CN110909911B - Aggregation method of multidimensional time series data considering space-time correlation - Google Patents

Aggregation method of multidimensional time series data considering space-time correlation Download PDF

Info

Publication number
CN110909911B
CN110909911B CN201910932946.4A CN201910932946A CN110909911B CN 110909911 B CN110909911 B CN 110909911B CN 201910932946 A CN201910932946 A CN 201910932946A CN 110909911 B CN110909911 B CN 110909911B
Authority
CN
China
Prior art keywords
photovoltaic
load
scene
power
wind power
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201910932946.4A
Other languages
Chinese (zh)
Other versions
CN110909911A (en
Inventor
叶林
李镓辰
李延和
马明顺
李湃
黄越辉
张舒捷
肖明
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.)
China Agricultural University
China Electric Power Research Institute Co Ltd CEPRI
State Grid Qinghai Electric Power Co Ltd
Electric Power Research Institute of State Grid Qinghai Electric Power Co Ltd
Original Assignee
China Agricultural University
China Electric Power Research Institute Co Ltd CEPRI
State Grid Qinghai Electric Power Co Ltd
Electric Power Research Institute of 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 China Agricultural University, China Electric Power Research Institute Co Ltd CEPRI, State Grid Qinghai Electric Power Co Ltd, Electric Power Research Institute of State Grid Qinghai Electric Power Co Ltd filed Critical China Agricultural University
Priority to CN201910932946.4A priority Critical patent/CN110909911B/en
Publication of CN110909911A publication Critical patent/CN110909911A/en
Application granted granted Critical
Publication of CN110909911B publication Critical patent/CN110909911B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • 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
    • 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

The invention discloses a method for aggregating multidimensional time series data considering space-time correlation. Aiming at the problem of multi-energy source season-crossing complementary year/month optimized scheduling in an electric power system containing wind power and photovoltaic power generation, the invention provides a Markov decision method for optimizing optimal action strategies of wind power, photovoltaic and loads under different initial state combinations. So as to achieve that the day scene combination selected under the initial state combination condition is closest to the original time sequence data in a numerical probability distribution mode. And then, a Markov Monte Carlo method is adopted to generate a 3 XN Markov state matrix with correlation in a sampling mode, an optimal strategy obtained by Markov decision is matched with a state column vector combination in the Markov state matrix to obtain a wind power, photovoltaic and load aggregation sequence with space-time correlation, and a typical power scene conforming to wind, light and load in a target area is obtained, so that guidance is provided for a year/month electric quantity plan in system optimization scheduling.

