CN112819216A - Wind power sequence scene set based generation method and system - Google Patents

Abstract

The invention provides a generation method based on a wind power sequence scene set, which comprises the following steps of 1, constructing a prediction box and numbering the prediction box; 2. generating a wind power static scene set; 3. constructing a wind power state transition matrix; 4. constructing an initial solution, a neighborhood solution and a fitness function of a tabu search algorithm by combining a wind power state transition matrix; 5. and iteratively generating a wind power sequence scene set. The invention also provides a generation system based on the wind power sequence scene set. The invention has the advantages that: the construction process of the initial solution of the tabu search algorithm is combined with the wind power state transfer matrix, unreasonable fluctuation of the wind power sequence multi-period transfer process is effectively filtered, the initial solution quality is improved by improving the time sequence of the initial solution, the iteration times of the tabu search algorithm are reduced, and the algorithm timeliness is improved; the construction process of the neighborhood solution of the tabu search algorithm is combined with the wind power state transition matrix, unreasonable fluctuation of the wind power sequence multi-period transition process is effectively filtered, and the time sequence of the sequence scene set is improved.

Description

Wind power sequence scene set based generation method and system

Technical Field

The invention relates to wind power output uncertainty modeling, in particular to a method for generating a scene set of a wind power output sequence, which is used for describing various possibilities of the wind power output sequence within a plurality of hours in the future so as to draw the uncertainty of the wind power output at the moment.

Background

With the gradual exhaustion of traditional fossil energy, new energy power generation represented by wind power and photovoltaic power generation occupies an increasingly important position in a power system. The output of the wind power generation is random, fluctuating and intermittent due to the influence of factors such as weather factors, geographical positions and the like, but the existing wind power prediction output technology only provides one condition of the wind power output, a bottleneck exists in prediction precision, the uncertainty of the wind power is not considered enough, and the unstable factor is brought to the safe and stable operation of a power system. The scene analysis technology describes various possibilities of wind power output by generating a sequence scene set of the wind power output, describes uncertainty information of the wind power output by using a deterministic scene, converts uncertainty problems into deterministic problems for analysis, and provides data support for planning and scheduling of a power grid.

The existing wind power sequence scene set generation methods can be summarized into three types: (1) the historical sequence clustering method is characterized in that historical wind power sequence samples are used as original scenes, clustering reduction is directly performed on the original samples by adopting a clustering algorithm, a small number of representative scenes are obtained, the method is simple in operation, only historical wind power data are processed, the change rule of historical output sequences can be reasonably expressed, the method is insufficient in predictability of wind power sequence development trends, and the effect is poor in the aspect of describing uncertainty characteristics of future output; (2) the method comprises the steps that a wind power sequence scene set meeting strong autocorrelation is generated by constructing a multi-standard normal distribution sequence and carrying out inverse transformation on the multi-standard normal distribution sequence, the scene set generated by the method has a good effect on describing the time sequence of wind power, but the difference between a small number of scenes and an actual output sequence is possibly overlarge, the variation trend of each sequence scene is approximately the same, and the description capacity of the randomness of the wind power output is insufficient; (3) the single-time-interval/multi-time-interval method comprises the steps that firstly, a static scene set is generated in a single-time-interval analysis stage; and then generating a sequence scene set capable of describing the wind power random characteristic at a multi-time period analysis stage, wherein the method has the advantages that the description of the wind power output random characteristic is accurate, but the timing sequence and the accuracy of the sequence scene set are difficult to be considered in the connection aspect of wind power scenes at adjacent time periods. Therefore, the existing wind power sequence scene set generation technology still needs to be improved in the aspect of the overall quality of the scene set.

The patent application with the publication number of CN 111934319A is proposed for the inventor team, and discloses a generating method and a generating system based on a wind power typical scene set, wherein the method comprises the steps of 1, constructing a self-adaptive prediction box and numbering the box; 2. fitting the probability distribution of the prediction error data in the prediction box; 3. generating a wind power static scene set; 4. generating a wind power dynamic scene set by adopting a tabu search algorithm; 5. carrying out weight assignment on the wind power dynamic scene set; 6. and clustering and reducing the wind power dynamic scene set to obtain a wind power typical scene set. The method adopts the self-adaptive prediction box, can describe various possibilities of wind power output, overcomes the problem of insufficient precision of the conventional wind power output prediction to a certain extent, and still needs to improve the time sequence, the accuracy and the overall quality of a sequence scene set.

Disclosure of Invention

The technical problems to be solved by the invention are how to improve the time sequence and the accuracy of a sequence scene set and how to make up the problem of insufficient prediction precision of the existing wind power output to a certain extent by describing the uncertainty of the wind power output, thereby providing data support for the safe and stable operation of a power grid.

