CN106529814B - Distributed photovoltaic ultra-short term prediction method based on Adaboost clustering and Markov chain - Google Patents

Distributed photovoltaic ultra-short term prediction method based on Adaboost clustering and Markov chain Download PDF

Info

Publication number
CN106529814B
CN106529814B CN201611024358.3A CN201611024358A CN106529814B CN 106529814 B CN106529814 B CN 106529814B CN 201611024358 A CN201611024358 A CN 201611024358A CN 106529814 B CN106529814 B CN 106529814B
Authority
CN
China
Prior art keywords
solar irradiance
prediction
error
value
state
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201611024358.3A
Other languages
Chinese (zh)
Other versions
CN106529814A (en
Inventor
邓长虹
谭津
李丰君
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Wuhan University WHU
Original Assignee
Wuhan University WHU
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Wuhan University WHU filed Critical Wuhan University WHU
Priority to CN201611024358.3A priority Critical patent/CN106529814B/en
Publication of CN106529814A publication Critical patent/CN106529814A/en
Application granted granted Critical
Publication of CN106529814B publication Critical patent/CN106529814B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

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, in particular to a distributed photovoltaic ultra-short term prediction method based on Adaboost clustering and Markov chains, which comprises the following steps: 1. extracting the deterministic component of the illumination intensity sequence by adopting a moving average method, and carrying out statistical analysis to obtain illumination intensity attenuation factors of different weather types; 2. clustering and analyzing historical data by adopting an Adaboost improved KNN method, and establishing a classification model; 3. predicting the surface solar irradiance by adopting a multi-order weighted Markov chain method; 4. and establishing a photoelectric conversion model to finish the ultra-short-term prediction of the photovoltaic power. The invention provides a combined prediction method for feature extraction and data mining of input data, which is characterized in that after historical photovoltaic output data are classified according to typical weather types, states in a prediction process are refined by introducing a weather type attenuation factor, so that a better prediction effect can be obtained in sunny weather, and the prediction precision and accuracy in non-sunny weather are improved.