Description

Aggregation method of multidimensional time series data considering space-time correlation
Technical Field
The invention relates to the field of multi-energy power system complementary optimization scheduling, in particular to a method for aggregating multi-dimensional time sequence data considering space-time correlation.
Background
With the increasing exhaustion of non-renewable resources such as coal, petroleum and the like and the increasing severity of energy troubles, renewable energy sources such as wind energy, solar energy, tidal energy, biomass energy and the like are receiving more and more attention worldwide. The utilization of wind and light natural resources is two renewable energy sources with the most mature technology and the most development value in the renewable energy power generation technology. The development of wind power and photoelectricity has very important significance for guaranteeing energy safety, adjusting energy structures, reducing environmental pollution, realizing sustainable development and the like.
The renewable natural wind and light energy has high uncertainty, the characteristics determine that the power of wind power and photoelectric power has strong fluctuation, and with the large-scale wind power and photoelectric power connected into a power grid, the power fluctuation brings great challenges to the safe and economic operation of the power grid. By analyzing the wide-area time-space correlation of wind, light and other uncertain power generation, researching and considering a multi-energy season-crossing complementary year/month optimized scheduling method, fully utilizing the season-crossing complementary action among renewable energy resources, and analyzing the typical output scene of wind, light and a load electric field in a target area, the reasonable and scientific formulation of a year/month electric quantity plan is realized, and the consumption capacity of a power grid to renewable energy is effectively improved.
At present, the research on optimizing scheduling at home and abroad is more and more extensive and deep. In the process of planning and researching the annual/monthly electricity quantity of wind power/photovoltaic, the problems that the output time sequence data of wind power, photovoltaic and load is too huge and the data is redundant are found, so that the time sections are more, the processing time is too long, the problems that the time cannot be quickly solved and the timeliness of annual/monthly optimized scheduling is difficult to meet when annual/monthly optimized calculation is carried out, and the redundancy of the data is not beneficial to constructing a typical output scene of the wind power, the photovoltaic and the load to a great extent. The most basic method at present is to form a new output sequence by sampling data points of an original output time sequence at equal intervals; in addition, a segmentation aggregation approximation method based on the information entropy is adopted to calculate the distribution of the information entropy of the original output time sequence and carry out segmentation aggregation approximation, so that a new output sequence is formed, a principal component analysis method, a discrete Fourier transform method and the like are also adopted, and a typical scene is constructed by using a k-means clustering algorithm or a hierarchical clustering algorithm and the like on the basis. However, the above polymerization method does not reflect the tendency and fluctuation of the original sequence well. And the traditional k-means clustering algorithm has strong sensitivity to initial class center selection, has poor stability of multi-time clustering, and can not give the optimal classification number. The hierarchical clustering has the defect of large calculated amount, and in addition, because the hierarchical clustering uses a greedy algorithm, the obtained local optimization is obvious, and not necessarily, the global optimization is obtained. Therefore, the typical scenes obtained by the above methods are all insufficient.
Disclosure of Invention
In view of the defects in the prior art, the present invention aims to provide a method for aggregating multidimensional time series data considering spatio-temporal correlation. The method adopts a Markov decision method to optimize the optimal action strategy of wind power, photovoltaic and load under different initial state combinations. So as to achieve that the selected day scene combination under the initial state combination condition is closest to the original time sequence data in a numerical probability distribution mode. And then, a Markov Monte Carlo method is adopted to generate a 3 XN Markov state matrix with correlation in a sampling mode, an optimal action strategy obtained by Markov decision is matched with a state column vector combination in the Markov state matrix, and a wind power, photovoltaic and load aggregation sequence with space-time correlation is obtained.
In order to achieve the above purposes, the technical scheme adopted by the invention is as follows:
a method for aggregating multi-dimensional time series data considering spatio-temporal correlation, comprising the steps of:
s1, respectively acquiring wind power, photovoltaic and load medium and long term historical power sequences with the length of m at the same time, and carrying out extremum normalization processing on the sequences to obtain a wind power medium and long term power time sequence, a photovoltaic medium and long term power time sequence and a load medium and long term power time sequence with the length of m;
the wind power medium-long term power time sequence with the length of m is represented as follows:
Figure BDA0002220746900000031
in sequence
Figure BDA0002220746900000032
All represent a historical value of wind power;
the photovoltaic medium and long term power time series of length m is represented as:
Figure BDA0002220746900000033
in sequence
Figure BDA0002220746900000034
Each represents a photovoltaic power historical value;
