A scene generation method based on transverse-longitudinal fusion photovoltaic sunrise power sequence

Abstract

The application discloses a kind of based on horizontal longitudinal fusion photovoltaic sunrise power sequence scene generation method, it is related to new energy system production simulation technical field, solve the technical problem that accurate and reasonable new energy power sequence scene cannot be generated, its technical scheme main point is to state transition matrix is constructed with day as unit instead of with record period (15 minutes) as unit, can greatly reduce the number of Markov chain rolling and reduce the accumulation of error;Radiation index series and corresponding power scene are generated from two dimensions of horizontal and vertical, and through information fusion, optimization technology and regression analysis, obtain 96 point daily power sequence scene of each day in future one month, more fully utilize the random characteristics contained in historical data, so that the random characteristics of the generated scene also more comply with the statistical law of historical data, can be widely applied to the scene generation of new energy power sequence.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of new energy system production simulation, and particularly relates to a photovoltaic daily output sequence scene generation method based on horizontal and vertical fusion. BACKGROUND

[0002] With the comprehensive implementation of the "double carbon target" policy, the construction of new power systems, as an important carrier of this policy, will also be fully and rapidly promoted. Accordingly, there will be more and more new energy connected to the power grid in the provinces, cities and regions of the Southern Power Grid, so that the consumption of fossil energy in the power grid will be greatly reduced and renewable energy dominated by hydropower, photovoltaic and wind power. The available output of renewable energy depends on weather conditions, natural environment, etc., and has natural volatility and intermittency. The load demand is subject to high penetration of distributed generation, popularization of electric vehicles and load demand response, so that its uncertainty is greatly improved. Thus, it brings great challenges to the medium and long-term power balance of the power grid. In the new power system environment, for uncertain variables, the probability of a specific new energy deterministic output or load demand is zero, and the power balance made under the traditional deterministic conditions often deviates greatly and has lost its significance, so it is necessary to make power balance calculation and mode formulation considering uncertainty. How to accurately and reasonably generate new energy output sequence scenes is a problem that needs to be solved in power generation planning, dispatching and optimization of the power grid. SUMMARY

[0003] The present application provides a photovoltaic daily output sequence scene generation method based on horizontal and vertical fusion, which aims to generate photovoltaic output sequence in a future period of time, so as to provide suggestions for power generation planning, dispatching and optimization of the power grid.

[0004] The above technical purpose of the present application is realized by the following technical scheme:

[0005] A photovoltaic daily output sequence scene generation method based on horizontal and vertical fusion, comprising:

[0006] S1: constructing a daily irradiance feature index sequence of historical data by using daily irradiance records;

[0007] S2: dividing the historical data into extreme days and a plurality of different types of typical days according to days by statistical analysis and clustering calculation on the daily irradiance feature index sequence, to obtain historical daily types of historical daily irradiance;

[0008] S3: constructing a horizontal daily irradiance feature index sequence and a longitudinal daily irradiance feature index sequence from two dimensions of horizontal and longitudinal directions, constructing a daily irradiance daily type sequence of each day in a future month according to historical daily types and the horizontal daily irradiance feature index sequence and the longitudinal daily irradiance feature index sequence, and obtaining a horizontal daily irradiance daily type sequence and a longitudinal daily irradiance daily type sequence;

[0009] S4: obtaining a joint probability distribution function of the horizontal daily irradiance daily type sequence and the longitudinal daily irradiance daily type sequence through kernel density estimation and a Copula function, obtaining a daily irradiance feature index through random sampling according to the joint probability distribution function, obtaining a horizontal daily irradiance daily type sequence scene and a longitudinal daily irradiance daily type sequence scene of the future month according to the daily irradiance feature index, and performing information fusion, optimization and regression analysis on the horizontal daily irradiance daily type sequence scene and the longitudinal daily irradiance daily type sequence scene to obtain a 96-point photovoltaic daily output sequence scene of each day in the future month.

[0010] The application has the following advantages: (1) the application constructs a state transition matrix in units of days instead of recording periods (15 minutes), which can greatly reduce the number of Markov chain rolling and the accumulation of errors; (2) the application uses daily irradiance feature indexes to represent 96-point daily irradiance sequences, which improves the accuracy of clustering analysis and the accuracy of the generated scene; (3) the application generates irradiance index series and corresponding output scenes from two dimensions of horizontal and longitudinal directions, and fuses to generate the final scene, which more fully utilizes the random characteristics contained in the historical data, so that the random characteristics of the generated scene are more in line with the statistical rules of the historical data. The daily irradiance and daily output data of a photovoltaic power plant in the Southern Power Grid are used in the example part to calculate and generate 96-point photovoltaic daily output sequence scenes for the next 30 days, and are compared with the traditional Markov chain scene generation method (hereinafter referred to as the comparative method) to verify the effectiveness and advancement of the method. The example results show that the photovoltaic power plant daily output sequence scenes generated based on the horizontal and longitudinal fusion are very close to the historical output in terms of probability distribution characteristics, monthly characteristic output indicators, etc., proving the effectiveness thereof. BRIEF DESCRIPTION OF DRAWINGS

[0011] Figure 1 a flowchart of the method described in the application;

[0012] Figure 2 a flowchart of the algorithm of the application;

[0013] Figure 3 a PDF, CDF and historical data PDF, CDF curve diagram of the daily output sequence scenes generated by the method and the comparative method;

[0014] Figure 4Fig. 1 is a schematic diagram of ACF results of the method, comparative method and historical data of the present application. DETAILED DESCRIPTION

[0015] The technical solutions of the present application will be described in detail below with reference to the accompanying drawings.

