CN106529814A - Distributed photovoltaic ultra-short-term forecasting method based on Adaboost clustering and Markov chain - Google Patents
Distributed photovoltaic ultra-short-term forecasting method based on Adaboost clustering and Markov chain Download PDFInfo
- Publication number
- CN106529814A CN106529814A CN201611024358.3A CN201611024358A CN106529814A CN 106529814 A CN106529814 A CN 106529814A CN 201611024358 A CN201611024358 A CN 201611024358A CN 106529814 A CN106529814 A CN 106529814A
- Authority
- CN
- China
- Prior art keywords
- intensity
- illumination
- prediction
- state
- error
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0637—Strategic management or analysis, e.g. setting a goal or target of an organisation; Planning actions based on goals; Analysis or evaluation of effectiveness of goals
- G06Q10/06375—Prediction of business process outcome or impact based on a proposed change
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
Abstract
The invention relates to the technical field of distributed photovoltaic power generation systems, and particularly relates to a distributed photovoltaic ultra-short-term forecasting method based on Adaboost clustering and Markov chains. The method comprises the following steps: 1, extracting deterministic components of illumination intensity sequences by using a sliding average method, and implementing statistical analysis to obtain illumination intensity attenuation factors of different weather types; 2, implementing clustering analysis on historical data by using an Adaboost improved KNN method to establish a classification model; 3, forecasting the solar irradiance of the earth surface by using a multistage weighted Markov chain method; and 4, establishing a photoelectric conversion model to complete the ultra-short-term forecasting of photovoltaic power. According to a combination forecasting method for implementing feature extraction and data mining on input data, provided by the invention, after the historical photovoltaic output data is classified according to the typical weather type, the state of the forecasting process is refined by introducing the weather type attenuation factors, so that better forecasting effects can be achieved in sunny weather, and the forecasting precision and accuracy in non-sunny weather can also be increased.
Description
Technical field
The invention belongs to distributed photovoltaic power generation systems technology field, more particularly to it is a kind of based on Adaboost clusters and horse
The distributed photovoltaic ultra-short term prediction method of Er Kefu chains.
Background technology
Photovoltaic generation installed capacity is continuously increased in recent years, its fluctuation and intermittence Operation of Electric Systems is caused rush
Hit outstanding day by day.The ultra-short term prediction of distributed photovoltaic is significant to power system security economical operation, is mainly reflected in
Two aspects:First, control strategy is formulated based on pre- power scale, shadow of the power swing to power system when reducing grid-connected
Ring, improve security, reliability and the controllability of system;Second, contribute to electric power system dispatching department and determine unit output meter
Draw, spinning reserve in reduction system, reduce electric cost.Under the precondition for meeting safety and stability, dissolve to greatest extent
Regenerative resource is increasingly becoming the emphasis of photovoltaic generation research field, therefore power prediction to distribution and grid-connected photovoltaic generation
The extensive development of system has practical value.
Photovoltaic power output shows randomness and fluctuation, different weather type under the influence of earth's surface intensity of illumination
Under, there is obvious difference in photovoltaic power output Changing Pattern, therefore be difficult with unified forecast model in actual applications
Photovoltaic under each meteorological condition is exerted oneself and is predicted.Forecast is dug by feature extraction and data are carried out to input data
The combination forecasting method of pick, can improve precision of prediction compared with Individual forecast method.As high cost numerical weather forecast is difficult to
It is widely used in distributed photovoltaic forecasting system, current great majority research is adopted based on weather forecast information and cloud amount information to not
The combination forecasting method classified by same type data, does not fully excavate history intensity of solar radiation information, and precision of prediction exists
It is remarkably decreased in the case of weather forecast is inaccurate.
The content of the invention
It is an object of the invention to provide a kind of clustered based on Adaboost and markovian combination forecasting method, not only
Preferable prediction effect can be obtained under fair weather, the precision of prediction under non-fair weather is also improved.
For achieving the above object, the technical solution used in the present invention is:Clustered based on Adaboost and markovian
Distributed photovoltaic ultra-short term prediction method, comprises the following steps:
Step 1, the certainty component that intensity of illumination sequence is extracted using moving average method, statistical analysis obtain different weather
The intensity of illumination decay factor of type;
Step 2, cluster analysis is carried out using the improved KNN methods of Adaboost to historical data, set up classification prediction mould
Type;
Step 3, using multistage weighting Markov Chain method prediction earth's surface solar irradiance;
Step 4, opto-electronic conversion model is set up, complete the prediction of photovoltaic power ultra-short term.
It is above-mentioned based on Adaboost cluster and markovian distributed photovoltaic ultra-short term prediction method in, step
1 realization includes:
The 1.1 random fluctuation components that intensity of illumination is filtered using moving average method, are extracted certainty component and select this point
Measure the characteristic variable as cluster analysis;
(1) y in formulatIntensity measurement value is shone for primary light, m selects moving average time window, YtFor the cunning after smoothing processing
Dynamic mean value;
1.2 calculate earth's surface solar irradiance Hottel standard fine day theoretical value according to astronomical solar-terrestrial relationship, by historical data
Contrast statistics is carried out with corresponding standard fine day theoretical value, the intensity of illumination decay factor empirical value of different weather type defeated is obtained
Enter forecast model;
(2) in formula, Imea(i) i-th moment earth's surface intensity of illumination, IstdI () Hottel standard fine day theoretical values, m are decay
The factor.