the load medium and long term power time series with length m is represented as:
Figure BDA0002220746900000035
in sequence
Figure BDA0002220746900000036
Each represents a load power history value;
then, the daily scene segmentation is respectively carried out on the medium-long term power time sequence of the wind power, the photovoltaic power and the load to obtain the same quantity:
wind power day scene
Figure BDA0002220746900000037
Wherein
Figure BDA0002220746900000038
Figure BDA0002220746900000039
Representing the k-th wind power day scene after segmentation, k is more than or equal to 1 and less than or equal to dayn,
photovoltaic day scene
Figure BDA00022207469000000310
Wherein
Figure BDA00022207469000000311
Figure BDA00022207469000000312
Representing the kth photovoltaic day scene after division, k is more than or equal to 1 and less than or equal to dayn,
load day scene
Figure BDA00022207469000000313
Wherein
Figure BDA00022207469000000314
Figure BDA00022207469000000315
Representing a k-th load day scene after segmentation, wherein k is more than or equal to 1 and less than or equal to dayn, dayn is determined by the total days contained in the medium-long term historical power sequence, and the total days contained in the medium-long term historical power sequence of wind power, photovoltaic and load are the same;
wind power day scene SCwindPhotovoltaic day scene SCpvLoad day scene SCloadThe day scene elements contained in the method are respectively distributed to the wind power day scene set through a neighbor propagation clustering algorithm
Figure BDA0002220746900000041
Photovoltaic day scene set
Figure BDA0002220746900000042
Load day scene set
Figure BDA0002220746900000043
In the method, the optimal clustering numbers of the daily scenes of wind power, photovoltaic and load are obtained through Davies-Bouldin Index analysis
Figure BDA0002220746900000044
Figure BDA0002220746900000045
And simultaneously obtaining clustering results under respective optimal clustering numbers:
Figure BDA0002220746900000046
wherein the content of the first and second substances,
Figure BDA0002220746900000047
Figure BDA0002220746900000048
representswiWind-like power day scene set composed of nwiThe solar energy power generation system is composed of a plurality of wind power day scenes,
Figure BDA0002220746900000049
wherein
Figure BDA00022207469000000410
Figure BDA00022207469000000411
Represents the firstpClass i photovoltaic day scene set, from npiThe solar photovoltaic solar scene is formed by a plurality of photovoltaic solar scenes,
Figure BDA00022207469000000412
wherein
Figure BDA00022207469000000413
Figure BDA00022207469000000414
Represents the firstliClass load daily scene set, from nliThe solar photovoltaic solar scene is formed by a plurality of photovoltaic solar scenes,
wherein n iswi、npiAnd nliA value obtained by a neighbor propagation clustering algorithm through calculation;
s2, on the basis of the step S1, clustering results of the wind power daily scene, the photovoltaic daily scene and the load daily scene obtained by adopting a neighbor propagation clustering algorithm are used as state quantities to obtain a random state model of the wind power, the photovoltaic and the load, the random state model is represented by a Markov Monte Carlo process, and the high-order state transition probability of the Markov Monte Carlo process is represented as follows:
Figure BDA00022207469000000415
Figure BDA00022207469000000416
Figure BDA0002220746900000051
where N represents the nth column state in the Markov polymerization state matrix and N is [1, N ],
Figure BDA0002220746900000052
Figure BDA0002220746900000053
Figure BDA0002220746900000054
obtaining a state transition probability matrix of the Markov Monte Carlo process by counting historical data:
Figure BDA0002220746900000055
Figure BDA0002220746900000056
Figure BDA0002220746900000057
s3, Markov decision process is composed of five parts (S, A, { PSA},γ,R)，
S is the result of clustering
Figure BDA0002220746900000058
Figure BDA0002220746900000059
Set of constituent states, S ═<Swind,Spv,Sload>；
The state set for step n is represented as:
Figure BDA0002220746900000061
a represents action set, and is composed of daily scene elements contained in each daily scene set in clustering result
Figure BDA0002220746900000062
Wherein:
Figure BDA0002220746900000063
Figure BDA0002220746900000064
Figure BDA0002220746900000065
Figure BDA0002220746900000066
Figure BDA0002220746900000067
Figure BDA0002220746900000068
the set of actions for step n may be represented as:
Figure BDA0002220746900000069
{PSAdenotes the state transition probability, expressed as:
Figure BDA00022207469000000610
γ ∈ [0, 1)) represents a damping coefficient representing a discount of the rate of return over time, and γ ∈ [ 0.5 ] can be generally selected according to empirical values;
r represents the error function:
Figure BDA00022207469000000611
Figure BDA00022207469000000612
wherein the sort function indicates that the time series data are arranged from large to small in numerical value,
Figure BDA00022207469000000613
representing wind-solar scenes
Figure BDA00022207469000000614
After the multiplication (dayn) is carried out, the time sequence data are arranged from large to small according to the numerical value, so that the length is consistent with the length of the original sequence, the subsequent solution of Euclidean distance is convenient,
Figure BDA00022207469000000615
representing a photovoltaic daily scene
Figure BDA0002220746900000071
After the expansion (dayn) times, the time sequence data are arranged from large to small according to the numerical value, so that the length is consistent with the length of the original sequence,
Figure BDA0002220746900000072
indicating loading daily scenes
Figure BDA0002220746900000073
After expanding (dayn) times, arranging the time sequence data from large to small according to the numerical value, and enabling the length to be consistent with the length of the original sequence;
s4, defining a value function