[0016] As shown in Figure 1 and Figure 2 The scene generation method based on horizontal and vertical fusion photovoltaic daily power sequence of the present application comprises:

[0017] S1: constructing a daily irradiance feature index sequence of historical data by using daily irradiance records;

[0018] S2: dividing the historical data into extreme days and a plurality of different types of typical days according to statistical analysis and clustering calculation on the daily irradiance feature index sequence, and obtaining the historical daily type of the daily irradiance of the historical data;

[0019] S3: constructing a horizontal daily irradiance feature index sequence and a vertical daily irradiance feature index sequence from two dimensions of horizontal and vertical, constructing a daily irradiance daily type sequence of each day in the future one month according to the historical daily type and the horizontal daily irradiance feature index sequence and the vertical daily irradiance feature index sequence, and obtaining a horizontal daily irradiance daily type sequence and a vertical daily irradiance daily type sequence;

[0020] S4: obtaining a joint probability distribution function of the horizontal daily irradiance daily type sequence and the vertical daily irradiance daily type sequence by kernel density estimation and Copula function, obtaining daily irradiance feature indexes by random sampling according to the joint probability distribution function, obtaining a horizontal daily irradiance daily type sequence scene and a vertical daily irradiance daily type sequence scene of the future one month according to the daily irradiance feature indexes, performing information fusion, optimization and regression analysis on the horizontal daily irradiance daily type sequence scene and the vertical daily irradiance daily type sequence scene, and obtaining a 96-point photovoltaic daily power sequence scene of each day in the future one month.

[0021] As a specific embodiment, the daily irradiance feature index sequence of the historical data in step S1 is represented as: wherein, represents the daily irradiance mean of the nth day; σ n represents the daily irradiance variance of the nth day; X max,n represents the daily maximum irradiance of the nth day; X min,n represents the daily minimum irradiance of the nth day.

[0022] As a specific embodiment, the step S2 comprises:

[0023] S21: According to the sequence of daily irradiance characteristic indexes of historical data, the historical data is divided into a sequence of extreme days of sunshine amplitude and a sequence of typical days of sunshine amplitude, denoted as an extreme day sequence E-Days and a typical day sequence T-Days, including:

[0024] S211: The expected value and the variance value of the sequence of daily irradiance characteristic indexes of historical data are calculated, and the value obtained by subtracting 1.5 times the variance value from the expected value is the discrimination threshold value of the typical day and the extreme day;

[0025] S212: The daily type is judged according to the discrimination threshold value, if the mean value of the daily sunshine amplitude is greater than or equal to the discrimination threshold value, the day is a typical day, otherwise the day is an extreme day, so as to obtain the sequence of extreme day irradiance characteristic indexes E-Days and the sequence of typical day irradiance characteristic indexes T-Days, denoted as:

[0026]

[0027] Wherein, The mean value of the daily irradiance of the a-th day is denoted as The expected value of the sequence of daily irradiance characteristic indexes of historical data is denoted as X The variance value of the sequence of daily irradiance characteristic indexes of historical data is denoted as

[0028] S22: The K-means clustering method is used to cluster and divide the daily irradiance characteristic indexes of the typical days in the typical day sequence T-Days, the typical day sequence T-Days is subdivided into typical days T-Days-1, T-Days-2, …, T-Days-k*, and combined with the extreme day, to obtain k*+1 daily types of historical data irradiance including T-Days-1, T-Days-2, …, T-Days-k* and the extreme day, including:

[0029] S221: From the sequence of typical day irradiance characteristic indexes T-Days={Day1,Day2,...,Day T}, k daily irradiance characteristic index samples {Day1,Day2,...,Day k} are randomly extracted, and the k daily irradiance characteristic index samples {Day1,Day2,...,Day k} form an initial k typical day type set S j ={Day j}, j=1,2,...,k, the clustering center of S j is C j , and C j =Day j ;

[0030] S222: For {Day1, Day2, ..., Day...} T The i-th sample Day in} i With typical day type set S j The distance is calculated and assigned to the nearest set S. j In the middle, then S j The update is represented as:

[0031]

[0032] in, D(Day i C j ) represents the sample Day i With C j The distance between them;

[0033] S223: Regarding S' j The cluster centers are calculated and represented as follows:

[0034]