It is above-mentioned based on Adaboost cluster and markovian distributed photovoltaic ultra-short term prediction method in, step
2 realization includes:
2.1 are clustered using k-means, by arranging cluster centre number k, selecting initial cluster center, and are being changed every time
Sample is classified as into closest center generic in generation, again according in the Sample Refreshment cluster in classification after each iteration
The heart, until the change of two subcenters in front and back is less than arranges value or iterations is less than arranges value;
2.2 adopt KNN sorting algorithms, are made by classification maximum in the classification proportion of now k closest sample
For the classification of unknown object;Or according to the distance classification maximum to neighbouring class weights selection result as unknown object
Classification;
2.3 adopt Adaboost algorithm, obtain certain Weak Classifier G by training data studym(x), by weak point
The combination of class device obtains strong classifier, i.e. strong classifier and can be expressed as:
(3) α in formulamThe weighted value for being the m time calculated Weak Classifier in strong classifier.
It is above-mentioned based on Adaboost cluster and markovian distributed photovoltaic ultra-short term prediction method in, step
3 realization includes:
3.1 pairs of historical data pretreatments, obtain error percentage sequence, define relative prediction residual eitFor:
(4) in formulaFor intensity of illumination predicted value,For intensity of illumination actual value, i is daily prediction points;
N is sample number;
3.2 state demarcations, research object are the error between Hottel standard fine day model values and actual value, will pretreatment
In the error percentage that obtains be divided into 11 discrete states by practical operating experiences, wherein point of the error less than -5% is defined as
State 1, it is 10% that the error of adjacent two state is differential, the like, point of the error more than 85% is defined as state 11;
3.3 ask for state transition probability matrix;
It is S in the state of m moment solar radiationsiThe state for being transferred to m+k moment solar radiations is SjTransition probability
For:
Using the result of above-mentioned state demarcation, the transfer frequency square of each step-length (exponent number k is different) state change is counted
Battle array:
3.4 calculate each rank weighing factor;
Error sequence k rank auto-correlation coefficient computing formula are as follows:
(7) x in formulaiThe error of the i-th period is represented,For mean error, n is time span;
Each rank weighing factor is calculated by each rank auto-correlation coefficient that (7) formula is tried to achieve:
3.5 earth's surface intensities of illumination are predicted;
It is after the state transition probability matrix of the different step-lengths for obtaining, pre- by prediction period day error percentage place state
A certain shape probability of state that next sampling instant error percentage is located is measured, and obtains a state of maximum probability and missed for prediction
Difference state, so as to obtain corresponding relative prediction residual value e (t);Calculate earth's surface intensity of illumination predicted value:
Ipre(t)=Istd(t)×(1-e(t)) (9);
(9) in formula, IpreT () is t earth's surface intensity of illumination predicted value, IstdT () is t standard fine day theoretical value, e
T () is prediction calculating error percentage.
It is above-mentioned based on Adaboost cluster and markovian distributed photovoltaic ultra-short term prediction method in, step
4 realization includes:
Using indirect prediction method, after obtaining the prediction numerical value of intensity of illumination, photovoltaic output is calculated by opto-electronic conversion model
Power, computing formula is:
P=η SI [1-0.005 (t0+25)] (10);
(10) in formula, η is photoelectric transformation efficiency, and S is the effective gross area for photovoltaic generation, and I (t) is having of receiving
Effect intensity of illumination, t0For the operating temperature of photovoltaic cell.
The invention has the beneficial effects as follows:The present invention is proposed based on Adaboost clusters and markovian combined prediction
Method, the method not only can obtain preferable prediction effect under fair weather, also improve the prediction essence under non-fair weather
Degree.Adaboost improved k nearest neighbor (KNN) algorithms need not know error boundary in advance, with adaptivity, can strengthen weak point
Classifying quality of the class device to unbalanced dataset.Markov Chain (Markov Chain) is by division state and use state
Between probability transfer process forecasting system after possible development trend and determine unknown state, it is adaptable to random uncertain
Event is tracked prediction.Introduce different weather type light and shine strength retrogression's factor, the change that can amplify in predicated error is thin
Section, and amplitude is reflected relative to the less error change of clear sky model value, with significantly improving under rainy weather type
The precision of prediction of table intensity of illumination.Comprehensive multifactor multistep modeling, with the wider scope of application, sets up light using engineering experience
Electric transformation model simplifies power prediction problem.
Description of the drawings
Fig. 1 is the ultra-short term prediction method flow chart of one embodiment of the invention;
Fig. 2 is the extraction figure of one embodiment of the invention moving average certainty variable;
Fig. 3 is the interval distribution map of one embodiment of the invention intensity of illumination pad value;
Fig. 4 is the four classes cluster Centered Graphs of one embodiment of the invention earth's surface intensity of illumination;
Fig. 5 is one embodiment of the invention typical case day actual measurement intensity of illumination and standard fine day intensity of illumination correlation curve;
Fig. 6 is the model under one embodiment of the invention fine day and traditional markovian prediction correlation curve;
Fig. 7 be one embodiment of the invention it is cloudy under model and traditional markovian prediction correlation curve;
Fig. 8 is the model under the one embodiment of the invention rainy day and traditional markovian prediction correlation curve;
Fig. 9 is model and traditional markovian prediction correlation curve under the one embodiment of the invention cloudy day;
Specific embodiment
Below in conjunction with the accompanying drawings embodiments of the present invention are described in detail.
The example of the embodiment is shown in the drawings, wherein same or similar label represents identical or class from start to finish
As element or the element with same or like function.Below with reference to Description of Drawings embodiment be it is exemplary, only
For explaining the present invention, and it is not construed as limiting the claims.