Description

Distributed photovoltaic ultra-short term prediction method based on Adaboost clustering and Markov chain
Technical Field
The invention belongs to the technical field of distributed photovoltaic power generation systems, and particularly relates to a distributed photovoltaic ultra-short term prediction method based on Adaboost clustering and a Markov chain.
Background
In recent years, the installed capacity of photovoltaic power generation is increasing, and the impact of fluctuation and intermittence on the operation of a power system is more and more prominent. The ultrashort-term prediction of the distributed photovoltaic system has important significance on safe and economic operation of a power system, and mainly comprises the following two aspects: firstly, a control strategy is formulated based on predicted power, the influence of power fluctuation on an electric power system during photovoltaic grid connection is reduced, and the safety, reliability and controllability of the system are improved; and secondly, the method is beneficial to determining the output plan of the unit by a power system dispatching department, reduces the rotation reserve in the system and reduces the power consumption cost. Under the precondition of meeting safety and stability, the maximum consumption of renewable energy sources gradually becomes the key point in the research field of photovoltaic power generation, so that power prediction has practical value for the large-scale development of distributed and grid-connected photovoltaic power generation systems.
The photovoltaic output power shows randomness and volatility under the influence of surface solar irradiance, and the photovoltaic output power change rules have obvious difference under different weather types, so that the photovoltaic output power under various meteorological conditions is difficult to predict by using a uniform prediction model in practical application. The cluster prediction is a combined prediction method for performing feature extraction and data mining on input data, and compared with a single prediction method, the cluster prediction can improve the prediction accuracy. Because high-cost numerical weather forecast is difficult to be widely applied to a distributed photovoltaic prediction system, most of the current researches adopt a combined prediction method for classifying different types of data based on weather forecast information and cloud cover information, historical solar irradiance intensity information is not fully mined, and the prediction accuracy is remarkably reduced under the condition of inaccurate weather forecast.
Disclosure of Invention
The invention aims to provide a combined prediction method based on Adaboost clustering and a Markov chain, which not only can obtain a better prediction effect in sunny weather, but also can improve the prediction accuracy in non-sunny weather.
In order to achieve the purpose, the invention adopts the technical scheme that: a distributed photovoltaic ultra-short term prediction method based on Adaboost clustering and Markov chains comprises the following steps:
step 1, extracting the deterministic component of a solar irradiance sequence by adopting a sliding average method, and obtaining solar irradiance attenuation factors of different weather types through statistical analysis;
step 2, clustering analysis is carried out on historical data by adopting an Adaboost improved KNN method, and a classification prediction model is established;
step 3, predicting the surface solar irradiance by adopting a multi-order weighted Markov chain method;
and 4, establishing a photoelectric conversion model to finish the ultra-short-term prediction of the photovoltaic power.
In the distributed photovoltaic ultra-short term prediction method based on Adaboost clustering and markov chains, the implementation of step 1 includes:
1.1 filtering out random fluctuation components of solar irradiance by adopting a moving average method, extracting deterministic components and selecting the deterministic components as characteristic variables of cluster analysis;
(1) wherein t is the sampling time, h is the number of sampling intervals selected in the time window to the left and the right by taking the time t as the reference,
Figure GDA0002271027060000022
is t + i1Actual measurement value of original solar irradiance at sampling point at moment, YtThe sliding average value at the sampling point at the time t after the smoothing processing is carried out;
1.2, calculating a theoretical value of earth surface solar irradiance Hottel standard sunny day according to an astronomical day-ground relation, comparing historical data with the theoretical value of the corresponding standard sunny day, performing statistics to obtain empirical values of solar irradiance attenuation factors of different weather types, and inputting the empirical values into a prediction model;
(2) in the formula Imea(i) Measured value of surface solar irradiance at the ith moment, Istd(i) The theoretical value of the Hottel standard sunny day at the ith moment, alpha is an attenuation factor, and alphaiAnd the irradiance attenuation value at the ith moment is N, and the total number of sampling points in one day is N.
In the distributed photovoltaic ultra-short term prediction method based on Adaboost clustering and markov chains, the implementation of step 2 includes:
2.1 clustering by adopting k-means, selecting an initial clustering center by setting the number s of the clustering centers, classifying the sample into the class to which the center closest to the initial clustering center belongs in each iteration, and updating the clustering centers according to the samples in the class after each iteration until the change of the centers in the previous and subsequent times is not more than a set value or the iteration times is less than the set value;
2.2 selecting the category with the largest proportion in the l adjacent samples as the category of the unknown object by adopting a KNN classification algorithm; or selecting the category with the maximum result as the category of the unknown object according to the distance and the distance by weighting the adjacent categories;
2.3 obtaining a certain weak classifier G through training data learning by adopting Adaboost algorithmm(x) By pairing groups of weak classifiersThe strong classifiers are obtained, and are expressed as:
Figure GDA0002271027060000032
(3) in the formula of alphamCalculating the weighted value of the weak classifier in the strong classifier for the mth time; m is the number of Adaboost iterations;
in the distributed photovoltaic ultra-short term prediction method based on Adaboost clustering and markov chains, the implementation of step 3 includes:
3.1 preprocessing the historical data to obtain an error percentage sequence, and defining a relative prediction error e (i) as:
Figure GDA0002271027060000041
(4) in the formula Ipre(i) Is a predicted value of the surface solar irradiance at the ith moment, Imea(i) The measured value of the surface solar irradiance at the ith moment is N, and the total sampling point number in one day is N;