[0035] Among them, Day j,i' Describe the set S' j The i'th sample included in the sample. S' j The number of samples included;

[0036] S224: Repeat steps S222 and S223 until the cluster centers remain unchanged, output the clustering results with cluster number k, and go to step S225;

[0037] S225: The silhouette coefficient SC of the clustering results when the number of clusters is k is calculated and expressed as follows:

[0038]

[0039] Among them, C i Day i The cluster center of the set to which it belongs; C r Day r The cluster center of the set to which it belongs, r≠i; T represents the number of typical days;

[0040] S226: k = k + 1; if k = 9, the calculation ends. K opt This indicates that the corresponding day type classification is the optimal classification, and the corresponding optimal cluster number is k. * Otherwise, proceed to step S221 until the historical daily irradiance k is obtained. * +1 historical day type, this k* +1 historical day types include T-Days-1, T-Days-2, …, T-Days-k* and extreme days.

[0041] As a specific embodiment, the step S3 comprises:

[0042] S31: taking a sequence of daily irradiance feature indicators Days X = {Day1, Day2, …, Day3650} of 3650 days in 10 consecutive years as a sequence of horizontal daily irradiance feature indicators, performing daily type division on the sequence of horizontal daily irradiance feature indicators by using the historical day types to obtain horizontal daily types of the sequence of horizontal daily irradiance feature indicators, and all the horizontal daily types constitute a sequence of horizontal daily irradiance daily types. N X = {Day1, Day2, …, Day3650} of 3650 days in 10 consecutive years as a sequence of horizontal daily irradiance feature indicators, performing daily type division on the sequence of horizontal daily irradiance feature indicators by using the historical day types to obtain horizontal daily types of the sequence of horizontal daily irradiance feature indicators, and all the horizontal daily types constitute a sequence of horizontal daily irradiance daily types.

[0043] Specifically, after obtaining the daily types of the sequence of horizontal daily irradiance feature indicators, each daily feature indicator is expanded to 5 dimensions, and the 5th dimension is the daily type of the day, to obtain a sequence of horizontal daily irradiance feature indicators including daily types Days' N = {Day1, Day2, …, Day3650} of 3650 days in 10 consecutive years as a sequence of horizontal daily irradiance feature indicators, performing daily type division on the sequence of horizontal daily irradiance feature indicators by using the historical day types to obtain horizontal daily types of the sequence of horizontal daily irradiance feature indicators, and all the horizontal daily types constitute a sequence of horizontal daily irradiance daily types. n,5 = u, u ∈ {1, 2, …, k * +1}, u represents the n-th daily type. * +1}, u represents the n-th daily type.

[0044] S32: connecting the sequences of daily irradiance feature indicators of the e-th month in 10 consecutive years in a chronological order to form a sequence of longitudinal daily irradiance feature indicators Days Y = {Day1, Day2, …, Day3650} of 3650 days in 10 consecutive years as a sequence of horizontal daily irradiance feature indicators, performing daily type division on the sequence of horizontal daily irradiance feature indicators by using the historical day types to obtain horizontal daily types of the sequence of horizontal daily irradiance feature indicators, and all the horizontal daily types constitute a sequence of horizontal daily irradiance daily types. Z = {Day1, Day2, …, Day3650} of 3650 days in 10 consecutive years as a sequence of horizontal daily irradiance feature indicators, performing daily type division on the sequence of horizontal daily irradiance feature indicators by using the historical day types to obtain horizontal daily types of the sequence of horizontal daily irradiance feature indicators, and all the horizontal daily types constitute a sequence of horizontal daily irradiance daily types.

[0045] Similarly, after obtaining the daily types of the sequence of longitudinal daily irradiance feature indicators, each daily feature indicator is expanded to 5 dimensions, and the 5th dimension is the daily type of the day, to obtain a sequence of longitudinal daily irradiance feature indicators including daily types Days' Y = {Day1, Day2, …, Day3650} of 3650 days in 10 consecutive years as a sequence of horizontal daily irradiance feature indicators, performing daily type division on the sequence of horizontal daily irradiance feature indicators by using the historical day types to obtain horizontal daily types of the sequence of horizontal daily irradiance feature indicators, and all the horizontal daily types constitute a sequence of horizontal daily irradiance daily types. Z = h, h ∈ {1, 2, …, k Z,5 +1}, h represents the z-th daily type. * +1}, h represents the z-th daily type. * +1}, h represents the z-th daily type.

[0046] ​S33: The horizontal daily irradiance daily type sequence is processed by superimposed Markov chains to generate the horizontal daily type sequence scene for each day in the next month, and the vertical daily irradiance daily type sequence is processed by conditional transition to generate the vertical daily type sequence scene for each day in the e-th month of the next year.