Embodiment is adopted the following technical scheme that:Based on Adaboost clusters and markovian distributed photovoltaic ultra-short term
Forecasting Methodology, comprises the following steps:
Step 1, the certainty component that intensity of illumination sequence is extracted using moving average method, statistical analysis obtain different weather
The intensity of illumination decay factor of type;
Step 2, cluster analysis is carried out using the improved KNN methods of Adaboost to historical data, set up classification prediction mould
Type;
Step 3, using multistage weighting Markov Chain method prediction earth's surface solar irradiance;
Step 4, opto-electronic conversion model is set up, complete the prediction of photovoltaic power ultra-short term.
Further, the realization of step 1 includes:
The 1.1 random fluctuation components that intensity of illumination is filtered using moving average method, are extracted certainty component and select this point
Measure the characteristic variable as cluster analysis;
(1) y in formulatIntensity measurement value is shone for primary light, m selects moving average time window, YtFor the cunning after smoothing processing
Dynamic mean value;
1.2 calculate earth's surface solar irradiance Hottel standard fine day theoretical value according to astronomical solar-terrestrial relationship, by historical data
Contrast statistics is carried out with corresponding standard fine day theoretical value, the intensity of illumination decay factor empirical value of different weather type defeated is obtained
Enter forecast model;
(2) in formula, Imea(i) i-th moment earth's surface intensity of illumination, IstdI () Hottel standard fine day theoretical values, m are decay
The factor.
Further, the realization of step 2 includes:
2.1 are clustered using k-means, by arranging cluster centre number k, selecting initial cluster center, and are being changed every time
Sample is classified as into closest center generic in generation, again according in the Sample Refreshment cluster in classification after each iteration
The heart, until the change of two subcenters in front and back is less than arranges value or iterations is less than arranges value;
2.2 adopt KNN sorting algorithms, are made by classification maximum in the classification proportion of now k closest sample
For the classification of unknown object;Or according to the distance classification maximum to neighbouring class weights selection result as unknown object
Classification;
2.3 adopt Adaboost algorithm, obtain certain Weak Classifier G by training data studym(x), by weak point
The combination of class device obtains strong classifier, i.e. strong classifier and can be expressed as:
(3) α in formulamThe weighted value for being the m time calculated Weak Classifier in strong classifier.
Further, the realization of step 3 includes:
3.1 pairs of historical data pretreatments, obtain error percentage sequence, define relative prediction residual eitFor:
(4) in formulaFor intensity of illumination predicted value;For intensity of illumination actual value;I is daily prediction points;
N is sample number;
3.2 state demarcations, research object are the error between Hottel standard fine day model values and actual value, will pretreatment
In the error percentage that obtains be divided into 11 discrete states by practical operating experiences, wherein point of the error less than -5% is defined as
State 1, it is 10% that the error of adjacent two state is differential, the like, point of the error more than 85% is defined as state 11;
3.3 ask for state transition probability matrix;
It is S in the state of m moment solar radiationsiThe state for being transferred to m+k moment solar radiations is SjTransition probability
For:
Using the result of above-mentioned state demarcation, the transfer frequency square of each step-length (exponent number k is different) state change is counted
Battle array:
3.4 calculate each rank weighing factor;
Error sequence k rank auto-correlation coefficient computing formula are as follows:
(7) x in formulaiThe error of the i-th period is represented,For mean error, n is time span;
Each rank weighing factor is calculated by each rank auto-correlation coefficient that (7) formula is tried to achieve:
3.5 earth's surface intensities of illumination are predicted;
It is after the state transition probability matrix of the different step-lengths for obtaining, pre- by prediction period day error percentage place state
A certain shape probability of state that next sampling instant error percentage is located is measured, and obtains a state of maximum probability and missed for prediction
Difference state, so as to obtain corresponding relative prediction residual value e (t);Calculate earth's surface intensity of illumination predicted value:
Ipre(t)=Istd(t)×(1-e(t)) (9);
(9) in formula, IpreT () is t earth's surface intensity of illumination predicted value, IstdT () is t standard fine day theoretical value, e
T () is prediction calculating error percentage.
Further, the realization of step 4 includes:
Using indirect prediction method, after obtaining the prediction numerical value of intensity of illumination, photovoltaic output is calculated by opto-electronic conversion model
Power, computing formula is:
P=η SI [1-0.005 (t0+25)] (10);
(10) in formula, η is photoelectric transformation efficiency, and S is the effective gross area for photovoltaic generation, and I (t) is having of receiving
Effect intensity of illumination, t0For the operating temperature of photovoltaic cell.
Embodiment 1
In order to improve the precision of prediction in the case of non-fine day, propose a kind of based on Adaboost clusters and markovian
Distributed photovoltaic ultra-short term prediction method, extracts the certainty component of intensity of illumination sequence, statistical analysis using moving average method
Obtain the intensity of illumination decay factor of different weather type;Using cluster of the improved KNN methods of Adaboost to historical data
Analysis, sets up disaggregated model;Realize that future time instance earth's surface illumination is strong with reference to based on the multistage weighting Markov Chain of error sequence
The prediction of degree, predicts finally by opto-electronic conversion model realization distributed photovoltaic power generation ultra-short term.
Embodiment 1 is addressed by the following technical programs:
It is a kind of to be clustered based on Adaboost and markovian distributed photovoltaic ultra-short term prediction method, including following step
Suddenly:
S1, extracts the certainty component of intensity of illumination sequence using moving average method, and statistical analysis obtains different weather class
The intensity of illumination decay factor of type;
S2, using cluster analysis of the improved KNN methods of Adaboost to historical data, sets up disaggregated model;
S3, using multistage weighting Markov Chain method prediction earth's surface solar irradiance;
S4, sets up opto-electronic conversion model, completes the prediction of photovoltaic power ultra-short term.