The invention solves the technical problems through the following technical means: a generation method based on a wind power sequence scene set comprises the following steps:

step 1, constructing a prediction box and numbering:

step 1.1, forming predicted output data and corresponding predicted error data of the predicted output data at the same moment in historical data into data pairs, arranging the data pairs according to the amplitude of the predicted output data in an ascending order, equally dividing the data pairs into H groups according to amplitude intervals, forming an initial prediction box by all data pairs in any group, and obtaining H initial prediction boxes and sequentially numbering the H initial prediction boxes;

step 1.2, fitting the probability distribution of all the prediction error data in each initial prediction box to obtain H fitting results;

step 2, generating a wind power static scene set:

step 2.1, predicting an output sequence E ═ E [ E ] according to the known future wind power with the sampling granularity t₁,E₂,…,E_g,…,E_G]；E_gThe predicted wind power output at the time of t multiplied by G is shown, G belongs to [1, G ]]Initializing g to be 1;

step 2.2, determining a self-adaptive prediction box corresponding to the txg moment so as to obtain a fitting result Fg of probability distribution of prediction error data at the txg moment;

step 2.3, fitting result F to the error at the t × g moment_gRandom sampling is carried out for M times to obtain an error sample sequence U at the time of t multiplied by g_g＝[U¹ _g,U² _g,…,U^m _g,…,U^M _g]]；U^m _gM error sample, representing the t × g time, M belonging to [1, M]；

Step 2.4, the error sample sequence U at the time of t × g_gEach element of (1) is respectively added with the predicted wind power output E at the time of t multiplied by g_gSo as to obtain a static scene set P with the scale of M at the time of t multiplied by g_g＝[P¹ _g,P² _g,…,P^m _g,…,P^M _g]，P^m _gRepresenting the mth wind power static scene at the time of the txg;

step 2.5, after G +1 is assigned to G, if G is less than G +1, executing step 2.2, otherwise, generating the wind power static scene sets at all moments;

step 3, counting output data at adjacent moments in the historical wind power output actual measurement data, and constructing a wind power state transfer matrix Q;

step 4, constructing an initial solution, a neighborhood solution and a fitness function of the tabu search algorithm:

step 4.1, combining the wind power state transfer matrix Q to construct an initial solution of a tabu search algorithm;

step 4.2, combining the wind power state transition matrix Q to construct a neighborhood solution of a tabu search algorithm;

4.3, constructing a fitness function of a tabu search algorithm;

and 5, iteratively generating a wind power sequence scene set.

According to the method, the construction process of the initial solution of the tabu search algorithm is combined with the wind power state transfer matrix, unreasonable fluctuation of the wind power sequence multi-period transfer process is effectively filtered, the initial solution quality is improved by improving the time sequence of the initial solution, the iteration times of the tabu search algorithm are reduced, and the algorithm timeliness is improved;

according to the method, the construction process of the neighborhood solution of the tabu search algorithm is combined with the wind power state transition matrix, unreasonable fluctuation of the wind power sequence multi-period transition process is effectively filtered, and the time sequence of the sequence scene set is improved;

the fitness function of the tabu search algorithm gives consideration to the distance and the autocorrelation coefficient of the sequence scene set, and the candidate solution with a larger function value is selected as the optimal solution through iteration, so that the inter-variability among scenes in the optimal solution is ensured, and the function of time sequence in the optimization process is reflected by introducing the autocorrelation coefficient.

Further, the step 4.1 comprises:

s411, acquiring the actual wind power output amplitude z in the time period t equal to 0₀Calculating the corresponding state b₀，i＝1，t＝1；

S412, randomly extracting a wind power static scene in the t period and recording the scene as p_tCalculating p_tCorresponding state b_t；

S413, recording the b-th in the state transition matrix Q_t-1Line b_tColumn corresponding elements of

If it is

Go to step S412, else ξⁱ _t,＝p_t，ξⁱ _tTurning to step S414 for the element in the ith row and the tth column;

s414, if t <96, then t equals t +1, go to step S412, otherwise go to step S415;

s415, if i <100, i equals i +1, go to step S412, otherwise, the process ends.

Further, the step 4.2 comprises:

s422, randomly determining d time periods in which values need to be changed, and recording as y ═ y₁,y₂…y_r…y_d](1≤r≤d,1≤y_r≤96)，r＝1；

S423 from y_rRandomly extracting a static scene from a static scene set corresponding to a time interval as a sequence scene y in the current solution_rThe value of the time interval is changed, if y is more than or equal to 1_rIf not more than 95, turning to the step S424, otherwise, turning to the step S425;

s424, calculating y_rTime period of-1, y_rTime period and y_rState corresponding to wind power output in +1 time period

And

note that in the state transition matrix Q

Go to the first

Column corresponding elements of

Note that in the state transition matrix Q

Go to the first

Column corresponding elements of

If it is

And

if not, turning to step S426, otherwise, turning to step S423;

s425, calculating y_r-1 period and y_rState corresponding to wind power output in time interval

If it is

If not, go to step S426, otherwise go to step S423;

s426, if r is equal to r +1, go to step 423 if r is equal to or less than d, otherwise go to step 427;

and S427, judging whether the neighborhood scene is repeated with the sequence scene in the tabu table, if so, turning to the step 423, otherwise, finishing the construction of the neighborhood scene of the sequence scene.