[0047] In a specific embodiment, step S33 involves using a superimposed Markov chain to generate a daily horizontal irradiance sequence for the next month, including:

[0048] S3311: Construct the state transition probability matrix P of the Markov chain with step sizes l of 1, 2, and 3. 1 P 2 P 3 Then P 1 P 2 P 3 elements in They are represented as follows:

[0049]

[0050]

[0051]

[0052] Where p,q=1,2,...,k * +1; This represents the total number of events in historical data that transition from day type p to day type q via step size l. This represents the event on day n where day type p is transferred to day type q via step size l.

[0053] S3312: Using the day types of day n, day n-1, and day n-2 as the initial day types, according to the state transition probability matrix P... 1 P 2 P 3 The day type for the (n+1)th future day is determined by sampling, including:

[0054] S33121: Denote the day type of day n, day n-1, and day n-2 as follows: and The day type is divided into and The row vectors representing the transition probabilities from day n, day n-1, and day n-2 to day n+1 are denoted as follows: and Then the probability row vector of each day type appearing on day n+1 (composed of the corresponding probabilities of each day type) is represented as: in, P represents1 The row vectors P represents 2 The row vectors P represents 3 The Row vectors;

[0055] S33122: Using a roulette wheel, a random number ω following a uniform distribution is randomly selected from [0,1]. If ω≤η n+1,1 This means that ω falls within the first probability interval, Day N+1+n,5 =1; if This means that ω falls within the (i+1)th probability interval, Day N+1+n,5 =i”+1; among them, η n+1,u Indicates η n+1 The u-th element; i”∈{1,2,...,k *};

[0056] S3313: n = n + 1. If n = 30, the calculation ends; otherwise, proceed to step S3312 until the horizontal daily type sequence scene Days-y = {Day N+1,5 Day N+2,5 ,...,Day N+30,5};

[0057] In step S33, the vertical daily irradiance daily type sequence scenario for the next year is generated through a conditional transition scenario, including:

[0058] S3321: The longitudinal daily irradiance daily type sequence includes one extreme daily type and k' typical daily types, and the frequency |S| in each daily type cluster is obtained. k | The frequency of each day type |S k Divide by the sequence length Z to obtain the frequency of each day type. With frequency Construct ZDay for probability z,5 If the cumulative probability distribution function T(x) is given, then T(x) can be expressed as:

[0059]

[0060] Where z = 1,...,Z; k = 1,...,k'+1;

[0061] S3322: Referring to the horizontal scene generation method, construct the allowed state transition matrix Q of the Markov chain with a step size of 1. 1 Then Q1 elements in Represented as:

[0062]

[0063] z≠30x,x=1,...,10;z=1,...,Z;

[0064]

[0065]

[0066] in, This indicates that day Z is a day type p' and transitions to day type q' with a step size of 1; x represents the indicator variable for the last day of month e in year 11-x; z≠30x represents any day type transition event in the sequence where the last day of month e of each year connects to the first day of month e of the following year; n' p',q' This represents the total number of events that transition from historical day type p' to historical day type q' via a step size of 1; n' p',q' / ∑ q' n' p',q' Let n' represent the probability of an event that transitions from day type p' to day type q', where n' is the probability of such an event. p',q' / ∑ q' n' p',q' If ≥γ, it indicates that the state transition from day type p' to day type q' is allowed, and the corresponding... Otherwise, it means that the state transition from type p' to type q' is not allowed, and the corresponding... Meanwhile, let n = 0; where γ represents the minimum probability of a given transition.

[0067] S3323: Randomly select a number ω' that follows a uniform distribution within [0,1]. If Let p' = 0, and the initial day type for the (n+1)th day be p'+1; if Then the initial day type for the (n+1)th day is p'+1; where p'∈{1,2,...,k'};

[0068] S3324: If n = 0, let the actual day type of the (n+1)th day be the initial day type, i.e., ZDay. Z+1,5 =p'+1, which gives the day type for the first day. Proceed to step S3325.

[0069] If n > 0, according to the allowed state transition matrix Q 1 Determine the day type ZDay for day n. Z+n,5 Allowable transition states up to the initial day type p'+1 on day n+1 like ZDay typeZ+n,5 The transition to the initial day type p'+1 is allowed, and the actual day type of the (n+1)th day is the initial day type, i.e., ZDay. Z+n+1,5 =p'+1; if ZDay type Z+n,5 The state transition to the initial day type p′+1 is not allowed, and resampling is required, returning to step S3322;

[0070] S3325: n = n + 1. If n = 30, the calculation ends; otherwise, proceed to step S3322 until the vertical daily type sequence scene ZDays-y = {ZDay} for each day in the e-th month of the next year is obtained. Z+1,5 ZDay Z+2,5 ,...,ZDay Z+30,5}

[0071] In a specific embodiment, step S4 includes:

[0072] S41: k in the horizontal and vertical daily irradiance type sequences * +1 day type and corresponding k * +1 dataset, using kernel density estimation to obtain the marginal distribution of four feature indicators for each daily type. F2(σ), F3(X) max ) and F4(X min ).