The certainty component of intensity of illumination sequence is extracted in step S1 using moving average method, statistical analysis is obtained not
Flow process with the intensity of illumination decay factor of weather pattern is as follows:
Select moving average method to filter the random fluctuation component of intensity of illumination, extract certainty component and select the component to make
For the characteristic variable of cluster analysis.If primary light is y according to intensity measurement valuet, select moving average time window m, smoothing processing
Sliding average afterwards is Yt。
Earth's surface intensity of illumination is carved with different degrees of decline, the i-th moment earth's surface intensity of illumination I in cloud covermea(i)
With Hottel standard fine day theoretical values IstdThere are different multiplying power relations between (i), be designated as decay factor m.Its expression formula is as follows
It is shown:
Under different weather type, earth's surface irradiation level is different compared with the attenuation degree of ground external irradiation degree.
Step S2 adopt the improved KNN methods of Adaboost to the pre- flow gauge of the cluster analysis of historical data for:
1) batch input training data, extracts the mathematical feature of data, determines Weak Classifier G according to KNNm(x)
Number;
2) determine the iterations k in Adaboost learning processes, initialize weight d of each sample datami。
3) sample point is selected according to weight, obtains new training subset, apparatus have the right Distribution value training data training it is weak
Grader GmX (), uses GmX () is classified to all samples, calculate Weak Classifier GmThe error rate e of (x) under training setm
And its factor alpha in final classification devicem。
4) sample weights correctly classified are improved, is reduced by the sample weights of mistake classification, be that next round iteration updates
The weight distribution of data sample:
After obtaining next round weight, return to step (3) carries out next round iteration, until iterations reaches arranges value or mistake
Rate reaches the sufficiently small value of certain setting.
Step S3 is as follows using multistage weighting Markov Chain method prediction earth's surface solar irradiance flow process:
I, to historical data pre-process, obtain error percentage sequence, define relative prediction residual eitFor:
WhereinFor intensity of illumination predicted value;AndFor intensity of illumination actual value;I is daily prediction points;
N is sample number.
II, the error percentage obtained in pretreatment is divided into into 11 discrete states by practical operating experiences, wherein missing
Point of the difference less than -5% is defined as state 1, and it is 10% that the error of adjacent two state is differential, the like, error is more than 85%
Point is defined as state 11.
III, the transfer frequency for using the result of above-mentioned state demarcation, counting each step-length (exponent number k is different) state change
Matrix number:
It is S in the state of m moment solar radiationsiThe state for being transferred to m+k moment solar radiations is SjTransition probability
For:
IV, error sequence k rank auto-correlation coefficient computing formula are as follows:
Wherein xiThe error of the i-th period is represented,For mean error, n is time span.By each rank auto-correlation system tried to achieve
Number calculates each rank weighing factor:
After V, the state transition probability matrix of the different step-lengths for obtaining, by prediction period day error percentage place state
A certain shape probability of state that next sampling instant error percentage is located is predicted, and a state of maximum probability is obtained to predict
Error state, so as to obtain corresponding prediction error value, completes the prediction at current time by following formula:
Ipre(t)=Istd(t)×(1-e(t)) (23)
Wherein, IpreT () is t earth's surface intensity of illumination predicted value, IstdT () is t standard fine day theoretical value, e (t)
To predict calculating error percentage.
It is as follows that the S4 sets up opto-electronic conversion model flow:
Using indirect prediction method, after obtaining the prediction numerical value of intensity of illumination, photovoltaic output is calculated by opto-electronic conversion model
Power, computing formula are as follows:
P=η SI [1-0.005 (t0+25)] (24)
Wherein, η is photoelectric transformation efficiency (%), and S is the effective gross area (m for photovoltaic generation2),
I (t) is the effective intensity of illumination (kW/m for receiving2), t0For the operating temperature (DEG C) of photovoltaic cell.
Embodiment 2
It is a kind of to be clustered based on Adaboost and markovian distributed photovoltaic ultra-short term prediction method, as shown in figure 1,
Comprise the following steps:
1. the certainty component of intensity of illumination sequence is extracted using moving average method, statistical analysis obtains different weather type
Intensity of illumination decay factor.
In example 2, be extract intensity of illumination sequence certainty component, reflect intensity of illumination mean change trend,
The time window of 30min is selected to be smoothed, by taking real data as an example, its treatment effect is as shown in Figure 2.Statistics 2015
~2016 years historical record datas and corresponding standard fine day model data, obtain under four kinds of typical weather types (fine day, it is cloudy,
Rainy day, cloudy day) pad value frequency distribution as shown in figure 3, distributed area be respectively [0,0.3), [0.3,0.6), [0.6,
0.8)、[0.8,1].Attenuation coefficient result of calculation under each typical weather type is as shown in table 1.
The attenuation coefficient of 1 typical weather type of table
2. cluster analysis of the improved KNN methods of Adaboost to historical data is adopted, disaggregated model is set up.
In example 2, it is (real with reference to Wuhan photovoltaic experiment porch weather monitoring device in March, -2016 in October, 2015
Valid data amount in border is 130 days) real-time sampling at intervals of 5min intensity of illumination data and temperature data, the illumination that will be extracted is strong
Degree series certainty component obtains cluster centre as shown in Figure 4 as the feature input quantity of Cluster Program;By KNN and
Adaboost programs complete classification, and iterations is 7 times, with reference to the weather history information inspection history that network weather record is obtained
The accuracy of data classification, it is as a result as shown in the table:
2 weather typing result of table
3. using multistage weighting Markov Chain method prediction earth's surface solar irradiance.
