CN113449471A - Wind power output simulation generation method for continuously improving MC (multi-channel) by utilizing AP (access point) clustering-skipping - Google Patents

Wind power output simulation generation method for continuously improving MC (multi-channel) by utilizing AP (access point) clustering-skipping Download PDF

Info

Publication number
CN113449471A
CN113449471A CN202110710544.7A CN202110710544A CN113449471A CN 113449471 A CN113449471 A CN 113449471A CN 202110710544 A CN202110710544 A CN 202110710544A CN 113449471 A CN113449471 A CN 113449471A
Authority
CN
China
Prior art keywords
wind power
power output
state
matrix
clustering
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
Application number
CN202110710544.7A
Other languages
Chinese (zh)
Other versions
CN113449471B (en
Inventor
肖白
李梦雪
董凌
杨森林
苟晓侃
王学斌
王茂春
杨洪志
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.)
Northeast Electric Power University
State Grid Qinghai Electric Power Co Ltd
Original Assignee
Northeast Dianli University
State Grid Qinghai Electric Power Co Ltd
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 Northeast Dianli University, State Grid Qinghai Electric Power Co Ltd filed Critical Northeast Dianli University
Priority to CN202110710544.7A priority Critical patent/CN113449471B/en
Publication of CN113449471A publication Critical patent/CN113449471A/en
Application granted granted Critical
Publication of CN113449471B publication Critical patent/CN113449471B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
    • 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
    • G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
    • 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
    • G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Electricity, gas or water supply
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2111/00—Details relating to CAD techniques
    • G06F2111/08—Probabilistic or stochastic CAD
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2111/00—Details relating to CAD techniques
    • G06F2111/10—Numerical modelling
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2113/00—Details relating to the application field
    • G06F2113/06—Wind turbines or wind farms
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
    • Y02E10/00—Energy generation through renewable energy sources
    • Y02E10/70—Wind energy
    • Y02E10/76—Power conversion electric or electronic aspects

Abstract

A wind power output simulation generation method utilizing AP clustering-skipping to continuously improve MC is characterized in that AP clustering is carried out on historical wind power output data, state transition matrixes of clustered wind power output are respectively established, and the state transition matrixes are converted into state skipping matrixes; forming a wind power output state skip sequence meeting the state skip characteristic based on the Markov chain; according to the wind power output state duration characteristic, fitting a wind power output state duration distribution histogram by adopting nuclear density estimation, and sampling to determine the duration of each state in a state skip sequence to obtain a wind power output state time sequence; the wind power output fluctuation quantity probability density distribution is fitted by adopting mixed Gaussian distribution, and the fluctuation components which accord with the mixed Gaussian distribution are superposed in the process of simulating and generating the wind power output so as to ensure the fluctuation characteristic of the wind power output; and generating a wind power output time sequence based on inter-class transfer matrix simulation. The method has the advantages of strong practicability, higher precision, better simulation effect and the like.