3.2, dividing the state, wherein the research object is the error between the theoretical value and the actual value of the Hottel standard sunny day, dividing the error percentage obtained in the preprocessing into 11 discrete states according to the actual operation experience, wherein the point with the error smaller than-5% is defined as the state 1, the error level difference between the two adjacent states is 10%, and the like, and the point with the error larger than 85% is defined as the state 11;
3.3 solving a state transition probability matrix;
the state of solar irradiance at time t is
Figure GDA0002271027060000042
The state of the solar irradiance transferred to the moment t + k is SjTransition probability of
Figure GDA0002271027060000043
Comprises the following steps:
Figure GDA0002271027060000044
x (t) is a random variable in the markov chain;
Figure GDA0002271027060000045
the state of solar irradiance at time t; k is the step length of state transition, and P { } is the probability of occurrence of an event in brackets;
obtaining transition probability among the states at the current moment by using the state division resultThe k-order state transition probability matrix is formed by the following steps:
(6) wherein Z is the total number of states divided by the system in the Markov chain;
3.4 calculating the influence weight of each order;
the k-order autocorrelation coefficient calculation formula of the error sequence is as follows:
Figure GDA0002271027060000051
(7) in the formula
Figure GDA0002271027060000052
Denotes the ith3The error in the time of day is,
Figure GDA0002271027060000053
is the average error, n is the time length;
calculating the influence weight of each order according to the autocorrelation coefficient of each order obtained by the formula (7):
Figure GDA0002271027060000054
(8) wherein K is the maximum step size in the Markov chain;
3.5 predicting the surface solar irradiance;
after the state transition probability matrixes of different step lengths are obtained, the probability of a certain state of the error percentage at the next sampling time is predicted according to the state of the error percentage at a certain time of the prediction day, and the state with the maximum probability is obtained as a prediction error state, so that a corresponding relative prediction error value e (i) is obtained; calculating a predicted value of the surface solar irradiance:
Ipre(i)＝Istd(i)×(1-e(i)) (9)；
(9) in the formula Ipre(i) Is a predicted value of the surface solar irradiance at the ith moment, Istd(i) The theoretical value of the Hottel standard sunny day at the ith moment, and e (i) the relative prediction error.
In the distributed photovoltaic ultra-short term prediction method based on Adaboost clustering and markov chains, the implementation of step 4 includes:
after the prediction value of the solar irradiance is obtained by adopting an indirect prediction method, the photovoltaic output power is calculated through a photoelectric conversion model, and the calculation formula is as follows:
P＝ηSI[1-0.005(t0+25)] (10)；
(10) where η is the photoelectric conversion efficiency, S is the effective total area for photovoltaic power generation, I is the effective solar irradiance received, t0Is the operating temperature of the photovoltaic cell.
The invention has the beneficial effects that: the invention provides a combined prediction method based on Adaboost clustering and a Markov chain, which not only can obtain a better prediction effect in sunny weather, but also improves the prediction precision in non-sunny weather. The improved K Nearest Neighbor (KNN) algorithm of Adaboost does not need to know an error boundary in advance, has self-adaptability, and can enhance the classification effect of a weak classifier on an unbalanced data set. Markov Chain (Markov Chain) is a method for predicting possible development trend and determining unknown state after a system by dividing states and using probability transition process between the states, and is suitable for tracking and predicting random uncertain events. Different weather type solar irradiance attenuation factors are introduced, change details in the prediction error can be amplified, and the error change of the amplitude value relative to the clear sky model value is reflected, so that the prediction precision of the surface solar irradiance in the rainy weather type is obviously improved. The multi-factor multi-step modeling is integrated, the application range is wide, and the power prediction problem is simplified by establishing a photoelectric conversion model by adopting engineering experience.
Drawings
FIG. 1 is a flow chart of a method of ultra-short term prediction according to an embodiment of the present invention;
FIG. 2 is a graph of the extraction of a running average deterministic variable according to one embodiment of the present invention;
FIG. 3 is a graph of interval distribution of solar irradiance attenuation values for one embodiment of the present invention;
FIG. 4 is a four class center plot of surface solar irradiance for one embodiment of the present invention;
FIG. 5 is a comparison of measured solar irradiance on a typical day versus solar irradiance on a standard sunny day in accordance with an embodiment of the present invention;
figure 6 is a predictive comparison curve of a model and a conventional markov chain under a sunny day in accordance with one embodiment of the present invention;
FIG. 7 is a predicted contrast curve for a model under cloudiness and a traditional Markov chain in accordance with an embodiment of the present invention;
FIG. 8 is a predicted contrast curve for a model and a conventional Markov chain in rainy weather for one embodiment of the present invention;
FIG. 9 is a predicted contrast curve for a model and a conventional Markov chain under cloudy days in accordance with an embodiment of the present invention;
Detailed Description
Embodiments of the present invention will be described in detail below with reference to the accompanying drawings.
Examples of which are illustrated in the accompanying drawings, wherein like reference numerals refer to the same or similar elements or elements having the same or similar function throughout. The embodiments described below with reference to the drawings are illustrative only and should not be construed as limiting the invention.
The embodiment adopts the following technical scheme: a distributed photovoltaic ultra-short term prediction method based on Adaboost clustering and Markov chains comprises the following steps:
step 1, extracting the deterministic component of a solar irradiance sequence by adopting a sliding average method, and obtaining solar irradiance attenuation factors of different weather types through statistical analysis;
step 2, clustering analysis is carried out on historical data by adopting an Adaboost improved KNN method, and a classification prediction model is established;
step 3, predicting the surface solar irradiance by adopting a multi-order weighted Markov chain method;
and 4, establishing a photoelectric conversion model to finish the ultra-short-term prediction of the photovoltaic power.
Further, the implementation of step 1 includes:
1.1 filtering out random fluctuation components of solar irradiance by adopting a moving average method, extracting deterministic components and selecting the deterministic components as characteristic variables of cluster analysis;