(1), historical data is pre-processed, error percentage sequence is obtained;
Research object is the difference between standard fine day model and actual value, and typical daylight is according to intensity such as Fig. 5 institutes
Show, show in figure that difference state mainly includes meadow, wave zone and cataclysm area, be calculated meadow difference in the research period
Within 10%, wave zone difference reaches 30%~40% to overall distribution, and cataclysm area difference is sometimes even super due to blocking for cloud
Cross 70%.
(2), state demarcation, calculates state transition probability matrix;
The error percentage obtained in pretreatment is divided into into 11 discrete states by practical operating experiences, wherein error is little
Point in -5% is defined as state 1, and it is 10% that the error of adjacent two state is differential, the like, point of the error more than 85% is fixed
Justice is state 11, and interval division is as shown in the table:
3 stochastic regime of table is interval
The transfer frequency matrix of each step-length (exponent number k is different) state change is counted, turning for each rank 11 × 11 can be calculated
Move probability matrix:
(3), predictably table intensity of illumination;
On the basis of step 1 and step 2, historical data is divided into fine day, cloudy, rainy day and cloudy four big class difference
It is predicted, the present embodiment is selected to predicting day 9:00 to 15:The intensity of illumination of 00 period enters the rolling forecast of between-line spacing 5min,
Here choose mean absolute percentage error to evaluate prediction effect:
Wherein, x is measured value;For predicted value;P is photovoltaic installed capacity.
Historical data is divided into fine day, cloudy, rainy day and cloudy day as similar day training sample set according to Fig. 2 methods,
Statistics determines that fine day intensity of illumination decay factor is 0.83, input fine day prediction submodel, predicts the outcome as shown in Figure 6;It is cloudy
Intensity of illumination decay factor is 0.69, is input into cloudy prediction submodel, is predicted the outcome as shown in Figure 7;Rainy day intensity of illumination decays
The factor is 0.35, input rainy day prediction submodel, is predicted the outcome as shown in Figure 8;Cloudy intensity of illumination decay factor is 0.12, defeated
Enter cloudy day prediction submodel, predict the outcome as shown in Figure 9.Its average absolute percent prediction error contrast is as shown in the table:
4 different model predictive error MAPE fiducial values of table
Using time series based on certain 3 months a few days ago data of photovoltaic experiment porch typical case in embodiment 2, to typical case
The earth's surface intensity of illumination of 6 hours days is predicted.Predict the outcome the carried clustering combination forecast model prediction effect of the display present invention
It is overall to be better than traditional Markov Chain forecast model.For many varieties of clouds weather and intensity of illumination that significantly high-frequency is shaken significantly
The rainy weather of decline, traditional Markov chain model predicated error decrease, it is difficult to regular tracking intensity of illumination width
Value quick change, and cloudy day change details can not be analyzed well, introduce attenuation coefficient m after forecast model to illumination
The mistake forecast fallen reaction more accurately, reduce peak value, valley of intensity, improves in the case of fine day situation and non-fine day
Precision of prediction.
4. opto-electronic conversion model is set up, the prediction of photovoltaic power ultra-short term is completed.
In embodiment 2, experiment porch adopts English profit YL175P-23b type polycrystal silicon cell plates, and conversion efficiency is 15%, is obtained
After the prediction numerical value of intensity of illumination, photovoltaic power output is calculated by opto-electronic conversion model, computing formula is as follows:
P=η SI [1-0.005 (t0+25)] (27)
Wherein, η is photoelectric transformation efficiency (%), and S is the effective gross area (m for photovoltaic generation2), I (t) is to receive
Effective intensity of illumination (kW/m2), t0For the operating temperature (DEG C) of photovoltaic cell.
It should be appreciated that the part that this specification is not elaborated belongs to prior art.
Although the specific embodiment of the present invention above in association with Description of Drawings, those of ordinary skill in the art should
Understand, these are merely illustrative of, and various deformation or modification can be made to these embodiments, without departing from the original of the present invention
Reason and essence.The scope of the present invention is only limited by the claims that follow.
Claims (5)
1. based on Adaboost clusters and markovian distributed photovoltaic ultra-short term prediction method, it is characterised in that include
Following steps:
Step 1, the certainty component that intensity of illumination sequence is extracted using moving average method, statistical analysis obtain different weather type
Intensity of illumination decay factor;
Step 2, cluster analysis is carried out using the improved KNN methods of Adaboost to historical data, set up classification forecast model;
Step 3, using multistage weighting Markov Chain method prediction earth's surface solar irradiance;
Step 4, opto-electronic conversion model is set up, complete the prediction of photovoltaic power ultra-short term.
2. as claimed in claim 1 based on Adaboost clusters and markovian distributed photovoltaic ultra-short term prediction side
Method, it is characterised in that the realization of step 1 includes:
The 1.1 random fluctuation components that intensity of illumination is filtered using moving average method, are extracted certainty component and select the component to make
For the characteristic variable of cluster analysis;
(1) y in formulatIntensity measurement value is shone for primary light, m selects moving average time window, YtPut down for the slip after smoothing processing
Average;
1.2 according to astronomical solar-terrestrial relationship calculate earth's surface solar irradiance Hottel standard fine day theoretical value, by historical data with it is right
Answer standard fine day theoretical value to carry out contrast statistics, obtain the intensity of illumination decay factor empirical value of different weather type and be input into pre-
Survey model;
(2) in formula, Imea(i) i-th moment earth's surface intensity of illumination, IstdI () Hottel standard fine day theoretical values, m is decay factor.