Description

Wind power output simulation generation method for continuously improving MC (multi-channel) by utilizing AP (access point) clustering-skipping
Technical Field
The invention relates to the field of wind power output time sequence simulation generation in wind power grid-connected power system planning, in particular to a wind power output simulation generation method for continuously improving MC (multi-core) by utilizing AP (access point) clustering-skipping.
Background
In the prior art, wind power output time sequence simulation generation methods are mainly divided into a wind speed method and a wind power method according to different research objects.
The wind speed method firstly simulates and generates a wind speed time sequence, and then converts the wind speed time sequence into a wind power output time sequence through an energy conversion model. The wind speed method can avoid errors caused by the phenomenon of wind abandon in the operation process of a power grid, and an energy conversion model is complex and is easily influenced by the characteristics of a unit. The wind power method directly utilizes the historical measured data to establish a wind power output time series model without an energy conversion process, thereby avoiding errors in the process. The wind power method can better simulate and generate a wind power output time sequence, but the wind power output time sequence is easy to fall into a certain state and is difficult to jump when the time sequence is generated, and the problem of strong subjectivity exists in category division.
In conclusion, the wind power output simulation generation method for continuously improving MC by utilizing AP clustering-skipping is provided by comprehensively considering wind power output daily statistical characteristics, state transition characteristics, time domain characteristics and fluctuation characteristics.
Disclosure of Invention
The invention aims to provide a wind power output simulation generation method for continuously improving MC by utilizing AP clustering-skipping, which has the advantages of strong practicability, higher precision and better simulation effect.
The technical scheme adopted for realizing the aim of the invention is that a wind power output simulation generation method for continuously improving MC by utilizing AP clustering-skipping is characterized by comprising the following steps: the method comprises the following steps of carrying out AP clustering on historical wind power output data, forming a wind power output state skip sequence under a certain clustering category, forming a wind power output state time sequence with the time sequence length of one day under a certain clustering category, sampling and superposing the wind power output fluctuation quantity on a primary wind power output time sequence formed under a certain clustering category, and generating a wind power output time sequence according to inter-category transfer probability matrix simulation, wherein the specific content is as follows:
1) AP clustering is carried out on historical wind power output data
An Affinity Propagation (AP) clustering algorithm is an unsupervised clustering algorithm based on "information transfer"; firstly, establishing a similarity matrix S of a sample set as formula (1), wherein elements S in the matrix area,b(a ≠ b) denotes sample xaAnd xbThe similarity between the two is expressed by negative Euclidean distance and is expressed as formula (2); sa,aThe reference degree of the clustering center is used as a judgment standard for judging whether the clustering center can be formed, a deviation parameter is set at the beginning, the deviation parameter directly influences the number of final clusters, and the larger the value of the deviation coefficient is, the more clusters are generated; constructing a vector capable of reflecting the characteristics of the corresponding solar wind power output according to the daily mean value and the daily minimum value of the actually measured historical wind power output at a set sampling interval in a unit time period, and carrying out AP clustering on the historical wind power output by using the vector in a day unit, wherein: the unit time period generally meansMonth, season, year, years; the sampling interval is typically 1 minute, 5 minutes or 10 minutes;
Figure BDA0003133540770000021
Sa,b＝-||xa－xb||2 (2)
in the formula: s is a similarity matrix of the sample set; sa,b(a ≠ b) denotes sample xaAnd xbThe similarity between them; sa,aThe reference degree of the clustering center; x is the number ofaAnd xbAre data sample points; a is 1,2, … …, n; b is 1,2, … …, n; n is the total number of samples used for AP clustering; | xa-xb||2Denotes xaAnd xbThe Euclidean distance between;
taking the similarity matrix S of the sample set as input, and then calculating the attraction r (x) of each sample pointa,xk) Degree of affiliation a (x)a,xk)；r(xa,xk) Denotes xkWhether a point is suitable as xaThe cluster center of the points, i.e. xkFor xaThe degree of attraction of (c); a (x)a,xk) Denotes xaWhether a point selects x or notkAs its clustering center, i.e. xaFor xkDegree of attribution of; r (x)a,xk) And a (x)a,xk) The larger, indicates xkThe more suitable as a clustering center; continuously updating the attraction degree matrix and the attribution degree matrix in the clustering process, and introducing a damping coefficient for avoiding oscillation in the iteration process until stable clustering centers and class attribution conditions are generated in the iteration process; thus, near the end of propagation, xaIs determined as xk(ii) a Wherein x iskIs represented by the formula (3):
xk＝arg max{a(xa,xk)+r(xa,xk)} (3)
in the formula: argmax is the argument for finding the maximum function, i.e. xkTo satisfy { a (x) }a,xk)+r(xa,xk) GetObtaining the value of the maximum value; a (x)a,xk) Is a sample point xaFor sample point xkDegree of attribution of; r (x)a,xk) Is a sample point xkFor sample point xaThe degree of attraction of (c); x is the number ofkAnd xaAre data sample points; k is 1,2, … …, n; a is 1,2, … …, n; n is the total number of samples used for AP clustering;
adopting a contour Coefficient (SC) as a clustering evaluation index; the larger the contour coefficient is, the better the clustering effect is; contour coefficient SC (x)k) Defined by formulae (4) to (6):
Figure BDA0003133540770000022
Figure BDA0003133540770000023
Figure BDA0003133540770000024
in the formula: x is the number ofk，xa，xbAre data sample points; c (x)k) Represents a sample point xkThe cluster category to which it belongs; c (x)a) Represents a sample point xaThe cluster category to which it belongs; c (x)b) Represents a sample point xbThe cluster category to which it belongs; contour coefficient SC (x)k) Has a value of [ -1,1 [)]Reflect xkAverage distance I (x) between cluster center and sample in class when serving as cluster centerk) Whether it is significantly different from its average distance O (x) to the out-of-class samplek)；I(xk) Representing the average distance between the cluster center and the samples in the class; d (x)k,xa) Represents a sample point xkAnd xaThe distance between them; m represents the sum of sample point xkThe number of data points belonging to the same category; o (x)k) Representing the average distance between the cluster center and the sample outside the cluster; d (x)k,xb) Represents a sample point xkAnd xbThe distance between them;g represents the sum of sample point xkThe number of data points not belonging to the same category;
by carrying out AP clustering on historical wind power output data in units of days, the solar wind power output sequences with output levels similar to fluctuation characteristics can be classified into the same class;
2) forming a wind power output state skip sequence under a certain cluster category
The historical wind power output data is expressed as (P'min，P′max) The wind power output range is divided into N data segments uniformly, each data segment corresponds to one state, and the wind power output range corresponding to each state is (P'max-P′min) N; wherein, P'minIs the minimum value of historical wind power output, P'maxThe output is the maximum value of the historical wind power, and N is a state number; aiming at the situation that the wind power output sequence is easy to fall into a certain state and is difficult to jump in the traditional Markov chain, so that the generated wind power output time sequence possibly has overlong duration in a certain state, the method adopts a state jump matrix to replace a state transition matrix for improvement;
when a state transition matrix is generated by a clustering reconstruction sequence, when the jump amplitude of the state number between two adjacent data is larger than 1/3 of the divided state number, the transition is considered to be not practical, the value of the state transition matrix is set to zero, and an improved state transition matrix is obtained;
secondly, setting all diagonal elements of the obtained state transition matrix to zero, calculating the proportion of each element in the sum of all elements of the row, and taking the proportion as a new probability value, wherein the obtained probability matrix is a state jump matrix; the state hopping matrix is of equations (7) - (8);
Figure BDA0003133540770000031
pij＝P(Et+1＝j|Et＝i) (8)
in the formula: the matrix P is a state jump matrix; the row of the matrix P corresponds to the current output state of the wind power output, and the column corresponds to the next momentThe engraved output state; n is the number of states; i and j are wind power output states; 1,2, … …, N; j ═ 1,2, … …, N; p is a radical ofijRepresenting the probability of the wind power output transferring from the state i to the state j; et、Et+1The wind power output values at the t moment and the t +1 moment are in corresponding states; p (a | B) is a conditional probability function representing the probability of a occurring under the condition of B;
the corresponding accumulated state transition matrix is of formula (9):
Figure BDA0003133540770000032
the value of the element in Q is represented by formula (10):
Figure BDA0003133540770000041
in the formula: the matrix Q is an accumulated state jump matrix; q. q.sulJumping to the element of the u row and l column in the matrix Q for the accumulated state; p is a radical ofujThe elements of the u-th row and the j-th column in the state jump matrix P; n is the number of states; u-1, 2, … …, N; 1,2, … …, N; j ═ 1,2, … …, N;
through the forming process of the accumulative state jump matrix Q, an accumulative state jump matrix under a certain cluster category, namely an intra-category accumulative state jump matrix, is obtained; then, sampling to form a state jump sequence under each cluster category according to the intra-category accumulated state jump matrix; under the cluster category, the state of the wind power output at the current moment is EtThe state at the next time is Et+1Generating random number xi following uniform distribution, if 0<ξ≤qEt,1Then E ist+11 is ═ 1; if q isEt,l<ξ≤qEt,l+1Then E ist+1L +1, wherein qEt,1E < th > representing intra-class accumulated state transition matrix QtRow, column 1 elements; q. q.sEt,lE < th > representing intra-class accumulated state transition matrix QtRow, column I elements; q. q.sEt,l+1E < th > representing intra-class accumulated state transition matrix QtRow, column l + 1;
since the diagonal elements of the state-hopping matrix are zero, the diagonal elements in the corresponding intra-class accumulated state-hopping matrix are equal to the left elements, i.e., qi,i-1Is equal to qiiSo [ q ]i,i-1,qii]For an empty set, when a wind power output state sequence is generated by using a Markov chain method, xi does not fall to [ q ]i,i-1,qii]Within the range; wherein q isiiThe elements of the ith row and the ith column in the intra-class accumulated state jump matrix Q are represented by i, which is 1,2, … …, N; q. q.si,i-1Jumping to the ith row and the (i-1) th column elements in the matrix Q for the accumulated state;
through the steps, a state skip sequence with different adjacent states when the wind power output is in a certain clustering category can be formed by sampling according to the intra-class accumulated state skip matrix in the clustering category;
3) forming a wind power output state time sequence with the time sequence length of one day under a certain cluster category
According to the method, the historical wind power output time sequence is traversed by a statistical method of recording the duration time T once from the first moment when the wind power output enters a certain state, and if the duration time is T, the times of occurrence of each duration time of the wind power output in the state are counted to obtain the distribution condition of the duration time in the state;
after a distribution histogram of the duration time of a certain state of the wind power output is obtained through statistics and drawing, a smooth curve is adopted to describe the histogram, so that the particularity existing in sampling the histogram can be avoided; when describing data distribution characteristics, Kernel Density Estimation (KDE) does not depend on selection of a parameter Estimation model, and dependence of histogram Estimation on a histogram group distance and a histogram position can be effectively avoided, so that the KDE is adopted to describe wind power output state duration distribution characteristics; the KDE expression is formula (11):
Figure BDA0003133540770000042
in the formula:
Figure BDA0003133540770000043
a kernel density estimation function for the duration of the wind power output state; k (-) is a kernel function; h is the bandwidth; x is the number of1,x2,……,xWThe data number is W; x is the number ofwThe W-th sample value of the duration time of the wind power output state is 1,2, … … and W; w represents that W different durations of the wind power output sequence appear in a certain state;
in KDE, the selection of the bandwidth determines the smoothness degree of a fitting curve, the larger the bandwidth is, the smoother the bandwidth is, but the poorer the fitting effect is; selecting a Gaussian function as a kernel function, and solving the bandwidth by adopting an empirical rule, wherein the formula is as follows (12):
h＝1.06σW-1/5 (12)
in the formula: h is the bandwidth; sigma is the normal distribution standard deviation of the duration time of the wind power output state; w represents that W different durations of the wind power output sequence appear in a certain state;
in order to obtain the state time sequence of the wind power output with the time sequence length of one day under a certain clustering category, after a random number set meeting the duration time of each state of a kernel density estimation function is generated, traversing the obtained wind power output state jump sequence under the clustering category, and performing repeated sampling: the current state is EtThen in state EtIs arbitrarily chosen as a value in the set of duration random numbers of (a) as state EtThe time duration is repeated until the length of the generated time sequence meets the requirement; 4) sampling and superposing wind power output fluctuation quantity on a preliminary wind power output time sequence formed under a certain cluster category
After a wind power output state time sequence under a certain cluster category is formed, determining the initial wind power output at each moment; the current state of wind power output is EtThen, the preliminary wind power output P 'at the moment can be randomly generated't·cbNumerical value of (1), P't·cb∈(PEt.min，PEt.max)，PEt.min、PEt.maxAre respectively in state EtThe minimum value and the maximum value of the corresponding wind power output range;
the wind power output fluctuation is that the output values at adjacent moments are different, and the fluctuation characteristic is described by adopting a first-order difference fluctuation quantity, namely the output difference value in adjacent 2 unit times; the first-order difference fluctuation quantity expression of the wind power output is as shown in formula (13):
ΔPt′＝P′t+1-Pt′ (13)
in the formula: t represents the wind power output moment; delta Pt' represents the fluctuation amount of wind power output at the moment t; pt'and P't+1Respectively representing wind power output values at the time t and the time t + 1;
the probability density distribution fitting function of the first-order difference fluctuation quantity of the wind power output is as follows: normal distribution, t-location-scale distribution and logistic distribution; the wind power output fluctuation characteristic is characterized by mixed Gaussian distribution with better fitting precision, the probability density function is the weighting of a plurality of Gaussian probability density functions, when the data distribution is complex, the mixed Gaussian distribution function can solve the problem of the insufficient fitting precision of a single Gaussian distribution function, and the mathematical model expression is as the following formulas (14) - (15):
Figure BDA0003133540770000051
Figure BDA0003133540770000052
in the formula: (x) is a Gaussian mixture distribution function; v ═ 1,2, … …, V; v is the number of single Gaussian distributions in the mixed Gaussian distribution; a isvIs the mixing coefficient; p is a radical ofgs(x|bv，cv) A probability density function that is a v-th single gaussian distribution, where x represents a random number that follows the single gaussian distribution; bvAnd cvRespectively represent the mean and standard deviation of the v-th single gaussian distribution;
solving the distribution parameters corresponding to the Gaussian mixture distribution function by adopting an expectation-maximization (EM) algorithm, wherein the EM algorithm is divided into two steps: step 1, called the expectation (E), calculates the expectation of the likelihood function based on the parameters of the initial values or the previous iteration values; step 2 is a maximization (M) step, which maximizes the likelihood function and converts it into new parameter values that can be obtained, and the EM algorithm is a process of repeated iteration until the above two steps converge;
after a fluctuation amount random number set meeting the mixed Gaussian distribution function is generated, randomly extracting the fluctuation amount, and superposing the fluctuation amount on the formed initial wind power output value, as shown in a formula (16); traversing the generated preliminary wind power output time sequence, and completing superposition of fluctuation components to form a wind power output time sequence with the time sequence length of one day under a certain cluster type;
P′t·fd＝P′t·cb+ΔPt′ (16)
in the formula: p't·fdThe wind power output value at the moment t of the wind power output time sequence under a certain clustering category is obtained; p't·cbThe preliminary wind power output value at the moment t of the wind power output time sequence under a certain clustering category is obtained; delta Pt' represents the fluctuation amount of wind power output at the moment t;
5) generating wind power output time sequence according to inter-class transition probability matrix simulation
Determining the time length of the wind power output time sequence generated by simulation according to the requirement of medium-long term optimization scheduling of the power system containing wind power generation, wherein the time length required by the medium-long term optimization scheduling is generally 2 days or more;
respectively counting the transfer times between the classes of the wind power output on two adjacent days, and calculating the corresponding transfer probability to obtain inter-class transfer probability matrixes, such as formulas (17) to (18), so as to obtain accumulated inter-class transfer probability matrixes, such as formulas (19) to (20);
Figure BDA0003133540770000061
plj·rz＝P(He+1＝z|He＝r) (18)
in the formula: matrix PljMove between classesA rate matrix; matrix PljThe row of the corresponding cluster type of the wind power output of the current day, and the column of the corresponding cluster type of the wind power output of the next day; l is the number of categories generated by AP clustering; r and z are clustering categories to which the wind power output belongs; r is 1,2, … …, L; z is 1,2, … …, L; p is a radical oflj·rzRepresenting the probability that the cluster category to which the wind power output belongs is transferred from the category r to the category z; he、He+1The cluster type of the wind power output on the current day and the next day; p (a | B) is a conditional probability function representing the probability of a occurring under the condition of B;
Figure BDA0003133540770000062
Figure BDA0003133540770000071
in the formula: matrix QljIs an accumulative inter-class transition probability matrix; q. q.slj·rgFor accumulating inter-class transition probability matrix QljThe elements of the r-th row and the g-th column; p is a radical oflj·rzAs an inter-class transition probability matrix PljThe elements of the r-th row and the z-th column; l is the number of categories generated by AP clustering; r is 1,2, … …, L; z is 1,2, … …, L; g is 1,2, … …, L;
and sampling the cluster type to which the wind power output of the current day belongs according to the accumulated inter-class transition probability matrix to obtain the AP cluster type to which the wind power output of the next day belongs: the cluster type of the wind power output on the same day is HeThe cluster category to which the next-day wind power output belongs is He+1Generating random number epsilon subject to uniform distribution if 0<ε≤qHe,1Then H ise+11 is ═ 1; if q isHe,g<ε≤qHe,g+1Then H ise+1G +1, wherein qHe,1Representing transition probability matrix Q between cumulative classesljH ofeRow, column 1 elements; q. q.sHe,gRepresenting transition probability matrix Q between cumulative classesljH ofeRow, column g elements; q. q.sHe,g+1Representing transition probability matrix Q between cumulative classesljTo (1) aHeRow, column g +1 elements;
the method comprises the following steps of generating a wind power output time sequence within R days in a simulated mode at a set sampling interval, wherein R is a positive integer:
judging the AP clustering category to which the wind power output of the 1 st day belongs;
secondly, forming a wind power output simulation generation time sequence of the day 1 by executing the steps 2-4), and recording the time sequence as omega1，Ω1＝{P′1·fd(1),P′2·fd(1),……,P′t·fd(1)In which Ω is1Representing a wind power output simulation generation sequence of the 1 st day; wherein, P't·fd(1)The simulation generation wind power output value of the tth moment of the 1 st day is shown, and t shows the wind power output moment;
thirdly, calculating to obtain a wind power output accumulated inter-class transfer probability matrix in the step 5), and sampling to determine the AP cluster class to which the wind power output belongs on the day tau, wherein tau is 2,3, … … and R;
fourthly, repeating the steps 2) to 4) to obtain a wind power output simulation generation time sequence on the tau day, and recording the time sequence as omegaτ，Ωτ＝{P′1·fd(τ),P′2·fd(τ),……,P′t·fd(τ)In which Ω isτRepresenting a wind power output simulation generation sequence on the tau day; p't·fd(τ)Wherein, the simulation generation wind power output value at the tth moment on the Tth day is shown;
fifthly, repeatedly executing the third step and the fourth step until a wind power output time sequence of R days in total is generated and is recorded as omega, and omega is { omega }1,……,Ωτ,……,ΩR}。
The invention discloses a wind power output simulation generation method for continuously improving MC by utilizing AP clustering-skipping, which is characterized in that actually-measured wind power output data are classified by adopting AP clustering, the problem of poor modeling stability caused by artificial subjectivity is effectively solved, the skipping characteristic and the time sequence characteristic of a wind power output state are comprehensively considered, the problem of long duration time of a certain state in the simulation generation process is improved to a certain extent, a mixed Gaussian distribution is adopted to fit a wind power output fluctuation quantity probability density distribution histogram, the wind power output fluctuation characteristic can be better described, and the superiority of the method can be verified by comparing the probability characteristic, the autocorrelation characteristic and the statistical characteristic of a wind power output time sequence and a historical time sequence obtained by different simulation generation methods.
Drawings
FIG. 1 is a flow chart of wind power output time series simulation generation;
FIG. 2 is a wind power output AP cluster diagram in a certain area;
FIG. 3 is a wind power output state duration time fitting graph;
FIG. 4 is a schematic diagram of comparison of wind power output fluctuation amount fitted by normal distribution, Logistic distribution, t Location-scale distribution and mixed Gaussian distribution;
FIG. 5 is a schematic diagram comparing a partial length historical wind power output time series, a wind power output time series generated based on conventional Markov chain simulation, a wind power output time series generated based on continuous Markov chain simulation, and a wind power output time series generated by AP clustering-skipping continuous improvement MC simulation;
FIG. 6 is a schematic diagram of a comparison of probability distributions of a historical wind power output time series, a wind power output time series generated based on conventional Markov chain simulation, a wind power output time series generated based on continuous Markov chain simulation, and a wind power output time series generated by utilizing AP clustering-skipping continuous improvement MC simulation;
FIG. 7 is a schematic diagram of the comparison of the autocorrelation coefficients of a historical wind power output time series, a wind power output time series generated based on conventional Markov chain simulation, a wind power output time series generated based on continuous Markov chain simulation, and a wind power output time series generated by utilizing AP clustering-skipping continuous improvement MC simulation.
Detailed Description
The invention will be further illustrated with reference to the accompanying figures 1-7 and examples.
The invention discloses a wind power output simulation generation method for continuously improving MC (multi-core) by utilizing AP (access point) clustering-skipping, which is characterized by comprising the following steps of: the method comprises the following steps of carrying out AP clustering on historical wind power output data, forming a wind power output state skip sequence under a certain clustering category, forming a wind power output state time sequence with the time sequence length of one day under a certain clustering category, sampling and superposing the wind power output fluctuation quantity on a primary wind power output time sequence formed under a certain clustering category, and generating a wind power output time sequence according to inter-category transfer probability matrix simulation, wherein the specific content is as follows:
1) AP clustering is carried out on historical wind power output data
An Affinity Propagation (AP) clustering algorithm is an unsupervised clustering algorithm based on "information transfer"; firstly, establishing a similarity matrix S of a sample set as formula (1), wherein elements S in the matrix area,b(a ≠ b) denotes sample xaAnd xbThe similarity between the two is expressed by negative Euclidean distance and is expressed as formula (2); sa,aThe reference degree of the clustering center is used as a judgment standard for judging whether the clustering center can be formed, a deviation parameter is set at the beginning, the deviation parameter directly influences the number of final clusters, and the larger the value of the deviation coefficient is, the more clusters are generated; constructing a vector capable of reflecting the characteristics of the corresponding solar wind power output according to the daily mean value and the daily minimum value of the actually measured historical wind power output at a set sampling interval in a unit time period, and carrying out AP clustering on the historical wind power output by using the vector in a day unit, wherein: the unit time period generally refers to month, season, year and years; the sampling interval is typically 1 minute, 5 minutes or 10 minutes;
Figure BDA0003133540770000091
Sa,b＝-||xa－xb||2 (2)
in the formula: s is a similarity matrix of the sample set; sa,b(a ≠ b) denotes sample xaAnd xbThe similarity between them; sa,aThe reference degree of the clustering center; x is the number ofaAnd xbAre data sample points; a is 1,2, … …, n; b is 1,2, … …, n; n is the total number of samples used for AP clustering; | xa-xb||2Denotes xaAnd xbThe Euclidean distance between;
taking a similarity matrix S of the sample set as an input, and thenCalculate the attraction r (x) for each sample pointa,xk) Degree of affiliation a (x)a,xk)；r(xa,xk) Denotes xkWhether a point is suitable as xaThe cluster center of the points, i.e. xkFor xaThe degree of attraction of (c); a (x)a,xk) Denotes xaWhether a point selects x or notkAs its clustering center, i.e. xaFor xkDegree of attribution of; r (x)a,xk) And a (x)a,xk) The larger, indicates xkThe more suitable as a clustering center; continuously updating the attraction degree matrix and the attribution degree matrix in the clustering process, and introducing a damping coefficient for avoiding oscillation in the iteration process until stable clustering centers and class attribution conditions are generated in the iteration process; thus, near the end of propagation, xaIs determined as xk(ii) a Wherein x iskIs represented by the formula (3):
xk＝arg max{a(xa,xk)+r(xa,xk)} (3)
in the formula: argmax is the argument for finding the maximum function, i.e. xkTo satisfy { a (x) }a,xk)+r(xa,xk) Taking the value of the maximum value; a (x)a,xk) Is a sample point xaFor sample point xkDegree of attribution of; r (x)a,xk) Is a sample point xkFor sample point xaThe degree of attraction of (c); x is the number ofkAnd xaAre data sample points; k is 1,2, … …, n; a is 1,2, … …, n; n is the total number of samples used for AP clustering;
adopting a contour Coefficient (SC) as a clustering evaluation index; the larger the contour coefficient is, the better the clustering effect is; contour coefficient SC (x)k) Defined by formulae (4) to (6):
Figure BDA0003133540770000092
Figure BDA0003133540770000093
Figure BDA0003133540770000094
in the formula: x is the number ofk，xa，xbAre data sample points; c (x)k) Represents a sample point xkThe cluster category to which it belongs; c (x)a) Represents a sample point xaThe cluster category to which it belongs; c (x)b) Represents a sample point xbThe cluster category to which it belongs; contour coefficient SC (x)k) Has a value of [ -1,1 [)]Reflect xkAverage distance I (x) between cluster center and sample in class when serving as cluster centerk) Whether it is significantly different from its average distance O (x) to the out-of-class samplek)；I(xk) Representing the average distance between the cluster center and the samples in the class; d (x)k,xa) Represents a sample point xkAnd xaThe distance between them; m represents the sum of sample point xkThe number of data points belonging to the same category; o (x)k) Representing the average distance between the cluster center and the sample outside the cluster; d (x)k,xb) Represents a sample point xkAnd xbThe distance between them; g represents the sum of sample point xkThe number of data points not belonging to the same category;
by carrying out AP clustering on historical wind power output data in units of days, the solar wind power output sequences with output levels similar to fluctuation characteristics can be classified into the same class;
2) forming a wind power output state skip sequence under a certain cluster category
The historical wind power output data is expressed as (P'min，P′max) The wind power output range is divided into N data segments uniformly, each data segment corresponds to one state, and the wind power output range corresponding to each state is (P'max-P′min) N; wherein, P'minIs the minimum value of historical wind power output, P'maxThe output is the maximum value of the historical wind power, and N is a state number; aiming at the problem that the wind power output sequence is easy to fall into a certain state and is difficult to jump in the traditional Markov chain, so that the generated wind power output time sequence canThe situation that the duration time is too long in a certain state can occur, and the state transition matrix is replaced by the state jump matrix for improvement;
when a state transition matrix is generated by a clustering reconstruction sequence, when the jump amplitude of the state number between two adjacent data is larger than 1/3 of the divided state number, the transition is considered to be not practical, the value of the state transition matrix is set to zero, and an improved state transition matrix is obtained;
secondly, setting all diagonal elements of the obtained state transition matrix to zero, calculating the proportion of each element in the sum of all elements of the row, and taking the proportion as a new probability value, wherein the obtained probability matrix is a state jump matrix; the state hopping matrix is of equations (7) - (8);
Figure BDA0003133540770000101
pij＝P(Et+1＝j|Et＝i) (8)
in the formula: the matrix P is a state jump matrix; the row of the matrix P corresponds to the current output state of the wind power output, and the column corresponds to the output state at the next moment; n is the number of states; i and j are wind power output states; 1,2, … …, N; j ═ 1,2, … …, N; p is a radical ofijRepresenting the probability of the wind power output transferring from the state i to the state j; et、Et+1The wind power output values at the t moment and the t +1 moment are in corresponding states; p (a | B) is a conditional probability function representing the probability of a occurring under the condition of B;
the corresponding accumulated state transition matrix is of formula (9):
Figure BDA0003133540770000102
the value of the element in Q is represented by formula (10):
Figure BDA0003133540770000103
in the formula: the matrix Q is an accumulated state jump matrix; q. q.sulJumping to the element of the u row and l column in the matrix Q for the accumulated state; p is a radical ofujThe elements of the u-th row and the j-th column in the state jump matrix P; n is the number of states; u-1, 2, … …, N; 1,2, … …, N; j ═ 1,2, … …, N;
through the forming process of the accumulative state jump matrix Q, an accumulative state jump matrix under a certain cluster category, namely an intra-category accumulative state jump matrix, is obtained; then, sampling to form a state jump sequence under each cluster category according to the intra-category accumulated state jump matrix; under the cluster category, the state of the wind power output at the current moment is EtThe state at the next time is Et+1Generating random number xi following uniform distribution, if 0<ξ≤qEt,1Then E ist+11 is ═ 1; if q isEt,l<ξ≤qEt,l+1Then E ist+1L +1, wherein qEt,1E < th > representing intra-class accumulated state transition matrix QtRow, column 1 elements; q. q.sEt,lE < th > representing intra-class accumulated state transition matrix QtRow, column I elements; q. q.sEt,l+1E < th > representing intra-class accumulated state transition matrix QtRow, column l + 1;
since the diagonal elements of the state-hopping matrix are zero, the diagonal elements in the corresponding intra-class accumulated state-hopping matrix are equal to the left elements, i.e., qi,i-1Is equal to qiiSo [ q ]i,i-1,qii]For an empty set, when a wind power output state sequence is generated by using a Markov chain method, xi does not fall to [ q ]i,i-1,qii]Within the range; wherein q isiiThe elements of the ith row and the ith column in the intra-class accumulated state jump matrix Q are represented by i, which is 1,2, … …, N; q. q.si,i-1Jumping to the ith row and the (i-1) th column elements in the matrix Q for the accumulated state;
through the steps, a state skip sequence with different adjacent states when the wind power output is in a certain clustering category can be formed by sampling according to the intra-class accumulated state skip matrix in the clustering category;
3) forming a wind power output state time sequence with the time sequence length of one day under a certain cluster category
According to the method, the historical wind power output time sequence is traversed by a statistical method of recording the duration time T once from the first moment when the wind power output enters a certain state, and if the duration time is T, the times of occurrence of each duration time of the wind power output in the state are counted to obtain the distribution condition of the duration time in the state;
after a distribution histogram of the duration time of a certain state of the wind power output is obtained through statistics and drawing, a smooth curve is adopted to describe the histogram, so that the particularity existing in sampling the histogram can be avoided; when describing data distribution characteristics, Kernel Density Estimation (KDE) does not depend on selection of a parameter Estimation model, and dependence of histogram Estimation on a histogram group distance and a histogram position can be effectively avoided, so that the KDE is adopted to describe wind power output state duration distribution characteristics; the KDE expression is formula (11):
Figure BDA0003133540770000111
in the formula:
Figure BDA0003133540770000112
a kernel density estimation function for the duration of the wind power output state; k (-) is a kernel function; h is the bandwidth; x is the number of1,x2,……,xWThe data number is W; x is the number ofwThe W-th sample value of the duration time of the wind power output state is 1,2, … … and W; w represents that W different durations of the wind power output sequence appear in a certain state;
in KDE, the selection of the bandwidth determines the smoothness degree of a fitting curve, the larger the bandwidth is, the smoother the bandwidth is, but the poorer the fitting effect is; selecting a Gaussian function as a kernel function, and solving the bandwidth by adopting an empirical rule, wherein the formula is as follows (12):
h＝1.06σW-1/5 (12)
in the formula: h is the bandwidth; sigma is the normal distribution standard deviation of the duration time of the wind power output state; w represents that W different durations of the wind power output sequence appear in a certain state;
in order to obtain the state time sequence of the wind power output with the time sequence length of one day under a certain clustering category, after a random number set meeting the duration time of each state of a kernel density estimation function is generated, traversing the obtained wind power output state jump sequence under the clustering category, and performing repeated sampling: the current state is EtThen in state EtIs arbitrarily chosen as a value in the set of duration random numbers of (a) as state EtThe time duration is repeated until the length of the generated time sequence meets the requirement; 4) sampling and superposing wind power output fluctuation quantity on a preliminary wind power output time sequence formed under a certain cluster category
After a wind power output state time sequence under a certain cluster category is formed, determining the initial wind power output at each moment; the current state of wind power output is EtThen, the preliminary wind power output P 'at the moment can be randomly generated't·cbNumerical value of (1), P't·cb∈(PEt.min，PEt.max)，PEt.min、PEt.maxAre respectively in state EtThe minimum value and the maximum value of the corresponding wind power output range;
the wind power output fluctuation is that the output values at adjacent moments are different, and the fluctuation characteristic is described by adopting a first-order difference fluctuation quantity, namely the output difference value in adjacent 2 unit times; the first-order difference fluctuation quantity expression of the wind power output is as shown in formula (13):
ΔPt′＝P′t+1-Pt′ (13)
in the formula: t represents the wind power output moment; delta Pt' represents the fluctuation amount of wind power output at the moment t; pt'and P't+1Respectively representing wind power output values at the time t and the time t + 1;
the probability density distribution fitting function of the first-order difference fluctuation quantity of the wind power output is as follows: normal distribution, t-location-scale distribution and logistic distribution; the wind power output fluctuation characteristic is characterized by mixed Gaussian distribution with better fitting precision, the probability density function is the weighting of a plurality of Gaussian probability density functions, when the data distribution is complex, the mixed Gaussian distribution function can solve the problem of the insufficient fitting precision of a single Gaussian distribution function, and the mathematical model expression is as the following formulas (14) - (15):
Figure BDA0003133540770000121
Figure BDA0003133540770000122
in the formula: (x) is a Gaussian mixture distribution function; v ═ 1,2, … …, V; v is the number of single Gaussian distributions in the mixed Gaussian distribution; a isvIs the mixing coefficient; p is a radical ofgs(x|bv，cv) A probability density function that is a v-th single gaussian distribution, where x represents a random number that follows the single gaussian distribution; bvAnd cvRespectively represent the mean and standard deviation of the v-th single gaussian distribution;
solving the distribution parameters corresponding to the Gaussian mixture distribution function by adopting an expectation-maximization (EM) algorithm, wherein the EM algorithm is divided into two steps: step 1, called the expectation (E), calculates the expectation of the likelihood function based on the parameters of the initial values or the previous iteration values; step 2 is a maximization (M) step, which maximizes the likelihood function and converts it into new parameter values that can be obtained, and the EM algorithm is a process of repeated iteration until the above two steps converge;
after a fluctuation amount random number set meeting the mixed Gaussian distribution function is generated, randomly extracting the fluctuation amount, and superposing the fluctuation amount on the formed initial wind power output value, as shown in a formula (16); traversing the generated preliminary wind power output time sequence, and completing superposition of fluctuation components to form a wind power output time sequence with the time sequence length of one day under a certain cluster type;
P′t·fd＝P′t·cb+ΔPt′ (16)
in the formula: p't·fdWind power of a wind power output time sequence t moment under a certain clustering categoryOutputting a force value; p't·cbThe preliminary wind power output value at the moment t of the wind power output time sequence under a certain clustering category is obtained; delta Pt' represents the fluctuation amount of wind power output at the moment t;
5) generating wind power output time sequence according to inter-class transition probability matrix simulation
Determining the time length of the wind power output time sequence generated by simulation according to the requirement of medium-long term optimization scheduling of the power system containing wind power generation, wherein the time length required by the medium-long term optimization scheduling is generally 2 days or more;
respectively counting the transfer times between the classes of the wind power output on two adjacent days, and calculating the corresponding transfer probability to obtain inter-class transfer probability matrixes, such as formulas (17) to (18), so as to obtain accumulated inter-class transfer probability matrixes, such as formulas (19) to (20);
Figure BDA0003133540770000131
plj·rz＝P(He+1＝z|He＝r) (18)
in the formula: matrix PljIs an inter-class transition probability matrix; matrix PljThe row of the corresponding cluster type of the wind power output of the current day, and the column of the corresponding cluster type of the wind power output of the next day; l is the number of categories generated by AP clustering; r and z are clustering categories to which the wind power output belongs; r is 1,2, … …, L; z is 1,2, … …, L; p is a radical oflj·rzRepresenting the probability that the cluster category to which the wind power output belongs is transferred from the category r to the category z; he、He+1The cluster type of the wind power output on the current day and the next day; p (a | B) is a conditional probability function representing the probability of a occurring under the condition of B;
Figure BDA0003133540770000132
Figure BDA0003133540770000133
in the formula: matrix QljIs an accumulative inter-class transition probability matrix; q. q.slj·rgFor accumulating inter-class transition probability matrix QljThe elements of the r-th row and the g-th column; p is a radical oflj·rzAs an inter-class transition probability matrix PljThe elements of the r-th row and the z-th column; l is the number of categories generated by AP clustering; r is 1,2, … …, L; z is 1,2, … …, L; g is 1,2, … …, L;
and sampling the cluster type to which the wind power output of the current day belongs according to the accumulated inter-class transition probability matrix to obtain the AP cluster type to which the wind power output of the next day belongs: the cluster type of the wind power output on the same day is HeThe cluster category to which the next-day wind power output belongs is He+1Generating random number epsilon subject to uniform distribution if 0<ε≤qHe,1Then H ise+11 is ═ 1; if q isHe,g<ε≤qHe,g+1Then H ise+1G +1, wherein qHe,1Representing transition probability matrix Q between cumulative classesljH ofeRow, column 1 elements; q. q.sHe,gRepresenting transition probability matrix Q between cumulative classesljH ofeRow, column g elements; q. q.sHe,g+1Representing transition probability matrix Q between cumulative classesljH ofeRow, column g +1 elements;
the method comprises the following steps of generating a wind power output time sequence within R days in a simulated mode at a set sampling interval, wherein R is a positive integer:
judging the AP clustering category to which the wind power output of the 1 st day belongs;
secondly, forming a wind power output simulation generation time sequence of the day 1 by executing the steps 2-4), and recording the time sequence as omega1，Ω1＝{P′1·fd(1),P′2·fd(1),……,P′t·fd(1)In which Ω is1Representing a wind power output simulation generation sequence of the 1 st day; wherein, P't·fd(1)The simulation generation wind power output value of the tth moment of the 1 st day is shown, and t shows the wind power output moment;
thirdly, calculating to obtain a wind power output accumulated inter-class transfer probability matrix in the step 5), and sampling to determine the AP cluster class to which the wind power output belongs on the day tau, wherein tau is 2,3, … … and R;
fourthly, repeating the steps 2) to 4) to obtain a wind power output simulation generation time sequence on the tau day, and recording the time sequence as omegaτ，Ωτ＝{P′1·fd(τ),P′2·fd(τ),……,P′t·fd(τ)In which Ω isτRepresenting a wind power output simulation generation sequence on the tau day; p't·fd(τ)Wherein, the simulation generation wind power output value at the tth moment on the Tth day is shown;
fifthly, repeatedly executing the third step and the fourth step until a wind power output time sequence of R days in total is generated and is recorded as omega, and omega is { omega }1,……,Ωτ,……,ΩR}。
Specific examples are as follows: the invention provides a wind power output simulation generation method for continuously improving MC by utilizing AP clustering-skipping, which comprises the following steps:
analyzing the wind power output data of the whole year from 2018, month 1 and year 2019, month 9 and month 30 of a wind power plant in a certain area as a sample, wherein the total installed capacity is 1904MW, the data sampling time interval is 1min, and the wind power output time sequence from 2019, month 1 to year 2019, month 10 and day 31 is generated by performing normalization processing on the basis of the total installed capacity.
Using Residual Sum of Squares (RSS), Root Mean Square Error (RMSE), coefficient R-square, Mean Error epsilonmeanError of sum standard deviation epsilonstdAs an evaluation index of the effect of the simulation generation method. RSS, RMSE, εmeanAnd εstdThe smaller the R-square is, the closer the R-square is to 1, which shows that the more consistent the statistical property, the probability distribution property and the autocorrelation property of the time sequence generated by simulation and the historical measured sequence are, the better the simulation generation effect is.
The measured wind power output data of a certain area in one year is constructed into a vector capable of reflecting the characteristics of the wind power output corresponding to each day according to the daily mean value and the daily minimum difference of the measured wind power output data, the vector is utilized to perform AP clustering on the historical wind power output in units of days, the data sampling time interval is 1min, the historical measured wind power output data can be clustered into 5 classes through an AP clustering algorithm, and the clustering effect is shown in figure 2.
And respectively fitting the duration time distribution histogram of each state of the wind power output by utilizing an inverse Gaussian function, a t-location scale function and kernel density estimation, taking the duration time of the state 4 in the wind power output of the category 2 as an example, as shown in FIG. 3 (the probability that the state duration time is more than 50min is extremely small, so that only the state duration time is drawn in the portion of less than 50min in FIG. 3). And comparing the fitting curves with the wind power output state duration distribution histogram, and finding out that the fitting effect of the kernel density estimation is better and the wind power output state duration histogram can be better approximated.
The mixed gaussian distribution function is solved by using the EM algorithm to obtain the estimated values of each parameter, as shown in table 1.
TABLE 1 estimation of three-component Gaussian mixture distribution parameters
Figure BDA0003133540770000151
The wind power output fluctuation amount probability density distribution histogram is fitted by utilizing normal distribution, Logistic distribution, t Location-scale distribution and mixed Gaussian distribution, as shown in fig. 4.
As can be seen from FIG. 4, the fluctuation amount of the wind power output is concentrated at the low value end, the per unit value of the maximum fluctuation amount is less than 0.05, the probability density distribution has the characteristic of thick tail, and the whole has symmetry. The evaluation index results of the wind power output fluctuation amount probability density distribution fitting functions are shown in table 2.
TABLE 2 wind power output fluctuation quantity probability density distribution fitting function evaluation index comparison
Figure BDA0003133540770000152
In the distribution, 3 indexes of the mixed Gaussian distribution model are optimal, wherein the RSS and the RMSE of the mixed Gaussian distribution are respectively reduced by 98% and 86% compared with normal distribution, the R-square is improved by 33% compared with normal distribution, and the fitting effect of the mixed Gaussian distribution on the fluctuation amount probability density distribution histogram is verified to be the best.
The statistical inter-class transition probability matrix of the wind power output of the wind power plant in a certain area from 2018, 10 months and 1 days to 2019, 9 months and 30 days is as follows:
Figure BDA0003133540770000161
FIG. 5 compares the historical sequence with the time sequences generated by the wind power output simulation generation method, the continuous Markov chain method and the traditional Markov chain method using AP clustering-skipping continuous improvement MC. The method for simulating and generating the wind power output by continuously improving the MC through the AP clustering-skipping better solves the problem that the wind power output sequence in the traditional Markov chain method has overlong duration time in a low-output section, and the method for simulating and generating the wind power output by continuously improving the MC through the AP clustering-skipping considers the time domain characteristic of the wind power output and solves the problem that the generated time sequence fluctuates too frequently in a certain state to a certain extent.
Fig. 6 compares Probability Distribution Function (PDF) of the wind power output time series generated by different methods with the historical time series. As can be seen from fig. 6, the probability density distribution of the wind power output shows the characteristics of high probability density of the low-output part and low probability density of the high-output part, and the probability density distribution of the wind power output time sequence generated by the wind power output simulation generation method of the AP clustering-skipping continuous improvement MC is closer to the probability density distribution corresponding to the historical sequence, and further shows the superiority of the wind power output simulation generation method of the MC continuous improvement by the AP clustering-skipping from the aspect of the probability density distribution.
Autocorrelation calculation is performed on time sequences generated by the wind power output simulation generation method, the continuous Markov chain method and the traditional Markov chain method which utilize the AP clustering-skipping continuous improvement MC, and the result is shown in FIG. 7. The abscissa represents the lag time, and the ordinate represents the Auto Correlation Function (ACF) corresponding to different lag times, and the lag time is calculated to be 200min at intervals of 1 min. It can be seen from fig. 7 that, when the lag time is less than 50min, the autocorrelation coefficients of the wind power output time series generated by the three methods are relatively close to the historical sequence, but as the lag time increases, the autocorrelation coefficients of the wind power output time series obtained by the wind power output simulation generation method of continuously improving MC by AP clustering-skipping are closer to the historical sequence, i.e., it is verified that the wind power output simulation generation method of continuously improving MC by AP clustering-skipping can better retain the autocorrelation of the original wind power output time series.
TABLE 3 comparison of probability distribution characteristics and autocorrelation of sequences obtained by different simulation generation methods
Figure BDA0003133540770000162
Table 3 shows that the probability density distribution values of different model sequences and RSS, RMSE and R-square of autocorrelation function values are quantitatively compared, and the three indexes of the method are superior to those of a continuous Markov chain method and a traditional Markov chain method. Wherein, the RSS and RMSE of the probability density distribution value are respectively reduced by 64.3 percent and 94.2 percent compared with the sequences generated by the continuous Markov chain method in a simulation way, and the R-square is improved by 0.9 percent compared with the continuous Markov chain sequences; the RSS and RMSE of the autocorrelation function values are respectively reduced by 77.9% and 53.5% compared with the sequences generated by the continuous Markov chain method, and the R-square is improved by 0.3% compared with the continuous Markov chain sequences, namely the probability density distribution characteristics and the autocorrelation characteristics of the wind power output time sequence obtained by the method are verified to be closer to the historical wind power output time sequence compared with other two methods, and the simulation effect is better.
TABLE 4 statistical index comparisons of sequences obtained by different simulation Generation methods
Figure BDA0003133540770000171
To verify the superiority and inferiority of the time series obtained by each simulation generation method in the statistical properties, table 4 lists the average error (epsilon) between the sequence generated by each method and the historical wind power output time seriesmean) And standard deviation error (ε)std). Can be used forSo that the epsilon of the wind power output time sequence generated by the simulation of the methodmeanIndex and epsilonstdIndexes are respectively reduced by 76.8% and 53.1% compared with a continuous Markov chain, so that the wind power output simulation generation method for continuously improving MC by utilizing AP clustering-skipping has obvious superiority in simulation generation precision.
While the present invention has been described in detail and with reference to specific embodiments thereof, it will be apparent to one skilled in the art that various changes and modifications can be made therein without departing from the spirit and scope thereof as defined in the appended claims.

Claims (1)

1. A wind power output simulation generation method for continuously improving MC by utilizing AP clustering-skipping is characterized by comprising the following steps: the method comprises the following steps of carrying out AP clustering on historical wind power output data, forming a wind power output state skip sequence under a certain clustering category, forming a wind power output state time sequence with the time sequence length of one day under a certain clustering category, sampling and superposing the wind power output fluctuation quantity on a primary wind power output time sequence formed under a certain clustering category, and generating a wind power output time sequence according to inter-category transfer probability matrix simulation, wherein the specific content is as follows:
1) AP clustering is carried out on historical wind power output data
An Affinity Propagation (AP) clustering algorithm is an unsupervised clustering algorithm based on "information transfer"; firstly, establishing a similarity matrix S of a sample set as formula (1), wherein elements S in the matrix area,b(a ≠ b) denotes sample xaAnd xbThe similarity between the two is expressed by negative Euclidean distance and is expressed as formula (2); sa,aThe reference degree of the clustering center is used as a judgment standard for judging whether the clustering center can be formed, a deviation parameter is set at the beginning, the deviation parameter directly influences the number of final clusters, and the larger the value of the deviation coefficient is, the more clusters are generated; constructing a vector capable of reflecting the characteristics of the corresponding solar wind power output according to the daily mean value and the daily maximum difference of the actually measured historical wind power output at a set sampling interval in a unit time period, and carrying out AP aggregation on the historical wind power output by using the vector in a day unitClass (iv), wherein: the unit time period generally refers to month, season, year and years; the sampling interval is typically 1 minute, 5 minutes or 10 minutes;
Figure FDA0003133540760000011
Sa,b＝-||xa－xb||2 (2)
in the formula: s is a similarity matrix of the sample set; sa,b(a ≠ b) denotes sample xaAnd xbThe similarity between them; sa,aThe reference degree of the clustering center; x is the number ofaAnd xbAre data sample points; a is 1,2, … …, n; b is 1,2, … …, n; n is the total number of samples used for AP clustering; | xa-xb||2Denotes xaAnd xbThe Euclidean distance between;
taking the similarity matrix S of the sample set as input, and then calculating the attraction r (x) of each sample pointa,xk) Degree of affiliation a (x)a,xk)；r(xa,xk) Denotes xkWhether a point is suitable as xaThe cluster center of the points, i.e. xkFor xaThe degree of attraction of (c); a (x)a,xk) Denotes xaWhether a point selects x or notkAs its clustering center, i.e. xaFor xkDegree of attribution of; r (x)a,xk) And a (x)a,xk) The larger, indicates xkThe more suitable as a clustering center; continuously updating the attraction degree matrix and the attribution degree matrix in the clustering process, and introducing a damping coefficient for avoiding oscillation in the iteration process until stable clustering centers and class attribution conditions are generated in the iteration process; thus, near the end of propagation, xaIs determined as xk(ii) a Wherein x iskIs represented by the formula (3):
xk＝argmax{a(xa,xk)+r(xa,xk)} (3)
in the formula: argmax is the argument for finding the maximum function, i.e. xkTo satisfy { a (x) }a,xk)+r(xa,xk) Taking the value of the maximum value; a (x)a,xk) Is a sample point xaFor sample point xkDegree of attribution of; r (x)a,xk) Is a sample point xkFor sample point xaThe degree of attraction of (c); x is the number ofkAnd xaAre data sample points; k is 1,2, … …, n; a is 1,2, … …, n; n is the total number of samples used for AP clustering;
adopting a contour Coefficient (SC) as a clustering evaluation index; the larger the contour coefficient is, the better the clustering effect is; contour coefficient SC (x)k) Defined by formulae (4) to (6):
Figure FDA0003133540760000021
Figure FDA0003133540760000022
Figure FDA0003133540760000023
in the formula: x is the number ofk，xa，xbAre data sample points; c (x)k) Represents a sample point xkThe cluster category to which it belongs; c (x)a) Represents a sample point xaThe cluster category to which it belongs; c (x)b) Represents a sample point xbThe cluster category to which it belongs; contour coefficient SC (x)k) Has a value of [ -1,1 [)]Reflect xkAverage distance I (x) between cluster center and sample in class when serving as cluster centerk) Whether it is significantly different from its average distance O (x) to the out-of-class samplek)；I(xk) Representing the average distance between the cluster center and the samples in the class; d (x)k,xa) Represents a sample point xkAnd xaThe distance between them; m represents the sum of sample point xkThe number of data points belonging to the same category; o (x)k) Representing the center of the cluster andaverage distance of out-of-class samples; d (x)k,xb) Represents a sample point xkAnd xbThe distance between them; g represents the sum of sample point xkThe number of data points not belonging to the same category;
by carrying out AP clustering on historical wind power output data in units of days, the solar wind power output sequences with output levels similar to fluctuation characteristics can be classified into the same class;
2) forming a wind power output state skip sequence under a certain cluster category
The historical wind power output data is expressed as (P'min，P′max) The wind power output range is divided into N data segments uniformly, each data segment corresponds to one state, and the wind power output range corresponding to each state is (P'max-P′min) N; wherein, P'minIs the minimum value of historical wind power output, P'maxThe output is the maximum value of the historical wind power, and N is a state number; aiming at the situation that the wind power output sequence is easy to fall into a certain state and is difficult to jump in the traditional Markov chain, so that the generated wind power output time sequence possibly has overlong duration in a certain state, the method adopts a state jump matrix to replace a state transition matrix for improvement;
when a state transition matrix is generated by a clustering reconstruction sequence, when the jump amplitude of the state number between two adjacent data is larger than 1/3 of the divided state number, the transition is considered to be not practical, the value of the state transition matrix is set to zero, and an improved state transition matrix is obtained;
secondly, setting all diagonal elements of the obtained state transition matrix to zero, calculating the proportion of each element in the sum of all elements of the row, and taking the proportion as a new probability value, wherein the obtained probability matrix is a state jump matrix; the state hopping matrix is of equations (7) - (8);
Figure FDA0003133540760000031
pij＝P(Et+1＝j|Et＝i) (8)
in the formula: the matrix P is a state jump matrix; the row of the matrix P corresponds to the current output state of the wind power output, and the column corresponds to the output state at the next moment; n is the number of states; i and j are wind power output states; 1,2, … …, N; j ═ 1,2, … …, N; p is a radical ofijRepresenting the probability of the wind power output transferring from the state i to the state j; et、Et+1The wind power output values at the t moment and the t +1 moment are in corresponding states; p (a | B) is a conditional probability function representing the probability of a occurring under the condition of B;
the corresponding accumulated state transition matrix is of formula (9):
Figure FDA0003133540760000032
the value of the element in Q is represented by formula (10):
Figure FDA0003133540760000033
in the formula: the matrix Q is an accumulated state jump matrix; q. q.sulJumping to the element of the u row and l column in the matrix Q for the accumulated state; p is a radical ofujThe elements of the u-th row and the j-th column in the state jump matrix P; n is the number of states; u-1, 2, … …, N; 1,2, … …, N; j ═ 1,2, … …, N;
through the forming process of the accumulative state jump matrix Q, an accumulative state jump matrix under a certain cluster category, namely an intra-category accumulative state jump matrix, is obtained; then, sampling to form a state jump sequence under each cluster category according to the intra-category accumulated state jump matrix; under the cluster category, the state of the wind power output at the current moment is EtThe state at the next time is Et+1Generating random number xi following uniform distribution, if 0<ξ≤qEt,1Then E ist+11 is ═ 1; if q isEt,l<ξ≤qEt,l+1Then E ist+1L +1, wherein qEt,1E < th > representing intra-class accumulated state transition matrix QtRow, column 1 elements; q. q.sEt,lE < th > representing intra-class accumulated state transition matrix QtRow, column I elements; q. q.sEt,l+1E < th > representing intra-class accumulated state transition matrix QtRow, column l + 1;
since the diagonal elements of the state-hopping matrix are zero, the diagonal elements in the corresponding intra-class accumulated state-hopping matrix are equal to the left elements, i.e., qi,i-1Is equal to qiiSo [ q ]i,i-1,qii]For an empty set, when a wind power output state sequence is generated by using a Markov chain method, xi does not fall to [ q ]i,i-1,qii]Within the range; wherein q isiiThe elements of the ith row and the ith column in the intra-class accumulated state jump matrix Q are represented by i, which is 1,2, … …, N; q. q.si,i-1Jumping to the ith row and the (i-1) th column elements in the matrix Q for the accumulated state;
through the steps, a state skip sequence with different adjacent states when the wind power output is in a certain clustering category can be formed by sampling according to the intra-class accumulated state skip matrix in the clustering category;
3) forming a wind power output state time sequence with the time sequence length of one day under a certain cluster category
According to the method, the historical wind power output time sequence is traversed by a statistical method of recording the duration time T once from the first moment when the wind power output enters a certain state, and if the duration time is T, the times of occurrence of each duration time of the wind power output in the state are counted to obtain the distribution condition of the duration time in the state;
after a distribution histogram of the duration time of a certain state of the wind power output is obtained through statistics and drawing, a smooth curve is adopted to describe the histogram, so that the particularity existing in sampling the histogram can be avoided; when describing data distribution characteristics, Kernel Density Estimation (KDE) does not depend on selection of a parameter Estimation model, and dependence of histogram Estimation on a histogram group distance and a histogram position can be effectively avoided, so that the KDE is adopted to describe wind power output state duration distribution characteristics; the KDE expression is formula (11):
Figure FDA0003133540760000041
in the formula:
Figure FDA0003133540760000042
a kernel density estimation function for the duration of the wind power output state; k (-) is a kernel function; h is the bandwidth; x is the number of1,x2,……,xWThe data number is W; x is the number ofwThe W-th sample value of the duration time of the wind power output state is 1,2, … … and W; w represents that W different durations of the wind power output sequence appear in a certain state;
in KDE, the selection of the bandwidth determines the smoothness degree of a fitting curve, the larger the bandwidth is, the smoother the bandwidth is, but the poorer the fitting effect is; selecting a Gaussian function as a kernel function, and solving the bandwidth by adopting an empirical rule, wherein the formula is as follows (12):
h＝1.06σW-1/5 (12)
in the formula: h is the bandwidth; sigma is the normal distribution standard deviation of the duration time of the wind power output state; w represents that W different durations of the wind power output sequence appear in a certain state;
in order to obtain the state time sequence of the wind power output with the time sequence length of one day under a certain clustering category, after a random number set meeting the duration time of each state of a kernel density estimation function is generated, traversing the obtained wind power output state jump sequence under the clustering category, and performing repeated sampling: the current state is EtThen in state EtIs arbitrarily chosen as a value in the set of duration random numbers of (a) as state EtThe time duration is repeated until the length of the generated time sequence meets the requirement;
4) sampling and superposing wind power output fluctuation quantity on a preliminary wind power output time sequence formed under a certain cluster category
After a wind power output state time sequence under a certain cluster category is formed, determining the primary wind at each momentElectrical output; the current state of wind power output is EtThen, the preliminary wind power output P 'at the moment can be randomly generated't·cbNumerical value of (1), P't·cb∈(PEt.min，PEt.max)，PEt.min、PEt.maxAre respectively in state EtThe minimum value and the maximum value of the corresponding wind power output range;
the wind power output fluctuation is that the output values at adjacent moments are different, and the fluctuation characteristic is described by adopting a first-order difference fluctuation quantity, namely the output difference value in adjacent 2 unit times; the first-order difference fluctuation quantity expression of the wind power output is as shown in formula (13):
ΔP′t＝P′t+1-P′t (13)
in the formula: t represents the wind power output moment; delta P'tRepresenting the fluctuation quantity of the wind power output at the time t; p'tAnd P't+1Respectively representing wind power output values at the time t and the time t + 1;
the probability density distribution fitting function of the first-order difference fluctuation quantity of the wind power output is as follows: normal distribution, t-location-scale distribution and logistic distribution; the wind power output fluctuation characteristic is characterized by mixed Gaussian distribution with better fitting precision, the probability density function is the weighting of a plurality of Gaussian probability density functions, when the data distribution is complex, the mixed Gaussian distribution function can solve the problem of the insufficient fitting precision of a single Gaussian distribution function, and the mathematical model expression is as the following formulas (14) - (15):
Figure FDA0003133540760000051
Figure FDA0003133540760000052
in the formula: (x) is a Gaussian mixture distribution function; v ═ 1,2, … …, V; v is the number of single Gaussian distributions in the mixed Gaussian distribution; a isvIs the mixing coefficient; p is a radical ofgs(x|bv，cv) Probability of being the v-th single Gaussian distributionA density function, wherein x represents a random number obeying a single gaussian distribution; bvAnd cvRespectively represent the mean and standard deviation of the v-th single gaussian distribution;
solving the distribution parameters corresponding to the Gaussian mixture distribution function by adopting an expectation-maximization (EM) algorithm, wherein the EM algorithm is divided into two steps: step 1, called the expectation (E), calculates the expectation of the likelihood function based on the parameters of the initial values or the previous iteration values; step 2 is a maximization (M) step, which maximizes the likelihood function and converts it into new parameter values that can be obtained, and the EM algorithm is a process of repeated iteration until the above two steps converge;
after a fluctuation amount random number set meeting the mixed Gaussian distribution function is generated, randomly extracting the fluctuation amount, and superposing the fluctuation amount on the formed initial wind power output value, as shown in a formula (16); traversing the generated preliminary wind power output time sequence, and completing superposition of fluctuation components to form a wind power output time sequence with the time sequence length of one day under a certain cluster type;
P′t·fd＝P′t·cb+ΔP′t (16)
in the formula: p't·fdThe wind power output value at the moment t of the wind power output time sequence under a certain clustering category is obtained; p't·cbThe preliminary wind power output value at the moment t of the wind power output time sequence under a certain clustering category is obtained; delta P'tRepresenting the fluctuation quantity of the wind power output at the time t;
5) generating wind power output time sequence according to inter-class transition probability matrix simulation
Determining the time length of the wind power output time sequence generated by simulation according to the requirement of medium-long term optimization scheduling of the power system containing wind power generation, wherein the time length required by the medium-long term optimization scheduling is generally 2 days or more;
respectively counting the transfer times between the classes of the wind power output on two adjacent days, and calculating the corresponding transfer probability to obtain inter-class transfer probability matrixes, such as formulas (17) to (18), so as to obtain accumulated inter-class transfer probability matrixes, such as formulas (19) to (20);
Figure FDA0003133540760000061
plj·rz＝P(He+1＝z|He＝r) (18)
in the formula: matrix PljIs an inter-class transition probability matrix; matrix PljThe row of the corresponding cluster type of the wind power output of the current day, and the column of the corresponding cluster type of the wind power output of the next day; l is the number of categories generated by AP clustering; r and z are clustering categories to which the wind power output belongs; r is 1,2, … …, L; z is 1,2, … …, L; p is a radical oflj·rzRepresenting the probability that the cluster category to which the wind power output belongs is transferred from the category r to the category z; he、He+1The cluster type of the wind power output on the current day and the next day; p (a | B) is a conditional probability function representing the probability of a occurring under the condition of B;
Figure FDA0003133540760000062
Figure FDA0003133540760000063
in the formula: matrix QljIs an accumulative inter-class transition probability matrix; q. q.slj·rgFor accumulating inter-class transition probability matrix QljThe elements of the r-th row and the g-th column; p is a radical oflj·rzAs an inter-class transition probability matrix PljThe elements of the r-th row and the z-th column; l is the number of categories generated by AP clustering; r is 1,2, … …, L; z is 1,2, … …, L; g is 1,2, … …, L;
and sampling the cluster type to which the wind power output of the current day belongs according to the accumulated inter-class transition probability matrix to obtain the AP cluster type to which the wind power output of the next day belongs: the cluster type of the wind power output on the same day is HeThe cluster category to which the next-day wind power output belongs is He+1Generating random number epsilon subject to uniform distribution if 0<ε≤qHe,1Then H ise+11 is ═ 1; if q isHe,g<ε≤qHe,g+1Then H ise+1G +1, wherein qHe,1Representing transition probability matrix Q between cumulative classesljH ofeRow, column 1 elements; q. q.sHe,gRepresenting transition probability matrix Q between cumulative classesljH ofeRow, column g elements; q. q.sHe,g+1Representing transition probability matrix Q between cumulative classesljH ofeRow, column g +1 elements;
the method comprises the following steps of generating a wind power output time sequence within R days in a simulated mode at a set sampling interval, wherein R is a positive integer:
judging the AP clustering category to which the wind power output of the 1 st day belongs;
secondly, forming a wind power output simulation generation time sequence of the day 1 by executing the steps 2-4), and recording the time sequence as omega1，Ω1＝{P′1·fd(1),P′2·fd(1),……,P′t·fd(1)In which Ω is1Representing a wind power output simulation generation sequence of the 1 st day; wherein, P't·fd(1)The simulation generation wind power output value of the tth moment of the 1 st day is shown, and t shows the wind power output moment;
thirdly, calculating to obtain a wind power output accumulated inter-class transfer probability matrix in the step 5), and sampling to determine the AP cluster class to which the wind power output belongs on the day tau, wherein tau is 2,3, … … and R;
fourthly, repeating the steps 2) to 4) to obtain a wind power output simulation generation time sequence on the tau day, and recording the time sequence as omegaτ，Ωτ＝{P′1·fd(τ),P′2·fd(τ),……,P′t·fd(τ)In which Ω isτRepresenting a wind power output simulation generation sequence on the tau day; p't·fd(τ)Wherein, the simulation generation wind power output value at the tth moment on the Tth day is shown;
fifthly, repeatedly executing the third step and the fourth step until a wind power output time sequence of R days in total is generated and is recorded as omega, and omega is { omega }1,……,Ωτ,……,ΩR}。
CN202110710544.7A 2021-06-25 2021-06-25 Wind power output simulation generation method for continuously improving MC (multi-channel) by utilizing AP (access point) clustering-skipping Active CN113449471B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110710544.7A CN113449471B (en) 2021-06-25 2021-06-25 Wind power output simulation generation method for continuously improving MC (multi-channel) by utilizing AP (access point) clustering-skipping

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110710544.7A CN113449471B (en) 2021-06-25 2021-06-25 Wind power output simulation generation method for continuously improving MC (multi-channel) by utilizing AP (access point) clustering-skipping

Publications (2)

Publication Number Publication Date
CN113449471A true CN113449471A (en) 2021-09-28
CN113449471B CN113449471B (en) 2022-08-30

Family

ID=77812778

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110710544.7A Active CN113449471B (en) 2021-06-25 2021-06-25 Wind power output simulation generation method for continuously improving MC (multi-channel) by utilizing AP (access point) clustering-skipping

Country Status (1)

Country Link
CN (1) CN113449471B (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117150786A (en) * 2023-09-06 2023-12-01 北京航空航天大学 Method and system for representing randomness of rotational speed spectrum rotational speed state duration
CN117150786B (en) * 2023-09-06 2024-04-30 北京航空航天大学 Method and system for representing randomness of rotational speed spectrum rotational speed state duration

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20150186535A1 (en) * 2013-12-27 2015-07-02 Quixey, Inc. Determining an Active Persona of a User Device
CN107103411A (en) * 2017-04-08 2017-08-29 东北电力大学 Based on the markovian simulation wind power time series generation method of improvement
CN109344875A (en) * 2018-08-31 2019-02-15 中国南方电网有限责任公司电网技术研究中心 Based on clustering day wind power output timing generation method, device and medium
CN109389143A (en) * 2018-06-19 2019-02-26 北京九章云极科技有限公司 A kind of Data Analysis Services system and method for automatic modeling
CN109524993A (en) * 2018-12-19 2019-03-26 中国农业大学 The typical week power output scene generating method of wind-powered electricity generation photovoltaic for Mid-long Term Optimized Scheduling
CN109783841A (en) * 2018-11-28 2019-05-21 河海大学 A kind of photovoltaic power output time series analogy method based on more scene state transfer matrixes and conditional probability sampling
CN110222361A (en) * 2019-04-23 2019-09-10 中国电力科学研究院有限公司 A kind of Power Output for Wind Power Field analogy method and device

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20150186535A1 (en) * 2013-12-27 2015-07-02 Quixey, Inc. Determining an Active Persona of a User Device
CN107103411A (en) * 2017-04-08 2017-08-29 东北电力大学 Based on the markovian simulation wind power time series generation method of improvement
CN109389143A (en) * 2018-06-19 2019-02-26 北京九章云极科技有限公司 A kind of Data Analysis Services system and method for automatic modeling
CN109344875A (en) * 2018-08-31 2019-02-15 中国南方电网有限责任公司电网技术研究中心 Based on clustering day wind power output timing generation method, device and medium
CN109783841A (en) * 2018-11-28 2019-05-21 河海大学 A kind of photovoltaic power output time series analogy method based on more scene state transfer matrixes and conditional probability sampling
CN109524993A (en) * 2018-12-19 2019-03-26 中国农业大学 The typical week power output scene generating method of wind-powered electricity generation photovoltaic for Mid-long Term Optimized Scheduling
CN110222361A (en) * 2019-04-23 2019-09-10 中国电力科学研究院有限公司 A kind of Power Output for Wind Power Field analogy method and device

Non-Patent Citations (6)

* Cited by examiner, † Cited by third party
Title
SUN M 等: "Conditional aggregated probabilistic wind power forecasting based on spatio-temporal correlation", 《APPLIED ENERGY》 *
XIAO B 等: "An Optimization and Planning Method Research to Determine Power Transmission Capacity of Clustering wind Farms Accounting Risk", 《APPLIED MECHANICS AND MATERIALS. TRANS TECH PUBLICATIONS LTD》 *
丁明等: "应用改进马尔科夫链的光伏出力时间序列模拟", 《电网技术》 *
叶林 等: "基于近邻传播聚类与MCMC算法的风电时序数据聚合方法", 《中国电机工程学报》 *
管霖 等: "多风电场功率时间序列的时空相关性统计建模和运行模拟方法", 《电网技术》 *
肖白 等: "基于Markov链和Copula理论的风光联合输出功率时间序列模拟生成方法", 《现代电力》 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117150786A (en) * 2023-09-06 2023-12-01 北京航空航天大学 Method and system for representing randomness of rotational speed spectrum rotational speed state duration
CN117150786B (en) * 2023-09-06 2024-04-30 北京航空航天大学 Method and system for representing randomness of rotational speed spectrum rotational speed state duration

Also Published As

Publication number Publication date
CN113449471B (en) 2022-08-30

Similar Documents

Publication Publication Date Title
CN110110930B (en) Recurrent neural network short-term power load prediction method for improving whale algorithm
CN111353582B (en) Particle swarm algorithm-based distributed deep learning parameter updating method
CN114488140B (en) Small sample radar one-dimensional image target recognition method based on deep migration learning
CN109492748B (en) Method for establishing medium-and-long-term load prediction model of power system based on convolutional neural network
CN111401755B (en) Multi-new-energy output scene generation method, device and system based on Markov chain
CN111736084B (en) Valve-regulated lead-acid storage battery health state prediction method based on improved LSTM neural network
CN110543929A (en) wind speed interval prediction method and system based on Lorenz system
CN113344288B (en) Cascade hydropower station group water level prediction method and device and computer readable storage medium
CN114861533A (en) Wind power ultra-short-term prediction method based on time convolution network
CN109583588B (en) Short-term wind speed prediction method and system
CN112149883A (en) Photovoltaic power prediction method based on FWA-BP neural network
CN113191918A (en) Lunar wind-solar power generation power scene analysis method based on time sequence generation countermeasure network
CN110880062A (en) Method for determining state maintenance time of power distribution equipment
CN114298377A (en) Photovoltaic power generation prediction method based on improved extreme learning machine
CN115115090A (en) Wind power short-term prediction method based on improved LSTM-CNN
CN113657678A (en) Power grid power data prediction method based on information freshness
CN113095477B (en) Wind power prediction method based on DE-BP neural network
CN113449471B (en) Wind power output simulation generation method for continuously improving MC (multi-channel) by utilizing AP (access point) clustering-skipping
CN115392387B (en) Low-voltage distributed photovoltaic power generation output prediction method
CN110276478B (en) Short-term wind power prediction method based on segmented ant colony algorithm optimization SVM
CN116167456A (en) Federal learning method based on code comparison and classification correction
CN115936062A (en) Photovoltaic power day-ahead interval prediction method based on scene generation and deep learning
CN114897260A (en) Short-term wind speed prediction model modeling method and prediction method based on LSTM neural network
CN115983095A (en) Photovoltaic power generation prediction method based on clustering algorithm, neural network and genetic algorithm
CN111506868B (en) Ultra-short-term wind speed prediction method based on HHT weight optimization

Legal Events

Date Code Title Description
PB01 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