Further, the step 4.2 further comprises:

S421，i＝1；

in S427, if the solution is repeated, r is 1, go to step S423, otherwise, the neighborhood scene of the ith sequence scene in the current solution is constructed, go to step S428;

s428, if i is less than 100, i equals i +1, go to step S422, otherwise, the neighborhood solution is complete.

Further, in the step 4.3, constructing a fitness function of the tabu search algorithm specifically includes:

let us note that any one candidate solution is λ, λ is a matrix of 100 × 96, each row is a sequence scene,

using formulas

And calculating the distance between any two sequence scenes in the candidate solution, wherein,

λⁱand λ^j(i is more than or equal to 1, and j is more than or equal to S) are respectively the ith sequence scene and the jth sequence scene in the candidate solution;

using formulas

And calculating the autocorrelation coefficient of any sequence scene in the candidate solution, wherein,

A^kfor the autocorrelation coefficient of the kth sequence scene in the candidate solution, k is more than or equal to 1 and less than or equal to 100, c and v are covariance and variance respectively, and lambda^k _e(e is more than or equal to 1 and less than or equal to 96) is an element of the kth row and the e column of the candidate solution lambda;

using formulas

And calculating a fitness function f of the candidate solution, wherein S is the scale of the sequence scene set in the candidate solution.

Further, in step 5, the iterative generation of the wind power sequence scene set includes the following steps:

s51, the scale of the wind power sequence scene set is given as 100, and the termination criterion theta is 10^-4The number of neighborhood solutions of each iteration is 1000, the number of time segments of the current solution change value is 24, and an initial solution H is constructed according to the step 4.1₀Step 4.3, calculate the fitness function value f of the initial solution₀ContraindicationThe table is set to be null, and the iteration number IP is 1;

s52, repeating the step 4.2 for 1000 times, constructing 1000 neighborhood solutions of the IP iteration, calculating the fitness function value of each neighborhood solution, and recording the maximum value as f^IPCorresponding neighborhood solution is H^IP，

S53, recording the maximum fitness function value of the candidate solution as f_IPTake f_IP＝max{f_IP-1，f^IPThe current neighborhood solution corresponding to the current solution is marked as H_IP；

S54, if | f_IP-f_IP-1|/f_IP>And theta, if the IP is equal to IP +1, adding sequence scenes in all the candidate solutions into a tabu table, turning to the step S52, and otherwise, outputting the optimal solution H_best＝H_IPAnd the iteration is ended.

The invention also discloses a generating system based on the wind power sequence scene set, which comprises the following modules:

the prediction box construction module is used for constructing prediction boxes and numbering the prediction boxes, and comprises the following units:

the initial prediction box unit is used for forming each data pair by the predicted output data and the corresponding predicted error data at the same moment in the historical data, equally dividing each data pair into H groups according to amplitude intervals after the data pairs are arranged in an ascending order according to the amplitude of the predicted output data, and forming an initial prediction box by all the data pairs in any group so as to obtain H initial prediction boxes and sequentially numbering the H initial prediction boxes;

the fitting unit is used for fitting the probability distribution of all the prediction error data in each initial prediction box so as to obtain H fitting results;

the wind power static scene set generation module is used for generating a wind power static scene set and comprises the following units:

a processing sequence prediction unit for predicting an output sequence E ═ E [ E ] according to the known future wind power with the sampling granularity t₁,E₂,…,E_g,…,E_G]；E_gThe predicted wind power output at the time of t multiplied by G is shown, G belongs to [1, G ]]Initializing g to be 1;

the fitting unit is used for determining an adaptive prediction box corresponding to the txg moment so as to obtain a fitting result Fg of probability distribution of prediction error data at the txg moment;

a sampling unit for fitting the result F to the error at the t × g time_gRandom sampling is carried out for M times to obtain an error sample sequence U at the time of t multiplied by g_g＝[U¹ _g,U² _g,…,U^m _g,…,U^M _g]]；U^m _gM error sample, representing the t × g time, M belonging to [1, M]；

A static scene set unit for sampling the error sample sequence U at the time of t × g_gEach element of (1) is respectively added with the predicted wind power output E at the time of t multiplied by g_gSo as to obtain a static scene set P with the scale of M at the time of t multiplied by g_g＝[P¹ _g,P² _g,…,P^m _g,…,P^M _g]，P^m _gRepresenting the mth wind power static scene at the time of the txg;

the assignment unit is used for assigning G +1 to G, if G is less than G +1, executing the step 2.2, otherwise, generating the wind power static scene sets at all the moments;

the wind power state transfer matrix building module is used for counting output data at adjacent moments in the historical wind power output actual measurement data and building a wind power state transfer matrix Q;

the tabu search algorithm module is used for constructing an initial solution, a neighborhood solution and a fitness function of the tabu search algorithm and comprises the following units:

the initial solution construction unit is used for constructing an initial solution of a tabu search algorithm by combining the wind power state transition matrix Q;

the neighborhood solution construction unit is used for constructing a neighborhood solution of a tabu search algorithm by combining the wind power state transition matrix Q;

a fitness function constructing unit for constructing a fitness function of the tabu search algorithm;

and the generation wind power sequence scene set iteration module is used for generating a wind power sequence scene set in an iteration mode.

Further, the working process of the initial solution construction unit comprises the following steps:

s411, acquiring the actual wind power output amplitude z in the time period t equal to 0₀Calculating the corresponding state b₀，i＝1，t＝1；

S412, randomly extracting a wind power static scene in the t period and recording the scene as p_tCalculating p_tCorresponding state b_t；

S413, recording the b-th in the state transition matrix Q_t-1Line b_tColumn corresponding elements of

If it is

Go to step S412, otherwise

ξⁱ _tTurning to step S414 for the element in the ith row and the tth column;

s414, if t <96, then t equals t +1, go to step S412, otherwise go to step S415;

s415, if i <100, i equals i +1, go to step S412, otherwise, the process ends.

Further, the neighborhood solution construction unit working process comprises:

s422, randomly determining d time periods in which values need to be changed, and recording as y ═ y₁,y₂…y_r…y_d](1≤r≤d,1≤y_r≤96)，r＝1；

S423 from y_rRandomly extracting a static scene from a static scene set corresponding to a time interval as a sequence scene y in the current solution_rThe value of the time interval is changed, if y is more than or equal to 1_rIf not more than 95, turning to the step S424, otherwise, turning to the step S425;

s424, calculating y_rTime period of-1, y_rTime period and y_rState corresponding to wind power output in +1 time period

And

note that in the state transition matrix Q

Go to the first

Column corresponding elements of

Note that in the state transition matrix Q

Go to the first

Column corresponding elements of

If it is

And

if not, turning to step S426, otherwise, turning to step S423;

s425, calculating y_r-1 period and y_rState corresponding to wind power output in time interval

If it is

If not, go to step S426, otherwise go to step S423;

s426, if r is equal to r +1, go to step 423 if r is equal to or less than d, otherwise go to step 427;

and S427, judging whether the neighborhood scene is repeated with the sequence scene in the tabu table, if so, turning to the step 423, otherwise, finishing the construction of the neighborhood scene of the sequence scene.

Further, the working process of the fitness function building unit includes:

using formulas

And calculating the distance between any two sequence scenes in the candidate solution, wherein,

λⁱand λ^j(i is more than or equal to 1, and j is more than or equal to S) are respectively the ith sequence scene and the jth sequence scene in the candidate solution;

using formulas

And calculating the autocorrelation coefficient of any sequence scene in the candidate solution, wherein,

A^kfor the autocorrelation coefficient of the kth sequence scene in the candidate solution, c and v are covariance and variance, respectively, lambda^k _e(e is more than or equal to 1 and less than or equal to 96) is an element of the kth row and the e column of the candidate solution lambda;

using formulas

And calculating a fitness function f of the candidate solution, wherein S is the scale of the sequence scene set in the candidate solution.

The invention has the advantages that:

(1) according to the method, the construction process of the initial solution of the tabu search algorithm is combined with the wind power state transfer matrix, unreasonable fluctuation of the wind power sequence multi-period transfer process is effectively filtered, the initial solution quality is improved by improving the time sequence of the initial solution, the iteration times of the tabu search algorithm are reduced, and the algorithm timeliness is improved;

(2) according to the method, the construction process of the neighborhood solution of the tabu search algorithm is combined with the wind power state transition matrix, unreasonable fluctuation of the wind power sequence multi-period transition process is effectively filtered, and the time sequence of the sequence scene set is improved;

(3) the fitness function of the tabu search algorithm gives consideration to the distance and the autocorrelation coefficient of the sequence scene set, and the candidate solution with a larger function value is selected as the optimal solution through iteration, so that the inter-variability among scenes in the optimal solution is ensured, and the function of time sequence in the optimization process is reflected by introducing the autocorrelation coefficient.

Drawings

FIG. 1 is a schematic flow chart of a method for generating a wind power sequence scene set according to an embodiment of the present invention;

fig. 2 is a flow chart of the tabu search algorithm of the present invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the embodiments of the present invention, and it is obvious that the described embodiments are some embodiments of the present invention, but not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

In this embodiment, as shown in fig. 1, a method for generating a scene set based on a wind power sequence includes the following steps:

step 1, constructing a prediction box and numbering;

the wind power prediction error has better statistical characteristics than wind power prediction output, the prediction error is related to wind power prediction output amplitude, the prediction box technology can describe probability distribution conditions of the prediction error under different prediction output amplitudes, and the condition correlation of the prediction error and the wind power prediction output is established.

Step 1.1, forming each data pair by the predicted output data of each moment in the historical data and the corresponding predicted error data, arranging each data pair according to the amplitude of the predicted output data in an ascending order, equally dividing the data pairs into 50 groups according to the amplitude interval, forming an initial prediction box by all data pairs in any group, and obtaining 50 initial prediction boxes and sequentially numbering the data pairs;

step 1.2, fitting the probability distribution of all the prediction error data in each prediction box by adopting a ksDensity function in MATLAB to obtain 50 fitting results;

the ksDensity function in the MATLAB belongs to an nonparametric kernel density estimation function, and in view of different probability distributions of prediction errors in different output prediction intervals, the effect of uniformly fitting a certain function is poor, and the applicability of fitting the wind power prediction error can be improved by adopting a nonparametric kernel density estimation method.

Step 2, generating a wind power static scene set;

step 2.1, predicting an output sequence E ═ E [ E ] according to the known future wind power with the sampling granularity of 15min₁,E₂,…,E_g,…,E₉₆]；E_gThe predicted wind power output at the 15 Xg min moment is shown, and g belongs to [1,96 ]](ii) a Initializing g to 1; (24 hours)

Step 2.2, determining a self-adaptive prediction box corresponding to the 15 Xg min moment so as to obtain a fitting result Fg of probability distribution of prediction error data at the moment;

step 2.3, error fitting result F at the 15 Xg min moment_gRandom sampling is carried out for 100 times to obtain an error sample sequence U at the moment_g＝[U¹ _g,U² _g,…,U^m _g,…,U¹⁰⁰ _g]]；U^m _gThe m-th error sample at time 15 xg min, m belonging to [1,100 ]]；

Step 2.4, the error sample sequence U at the 15 Xg min moment_gRespectively adding the predicted wind power output E at the moment to each element_gSo as to obtain a static scene set P with the scale of 100 at the 15 Xg min time_g＝[P¹ _g,P² _g,…,P^m _g,…,P¹⁰⁰ _g]，P^m _gRepresenting the mth wind power static scene at the 15 Xg min moment;

step 2.5, after g +1 is assigned to g, if g is less than 101, executing step 2.2, otherwise, generating the wind power static scene sets at all moments;

step 3, constructing a wind power state transition matrix;

counting output data at adjacent moments in the historical wind power output actual measurement data, dividing the output data into 4 states, and constructing a state transition matrix of wind power output at the adjacent moments as follows:

in the formula (4), q_i,j(i is more than or equal to 1, and j is less than or equal to 4) is the probability that the wind power is transferred to the state j after 15min from the state i.

The wind power state transition matrix counts the probability of wind power output transferring from one state to another state after 15min, if the probability is 0, the transition does not exist in history, and the transition is not allowed to appear in a sequence scene.

Step 4, constructing an initial solution, a neighborhood solution and a fitness function of the tabu search algorithm;

step 4.1, constructing an initial solution of a tabu search algorithm;

the initial solution is a matrix of 100 x 96, ξⁱ _tIs the ith (1 ≦ i)<100) Line t (1. ltoreq. t)<96) The elements of the column.

S411, acquiring the actual wind power output amplitude z in the time period t equal to 0₀Calculating the corresponding state b₀，i＝1，t＝1；

S412, randomly extracting a wind power static scene in the t period and recording the scene as p_tCalculating p_tCorresponding state b_t；

S413, recording the b-th in the state transition matrix Q_t-1Line b_tColumn corresponding elements of

If it is

Go to step S412, else ξⁱ _t,＝p_tGo to step S414；

S414, if t <96, then t equals t +1, go to step S412, otherwise go to step S415;

s415, if i is less than 100, i is i +1, go to step S412, otherwise, end;

the initial solution is the current solution of the first iteration of the tabu search algorithm, and its quality often affects the time cost of the algorithm. Wind power fluctuation in the initial solution is limited through the state transition matrix, the quality of the initial solution is effectively improved, and the time used by the algorithm is reduced.

Step 4.2, constructing a neighborhood solution of a tabu search algorithm;

S421，i＝1；

s422, randomly determining 24 time periods in which the value needs to be changed, and recording as y ═ y₁,y₂…y_r…y₂₄](1≤r≤24,1≤y_r≤96)，r＝1；

S423 from y_rRandomly extracting one static scene from the static scene set corresponding to the time interval as the current solution y_rThe value of the time interval is changed, if y is more than or equal to 1_rIf not more than 95, turning to the step S424, otherwise, turning to the step S425;

s424, calculating y_rTime period of-1, y_rTime period and y_rState corresponding to wind power output in +1 time period

And

note that in the state transition matrix Q

Go to the first

Column corresponding elements of

Note that in the state transition matrix Q

Go to the first

Column corresponding elements of

If it is

And

if not, turning to step S426, otherwise, turning to step S423;

s425, calculating y_r-1 period and y_rState corresponding to wind power output in time interval

If it is

If not, go to step S426, otherwise go to step S423;

s426, where r is r +1, if r is less than or equal to 24, go to step 42S3, otherwise go to step S427;

s427, determining whether the neighborhood scene is repeated with the sequence scene in the tabu table, if so, r is 1, and going to step S423, otherwise, completing the construction of the neighborhood scene of the ith sequence scene in the current solution, and going to step S428;

s428, if i is less than 100, i is i +1, go to step S422, otherwise, the neighborhood solution is complete;

the neighborhood solution and the current solution are collectively called as a candidate solution of current iteration, the construction process of the neighborhood solution meets the fluctuation limitation of wind power, the time sequence of the neighborhood solution is effectively improved, the quality of the candidate solution is greatly improved, and the effects of improving the iteration speed and enhancing the time sequence of the sequence scene set are obvious.

4.3, constructing a fitness function of a tabu search algorithm;

let us note that any one candidate solution is λ, which is a matrix of 100 × 96, each row representing a sequence scenario.

Using formulas

And calculating the distance between any two sequence scenes in the candidate solution, wherein,

λⁱand λ^j(i is more than or equal to 1, j is less than or equal to 100) are respectively the ith sequence scene and the jth sequence scene in the candidate solution;

using formulas

And calculating the autocorrelation coefficient of any sequence scene in the candidate solution, wherein,

A^k(k is more than or equal to 1 and less than or equal to 100) is the autocorrelation coefficient of the kth sequence scene in the candidate solution, c and v are covariance and variance respectively, and lambda is^k _e(e is more than or equal to 1 and less than or equal to 96) is an element of the kth row and the e column of the candidate solution lambda;

using formulas

A fitness function of the candidate solution is calculated, wherein,

f is a fitness function value, and S is the scale of a sequence scene set in a candidate solution;

the fitness function gives consideration to the distance and the autocorrelation coefficient of the sequence scene set, and selects a candidate solution with a larger function value as an optimal solution through iteration, so that the inter-variability among scenes in the optimal solution is ensured, and the function of time sequence in the optimization process is reflected by introducing the autocorrelation coefficient.

Step 5, generating a wind power sequence scene set in an iteration mode;

s51, the scale of the wind power sequence scene set is given as 100, and the termination criterion theta is 10^-4The number of neighborhood solutions in each iteration is 1000, and the current solution is changedThe time period number of the variable value is 24, and an initial solution H is constructed according to the step 4.1₀Step 4.3, calculate the fitness function value f of the initial solution₀Setting a tabu table to be null, and setting the iteration number IP to be 1;

s52, repeating the step 4.2 for 1000 times, constructing 1000 neighborhood solutions of the IP iteration, calculating the fitness function value of each neighborhood solution, and recording the maximum value as f^IPCorresponding neighborhood solution is H^IP，

S53, recording the maximum fitness function value of the candidate solution as f_IPTake f_IP＝max{f_IP-1，f^IPThe current neighborhood solution corresponding to the current solution is marked as H_IP；

S54, if | f_IP-f_IP-1|/f_IP>And theta, if the IP is equal to IP +1, adding sequence scenes in all the candidate solutions into a tabu table, turning to the step S52, and otherwise, outputting the optimal solution H_best＝H_IPAnd the iteration is ended.

Compared with meta-heuristic algorithms such as a particle swarm algorithm, the taboo search algorithm can effectively jump out local optimal solutions through the taboo table, and search in a larger area is realized, so that a wind power sequence scene set with more comprehensive timeliness and stronger orderliness is obtained.

The above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.

Claims (10)

1. A generation method based on a wind power sequence scene set is characterized by comprising the following steps:

step 1, constructing a prediction box and numbering:

step 1.1, forming predicted output data and corresponding predicted error data of the predicted output data at the same moment in historical data into data pairs, arranging the data pairs according to the amplitude of the predicted output data in an ascending order, equally dividing the data pairs into H groups according to amplitude intervals, forming an initial prediction box by all data pairs in any group, and obtaining H initial prediction boxes and sequentially numbering the H initial prediction boxes;

step 1.2, fitting the probability distribution of all the prediction error data in each initial prediction box to obtain H fitting results;

step 2, generating a wind power static scene set:

step 2.1, predicting an output sequence E ═ E [ E ] according to the known future wind power with the sampling granularity t₁,E₂,…,E_g,…,E_G]；E_gThe predicted wind power output at the time of t multiplied by G is shown, G belongs to [1, G ]]Initializing g to be 1;

step 2.2, determining a self-adaptive prediction box corresponding to the txg moment so as to obtain a fitting result Fg of probability distribution of prediction error data at the txg moment;

step 2.3, fitting result F to the error at the t × g moment_gRandom sampling is carried out for M times to obtain an error sample sequence U at the time of t multiplied by g_g＝[U¹ _g,U² _g,…,U^m _g,…,U^M _g]]；U^m _gM error sample, representing the t × g time, M belonging to [1, M]；

Step 2.4, the error sample sequence U at the time of t × g_gEach element of (1) is respectively added with the predicted wind power output E at the time of t multiplied by g_gSo as to obtain a static scene set P with the scale of M at the time of t multiplied by g_g＝[P¹ _g,P² _g,…,P^m _g,…,P^M _g]，P^m _gRepresenting the mth wind power static scene at the time of the txg;

step 2.5, after G +1 is assigned to G, if G is less than G +1, executing step 2.2, otherwise, generating the wind power static scene sets at all moments;

step 3, counting output data at adjacent moments in the historical wind power output actual measurement data, and constructing a wind power state transfer matrix Q;

step 4, constructing an initial solution, a neighborhood solution and a fitness function of the tabu search algorithm:

step 4.1, combining the wind power state transfer matrix Q to construct an initial solution of a tabu search algorithm;

step 4.2, combining the wind power state transition matrix Q to construct a neighborhood solution of a tabu search algorithm;

4.3, constructing a fitness function of a tabu search algorithm;

and 5, iteratively generating a wind power sequence scene set.

2. The generation method according to claim 1, characterized in that said step 4.1 comprises:

s411, acquiring the actual wind power output amplitude z in the time period t equal to 0₀Calculating the corresponding state b₀，i＝1，t＝1；

S412, randomly extracting a wind power static scene in the t period and recording the scene as p_tCalculating p_tCorresponding state b_t；

S413, recording the b-th in the state transition matrix Q_t-1Line b_tColumn corresponding elements of

If it is

Go to step S412, else ξⁱ _t,＝p_t，ξⁱ _tTurning to step S414 for the element in the ith row and the tth column;

s414, if t <96, then t equals t +1, go to step S412, otherwise go to step S415;

s415, if i <100, i equals i +1, go to step S412, otherwise, the process ends.

3. The generation method according to claim 1, characterized in that said step 4.2 comprises:

s422, randomly determining d time periods in which values need to be changed, and recording as y ═ y₁,y₂…y_r…y_d](1≤r≤d,1≤y_r≤96)，r＝1；

S423 from y_rRandomly extracting a static scene from a static scene set corresponding to a time interval as a sequence scene y in the current solution_rThe value of the time interval is changed, if y is more than or equal to 1_rIf not more than 95, turning to the step S424, otherwise, turning to the step S425;

s424, calculating y_rTime period of-1, y_rTime period and y_rState corresponding to wind power output in +1 time period

And

note that in the state transition matrix Q

Go to the first

Column corresponding elements of

Note that in the state transition matrix Q

Go to the first

Column corresponding elements of

If it is

And

all are allIf yes, go to step S426, otherwise go to step S423;

s425, calculating y_r-1 period and y_rState corresponding to wind power output in time interval

If it is

If not, go to step S426, otherwise go to step S423;

s426, if r is equal to r +1, go to step 423 if r is equal to or less than d, otherwise go to step 427;

and S427, judging whether the neighborhood scene is repeated with the sequence scene in the tabu table, if so, turning to the step 423, otherwise, finishing the construction of the neighborhood scene of the sequence scene.

4. The generation method according to claim 3, characterized in that said step 4.2 further comprises:

S421，i＝1；

in S427, if the solution is repeated, r is 1, go to step S423, otherwise, the neighborhood scene of the ith sequence scene in the current solution is constructed, go to step S428;

s428, if i is less than 100, i equals i +1, go to step S422, otherwise, the neighborhood solution is complete.

5. The generation method according to claim 1, wherein the step 4.3 of constructing a fitness function of a tabu search algorithm specifically comprises:

let us note that any one candidate solution is λ, λ is a matrix of 100 × 96, each row is a sequence scene,

using formulas

And calculating the distance between any two sequence scenes in the candidate solution, wherein,

λⁱand λ^j(i is more than or equal to 1, and j is more than or equal to S) are respectively the ith sequence scene and the jth sequence scene in the candidate solution;

using formulas

And calculating the autocorrelation coefficient of any sequence scene in the candidate solution, wherein,

A^kfor the autocorrelation coefficient of the kth sequence scene in the candidate solution, k is more than or equal to 1 and less than or equal to 100, c and v are covariance and variance, lambda, respectively^k _eE is more than or equal to 1 and less than or equal to 96 and is an element of the kth row and the e column of the candidate solution lambda;

using formulas

And calculating a fitness function f of the candidate solution, wherein S is the scale of the sequence scene set in the candidate solution.

6. The generation method according to claim 1, wherein, in step 5, the iterative generation of the wind power sequence scene set comprises the following steps:

s51, the scale of the wind power sequence scene set is given as 100, and the termination criterion theta is 10^-4The number of neighborhood solutions of each iteration is 1000, the number of time segments of the current solution change value is 24, and an initial solution H is constructed according to the step 4.1₀Step 4.3, calculate the fitness function value f of the initial solution₀Setting a tabu table to be null, and setting the iteration number IP to be 1;

s52, repeating the step 4.2 for 1000 times, constructing 1000 neighborhood solutions of the IP iteration, calculating the fitness function value of each neighborhood solution, and recording the maximum value as f^IPCorresponding neighborhood solution is H^IP，

S53, recording the maximum fitness function value of the candidate solution as f_IPTake f_IP＝max{f_IP-1，f^IPAdaptation of current solution as IP iterationDegree function value, its corresponding current neighborhood solution is H_IP；

S54, if | f_IP-f_IP-1|/f_IP>And theta, if the IP is equal to IP +1, adding sequence scenes in all the candidate solutions into a tabu table, turning to the step S52, and otherwise, outputting the optimal solution H_best＝H_IPAnd the iteration is ended.

7. A generation system based on a wind power sequence scene set is characterized by comprising the following modules:

the prediction box construction module is used for constructing prediction boxes and numbering the prediction boxes, and comprises the following units:

the initial prediction box unit is used for forming each data pair by the predicted output data and the corresponding predicted error data at the same moment in the historical data, equally dividing each data pair into H groups according to amplitude intervals after the data pairs are arranged in an ascending order according to the amplitude of the predicted output data, and forming an initial prediction box by all the data pairs in any group so as to obtain H initial prediction boxes and sequentially numbering the H initial prediction boxes;

the fitting unit is used for fitting the probability distribution of all the prediction error data in each initial prediction box so as to obtain H fitting results;

the wind power static scene set generation module is used for generating a wind power static scene set and comprises the following units:

a processing sequence prediction unit for predicting an output sequence E ═ E [ E ] according to the known future wind power with the sampling granularity t₁,E₂,…,E_g,…,E_G]；E_gThe predicted wind power output at the time of t multiplied by G is shown, G belongs to [1, G ]]Initializing g to be 1;

the fitting unit is used for determining an adaptive prediction box corresponding to the txg moment so as to obtain a fitting result Fg of probability distribution of prediction error data at the txg moment;

a sampling unit for fitting the result F to the error at the t × g time_gRandom sampling is carried out for M times to obtain an error sample sequence U at the time of t multiplied by g_g＝[U¹ _g,U² _g,…,U^m _g,…,U^M _g]]；U^m _gM error sample, representing the t × g time, M belonging to [1, M]；

A static scene set unit for sampling the error sample sequence U at the time of t × g_gEach element of (1) is respectively added with the predicted wind power output E at the time of t multiplied by g_gSo as to obtain a static scene set P with the scale of M at the time of t multiplied by g_g＝[P¹ _g,P² _g,…,P^m _g,…,P^M _g]，P^m _gRepresenting the mth wind power static scene at the time of the txg;

the assignment unit is used for assigning G +1 to G, if G is less than G +1, executing the step 2.2, otherwise, generating the wind power static scene sets at all the moments;

the wind power state transfer matrix building module is used for counting output data at adjacent moments in the historical wind power output actual measurement data and building a wind power state transfer matrix Q;

the tabu search algorithm module is used for constructing an initial solution, a neighborhood solution and a fitness function of the tabu search algorithm and comprises the following units:

the initial solution construction unit is used for constructing an initial solution of a tabu search algorithm by combining the wind power state transition matrix Q;

the neighborhood solution construction unit is used for constructing a neighborhood solution of a tabu search algorithm by combining the wind power state transition matrix Q;

a fitness function constructing unit for constructing a fitness function of the tabu search algorithm;

and the generation wind power sequence scene set iteration module is used for generating a wind power sequence scene set in an iteration mode.

8. The generation system of claim 7, wherein the initial solution construction unit work process comprises:

s411, acquiring the actual wind power output amplitude z in the time period t equal to 0₀Calculating the corresponding state b₀，i＝1，t＝1；

S412, randomly extracting a wind power static scene in the t period and recording the scene as p_tMeter for measuringCalculation of p_tCorresponding state b_t；

S413, recording the b-th in the state transition matrix Q_t-1Line b_tColumn corresponding elements of

If it is

Go to step S412, else ξⁱ _t,＝p_t，ξⁱ _tTurning to step S414 for the element in the ith row and the tth column;

s414, if t <96, then t equals t +1, go to step S412, otherwise go to step S415;

s415, if i <100, i equals i +1, go to step S412, otherwise, the process ends.

9. The generation system of claim 7, wherein the neighborhood solution construction unit work process comprises:

s422, randomly determining d time periods in which values need to be changed, and recording as y ═ y₁,y₂…y_r…y_d](1≤r≤d,1≤y_r≤96)，r＝1；

S423 from y_rRandomly extracting a static scene from a static scene set corresponding to a time interval as a sequence scene y in the current solution_rThe value of the time interval is changed, if y is more than or equal to 1_rIf not more than 95, turning to the step S424, otherwise, turning to the step S425;

s424, calculating y_rTime period of-1, y_rTime period and y_rState corresponding to wind power output in +1 time period

And

note that in the state transition matrix Q

Go to the first

Column corresponding elements of

Note that in the state transition matrix Q

Go to the first

Column corresponding elements of

If it is

And

if not, turning to step S426, otherwise, turning to step S423;

s425, calculating y_r-1 period and y_rState corresponding to wind power output in time interval

If it is

If not, go to step S426, otherwise go to step S423;

s426, if r is equal to r +1, go to step 423 if r is equal to or less than d, otherwise go to step 427;

and S427, judging whether the neighborhood scene is repeated with the sequence scene in the tabu table, if so, turning to the step 423, otherwise, finishing the construction of the neighborhood scene of the sequence scene.

10. The generation system according to claim 1, wherein the fitness function building unit work process comprises:

using formulas

And calculating the distance between any two sequence scenes in the candidate solution, wherein,

λⁱand λ^j(i is more than or equal to 1, and j is more than or equal to S) are respectively the ith sequence scene and the jth sequence scene in the candidate solution;

using formulas

And calculating the autocorrelation coefficient of any sequence scene in the candidate solution, wherein,

A^kfor the autocorrelation coefficient of the kth sequence scene in the candidate solution, c and v are covariance and variance, respectively, lambda^k _e(e is more than or equal to 1 and less than or equal to 96) is an element of the kth row and the e column of the candidate solution lambda;

using formulas

And calculating a fitness function f of the candidate solution, wherein S is the scale of the sequence scene set in the candidate solution.

Images