3. as claimed in claim 1 based on Adaboost clusters and markovian distributed photovoltaic ultra-short term prediction side
Method, it is characterised in that the realization of step 2 includes:
2.1 are clustered using k-means, by setting cluster centre number k, select initial cluster center, and in each iteration
Sample is classified as into closest center generic, again according to the Sample Refreshment cluster centre in classification after each iteration,
Until the change of two subcenters in front and back is less than arranges value or iterations is less than arranges value;
2.2 adopt KNN sorting algorithms, are used as not by maximum classification in the classification proportion of now k closest sample
Know the classification of object;Or according to the distance classification maximum to neighbouring class weights selection result as unknown object class
Not;
2.3 adopt Adaboost algorithm, obtain certain Weak Classifier G by training data studym(x), by Weak Classifier
Combination obtains strong classifier, i.e. strong classifier and can be expressed as:
(3) α in formulamThe weighted value for being the m time calculated Weak Classifier in strong classifier.
4. as claimed in claim 1 based on Adaboost clusters and markovian distributed photovoltaic ultra-short term prediction side
Method, it is characterised in that the realization of step 3 includes:
3.1 pairs of historical data pretreatments, obtain error percentage sequence, define relative prediction residual eitFor:
(4) in formulaFor intensity of illumination predicted value,For intensity of illumination actual value, i is that daily prediction is counted, and N is
Sample number;
3.2 state demarcations, research object are the error between Hottel standard fine day model values and actual value, will obtain in pretreatment
To error percentage be divided into 11 discrete states by practical operating experiences, wherein point of the error less than -5% is defined as state
1, it is 10% that the error of adjacent two state is differential, the like, point of the error more than 85% is defined as state 11;
3.3 ask for state transition probability matrix;
It is S in the state of m moment solar radiationsiThe state for being transferred to m+k moment solar radiations is SjTransition probabilityFor:
Using the result of above-mentioned state demarcation, the transfer frequency matrix of each step-length (exponent number k is different) state change is counted:
3.4 calculate each rank weighing factor;
Error sequence k rank auto-correlation coefficient computing formula are as follows:
(7) x in formulaiThe error of the i-th period is represented,For mean error, n is time span;
Each rank weighing factor is calculated by each rank auto-correlation coefficient that (7) formula is tried to achieve:
3.5 earth's surface intensities of illumination are predicted;
After the state transition probability matrix of the different step-lengths for obtaining, gone out by prediction period day error percentage place status predication
A certain shape probability of state that next sampling instant error percentage is located, and to obtain a state of maximum probability be predicated error shape
State, so as to obtain corresponding relative prediction residual value e (t);Calculate earth's surface intensity of illumination predicted value:
Ipre(t)=Istd(t)×(1-e(t)) (9);
(9) in formula, IpreT () is t earth's surface intensity of illumination predicted value, IstdT () is t standard fine day theoretical value, e (t)
To predict calculating error percentage.
5. as claimed in claim 1 based on Adaboost clusters and markovian distributed photovoltaic ultra-short term prediction side
Method, it is characterised in that the realization of step 4 includes:
Using indirect prediction method, after obtaining the prediction numerical value of intensity of illumination, photovoltaic power output is calculated by opto-electronic conversion model,
Computing formula is:
P=η SI [1-0.005 (t0+25)] (10);
(10), in formula, η is photoelectric transformation efficiency, and S is the effective gross area for photovoltaic generation, and I (t) is the effective light for receiving
According to intensity, t0For the operating temperature of photovoltaic cell.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201611024358.3A CN106529814B (en) | 2016-11-21 | 2016-11-21 | Distributed photovoltaic ultra-short term prediction method based on Adaboost clustering and Markov chain |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201611024358.3A CN106529814B (en) | 2016-11-21 | 2016-11-21 | Distributed photovoltaic ultra-short term prediction method based on Adaboost clustering and Markov chain |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106529814A true CN106529814A (en) | 2017-03-22 |
CN106529814B CN106529814B (en) | 2020-01-07 |
Family
ID=58352480
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201611024358.3A Active CN106529814B (en) | 2016-11-21 | 2016-11-21 | Distributed photovoltaic ultra-short term prediction method based on Adaboost clustering and Markov chain |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106529814B (en) |
Cited By (19)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107194495A (en) * | 2017-04-21 | 2017-09-22 | 北京信息科技大学 | A kind of longitudinal Forecasting Methodology of photovoltaic power excavated based on historical data |
CN108364236A (en) * | 2018-02-02 | 2018-08-03 | 西安交通大学 | A kind of data Method of Stochastic of sunlight irradiation intensity |
CN109217357A (en) * | 2018-07-20 | 2019-01-15 | 河海大学 | A kind of grid-connected photovoltaic system MPPT method based on Markov model |