[0073] S42: By concatenating the marginal distributions of the four feature indicators using the copula function, we obtain the joint probability distribution function of the four feature indicators, thus obtaining k. * +1 joint probability distribution functions

[0074] S43: Regarding the joint probability distribution function Calculate the partial derivative to obtain the conditional probability cumulative distribution, and obtain the solar radiation amplitude characteristic index under the corresponding day type through random sampling. Based on the solar radiation amplitude characteristic index, obtain the horizontal solar radiation characteristic index sequence scene and the vertical solar radiation characteristic index sequence scene for the next month.

[0075] Specifically, step S43 includes:

[0076] S431: Joint probability distribution function for given day types in the horizontal and vertical solar irradiance day type sequences. Find the partial differential to obtain the definite value. The conditional cumulative distribution function of σ Sure and σ after X maxthe conditional probability cumulative distribution function of and determining σ and X max post-X min the conditional probability cumulative distribution function of

[0077] S432: four random numbers ω1, ω2, ω3 and ω4 obeying uniform distribution are randomly extracted within [0, 1]; let The four equations are sequentially solved to obtain the sampling values of the four daily irradiance characteristic indicators of a given day type, thereby obtaining the horizontal daily irradiance characteristic indicator sequence scene and the vertical daily irradiance characteristic indicator sequence scene of the future month, respectively denoted as: wherein, represents the horizontal daily irradiance characteristic indicator sequence scene; represents the vertical daily irradiance characteristic indicator sequence scene; H p”,q” = Day N+p”,q” , Z p”,q” = ZDay N+p”,q” , p" = 1, 2,..., 30, q" = 1, 2, 3, 4 represent the four dimensions of the characteristic indicators.

[0078] S44: the horizontal daily irradiance characteristic indicator sequence scene and the vertical daily irradiance characteristic indicator sequence scene of the future month are fused based on the Kalman gain to obtain a fused daily irradiance characteristic indicator sequence scene, including: let the optimal daily irradiance characteristic indicator vector of the i-th day be R i , then R i = H i + α (H i - Z i ), and the fused daily irradiance characteristic indicator sequence scene after fusion by and is wherein, represents the Kalman gain.

[0079] S45: the daily irradiance characteristic indicator vector of each day in the future month is optimized and regression analyzed according to the fused daily irradiance characteristic indicator sequence scene, thereby generating the 96-point daily power sequence scene of each day in the future month.

[0080] Specifically, the step S45 includes:

[0081] If the daily irradiance characteristic indicator sequence of the historical data is the corresponding historical daily irradiance sequence is wherein, IDay n = (I n,1 , In,2 ,...,I n,96 ) T ,I n,t denotes the daily irradiance value at the nth day and the tth time, t = 1, 2,..., 96.

[0082] S451: Selects the daily irradiance feature indicators from the sequence of daily irradiance feature indicators of historical data and the sequence of fused daily irradiance feature indicators to generate the nth day feature indicators R n Most similar daily irradiance sequence IDay m as the reference sequence a(t), and the 96-point daily irradiance sequence scene to be generated is Z n' (t); wherein,

[0083] S452: Uses an optimization method to make the four feature indicators of Z n' (t) respectively equal to σ' n , X' n,max and X' n,min ;

[0084] S453: Converts the 96-point daily irradiance sequence scene Z n' (t) of each day in the next 30 days into a 96-point photovoltaic daily power sequence scene P n',t by a polynomial regression method; wherein, t = 1, 2,..., 96; n' = 1, 2,..., 30.

[0085] Specifically, step S453 includes:

[0086] (1) According to the historical daily irradiance value I n,t and the photovoltaic power sequence P n',t , a m-degree polynomial regression model of daily irradiance and photovoltaic daily power is constructed, denoted as: P n',t = G m (I n,t ); wherein, the monomial m-degree polynomial regression equation is

[0087]

[0088] (2) The cumulative error SS m of the monomial m-degree polynomial regression result is calculated, denoted as:

[0089]

[0090] (3) m = m + 1; if m = 9, the calculation is ended, and m opt= MinSS m ; m opt The corresponding monomial polynomial regression of order m is the optimal polynomial regression; otherwise, go to step (1);

[0091] (4) Obtain the optimal G m (I n,t , and substitute the 96-point daily irradiance sequence scene Z n' (t) of each day in the next 30 days, i.e. the 96-point photovoltaic daily power sequence scene in the next 30 days can be obtained.

[0092] In summary, the daily irradiance and daily power data of a certain photovoltaic power plant of the Southern Power Grid are used to calculate and generate the 96-point photovoltaic daily power sequence scene in the next 30 days, and the traditional Markov chain scene generation method (hereinafter referred to as the comparative method) is used for comparison.

[0093] According to the above method, the comparison results of the generated sequence scene are shown in Figures 3-4 and Tables 1-3, wherein, Figure 3 is the PDF, CDF and PDF, CDF curve diagram of the historical data itself of the daily power sequence scene generated by the method and the comparative method; Figure 4 is the ACF result diagram of the method, the comparative method and the historical data; Table 1 is the RSS calculation result comparison of the 96-point photovoltaic daily power sequence scene of each day in the next 30 days by the method and the comparative method, Table 2 is the monthly average power and extreme scene occurrence ratio comparison of the power sequence generated by the method and the comparative method in the next month, and Table 3 is the daily difference distribution result comparison of the power sequence, the power sequence generated by the method and the comparative method.

[0094] The above is a demonstrative embodiment of the present application, and the protection scope of the present application is defined by the claims and their equivalents.

[0095] Table 1

[0096] PDF CDF ACF The method herein 9.23e-04 6.45e-03 2.15e-02 The comparative method 1.73e-3 1.1e-02 2.07

[0097] Table 2

[0098]

[0099] Table 3

[0100]

[0101] The above is a demonstrative embodiment of the present application, and the protection scope of the present application is defined by the claims and their equivalents.

Claims

1. A method for generating a scene based on a transverse-longitudinal fusion photovoltaic power sequence, characterized in that, The method comprises the following steps: S1: constructing a historical data daily irradiance feature index sequence by using a daily irradiance record; S2: dividing the historical data into an extreme day and a plurality of different types of typical days according to statistical analysis and clustering calculation on the daily irradiance feature index sequence, and obtaining a historical day type of the historical data daily irradiance; S3: constructing a horizontal daily irradiance feature index sequence and a vertical daily irradiance feature index sequence from two dimensions of horizontal and vertical, constructing a daily irradiance day type sequence of each day in a future month according to the historical day type and the horizontal daily irradiance feature index sequence and the vertical daily irradiance feature index sequence, and obtaining a horizontal daily irradiance day type sequence and a vertical daily irradiance day type sequence; S4: obtaining a joint probability distribution function of the horizontal daily irradiance day type sequence and the vertical daily irradiance day type sequence by kernel density estimation and Copula function, obtaining a daily irradiance feature index according to the joint probability distribution function and by random sampling, obtaining a horizontal daily irradiance day type sequence scene and a vertical daily irradiance day type sequence scene of the future month according to the daily irradiance feature index, performing information fusion, optimization and regression analysis on the horizontal daily irradiance day type sequence scene and the vertical daily irradiance day type sequence scene, and obtaining a 96-point photovoltaic daily output sequence scene of each day in the future month.

2. The method of claim 1, wherein, In step S1, the daily irradiance characteristic index sequence of the historical data is represented as: wherein, represents the daily irradiance mean value of the nth day; σ n represents the daily irradiance variance of the nth day; X max，n represents the daily maximum irradiance of the nth day; X min，n represents the daily minimum irradiance of the nth day.

3. The method of claim 1, wherein, The step S2 comprises: S21: dividing the historical data into a sunshine amplitude extreme day sequence and a sunshine amplitude typical day sequence according to the historical data daily irradiance feature index sequence, and denoting the extreme day sequence and the typical day sequence as E-Days and T-Days, respectively; S22: using K-moans clustering method to cluster and divide the daily irradiance characteristic indexes of the typical days in the typical day sequence T-Days, and subdividing the typical day sequence T-Days into typical days T-Days-1, T-Days-2, …, T-Days-k*, and then combining the extreme days to obtain k * +1 day types including T-Days-1, T-Days-2, …, T-Days-k* and the extreme days.

4. The method of claim 3, wherein, The step S21 comprises: S211: calculating an expected value and a variance value of the historical data daily irradiance feature index sequence, and obtaining a value obtained by subtracting 1.5 times the variance value from the expected value as a discrimination threshold value of the typical day and the extreme day; S212: judging the daily type according to the discrimination threshold value, if a mean value of the daily sunshine amplitude is greater than or equal to the discrimination threshold value, the day is a typical day, otherwise, the day is an extreme day, thereby obtaining an extreme day daily irradiance feature index sequence E-Days and a typical day daily irradiance feature index sequence T-Days, and denoted as: wherein, represents the daily irradiance mean value of the day a; represents the expected value of the daily irradiance characteristic indicator sequence of the historical data; σ X represents the variance value of the daily irradiance characteristic indicator sequence of the historical data.

5. The method of claim 4, wherein, The step S22 comprises: S221: From the typical daily irradiance characteristic index sequence T-Days={Day1, Day2, ..., Day...} T Randomly select k daily irradiance characteristic index samples {Day1, Day2, ..., Day} from the data. k }, for the k daily irradiance characteristic index samples {Day1, Day2, ..., Day k } This constitutes the initial set S of k typical day types. j ={Day j }, j = 1, 2, ..., k, S j The cluster center is C j And C j =Day j ; S222: Calculate the distance between the i-th sample Day T in {Day1, Day2,..., Day i and the set of typical day types S j and put the distance into the nearest set S j , then S j is updated as follows: wherein, D(Day i , C j ) denotes the distance between sample Day i and C j ; S223: Regarding S′ j The cluster centers are calculated and represented as follows: where Day j，i′ represents the i'th sample included in the set S' j represents the number of samples included in S' j ​​ S224: repeating the step S222 and the step S223 until the clustering center does not change, outputting a clustering result of the clustering number k, and turning to the step S225; S225: calculating a silhouette coefficient SC of the clustering result when the clustering number is k, and denoted as: where C i represents Day i the cluster center of the belonging set; C r represents Day r the cluster center of the belonging set, r≠i; T represents the number of typical days; S226: k=k+1; if k=9, the calculation is finished, K opt indicates that the corresponding day type classification is the optimal classification, and the optimal cluster number obtained is k * ; otherwise, go to step S221 until k * +1 historical day types of the historical data daily irradiance are obtained, and the k * +1 historical day types include T-Days-1, T-Days-2, …, T-Days-k* and the extreme day.

6. The method of claim 1, wherein, The step S3 comprises: S31: A sequence of daily irradiance characteristics over 10 consecutive years (3650 days) Days X = {Day1, Day2, ..., Day} N } is a sequence of horizontal daily irradiance characteristic indicators. The horizontal daily irradiance characteristic indicator sequence is divided into daily types according to historical daily types to obtain the horizontal daily types of the horizontal daily irradiance characteristic indicator sequence. All horizontal daily types constitute the horizontal daily irradiance daily type sequence. S32: connecting the sequence of the e-th month's daily irradiance feature indicators of the consecutive 10 years in the order of years to form a longitudinal daily irradiance feature indicator sequence Days y = {Day1, Day2,..., Day z}, dividing the longitudinal daily irradiance feature indicator sequence by the historical daily type to obtain longitudinal daily types of the longitudinal daily irradiance feature indicator sequence, and all the longitudinal daily types constitute a longitudinal daily irradiance daily type sequence; wherein e also represents the month for which the scene needs to be generated in the future; S33: generating a horizontal daily type sequence scene of each day in the future month by using a superimposed Markov chain on the horizontal daily irradiance day type sequence, and generating a vertical daily type sequence scene of each day in the e-th month of the future year by using a conditional transition scene on the vertical daily irradiance day type sequence.

7. The method of claim 6, wherein, In the step S33, the horizontal daily type sequence scene of each day in the future month is generated by using the superimposed Markov chain on the horizontal daily irradiance day type sequence, comprising: S3311: Construct the state transition probability matrix P of the Markov chain with step sizes l of 1, 2, and 3. 1 P 2 P 3 Then P 1 P 2 P 3 elements in They are represented as follows: where p, q = 1, 2,..., k * +1; represents the total number of events in the historical data in which day type p moves to day type q by step length l, represents the number of events in which day type p moves to day type q by step length l on the nth day. S3312: Taking the day type of the n-th day, the n-1-th day and the n-2-th day as initial day types, determining the day type of the future n+1-th day according to the state transition probability matrix P 1 2 3 the day type of the future n+1-th day is determined by sampling, comprising:​​ S33121: The day types for day n, day n-1, and day n-2 are respectively denoted as... and The day type is divided into and The row vectors representing the transition probabilities from day n, day n-1, and day n-2 to day n+1 are denoted as follows: and The probability row vector of each day type appearing on day n+1 is represented as follows: in, P represents 1 The row vectors P represents 2 The row vectors P represents 3 The Row vectors; S33122: A random number ω obeying uniform distribution in [0, 1] is randomly drawn by roulette method, if ω≤η n+1，1 , it indicates that ω falls in the first probability interval, Day N+1+n，5 =1; if , it indicates that ω falls in the i"+1th probability interval, Day N+1+n，5 =i"+1; wherein, η n+1，u represents the u-th element of η n+1 ; i"∈{1, 2,..., k *} S3313: n = n + 1. If n = 30, the calculation ends; otherwise, proceed to step S3312 until the horizontal daily type sequence scene Days-y = {Day N+1，5 Day N+2，5 Day N+30，5 }; In step S33, the longitudinal daily irradiance daily type sequence is used to generate the daily longitudinal daily type sequence scene in the e th month of the future year through the conditional transfer scenario, including: S3321: the longitudinal day type of the longitudinal day irradiance day type sequence includes one extreme day type and k' typical day day types, and the frequency |S k of each day type cluster is obtained k |S , divide the frequency of each day type by the sequence length Z to obtain the frequency of each day type ，5 , and the cumulative probability distribution function T(x) of ZDayz ，5 is constructed with the frequency as the probability, and T(x) is expressed as: Wherein, z = 1,..., Z; k = 1,..., k' + 1; S3322: Constructing the state transition matrix Q allowed by the Markov chain with a step size of 1 1 The elements in Q 1 The elements in Q is expressed as: z≠30x, x = 1,..., 10; z = 1,..., Z; wherein, represents an event that day type p' on day Z is shifted to day type q' by a step of 1; x represents an indicator variable for the last day of the e-th month of the x-th year; z≠30x represents any day type shift event that connects the last day of the e-th month of a year to the first day of the e-th month of the next year; n' p′，q′ represents the total number of events that historical day type p' is shifted to historical day type q' by a step of 1; n' p′，q′ / ∑ q′ n p′，q′ represents the probability that an event occurs that shifts day type p' to day type q', if n' p′，q′ / ∑ q′ n' p′，q′ ≥γ, then it represents that day type p' to day type q' is an allowable state transition, and correspondingly otherwise it represents that day type p' to day type q' is an unallowable state transition, and correspondingly while n = 0; wherein γ represents a minimum probability of allowable transition given beforehand; S3323: Randomly draw a number ω' obeying uniform distribution in [0, 1], if then let p' = 0, and the initial day type of the (n+1)th day is p'+1; if then the initial day type of the (n+1)th day is p'+1; wherein, p' ∈ {1, 2,..., k'}; S3324: If n = 0, let the actual day type of the (n+1)th day be the initial day type, i.e. ZDay = ZDay0 Z+1，5 = p' + 1, and the day type of the first day is obtained, go to step S3325. If n > 0, determine the day type ZDay on day n according to the allowed state transition matrix Q1 Z+n，5 allowed transition state to the initial day type p'+1 on day n+1 If then the day type ZDay Z+n，5 The transition to the initial day type p'+1 is allowed, and the actual day type on day n+1 is the initial day type, i.e. ZDay Z+n+1，5 = p'+1; if then the day type ZDay Z+n，5 The state transition to the initial day type p'+1 is not allowed, and resampling is needed, returning to step S3322; S3325: n = n + 1. If n = 30, the calculation ends; otherwise, proceed to step S3322 until the vertical daily type sequence scene ZDays-y = {ZDay} for each day in the e-th month of the next year is obtained. Z+1，5 ZDay Z+2，s ..., ZDay Z+30，5 } 8. The method of claim 4, wherein, Step S4 includes: S41: k * +1 day types and corresponding k * +1 data sets, the edge distribution of the 4 feature indicators of each day type is obtained by kernel density estimation F2(σ), F3(X max ) and F4(X min ); S42: Connect the marginal distribution of the four feature indicators as the joint probability distribution function of the four feature indicators through the copula function, and a total of k * +1 joint probability distribution functions F2(σ), F3(X max ), F4(X min )); S43: obtaining the conditional probability distribution function F2(σ), F3(X max ), F4(X min )) to obtain the conditional probability cumulative distribution, and through random sampling, obtaining the sunshine amplitude characteristic index under the corresponding day type, and obtaining the future one-month horizontal daily irradiance characteristic index sequence scene and vertical daily irradiance characteristic index sequence scene according to the sunshine amplitude characteristic index. S44: based on the Kalman gain, the transverse daily irradiance feature index sequence scene and the longitudinal daily irradiance feature index sequence scene in the future one month are fused to obtain a fused daily irradiance feature index sequence scene; S45: according to the fused daily irradiance feature index sequence scene, the daily irradiance feature index vector in each day in the future one month is optimized and regression analyzed, so as to generate a 96-point daily output sequence scene in each day in the future one month.

9. The method of claim 8, wherein, The step S43 includes: S431: Joint probability distribution function for given day types in the horizontal and vertical solar irradiance day type sequences. F2(σ), F3(X) max ), F4(X min Find the partial differential to obtain the definite value. The conditional cumulative distribution function of σ Sure and σ after X max conditional probability cumulative distribution function and determine σ and X max After X min conditional probability cumulative distribution function S432: four random numbers ω1, ω2, ω3 and ω4 obeying uniform distribution in [0, 1] are randomly extracted; let Sequentially solving the four equations can obtain the sampling values of the four daily irradiance characteristic indexes of a given day type, thereby obtaining the horizontal daily irradiance characteristic index sequence scene and the vertical daily irradiance characteristic index sequence scene of the next month, respectively denoted as: wherein, represents the horizontal daily irradiance characteristic index sequence scene; represents the vertical daily irradiance characteristic index sequence scene; H p″，q″ = Day N+p″，q″ , Z p″，q″ = ZDay N+p″，q″ , p" = 1, 2,..., 30, q" = 1, 2, 3, 4 represent the four dimensions of the characteristic indexes.

10. The method of claim 9, wherein, The step S45 includes: S451: Selecting the day irradiance feature index from the historical data day irradiance feature index sequence the day feature index R generated in the scene of the n-th day of the selected day irradiance feature index and the fused day irradiance feature index sequence n the most similar day irradiance feature sequence Day m As the reference sequence a(t), the 96-point day irradiance sequence scene to be generated is Z n′ (t); wherein, S452: using an optimization method to make Z n′ The four characteristic indexes of (t) are respectively equal to σ′ n , X′ n，max and X′ n，min ; S453: 96-point daily irradiance sequence scenario Z for each day of the next 30 days by polynomial regression n′ (t) converting to 96-point photovoltaic daily power sequence scenario P n′，t (t); where t = 1, 2,..., 96; n' = 1, 2,..., 30.

Images