural network-based control strategy for battery energy storage in generation systems with intermittent renewable energy sources | |
CN105633948B (en) | A kind of distributed energy accesses electric system Random-fuzzy power flow algorithm | |
CN105527965A (en) | Route planning method and system based on genetic ant colony algorithm | |
CN102478584B (en) | Wind power station wind speed prediction method based on wavelet analysis and system thereof | |
CN107516145A (en) | A kind of multichannel photovoltaic power generation output forecasting method based on weighted euclidean distance pattern classification | |
CN104732295A (en) | Power load predicating model based on big data technology | |
CN101888087A (en) | Method for realizing distributed super-short-term area load forecasting in distribution network terminal | |
CN105160423A (en) | Photovoltaic power generation prediction method based on Markov residual error correction | |
CN102938562B (en) | Prediction method of total wind electricity power in area | |
CN105826944A (en) | Method and system for predicting power of microgrid group | |
CN112215428B (en) | Photovoltaic power generation power prediction method and system based on error correction and fuzzy logic | |
CN105337310A (en) | Series-structure light storage type multi-microgrid economic operation system and method | |
CN107292502A (en) | A kind of distribution network reliability evaluation method | |
CN109345009A (en) | Second order grey neural network and markovian short-term wind speed interval prediction technique | |
CN103927460A (en) | Wind power plant short-term wind speed prediction method based on RBF | |
CN104732300A (en) | Neural network wind power short-term forecasting method based on fuzzy partition theory | |
CN115374692B (en) | Double-layer optimization scheduling decision method for regional comprehensive energy system | |
Alam et al. | A new subtractive clustering based ANFIS system for residential load forecasting | |
CN110363334A (en) | Grid-connected grid line loss prediction technique based on Grey Neural Network Model | |
CN104021315A (en) | Method for calculating station service power consumption rate of power station on basis of BP neutral network | |
Chatziagorakis et al. | Enhancement of hybrid renewable energy systems control with neural networks applied to weather forecasting: the case of Olvio |
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 | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20180216 |
|
CF01 | Termination of patent right due to non-payment of annual fee |