CN109492047A (en) * | 2018-11-22 | 2019-03-19 | 河南财经政法大学 | A kind of dissemination method of the accurate histogram based on difference privacy |
CN109494792A (en) * | 2018-11-21 | 2019-03-19 | 国网青海省电力公司 | Method and device for determining light abandoning electric quantity of photovoltaic power station |
CN109829572A (en) * | 2019-01-14 | 2019-05-31 | 国网江苏省电力有限公司苏州供电分公司 | Photovoltaic power generation power prediction method under thunder and lightning weather |
CN110188964A (en) * | 2019-06-06 | 2019-08-30 | 河北工业大学 | A kind of photovoltaic power generation output forecasting method based on correlation |
CN110852492A (en) * | 2019-10-25 | 2020-02-28 | 东北电力大学 | Photovoltaic power ultra-short-term prediction method for finding similarity based on Mahalanobis distance |
CN111064617A (en) * | 2019-12-16 | 2020-04-24 | 重庆邮电大学 | Network flow prediction method and device based on empirical mode decomposition clustering |
JPWO2019130718A1 (en) * | 2017-12-28 | 2020-12-24 | 住友電気工業株式会社 | Judgment device, photovoltaic power generation system, judgment method and judgment program |
CN112153000A (en) * | 2020-08-21 | 2020-12-29 | 杭州安恒信息技术股份有限公司 | Method and device for detecting network flow abnormity, electronic device and storage medium |
CN112328851A (en) * | 2020-11-10 | 2021-02-05 | 国能日新科技股份有限公司 | Distributed power supply monitoring method and device and electronic equipment |
CN113255985A (en) * | 2021-05-18 | 2021-08-13 | 国网山东省电力公司青州市供电公司 | Method and system for predicting power generation capacity of photovoltaic power station |
CN114169627A (en) * | 2021-12-14 | 2022-03-11 | 湖南工商大学 | Deep reinforcement learning distributed photovoltaic power generation excitation method |
TWI783605B (en) * | 2021-08-02 | 2022-11-11 | 崑山科技大學 | Solar power generation prediction method |
CN115498750A (en) * | 2022-11-21 | 2022-12-20 | 深圳市恒生智能科技有限公司 | Intelligent charging management method and system for photovoltaic power supply |
CN115829165A (en) * | 2023-02-07 | 2023-03-21 | 佰聆数据股份有限公司 | Distributed photovoltaic operation condition analysis method and device based on power generation performance difference |
CN116050666A (en) * | 2023-03-20 | 2023-05-02 | 中国电建集团江西省电力建设有限公司 | Photovoltaic power generation power prediction method for irradiation characteristic clustering |
CN117742135A (en) * | 2024-02-09 | 2024-03-22 | 石家庄学院 | Photovoltaic energy-saving control method and system for communication machine room |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102521670A (en) * | 2011-11-18 | 2012-06-27 | 中国电力科学研究院 | Power generation output power prediction method based on meteorological elements for photovoltaic power station |
CN103390199A (en) * | 2013-07-18 | 2013-11-13 | 国家电网公司 | Photovoltaic power generation capacity/power prediction device |
CN103617461A (en) * | 2013-12-10 | 2014-03-05 | 中国矿业大学 | Photovoltaic power station generated power predicting method |
CN105160423A (en) * | 2015-09-14 | 2015-12-16 | 河海大学常州校区 | Photovoltaic power generation prediction method based on Markov residual error correction |
CN105426989A (en) * | 2015-11-03 | 2016-03-23 | 河海大学 | EEMD and combined kernel RVM-based photovoltaic power short-term prediction method |
-
2016
- 2016-11-21 CN CN201611024358.3A patent/CN106529814B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102521670A (en) * | 2011-11-18 | 2012-06-27 | 中国电力科学研究院 | Power generation output power prediction method based on meteorological elements for photovoltaic power station |
CN103390199A (en) * | 2013-07-18 | 2013-11-13 | 国家电网公司 | Photovoltaic power generation capacity/power prediction device |
CN103617461A (en) * | 2013-12-10 | 2014-03-05 | 中国矿业大学 | Photovoltaic power station generated power predicting method |
CN105160423A (en) * | 2015-09-14 | 2015-12-16 | 河海大学常州校区 | Photovoltaic power generation prediction method based on Markov residual error correction |
CN105426989A (en) * | 2015-11-03 | 2016-03-23 | 河海大学 | EEMD and combined kernel RVM-based photovoltaic power short-term prediction method |
Cited By (28)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107194495A (en) * | 2017-04-21 | 2017-09-22 | 北京信息科技大学 | A kind of longitudinal Forecasting Methodology of photovoltaic power excavated based on historical data |
CN107194495B (en) * | 2017-04-21 | 2020-05-12 | 北京信息科技大学 | Photovoltaic power longitudinal prediction method based on historical data mining |
JP7188399B2 (en) | 2017-12-28 | 2022-12-13 | 住友電気工業株式会社 | Determination device, photovoltaic power generation system, determination method and determination program |
JPWO2019130718A1 (en) * | 2017-12-28 | 2020-12-24 | 住友電気工業株式会社 | Judgment device, photovoltaic power generation system, judgment method and judgment program |
CN108364236A (en) * | 2018-02-02 | 2018-08-03 | 西安交通大学 | A kind of data Method of Stochastic of sunlight irradiation intensity |
CN109217357A (en) * | 2018-07-20 | 2019-01-15 | 河海大学 | A kind of grid-connected photovoltaic system MPPT method based on Markov model |
CN109494792A (en) * | 2018-11-21 | 2019-03-19 | 国网青海省电力公司 | Method and device for determining light abandoning electric quantity of photovoltaic power station |
CN109494792B (en) * | 2018-11-21 | 2022-05-13 | 国网青海省电力公司 | Method and device for determining light abandoning electric quantity of photovoltaic power station |
CN109492047A (en) * | 2018-11-22 | 2019-03-19 | 河南财经政法大学 | A kind of dissemination method of the accurate histogram based on difference privacy |
CN109829572A (en) * | 2019-01-14 | 2019-05-31 | 国网江苏省电力有限公司苏州供电分公司 | Photovoltaic power generation power prediction method under thunder and lightning weather |
CN109829572B (en) * | 2019-01-14 | 2021-03-12 | 国网江苏省电力有限公司苏州供电分公司 | Photovoltaic power generation power prediction method under thunder and lightning weather |