Figure GDA0002271027060000071
(1) wherein t is the sampling time, h is the number of sampling intervals selected in the time window to the left and the right by taking the time t as the reference,
Figure GDA0002271027060000072
is t + i1Actual measurement value of original solar irradiance at sampling point at moment, YtThe sliding average value at the sampling point at the time t after the smoothing processing is carried out;
1.2, calculating a theoretical value of earth surface solar irradiance Hottel standard sunny day according to an astronomical day-ground relation, comparing historical data with the theoretical value of the corresponding standard sunny day, performing statistics to obtain empirical values of solar irradiance attenuation factors of different weather types, and inputting the empirical values into a prediction model;
Figure GDA0002271027060000081
(2) in the formula Imea(i) Measured value of surface solar irradiance at the ith moment, Istd(i) The theoretical value of the Hottel standard sunny day at the ith moment, alpha is an attenuation factor, and alphaiAnd the irradiance attenuation value at the ith moment is N, and the total number of sampling points in one day is N.
Further, the implementation of step 2 includes:
2.1 clustering by adopting k-means, selecting an initial clustering center by setting the number s of the clustering centers, classifying the sample into the class to which the center closest to the initial clustering center belongs in each iteration, and updating the clustering centers according to the samples in the class after each iteration until the change of the centers in the previous and subsequent times is not more than a set value or the iteration times is less than the set value;
2.2 selecting the category with the largest proportion in the l adjacent samples as the category of the unknown object by adopting a KNN classification algorithm; or selecting the category with the maximum result as the category of the unknown object according to the distance and the distance by weighting the adjacent categories;
2.3 obtaining a certain weak classifier G through training data learning by adopting Adaboost algorithmm(x) The strong classifier is obtained by combining the weak classifiers, and is expressed as:
Figure GDA0002271027060000082
(3) in the formula of alphamCalculating the weighted value of the weak classifier in the strong classifier for the mth time, wherein M is Adaboost iteration times;
further, the implementation of step 3 includes:
3.1 preprocessing the historical data to obtain an error percentage sequence, and defining a relative prediction error e (i) as:
Figure GDA0002271027060000083
(4) in the formula Ipre(i) The predicted value of the surface solar irradiance at the ith moment is obtained; i ismea(i) The measured value of the surface solar irradiance at the ith moment is obtained; (ii) a N is the total sampling point number in one day;
3.2, dividing the state, wherein the research object is the error between the theoretical value and the actual value of the Hottel standard sunny day, dividing the error percentage obtained in the preprocessing into 11 discrete states according to the actual operation experience, wherein the point with the error smaller than-5% is defined as the state 1, the error level difference between the two adjacent states is 10%, and the like, and the point with the error larger than 85% is defined as the state 11;
3.3 solving a state transition probability matrix;
the state of solar irradiance at time t is
Figure GDA0002271027060000091
The state of the solar irradiance transferred to the moment t + k is SjTransition probability of
Figure GDA0002271027060000092
Comprises the following steps:
Figure GDA0002271027060000093
x (t) is a random variable in the markov chain;the state of solar irradiance at time t; k is the step length of state transition, and P { } is the probability of occurrence of an event in brackets;
obtaining transition probability among the states at the current moment by using the state division resultThe k-order state transition probability matrix is formed by the following steps:
Figure GDA0002271027060000096
(6) wherein Z is the total number of states divided by the system in the Markov chain;
3.4 calculating the influence weight of each order;
the k-order autocorrelation coefficient calculation formula of the error sequence is as follows:
Figure GDA0002271027060000097
(7) in the formula
Figure GDA0002271027060000098
Denotes the ith3The error in the time of day is,
Figure GDA0002271027060000099
is the average error, n is the time length;
calculating the influence weight of each order according to the autocorrelation coefficient of each order obtained by the formula (7):
Figure GDA0002271027060000101
(8) wherein K is the maximum step size in the Markov chain;
3.5 predicting the surface solar irradiance;
after the state transition probability matrixes of different step lengths are obtained, the probability of a certain state of the error percentage at the next sampling time is predicted according to the state of the error percentage at a certain time of the prediction day, and the state with the maximum probability is obtained as a prediction error state, so that a corresponding relative prediction error value e (i) is obtained; calculating a predicted value of the surface solar irradiance:
Ipre(i)＝Istd(i)×(1-e(i)) (9)；
(9) in the formula Ipre(i) Is a predicted value of the surface solar irradiance at the ith moment, Istd(i) The theoretical value of the Hottel standard sunny day at the ith moment, and e (i) the relative prediction error.
Further, the implementation of step 4 includes:
after the prediction value of the solar irradiance is obtained by adopting an indirect prediction method, the photovoltaic output power is calculated through a photoelectric conversion model, and the calculation formula is as follows:
P＝ηSI[1-0.005(t0+25)] (10)；
(10) where η is the photoelectric conversion efficiency, S is the effective total area for photovoltaic power generation, I is the effective solar irradiance received, t0Is the operating temperature of the photovoltaic cell.
Example 1
In order to improve the prediction accuracy under the condition of non-sunny days, a distributed photovoltaic ultra-short-term prediction method based on Adaboost clustering and a Markov chain is provided, the deterministic component of a solar irradiance sequence is extracted by adopting a moving average method, and solar irradiance attenuation factors of different weather types are obtained through statistical analysis; clustering analysis is carried out on historical data by adopting an Adaboost improved KNN method, and a classification model is established; and the prediction of the surface solar irradiance at the future moment is realized by combining the multi-order weighted Markov chain based on the error sequence, and finally, the ultrashort-term prediction of the distributed photovoltaic power generation is realized through a photoelectric conversion model.
The embodiment 1 is solved by the following technical scheme:
a distributed photovoltaic ultra-short term prediction method based on Adaboost clustering and Markov chains comprises the following steps:
s1, extracting the deterministic component of the solar irradiance sequence by adopting a moving average method, and obtaining solar irradiance attenuation factors of different weather types through statistical analysis;
s2, adopting an Adaboost improved KNN method to perform cluster analysis on historical data, and establishing a classification model;
s3, predicting the surface solar irradiance by adopting a multi-order weighted Markov chain method;
and S4, establishing a photoelectric conversion model to finish the ultra-short-term prediction of the photovoltaic power.
In the step S1, the deterministic component of the solar irradiance sequence is extracted by using a moving average method, and the flow of obtaining the solar irradiance attenuation factors of different weather types through statistical analysis is as follows:
and (3) filtering random fluctuation components of solar irradiance by selecting a moving average method, extracting a deterministic component and selecting the deterministic component as a characteristic variable of cluster analysis. t is the sampling time, h is the number of sampling intervals selected in the time window towards the left and the right by taking the time t as the reference,
Figure GDA0002271027060000111
is t + i1The actual measured value of the original solar irradiance at the sampling point at the moment, the sliding average value at the sampling point at the moment t after smoothing treatment is Yt。
Figure GDA0002271027060000112
The surface solar irradiance is reduced to different degrees at the cloud layer shielding moment, and the surface solar irradiance measured value I at the ith momentmea(i) Theoretical value I of Hottel standard in sunny daysstd(i) Different multiplying power relations exist between the two parts, and are marked as attenuation factors alpha and alphaiIrradiance attenuation value at the ith moment, wherein N is the total number of sampling points in one day; the expression is as follows:
Figure GDA0002271027060000121
the degree of attenuation of the terrestrial solar irradiance to the extraterrestrial solar irradiance is different for different weather types.
The prediction process of the step S2 for cluster analysis of the historical data by using the Adaboost improved KNN method is as follows:
1) inputting training data in batches, extracting mathematical characteristics of the data, and determining a basic classifier G obtained by mth iteration in the weak classifier Adaboost learning process according to KNNm(x) The number of (2);
2) initializing a weight d for each sample datami。
Figure GDA0002271027060000122
3) Selecting sample points according to the weights to obtain a new training subset, and training the weak classifier G by using training data with weight distributionm(x) By Gm(x) Classifying all samples, and calculating the weak classifier Gm(x) Error rate under training set emAnd its coefficient alpha in the final classifierm。
Figure GDA0002271027060000124
4) Improving the weight of the correctly classified samples, reducing the weight of the incorrectly classified samples, and updating the weight distribution of the data samples for the next iteration:
Figure GDA0002271027060000126
and (4) returning to the step (3) for the next iteration after the next round of weight is obtained until the number of iterations reaches a set value or the error rate reaches a set sufficiently small value.
The process of predicting the surface solar irradiance by the multi-order weighted Markov chain method in the step S3 is as follows:
I. preprocessing historical data to obtain an error percentage sequence, and defining a relative prediction error e (i) as:
Figure GDA0002271027060000131
wherein Ipre(i) The predicted value of the surface solar irradiance at the ith moment is obtained; and Imea(i) The measured value of the surface solar irradiance at the ith moment is obtained; n is the total number of sampling points in one day.
II. And dividing the error percentage obtained in the preprocessing into 11 discrete states according to the actual operation experience, wherein a point with the error smaller than-5% is defined as a state 1, the error level difference between two adjacent states is 10%, and by analogy, a point with the error larger than 85% is defined as a state 11.
And III, counting a state transition probability matrix of each step length k order by using the state division result:
Figure GDA0002271027060000132
the state of solar irradiance at time t is
Figure GDA0002271027060000133
The state of the solar irradiance transferred to the moment t + k is SjHas a transition probability of
Figure GDA0002271027060000134
Figure GDA0002271027060000135
X (t) is a random variable in the markov chain;
Figure GDA0002271027060000136
the state of solar irradiance at time t; k is the step length of state transition, and P { } is the probability of occurrence of an event in brackets;
IV, an error sequence k-order autocorrelation coefficient calculation formula is as follows:
Figure GDA0002271027060000137
wherein
Figure GDA0002271027060000138
Denotes the ith3The error of the time period is such that,
Figure GDA0002271027060000139
n is the length of time for the average error. Calculating influence weight of each order according to the obtained autocorrelation coefficient of each order:
Figure GDA0002271027060000141
wherein K is the maximum step size in the Markov chain;
v, after state transition probability matrixes of different step lengths are obtained, the probability of a state where the error percentage of the next sampling time is predicted according to the state where the error percentage of the certain time of the prediction day is located, the state with the maximum probability is obtained as a prediction error state, and therefore a corresponding prediction error value e (i) is obtained, and the prediction value of the surface solar irradiance is calculated according to the following formula:
Ipre(i)＝Istd(i)×(1-e(i)) (23)
wherein, Ipre(i) Is a predicted value of the surface solar irradiance at the ith moment, Istd(i) The theoretical value of the Hottel standard sunny day at the ith moment, and e (i) the relative prediction error.
The process of establishing the photoelectric conversion model in the step S4 is as follows:
after the prediction value of the solar irradiance is obtained by adopting an indirect prediction method, the photovoltaic output power is calculated through a photoelectric conversion model, and the calculation formula is as follows:
P＝ηSI[1-0.005(t0+25)] (24)
where η is photoelectric conversion efficiency (%), and S is the total effective area (m) for photovoltaic power generation2)，
I is the received effective solar irradiance (kW/m)2)，t0Is the operating temperature (deg.C) of the photovoltaic cell.
Example 2
A distributed photovoltaic ultra-short term prediction method based on Adaboost clustering and Markov chains is shown in FIG. 1, and comprises the following steps:
firstly, a deterministic component of a solar irradiance sequence is extracted by adopting a moving average method, and solar irradiance attenuation factors of different weather types are obtained through statistical analysis.
In embodiment 2, in order to extract the deterministic component of the solar irradiance sequence and reflect the average variation trend of the solar irradiance, a time window of 30min is selected for smoothing, and taking the actual data as an example, the processing effect is shown in fig. 2. The recorded data of the almanac history from 2015 to 2016 and the corresponding standard clear-sky model data are counted to obtain frequency distributions of attenuation values in four typical weather types (clear, cloudy, rainy and cloudy), which are shown in fig. 3, and the distribution intervals are [0,0.3 ], [0.3,0.6 ], [0.6,0.8 ] and [0.8,1], respectively. The attenuation coefficient calculation results for each typical weather type are shown in table 1.
TABLE 1 attenuation coefficient for typical weather types
Figure GDA0002271027060000151
And secondly, clustering analysis is carried out on the historical data by adopting an Adaboost improved KNN method, and a classification model is established.
In embodiment 2, solar irradiance data and temperature data with an interval of 5min are sampled in real time by combining with meteorological monitoring equipment 2015 of a certain photovoltaic experimental platform in wuhan from 10 months to 2016 (actual effective data amount is 130 days), and the extracted solar irradiance sequence certainty component is used as a characteristic input quantity of a clustering program to obtain a clustering center as shown in fig. 4; classification is completed by KNN and Adaboost programs, the iteration times are 7 times, the accuracy of classification of historical data is checked by combining historical weather information obtained by network weather recording, and the result is shown in the following table:
TABLE 2 weather Classification results
Figure GDA0002271027060000152
And thirdly, predicting the surface solar irradiance by adopting a multi-order weighted Markov chain method.
Preprocessing historical data to obtain an error percentage sequence;
the research object is the difference between the standard sunny model and the actual value, a typical solar irradiance curve graph is shown in fig. 5, the displayed difference state mainly comprises a stable region, a fluctuation region and a sudden change region, the stable region difference value is calculated and obtained in the research period and is generally distributed within 10%, the fluctuation region difference value reaches 30% -40%, and the sudden change region difference value sometimes even exceeds 70% due to cloud shielding.
Secondly, state division is carried out, and a state transition probability matrix is calculated;
dividing the error percentage obtained in the preprocessing into 11 discrete states according to the actual operation experience, wherein the point with the error less than-5% is defined as the state 1, the error level difference between the two adjacent states is 10%, and by analogy, the point with the error more than 85% is defined as the state 11, and the interval division is shown in the following table:
TABLE 3 random State Interval
Figure GDA0002271027060000161
Obtaining the transition probability among the states at the current moment
Figure GDA0002271027060000162
The aggregated k-step state transition probability matrix can be calculated as follows, and the 11 × 11 transition probability matrix of each order can be calculated:
Figure GDA0002271027060000171
(III) predicting the surface solar irradiance;
on the basis of the step 1 and the step 2, historical data are divided into four categories of sunny days, cloudy days, rainy days and cloudy days for prediction, rolling prediction with an interval of 5min is selected for solar irradiance in a time period of 9:00 to 15:00 of a prediction day, and an average absolute percentage error is selected to evaluate the prediction effect:
Figure GDA0002271027060000172
wherein x is an actual measurement value;
Figure GDA0002271027060000173
is a predicted value; and P is photovoltaic installed capacity.
Dividing historical data into sunny days, cloudy days, rainy days and cloudy days according to the method of FIG. 2 as training sample sets of similar days, statistically determining that the solar irradiance attenuation factor is 0.83 in sunny days, inputting a sunny day prediction submodel, and obtaining a prediction result shown in FIG. 6; the cloudy solar irradiance attenuation factor is 0.69, the cloudy predictor model is input, and the prediction result is shown in figure 7; the solar irradiance attenuation factor in rainy days is 0.35, the rain solar irradiance attenuation factor is input into a rain solar predictor model, and the prediction result is shown in figure 8; the cloudy solar irradiance attenuation factor is 0.12, and the input is input into the cloudy predictor model, and the prediction result is shown in fig. 9. The average absolute percentage prediction error pair is shown in the following table:
TABLE 4 comparison of prediction errors MAPE for different models
Figure GDA0002271027060000174
In example 2, the data of a typical day 3 months before a photovoltaic experimental platform is used as a basic time sequence, and the surface solar irradiance of 6 hours of the typical day is predicted. The prediction result shows that the prediction effect of the clustering combination prediction model provided by the invention is integrally superior to that of the traditional Markov chain prediction model. For cloudy and rainy days with large amplitude and high frequency oscillation and with large reduction of solar irradiance, the prediction error of the traditional Markov chain model is reduced, the rapid change of the solar irradiance amplitude is difficult to track regularly, the details of the change in cloudy days cannot be well analyzed, the drop reaction of the prediction model to the solar irradiance is more accurate after the attenuation coefficient alpha is introduced, the error prediction of peak value and valley value is reduced, and the prediction precision under the conditions of sunny days and non-sunny days is improved.
And fourthly, establishing a photoelectric conversion model to finish the ultra-short-term prediction of the photovoltaic power.
In embodiment 2, the experimental platform adopts an english YL175P-23b type polysilicon battery plate, the conversion efficiency is 15%, and after a predicted value of solar irradiance is obtained, the photovoltaic output power is calculated through a photoelectric conversion model, and the calculation formula is as follows:
P＝ηSI[1-0.005(t0+25)] (27)
where η is photoelectric conversion efficiency (%), and S is the total effective area (m) for photovoltaic power generation2) I is the received effective solar irradiance (kW/m)2)，t0Is the operating temperature (deg.C) of the photovoltaic cell.
It should be understood that parts of the specification not set forth in detail are well within the prior art.
Although specific embodiments of the present invention have been described above with reference to the accompanying drawings, it will be appreciated by those skilled in the art that these are merely illustrative and that various changes or modifications may be made to these embodiments without departing from the principles and spirit of the invention. The scope of the invention is only limited by the appended claims.

Claims (1)

1. A distributed photovoltaic ultra-short term prediction method based on Adaboost clustering and Markov chains is characterized by comprising the following steps:
step 1, extracting the deterministic component of a solar irradiance sequence by adopting a sliding average method, and obtaining solar irradiance attenuation factors of different weather types through statistical analysis;
step 2, clustering analysis is carried out on historical data by adopting an Adaboost improved KNN method, and a classification prediction model is established;
step 3, predicting the surface solar irradiance by adopting a multi-order weighted Markov chain method;
step 4, establishing a photoelectric conversion model to finish the ultra-short-term prediction of the photovoltaic power;
the implementation of step 1 comprises:
1.1 filtering out random fluctuation components of solar irradiance by adopting a moving average method, extracting deterministic components and selecting the deterministic components as characteristic variables of cluster analysis;
Figure FDA0002271027050000011
(1) wherein t is the sampling time, h is the number of sampling intervals selected in the time window towards the left and the right by taking the time t as the reference,
Figure FDA0002271027050000013
is t + i1Actual measurement value of original solar irradiance at sampling point at moment, YtThe sliding average value at the sampling point at the time t after the smoothing processing is carried out;
1.2, calculating a theoretical value of earth surface solar irradiance Hottel standard sunny day according to an astronomical day-ground relation, comparing historical data with the theoretical value of the corresponding standard sunny day, performing statistics to obtain empirical values of solar irradiance attenuation factors of different weather types, and inputting the empirical values into a prediction model;
(2) in the formula Imea(i) Measured value of surface solar irradiance at the ith moment, Istd(i) The theoretical value of the Hottel standard sunny day at the ith moment, alpha is an attenuation factor, and alphaiIrradiance attenuation value at the ith moment, wherein N is the total number of sampling points in one day;
the implementation of step 2 comprises:
2.1 clustering by adopting k-means, selecting an initial clustering center by setting the number s of the clustering centers, classifying the sample into the class to which the center closest to the initial clustering center belongs in each iteration, and updating the clustering centers according to the samples in the class after each iteration until the change of the centers in the previous and subsequent times is not more than a set value or the iteration times is less than the set value;
2.2 selecting the category with the largest proportion in the l adjacent samples as the category of the unknown object by adopting a KNN classification algorithm; or selecting the category with the maximum result as the category of the unknown object according to the distance and the distance by weighting the adjacent categories;
2.3 obtaining weak classifier G through training data learning by adopting Adaboost algorithmm(x) The strong classifier is obtained by combining the weak classifiers, and is expressed as:
(3) in the formula of alphamCalculating the weighted value of the weak classifier in the strong classifier for the mth time; m is the number of Adaboost iterations;
the implementation of step 3 comprises:
3.1 preprocessing the historical data to obtain an error percentage sequence, and defining a relative prediction error e (i) as:
Figure FDA0002271027050000022
(4) in the formula Ipre(i) Is a predicted value of the surface solar irradiance at the ith moment, Imea(i) The measured value of the surface solar irradiance at the ith moment is N, and the total sampling point number in one day is N;
3.2, dividing the state, wherein the research object is the error between the theoretical value and the actual value of the Hottel standard sunny day, dividing the error percentage obtained in the preprocessing into 11 discrete states according to the actual operation experience, wherein the point with the error smaller than-5% is defined as the state 1, the error level difference between the two adjacent states is 10%, and the like, and the point with the error larger than 85% is defined as the state 11;
3.3 solving a state transition probability matrix;
the state of solar irradiance at time t is
Figure FDA00022710270500000310
The state of the solar irradiance transferred to the moment t + k is SjTransition probability of
Figure FDA0002271027050000031
Comprises the following steps:
Figure FDA0002271027050000032
x (t) is a random variable in the markov chain;
Figure FDA0002271027050000033
the state of solar irradiance at time t; k is the step length of state transition, and P { } is the probability of occurrence of an event in brackets;
obtaining transition probability among the states at the current moment by using the state division result
Figure FDA0002271027050000034
The k-order state transition probability matrix is formed by the following steps:
Figure FDA0002271027050000035
(6) wherein Z is the total number of states divided by the system in the Markov chain;
3.4 calculating the influence weight of each order;
the k-order autocorrelation coefficient calculation formula of the error sequence is as follows:
Figure FDA0002271027050000036
(7) in the formula
Figure FDA0002271027050000037
Denotes the ith3The error in the time of day is,
Figure FDA0002271027050000038
is the average error, n is the time length;
calculating the influence weight of each order according to the autocorrelation coefficient of each order obtained by the formula (7):
Figure FDA0002271027050000039
(8) wherein K is the maximum step size in the Markov chain;
3.5 predicting the surface solar irradiance;
after the state transition probability matrixes of different step lengths are obtained, the probability of a certain state of the error percentage at the next sampling time is predicted according to the state of the error percentage at a certain time of the prediction day, and the state with the maximum probability is obtained as a prediction error state, so that a corresponding relative prediction error value e (i) is obtained; calculating a predicted value of the surface solar irradiance:
Ipre(i)＝Istd(i)×(1-e(i)) (9)；
(9) in the formula Ipre(i) Is a predicted value of the surface solar irradiance at the ith moment, Istd(i) The theoretical value of the Hottel standard sunny day at the ith moment, and e (i) the relative prediction error;
the implementation of the step 4 comprises the following steps:
after the prediction value of the solar irradiance is obtained by adopting an indirect prediction method, the photovoltaic output power is calculated through a photoelectric conversion model, and the calculation formula is as follows:
P＝ηSI[1-0.005(t0+25)] (10)；
(10) where η is the photoelectric conversion efficiency, S is the effective total area for photovoltaic power generation, I is the effective solar irradiance received, t0Is the operating temperature of the photovoltaic cell.
CN201611024358.3A 2016-11-21 2016-11-21 Distributed photovoltaic ultra-short term prediction method based on Adaboost clustering and Markov chain Active CN106529814B (en)

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 CN106529814A (en) 2017-03-22
CN106529814B true 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)

Families Citing this family (19)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
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
WO2019130718A1 (en) * 2017-12-28 2019-07-04 住友電気工業株式会社 Determination device, photovoltaic power generation system, determination method, and determination 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
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
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
CN110188964B (en) * 2019-06-06 2021-03-16 河北工业大学 Photovoltaic output prediction 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
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
CN112153000B (en) * 2020-08-21 2023-04-18 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
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
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
CN114169627B (en) * 2021-12-14 2022-11-18 湖南工商大学 System for deep reinforcement learning distributed photovoltaic power generation excitation
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
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
CN116050666B (en) * 2023-03-20 2023-07-18 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)

* Cited by examiner, † Cited by third party
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

Patent Citations (5)

* Cited by examiner, † Cited by third party
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

Also Published As

Publication number Publication date
CN106529814A (en) 2017-03-22

Similar Documents

Publication Publication Date Title
CN106529814B (en) Distributed photovoltaic ultra-short term prediction method based on Adaboost clustering and Markov chain
CN112949945B (en) Wind power ultra-short-term prediction method for improving bidirectional long-term and short-term memory network
CN107766990B (en) Method for predicting power generation power of photovoltaic power station
CN112257941B (en) Photovoltaic power station short-term power prediction method based on improved Bi-LSTM
CN110909919A (en) Photovoltaic power prediction method of depth neural network model with attention mechanism fused
CN112418346B (en) Numerical weather forecast total radiation system error classification calculation method
CN113344288B (en) Cascade hydropower station group water level prediction method and device and computer readable storage medium
CN109636054A (en) Solar energy power generating amount prediction technique based on classification and error combination prediction
CN111242355A (en) Photovoltaic probability prediction method and system based on Bayesian neural network
CN111626473A (en) Two-stage photovoltaic power prediction method considering error correction
CN112465251A (en) Short-term photovoltaic output probability prediction method based on simplest gated neural network
CN114861533A (en) Wind power ultra-short-term prediction method based on time convolution network
CN114792156A (en) Photovoltaic output power prediction method and system based on curve characteristic index clustering
CN113554466A (en) Short-term power consumption prediction model construction method, prediction method and device
CN115796004A (en) Photovoltaic power station ultra-short term power intelligent prediction method based on SLSTM and MLSTNet models
CN116341613A (en) Ultra-short-term photovoltaic power prediction method based on Informar encoder and LSTM
CN110852492A (en) Photovoltaic power ultra-short-term prediction method for finding similarity based on Mahalanobis distance
CN113516271A (en) Wind power cluster power day-ahead prediction method based on space-time neural network
CN105956708A (en) Grey correlation time sequence based short-term wind speed forecasting method
CN110796292A (en) Photovoltaic power short-term prediction method considering haze influence
CN112949938B (en) Wind power climbing event direct forecasting method for improving training sample class imbalance
CN115544895A (en) Photovoltaic power station annual output guarantee rate model optimization method
CN115456286A (en) Short-term photovoltaic power prediction method
CN114219122A (en) Ultra-short-term prediction method for generating power of small hydropower station based on grid meteorological data
CN113761023A (en) Photovoltaic power generation short-term power prediction method based on improved generalized neural network

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