Figure BDA0002220746900000074
S0Is an artificially set initial state, the initial state S0Is in an amount of
Figure BDA0002220746900000075
Defining an optimum function
Figure BDA0002220746900000076
Solving using Bellman equations
Figure BDA0002220746900000077
Obtaining an optimal strategy
Figure BDA0002220746900000078
S5, setting all initial states S0Respectively substitute for
Figure BDA0002220746900000079
In the equation, | S | equations are obtained in total, thereby obtaining S in different initial states0Day scene combination strategy pi of optimal wind power, photovoltaic and load*(S0)；
S6, generating a Markov aggregation state matrix of 3 XN order consisting of wind power, photovoltaic and load through Gibbs sampling by using a Markov Monte Carlo method:
Figure BDA00022207469000000710
wherein
Figure BDA00022207469000000711
S7, matching each group of state vectors in the Markov state matrix with the mapping of the optimal strategy to obtain a day scene matrix
Figure BDA0002220746900000081
Wherein
Figure BDA0002220746900000082
Figure BDA0002220746900000083
Figure BDA0002220746900000084
S8, judging whether the difference value of the head-tail connection position between the adjacent day scenes of each row of the day scene matrix is larger than the first-order difference maximum value of the medium-long term power time sequence, if so, performing wavelet filtering processing on the head-tail connection position until the difference value is smaller than the first-order difference maximum value, if not, directly connecting the head and the tail of the day scenes, and finally generating three aggregation sequences formed by N day scenes.
The method of the invention considers the problems of excessive time sections and excessive data volume when year/month electric quantity optimization calculation is carried out in an optimization scheduling level, and considers the problems that wind, light and load in the same area have great similarity to a certain extent due to the influence of natural factors such as terrain, latitude and the like, and a representative output curve needs to be extracted. The method overcomes the defect of overlarge calculated amount when annual/monthly electric quantity optimization calculation is carried out on an optimization scheduling level, and the problem of low calculation efficiency of a time sequence simulation method in the existing wind, light and load scene analysis method, and the problem of over conservative calculation results due to the fact that the annual, monthly and daily output characteristics of wind, light and load cannot be reflected in a typical daily method. The method gives consideration to the calculation efficiency and the data change characteristic, provides effective guidance for year/month optimal scheduling of the new energy power generation system, and promotes the consumption of wind and light uncertain power supplies.
Drawings
The invention has the following drawings:
FIG. 1 is a flow chart of a method for aggregating multidimensional time series data considering spatio-temporal correlation according to the present invention.
Detailed Description
The present invention is described in further detail below with reference to fig. 1.
A method for aggregating multi-dimensional time series data considering spatio-temporal correlation, comprising the steps of:
s1, respectively acquiring wind power, photovoltaic and load medium and long term historical power sequences with the length of m at the same time, and carrying out extremum normalization processing on the sequences to obtain a wind power medium and long term power time sequence, a photovoltaic medium and long term power time sequence and a load medium and long term power time sequence with the length of m;
the wind power medium-long term power time sequence with the length of m is represented as follows:
Figure BDA0002220746900000091
in sequence
Figure BDA0002220746900000092
All represent a historical value of wind power;
the photovoltaic medium and long term power time series of length m is represented as:
Figure BDA0002220746900000093
in sequence
Figure BDA0002220746900000094
Each represents a photovoltaic power historical value;
the load medium and long term power time series with length m is represented as:
Figure BDA0002220746900000095
in sequence
Figure BDA0002220746900000096
Each represents a load power history value;
then, the normalized medium-long term power time series of the wind power, the photovoltaic and the load are respectively subjected to daily scene segmentation to obtain the same quantity:
wind power day scene
Figure BDA0002220746900000097
Wherein
Figure BDA0002220746900000098
Figure BDA0002220746900000099
Representing the k-th wind power day scene after segmentation, k is more than or equal to 1 and less than or equal to dayn,
photovoltaic day scene
Figure BDA0002220746900000101
Wherein
Figure BDA0002220746900000102
Figure BDA0002220746900000103
Representing the kth photovoltaic day scene after division, k is more than or equal to 1 and less than or equal to dayn,
load day scene
Figure BDA0002220746900000104
Wherein
Figure BDA0002220746900000105
Figure BDA0002220746900000106
Respectively representing the k-th load day scene after segmentation, wherein k is more than or equal to 1 and less than or equal to dayn, the dayn is determined by the total days contained in the medium-long term historical power sequence, and the total days contained in the medium-long term historical power sequence of the wind power, the photovoltaic and the load are the same;
wind power day scene SCwindPhotovoltaic day scene SCpvLoad day scene SCloadThe day scene elements contained in the method are respectively distributed to the wind power day scene set through a neighbor propagation clustering algorithm
Figure BDA0002220746900000107
Photovoltaic day scene set
Figure BDA0002220746900000108
Load day scene set
Figure BDA0002220746900000109
In the method, the optimal clustering numbers of the daily scenes of wind power, photovoltaic and load are obtained through Davies-Bouldin Index analysis
Figure BDA00022207469000001010
Figure BDA00022207469000001011
And simultaneously obtaining clustering results under respective optimal clustering numbers:
Figure BDA00022207469000001012
wherein the content of the first and second substances,
Figure BDA00022207469000001013
Figure BDA00022207469000001014
representswiWind-like power day scene set composed of nwiThe solar energy power generation system is composed of a plurality of wind power day scenes,
Figure BDA00022207469000001015
wherein
Figure BDA00022207469000001016
Figure BDA00022207469000001017
Represents the firstpClass i photovoltaic day scene set, from npiThe solar photovoltaic solar scene is formed by a plurality of photovoltaic solar scenes,
Figure BDA00022207469000001018
wherein
Figure BDA00022207469000001019
Figure BDA00022207469000001020
Represents the firstliClass load daily scene set, by nliThe solar photovoltaic solar scene is formed by a plurality of solar photovoltaic solar scenes,
wherein n iswi、npiAnd nliIs a value obtained by calculation by a neighbor propagation clustering algorithm;
s2, on the basis of the step S1, clustering results of the wind power daily scene, the photovoltaic daily scene and the load daily scene obtained by adopting a neighbor propagation clustering algorithm are used as state quantities to obtain a random state model of the wind power, the photovoltaic and the load, the random state model is represented by a Markov Monte Carlo process, and the high-order state transition probability of the Markov Monte Carlo process is represented as follows:
Figure BDA0002220746900000111
Figure BDA0002220746900000112
Figure BDA0002220746900000113
where N represents the nth column state in the Markov polymerization state matrix and N is [1, N ],
Figure BDA0002220746900000114
Figure BDA0002220746900000115
Figure BDA0002220746900000116
obtaining a state transition probability matrix of the Markov Monte Carlo process by counting historical data:
Figure BDA0002220746900000117
Figure BDA0002220746900000118
Figure BDA0002220746900000121
s3, Markov decision process is composed of five parts (S, A, { PSA},γ,R)，
S is the result of clustering
Figure BDA0002220746900000122
Figure BDA0002220746900000123
Set of constituent states, S ═<Swind,Spv,Sload>；
The state set for step n is represented as:
Figure BDA0002220746900000124
a represents action set, and is composed of daily scene elements contained in each daily scene set in clustering result
Figure BDA0002220746900000125
Wherein:
Figure BDA0002220746900000126
Figure BDA0002220746900000127
Figure BDA0002220746900000128
Figure BDA0002220746900000129
Figure BDA00022207469000001210
Figure BDA00022207469000001211
the set of actions for step n may be represented as:
Figure BDA00022207469000001212
{PSAdenotes the state transition probability, expressed as:
Figure BDA00022207469000001213
γ ∈ [0, 1)) represents a damping coefficient representing a discount of the rate of return over time, and γ ∈ [ 0.5 ] can be generally selected according to empirical values;
r represents the error function:
Figure BDA0002220746900000131
Figure BDA0002220746900000132
wherein the sort function indicates that the time series data are arranged from large to small in numerical value,
Figure BDA0002220746900000133
indicating the day of windScene
Figure BDA0002220746900000134
After the multiplication (dayn) is carried out, the time sequence data are arranged from large to small according to the numerical value, so that the length is consistent with the length of the original sequence, the subsequent solution of Euclidean distance is convenient,
Figure BDA0002220746900000135
representing a photovoltaic daily scene
Figure BDA0002220746900000136
After the expansion (dayn) times, the time sequence data are arranged from large to small according to the numerical value, so that the length is consistent with the length of the original sequence,
Figure BDA0002220746900000137
indicating loading day scene
Figure BDA0002220746900000138
After expanding (dayn) times, arranging the time sequence data from large to small according to the numerical value, and enabling the length to be consistent with the length of the original sequence;
s4, defining a value function
Figure BDA0002220746900000139
S0Is an artificially set initial state, the initial state S0Is in an amount of
Figure BDA00022207469000001310
Defining an optimum function
Figure BDA00022207469000001311
Solving using Bellman equations
Figure BDA00022207469000001312
Obtaining an optimal strategy
Figure BDA00022207469000001313
S5, setting all initial states S0Respectively substitute for
Figure BDA00022207469000001314
In the equation, | S | equations are obtained in total, thereby obtaining S in different initial states0Day scene combination strategy pi of optimal wind power, photovoltaic and load*(S0)；
S6, generating a Markov aggregation state matrix of 3 XN order consisting of wind power, photovoltaic and load through Gibbs sampling by using a Markov Monte Carlo method:
Figure BDA0002220746900000141
wherein
Figure BDA0002220746900000142
S7, matching each group of state vectors in the Markov state matrix with the mapping of the optimal strategy to obtain a day scene matrix
Figure BDA0002220746900000143
Wherein
Figure BDA0002220746900000144
Figure BDA0002220746900000145
Figure BDA0002220746900000146
S8, judging whether the difference value of the head-tail connection position between the adjacent day scenes of each row of the day scene matrix is larger than the first-order difference maximum value of the medium-long term power time sequence, if so, performing wavelet filtering processing on the head-tail connection position until the difference value is smaller than the first-order difference maximum value, if not, directly connecting the head and the tail of the day scenes, and finally generating three aggregation sequences formed by N day scenes.
While the invention has been described with reference to a preferred embodiment, it will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention.
Those not described in detail in this specification are within the skill of the art.

Claims (3)

1. A method for aggregating multidimensional time series data considering spatio-temporal correlation, comprising the steps of:
s1, respectively obtaining medium and long term historical power sequences of wind power, photovoltaic power and load with the length of m at the same time, and carrying out extremum normalization processing on the sequences to obtain medium and long term power time sequences of wind power, medium and long term power time sequences of photovoltaic power and medium and long term power time sequences of load with the length of m;
the wind power medium-long term power time sequence with the length of m is represented as follows:
Figure FDA0003478305040000011
in a sequence
Figure FDA0003478305040000012
All represent a historical value of wind power;
the photovoltaic medium and long term power time series of length m is represented as:
Figure FDA0003478305040000013
in sequence
Figure FDA0003478305040000014
Each represents a photovoltaic power historical value;
the load medium and long term power time series with length m is represented as:
Figure FDA0003478305040000015
in sequence
Figure FDA0003478305040000016
Each represents a load power history value;
then, the daily scene segmentation is respectively carried out on the medium-term and long-term power time sequences of wind power, photovoltaic power and load to obtain the same quantity:
wind power day scene
Figure FDA0003478305040000017
Wherein
Figure FDA0003478305040000018
Figure FDA0003478305040000019
Representing the k-th wind power day scene after segmentation, k is more than or equal to 1 and less than or equal to dayn,
photovoltaic day scene
Figure FDA00034783050400000110
Wherein
Figure FDA00034783050400000111
Figure FDA00034783050400000112
Representing the kth photovoltaic day scene after division, k is more than or equal to 1 and less than or equal to dayn,
load day scene
Figure FDA0003478305040000021
Wherein
Figure FDA0003478305040000022
Figure FDA0003478305040000023
Representing the k-th load day scene after division, k is more than or equal to 1 and less than or equal to dayn,
wind power day scene SCwindPhotovoltaic solar scene SCpvLoad day scene SCloadThe day scene elements contained in the method are respectively distributed to the wind power day scene set through a neighbor propagation clustering algorithm
Figure FDA0003478305040000024
Photovoltaic day scene set
Figure FDA0003478305040000025
Load day scene set
Figure FDA0003478305040000026
In the method, the optimal clustering numbers of the daily scenes of wind power, photovoltaic and load are obtained through Davies-Bouldin Index analysis
Figure FDA0003478305040000027
Figure FDA0003478305040000028
And simultaneously obtaining clustering results under respective optimal clustering numbers:
Figure FDA0003478305040000029
wherein the content of the first and second substances,
Figure FDA00034783050400000210
Figure FDA00034783050400000211
representswiWind-like power day scene set composed of nwiThe solar energy power generation system is composed of a plurality of wind power day scenes,
Figure FDA00034783050400000212
wherein
Figure FDA00034783050400000213
Figure FDA00034783050400000214
Represents the firstpiSet of quasi-photovoltaic daily scenes, from npiThe solar photovoltaic solar scene is formed by a plurality of photovoltaic solar scenes,
Figure FDA00034783050400000215
wherein
Figure FDA00034783050400000216
Figure FDA00034783050400000217
Represents the firstliClass load daily scene set, from nliThe composition of the daily scene of each load,
s2, on the basis of the step S1, clustering results of the wind power daily scene, the photovoltaic daily scene and the load daily scene obtained by adopting a neighbor propagation clustering algorithm are used as state quantities to obtain a random state model of the wind power, the photovoltaic and the load, the random state model is represented by a Markov Monte Carlo process, and the high-order state transition probability of the Markov Monte Carlo process is represented as follows:
Figure FDA0003478305040000031
Figure FDA0003478305040000032
Figure FDA0003478305040000033
where N represents the nth column state in the Markov polymerization state matrix and N is [1, N ],
Figure FDA0003478305040000034
Figure FDA0003478305040000035
Figure FDA0003478305040000036
obtaining a state transition probability matrix of the Markov Monte Carlo process by counting historical data:
Figure FDA0003478305040000037
Figure FDA0003478305040000038
Figure FDA0003478305040000039
s3, Markov decision process is composed of five parts (S, A, { PSA},γ,R)，
S is the result of clustering
Figure FDA00034783050400000310
Figure FDA0003478305040000041
Set of constituent states, S ═<Swind,Spv,Sload>；
The state set for step n is represented as:
Figure FDA0003478305040000042
a represents action set, and is composed of daily scene elements contained in each daily scene set in clustering result
Figure FDA0003478305040000043
Wherein:
Figure FDA0003478305040000044
Figure FDA0003478305040000045
Figure FDA0003478305040000046
Figure FDA0003478305040000047
Figure FDA0003478305040000048
Figure FDA0003478305040000049
the set of actions in step n is represented as:
Figure FDA00034783050400000410
{PSAdenotes the state transition probability, expressed as:
Figure FDA00034783050400000411
Figure FDA00034783050400000412
γ ∈ [0, 1) denotes a damping coefficient;
r represents the error function:
Figure FDA00034783050400000413
Figure FDA00034783050400000414
wherein XwindRepresenting a wind power medium-long term power time sequence with the length of m; xpvRepresenting a photovoltaic medium and long term power time series with the length of m; xloadRepresenting a load medium and long term power time series with the length of m; the sort function represents sorting the time series data from large to small in value,
Figure FDA0003478305040000051
representing wind-solar scenes
Figure FDA0003478305040000052
After the dayn times are expanded, the time sequence data are arranged from large to small according to the numerical value, the length is consistent with the length of the original sequence,
Figure FDA0003478305040000053
representing a photovoltaic daily scene
Figure FDA0003478305040000054
After the dayn times are expanded, the time sequence data are arranged from large to small according to the numerical value, the length is consistent with the length of the original sequence,
Figure FDA0003478305040000055
indicating loading daily scenes
Figure FDA0003478305040000056
After the dayn times are expanded, arranging the time sequence data from large to small according to the numerical value, and enabling the length to be consistent with the length of the original sequence;
s4, defining a value function
Figure FDA0003478305040000057
S0Is an artificially set initial state, the initial state S0Is in an amount of
Figure FDA0003478305040000058
Defining an optimum function
Figure FDA0003478305040000059
Solving using Bellman equations
Figure FDA00034783050400000510
Obtaining an optimal strategy
Figure FDA00034783050400000511
S5, setting all initial states S0Respectively substitute for
Figure FDA00034783050400000512
In the equation, | S | equations are obtained in total, thereby obtaining S in different initial states0Day scene combination strategy pi of optimal wind power, photovoltaic and load*(S0)；
S6, generating a Markov aggregation state matrix of 3 XN order consisting of wind power, photovoltaic and load through Gibbs sampling by using a Markov Monte Carlo method:
Figure FDA00034783050400000513
wherein T represents a Markov polymerization state matrix of order 3 XN composed of wind power, photovoltaic and load,
Figure FDA00034783050400000514
s7, matching each group of state vectors in the Markov state matrix with the mapping of the optimal strategy to obtain a day scene matrix
Figure FDA0003478305040000061
Wherein
Figure FDA0003478305040000062
Figure FDA0003478305040000063
Figure FDA0003478305040000064
S8, judging whether the difference value of the head-tail connection position between the adjacent day scenes of each row of the day scene matrix is larger than the first-order difference maximum value of the medium-long term power time sequence, if so, performing wavelet filtering processing on the head-tail connection position until the difference value is smaller than the first-order difference maximum value, if not, directly connecting the head and the tail of the day scenes, and finally generating three aggregation sequences formed by N day scenes.
2. The method of claim 1, wherein the dayn is determined by a total number of days included in the medium and long term historical power sequences.
3. The method for aggregating multi-dimensional time-series data considering spatio-temporal correlation according to claim 1, wherein n iswi、npiAnd nliObtained by calculation through a neighbor propagation clustering algorithm.
CN201910932946.4A 2019-09-29 2019-09-29 Aggregation method of multidimensional time series data considering space-time correlation Active CN110909911B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910932946.4A CN110909911B (en) 2019-09-29 2019-09-29 Aggregation method of multidimensional time series data considering space-time correlation

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910932946.4A CN110909911B (en) 2019-09-29 2019-09-29 Aggregation method of multidimensional time series data considering space-time correlation

Publications (2)

Publication Number Publication Date
CN110909911A CN110909911A (en) 2020-03-24
CN110909911B true CN110909911B (en) 2022-05-20

Family

ID=69815220

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910932946.4A Active CN110909911B (en) 2019-09-29 2019-09-29 Aggregation method of multidimensional time series data considering space-time correlation

Country Status (1)

Country Link
CN (1) CN110909911B (en)

Families Citing this family (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111523230B (en) * 2020-04-22 2023-05-26 国网能源研究院有限公司 Self-adaptive clustering method for wind-light load composite typical scene
CN112149346B (en) * 2020-09-07 2024-04-26 华中科技大学 Equivalent modeling method and device for wind farm, electronic equipment and storage medium
CN113077166A (en) * 2021-04-16 2021-07-06 国网吉林省电力有限公司 Community energy storage scheduling method based on Markov decision process
CN113128064B (en) * 2021-04-27 2023-10-24 国网北京市电力公司 Thermoelectric data aggregation method, system, device and storage medium for simulation
CN113807019A (en) * 2021-09-24 2021-12-17 清华大学 MCMC wind power simulation method based on improved scene classification and coarse grain removal
CN117370825A (en) * 2023-10-11 2024-01-09 国网经济技术研究院有限公司 Long-term scene generation method and system for generating countermeasure network based on attention condition

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
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

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20090228516A1 (en) * 2008-03-10 2009-09-10 The Hong Kong Polytechnic University Harmful algae indexing (HaiDex) method

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
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

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
Quantifying parameter uncertainty in a coral reef model using Metropolis-Coupled Markov Chain Monte Carlo;Clancy,D 等;《ECOLOGICAL MODELLING》;20100524;第221卷(第10期);全文 *
考虑气象因素的短期光伏出力预测的奇异谱分析方法;黎静华 等;《电力自动化设备》;20180504;全文 *

Also Published As

Publication number Publication date
CN110909911A (en) 2020-03-24

Similar Documents

Publication Publication Date Title
CN110909911B (en) Aggregation method of multidimensional time series data considering space-time correlation
CN109524993B (en) Wind power photovoltaic typical cycle output scene generation method for medium-long term optimization scheduling
CN103049798B (en) A kind of short-term power generation power Forecasting Methodology being applied to photovoltaic generating system
CN108879793B (en) Off-grid hybrid energy system optimization method for wind-solar energy storage hydropower station complementation
CN106295899B (en) Wind power probability density Forecasting Methodology based on genetic algorithm Yu supporting vector quantile estimate
CN110889603A (en) Power system economic dispatching method considering wind power correlation based on PCA-Copula theory
CN108075471B (en) Multi-objective constraint optimization power grid scheduling strategy based on stochastic power output prediction
De Freitas et al. Wind speed forecasting: a review
CN110994606A (en) Multi-energy power supply capacity configuration method based on complex adaptive system theory
Jebli et al. The forecasting of solar energy based on Machine Learning
CN109586279B (en) Interconnected power grid planning method
CN113469412A (en) Real-time operation strategy optimization method and system for comprehensive energy system
Chen et al. Graph neural network-based wind farm cluster speed prediction
CN106655266B (en) It is a kind of access new energy area power grid can flexible modulation power configuration method
CN110991927B (en) Power supply planning method for improving intermittent power supply complementation effect of different areas of regional power grid
CN110336332B (en) Interval power flow typical scene construction method based on output curve aggregation
CN116502074A (en) Model fusion-based photovoltaic power generation power prediction method and system
CN116523240A (en) Micro-grid evaluation feedback scheduling method based on MCMC method prediction
CN111401659A (en) Ultra-short-term or short-term photovoltaic power generation power prediction method based on case reasoning
Liu et al. Reliability assessment of micro-grid based on Markov model
CN111404193A (en) Data-driven-based microgrid random robust optimization scheduling method
CN116154875A (en) Photovoltaic power station active power optimization distribution method based on TCN and error function
CN116128211A (en) Wind-light-water combined short-term optimization scheduling method based on wind-light uncertainty prediction scene
CN115021331A (en) Clean energy-based ground water wind-solar integrated compensation method based on contribution coefficient method
Wan et al. Photovoltaic generation scenario analysis considering irradiation uncertainty and output derating probability

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