CN110188964A (en) * | 2019-06-06 | 2019-08-30 | 河北工业大学 | A kind of photovoltaic power generation output forecasting method based on correlation |
CN110852492A (en) * | 2019-10-25 | 2020-02-28 | 东北电力大学 | Photovoltaic power ultra-short-term prediction method for finding similarity based on Mahalanobis distance |
CN111064617A (en) * | 2019-12-16 | 2020-04-24 | 重庆邮电大学 | Network flow prediction method and device based on empirical mode decomposition clustering |
CN111064617B (en) * | 2019-12-16 | 2022-07-22 | 重庆邮电大学 | Network flow prediction method and device based on empirical mode decomposition clustering |
CN112153000A (en) * | 2020-08-21 | 2020-12-29 | 杭州安恒信息技术股份有限公司 | Method and device for detecting network flow abnormity, electronic device and storage medium |
CN112328851A (en) * | 2020-11-10 | 2021-02-05 | 国能日新科技股份有限公司 | Distributed power supply monitoring method and device and electronic equipment |
CN112328851B (en) * | 2020-11-10 | 2023-11-03 | 国能日新科技股份有限公司 | Distributed power supply monitoring method and device and electronic equipment |
CN113255985A (en) * | 2021-05-18 | 2021-08-13 | 国网山东省电力公司青州市供电公司 | Method and system for predicting power generation capacity of photovoltaic power station |
CN113255985B (en) * | 2021-05-18 | 2023-04-25 | 国网山东省电力公司青州市供电公司 | Method and system for predicting generating capacity of photovoltaic power station |
TWI783605B (en) * | 2021-08-02 | 2022-11-11 | 崑山科技大學 | Solar power generation prediction method |
CN114169627A (en) * | 2021-12-14 | 2022-03-11 | 湖南工商大学 | Deep reinforcement learning distributed photovoltaic power generation excitation method |
CN115498750A (en) * | 2022-11-21 | 2022-12-20 | 深圳市恒生智能科技有限公司 | Intelligent charging management method and system for photovoltaic power supply |
CN115498750B (en) * | 2022-11-21 | 2023-01-31 | 深圳市恒生智能科技有限公司 | Intelligent charging management method and system for photovoltaic power supply |
CN115829165A (en) * | 2023-02-07 | 2023-03-21 | 佰聆数据股份有限公司 | Distributed photovoltaic operation condition analysis method and device based on power generation performance difference |
CN115829165B (en) * | 2023-02-07 | 2023-05-05 | 佰聆数据股份有限公司 | Distributed photovoltaic operation condition analysis method and device based on power generation performance difference |
CN116050666A (en) * | 2023-03-20 | 2023-05-02 | 中国电建集团江西省电力建设有限公司 | Photovoltaic power generation power prediction method for irradiation characteristic clustering |
CN117742135A (en) * | 2024-02-09 | 2024-03-22 | 石家庄学院 | Photovoltaic energy-saving control method and system for communication machine room |
Also Published As
Publication number | Publication date |
---|---|
CN106529814B (en) | 2020-01-07 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106529814A (en) | Distributed photovoltaic ultra-short-term forecasting method based on Adaboost clustering and Markov chain | |
CN108549929B (en) | A kind of photovoltaic power prediction technique based on deep layer convolutional neural networks | |
CN104573879A (en) | Photovoltaic power station output predicting method based on optimal similar day set | |
CN108022001A (en) | Short term probability density Forecasting Methodology based on PCA and quantile estimate forest | |
CN104463349A (en) | Photovoltaic generated power prediction method based on multi-period comprehensive similar days | |
CN106909933A (en) | A kind of stealing classification Forecasting Methodology of three stages various visual angles Fusion Features | |
CN109636054A (en) | Solar energy power generating amount prediction technique based on classification and error combination prediction | |
US20210326696A1 (en) | Method and apparatus for forecasting power demand | |
CN109492748B (en) | Method for establishing medium-and-long-term load prediction model of power system based on convolutional neural network | |
CN112100911B (en) | Solar radiation prediction method based on depth BILSTM | |
CN106778846A (en) | A kind of method for forecasting based on SVMs | |
CN112418346B (en) | Numerical weather forecast total radiation system error classification calculation method | |
CN107679687A (en) | A kind of photovoltaic output modeling method and Generation System Reliability appraisal procedure | |
CN112465251A (en) | Short-term photovoltaic output probability prediction method based on simplest gated neural network | |
CN111242355A (en) | Photovoltaic probability prediction method and system based on Bayesian neural network | |
CN105701572A (en) | Photovoltaic short-term output prediction method based on improved Gaussian process regression | |
CN113052469B (en) | Method for calculating wind-solar-water-load complementary characteristic of small hydropower area lacking measurement runoff | |
CN113344288B (en) | Cascade hydropower station group water level prediction method and device and computer readable storage medium | |
CN113554466A (en) | Short-term power consumption prediction model construction method, prediction method and device | |
CN114792156A (en) | Photovoltaic output power prediction method and system based on curve characteristic index clustering | |
CN111460001A (en) | Theoretical line loss rate evaluation method and system for power distribution network | |
CN111985719A (en) | Power load prediction method based on improved long-term and short-term memory network | |
CN116345555A (en) | CNN-ISCA-LSTM model-based short-term photovoltaic power generation power prediction method | |
CN105956708A (en) | Grey correlation time sequence based short-term wind speed forecasting method | |
CN110852492A (en) | Photovoltaic power ultra-short-term prediction method for finding similarity based on Mahalanobis distance |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |