Source-load time sequence matching typical scene identification and construction method

Abstract

The invention relates to a source load time sequence matching typical scene identification and construction method. The method comprises the following steps: step 1, a multi-time scale wind-light-load scene generation method of source load matching time sequence characteristics; 2, constructing a typical scene set and an extreme scene set oriented to supply and consumption; step 3, typical example analysis, including wind and light load characteristic analysis, typical scene set construction, extreme scene set construction and guarantee supply and consumption oriented scene analysis; the method has the advantages that multi-time-scale modeling is achieved, dynamic changes are better reflected, supply and consumption demands are effectively balanced, and stability and economical efficiency are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of power system operation and management, and specifically relates to a method for identifying and constructing a typical scenario of source-load timing matching. Background Art

[0002] At present, with the continuous improvement of the penetration rate of renewable energy in the power system, the demand for flexible scheduling and source-load matching of the power grid is becoming increasingly urgent. However, the existing power system management methods have the following shortcomings in terms of source-load timing matching and typical scenario construction: First, the existing typical scenario generation methods are mostly based on simple statistical analysis of historical data, and the generated scenarios lack sufficient representativeness and accuracy; the generation of typical scenarios usually only considers the state changes of key nodes, ignoring the time series characteristics of system load and multiple energy outputs, as well as the synergistic relationship between them. Especially in the case of a high proportion of renewable energy access, this simplified processing causes the generated typical scenarios to fail to fully reflect the complex situation brought about by the volatility of new energy, and it is difficult to meet the power system's demand for scheduling flexibility; in addition, the existing scenario generation methods are relatively extensive in terms of time and space precision control, and cannot effectively predict the subtle changes in load demand and new energy output, thus affecting the actual applicability of the scheduling scheme; Second, the existing source-load matching technologies mostly use fixed load forecasts and power generation plans, lacking in-depth analysis of the dynamic characteristics of sources and loads. This method is relatively effective when the penetration rate of new energy is low, but with the rapid development of renewable energy such as wind power and photovoltaics, the accuracy of source-load matching faces higher requirements; due to the intermittent and random nature of new energy output, the existing source-load matching methods are difficult to adapt to the timing mismatch between load demand and new energy output, and often require frequent adjustments to the output of the power generation side or the start-up of backup power supplies to balance the supply and demand of the power grid, resulting in increased system operating costs; at the same time, this static matching-based method is difficult to capture the fluctuation characteristics of new energy, and the actual effect of the source-load matching scheme deviates greatly from the plan, affecting the reliability and economy of the power grid; therefore, it is very necessary to provide a typical scenario identification and construction method for source-load timing matching that can model multiple time scales, better reflect dynamic changes, effectively balance supply and consumption needs, and improve stability and economy. Summary of the invention

[0003] 1. Technical issues

[0004] In view of the above-mentioned existing technical status, this application mainly addresses the following technical problems:

[0005] 1. Most existing typical scenario generation methods are based on simple statistical analysis of historical data, and the generated scenarios lack sufficient representativeness and accuracy;

[0006] 2. The existing scenario generation methods are relatively rough in terms of temporal and spatial precision control, and cannot effectively predict subtle changes in load demand and renewable energy output, thus affecting the practical applicability of the scheduling scheme;

[0007] 3. Existing source-load matching technologies mostly use fixed load forecasting and power generation plans, lacking in-depth analysis of the dynamic characteristics of source and load. However, with the rapid development of renewable energy such as wind power and photovoltaics, the accuracy of source-load matching faces higher requirements;

[0008] 4. The static matching method is difficult to capture the fluctuation characteristics of new energy, and the actual effect of the source-load matching scheme deviates greatly from the plan, affecting the reliability and economy of the power grid.

[0009] (II) Technical solution

[0010] The purpose of the present invention is to overcome the shortcomings of the prior art and to provide a method for identifying and constructing typical scenarios of source-load timing matching that can model multiple time scales, better reflect dynamic changes, effectively balance supply and consumption needs, and improve stability and economy.

[0011] The object of the present invention is achieved by: a method for identifying and constructing a typical scenario of source-load timing matching, the method comprising the following steps:

[0012] Step 1: Multi-time scale wind-solar-load scenario generation method with source-load matching temporal characteristics;

[0013] Step 2: Construction method of typical scenario sets and extreme scenario sets for supply and consumption;

[0014] Step 3: Typical case analysis: including analysis of wind and solar load characteristics, construction of typical scenario sets, construction of extreme scenario sets, and scenario analysis for supply and consumption.

[0015] Furthermore, the step 1 specifically includes the following steps:

[0016] Step 1.1: Wind power scenario generation method;

[0017] Step 1.2: Photovoltaic scene generation method;

[0018] Step 1.3: Load scenario generation method.

[0019] Furthermore, the step 2 specifically includes the following steps:

[0020] Step 2.1: Typical scene construction method;

[0021] Step 2.2: Extreme scenario construction method;

[0022] Step 2.3: Grid flexibility power aggregation methods.

[0023] (III) Beneficial effects

[0024] 1. Source-load matching lacks multi-time-scale temporal characteristics analysis: The present invention aims to propose a multi-time-scale wind-solar-load scenario generation method that can fully consider the dynamic characteristics of different energy outputs and load demands at different time scales, thereby achieving more accurate source-load matching and ensuring the stability and economy of the system;

[0025] 2. Solve the problem that the current construction of typical scenarios and extreme scenario sets lacks comprehensiveness and applicability: The present invention proposes a method for constructing typical scenario sets and extreme scenario sets for supply and consumption, which can generate a comprehensive scenario set that is suitable for both daily scheduling and responding to extreme events, thereby improving the applicability and flexibility of the system in responding to complex scenarios;

[0026] 3. The present invention can effectively improve the accuracy of source-load matching, realize dynamic scheduling at multiple time scales, and the generated scene set has higher representativeness and applicability;

[0027] 4. The data processing flow of the present invention significantly improves the efficiency of scene recognition and meets the real-time scheduling requirements; in practical applications, the flexible scheduling mechanism of the present invention ensures the stability and economy of the power grid when a high proportion of new energy is connected, and promotes the development of intelligent and efficient power systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 It is a normalized curve diagram of real-time ideal output generated by the present invention.

[0029] Figure 2 A flow chart is provided for generating a random component of the present invention.

[0030] Figure 3 A comparison chart of the results of random component generation for the present invention.

[0031] Figure 4 It is a schematic diagram of the box-type uncertain set of the present invention.

[0032] Figure 5 It is a schematic diagram of the ellipsoid uncertainty set of the present invention.

[0033] Figure 6 It is a schematic diagram of convex hull correction of the present invention. DETAILED DESCRIPTION

[0034] The present invention belongs to the technical field of power system operation and management, and particularly relates to an analysis and optimization method for source-load timing matching, and aims to provide support for optimal scheduling of various energy resources in power systems through typical scenario identification and construction; the present invention is applicable to multiple aspects such as new energy access, power load management, and demand response optimization, and is directly applied to flexibility management and supply-demand matching of power grids; by establishing a typical scenario for matching source-load timing characteristics, the scheduling flexibility and reliability of the power system under the condition of a high proportion of renewable energy access is improved.

[0035] The present invention is further described below in conjunction with embodiments and / or drawings.

[0036] Example 1

[0037] like Figure 1-6 As shown, a typical scenario identification and construction method for source-load timing matching includes the following steps:

[0038] Step 1: Multi-time scale wind-solar-load scenario generation method with source-load matching temporal characteristics;

[0039] In the present invention, specifically: ① Wind power scenario generation method: The present invention adopts the MCMC method that takes into account the duration characteristics and fluctuation characteristics of wind power to generate wind power output time series. This method is based on the traditional MCMC method, takes into account the duration distribution characteristics of wind power in various states, and adopts the strategy of first determining the state and then giving the time; this method can generate wind power sequences while retaining the statistical characteristics of the original data such as mean, standard deviation, probability density function, and autocorrelation coefficient, while also retaining time characteristics such as duration distribution characteristics, state transition characteristics, and fluctuation characteristics. The generated sequence can more realistically reflect the output of wind power.

[0040] The method mainly includes four steps: generating a state change matrix, generating a wind power state sequence based on the MCMC method, determining the duration of each state using the duration distribution characteristics, and superimposing the fluctuation component. The specific description is as follows:

[0041] (1) Generation of state change matrix: The traditional MCMC method is combined with the duration characteristics of wind power. The Markov chain is used to determine the state jump. The length of the state is determined by the distribution characteristics of the duration. Therefore, the state change matrix must be generated first. In order to achieve the state change, all the diagonal elements of the state transfer matrix can be set to zero (that is, the probability of the state remaining unchanged becomes zero), and then the ratio of each remaining element to the sum of all elements in the row is calculated as the new probability value;

[0042] (2) Generate wind power state sequence based on MCMC method: Based on the state change matrix, the cumulative state transition probability matrix P cumThe calculation method is shown in the formula: If P r The dimension of is N×N, then P cum The dimension is N×(N+1); the first column of the cumulative state transition probability matrix is ​​all zero, and starting from the second column, each element p cum,mn The value of (m rows and n columns) is the matrix P r The sum of the elements in the corresponding mth row and column whose number is less than n.

[0043] Calculate the cumulative state transition probability matrix P cum After that, the Monte Carlo method can be used to generate the wind power time series. The steps are as follows:

[0044] Step 1: The current wind power is in state m;

[0045] Step 2: Generate a random number u uniformly distributed between [0,1];

[0046] Step 3: Compare u and P cum The mth row element (the row number of the element is the same as the current wind power state number) is compared with the current wind power state number. If the relationship p is satisfied, cum,mn <u<p cum,m(n+1) , then the next state is n;

[0047] Step 4: If the generated wind power time series meets the length requirement, stop; otherwise, the current state becomes n and returns to step 2 to continue.

[0048] According to the above method, the state change matrix is ​​first transformed into a cumulative state change probability matrix; assuming that the initial state is s, a series of discrete states are generated using the Monte Carlo method to form a wind power state sequence; it is not difficult to see that since the diagonal elements of the state transfer matrix are set to zero, the elements on the two related diagonal columns in the cumulative state change matrix are equal; when the Monte Carlo method is used to generate the wind power time state, no random numbers will fall between two equal values, thus ensuring that the wind power state does not remain unchanged; that is, in the wind power state sequence generated at this time, adjacent states are different from each other.

[0049] (3) Determine the duration of each state using the duration distribution characteristics: The wind power state sequence generated in the previous step only gives the jump of wind power between various states. In this step, it is necessary to determine the duration of each state based on the aforementioned wind power duration characteristics of each state.

[0050] First, according to the fitted inverse Gaussian distribution, a random variable set of the duration of each state is generated; the wind power state sequence generated in (2) is traversed. If the current state is m, an element x is randomly selected from the duration variable set of state m and is set to the length of m; if the length of the generated time series Σx i If it is greater than or equal to the required wind power time series length l, the simulation ends.

[0051] (4) Superimposed fluctuation component: As mentioned above, the wind power fluctuation value (here defined as the difference between two wind power values ​​at a fixed time interval) satisfies the tLocation-Scale distribution. Based on this feature, random variables that satisfy the tLocation-Scale distribution with specific parameters can be superimposed on the basic discrete sequence. Therefore, it is necessary to first generate random variables that satisfy the tLocation-Scale distribution with specific parameters. The fluctuation component in the wind power time series to be generated is generated from this part of random variables.

[0052] According to the definition of wind power fluctuation in the previous article, the generated fluctuation component cannot be directly superimposed on the discrete state point, but the wind power value at the next moment must be determined based on the previous wind power value. Therefore, the algorithm is as follows:

[0053] Step 1: Assume that the current wind power value is x w (i), the state it belongs to is x(i), the next state is x(i+1), and the wind power range represented by each state is CF;

[0054] Step 2: The next wind power value range is ([x(i)-1]×CF,x(i)×CF), and the corresponding fluctuation value range is ([x(i)-1]×CF-x w (i),x(i)×CF-x w (i));

[0055] Step 3: According to the determined value range, a fluctuation value is randomly generated in the generated fluctuation component and superimposed on the current wind power value x w (i) on the right, we get x w (i+1);

[0056] Step 4: Repeat the above process until the length of the generated wind power time series meets the requirements.

[0057] According to the above steps, multiple wind power time series can be generated that retain the statistical characteristics of the original data such as mean, standard deviation, probability density function, autocorrelation coefficient, while also retaining the time characteristics such as duration distribution characteristics, state transition characteristics, and fluctuation characteristics, so as to more reasonably describe the uncertainty of wind power.

[0058] ② Photovoltaic scene generation method: Photovoltaic output time series generation refers to the establishment of an appropriate mathematical model based on a certain length of photovoltaic output measured data, so that new photovoltaic output sequences can be infinitely generated and can well inherit the statistical characteristics of the original sequence (commonly used random sequence generation mathematical models include Markov chain-Monte Carlo method MCMC, support vector machine SVM and autoregressive moving average ARMA model, etc.), which is specifically described as follows:

[0059] (1) Ideal output curve generation method: In order to generate a photovoltaic sequence of arbitrary length, it is necessary to obtain the ideal output normalized curve of all day sequences in an annual cycle. The ideal output normalized curve of all typical days can be obtained by an extraction method based on measured data. The following will introduce the ideal output normalized curve generation method for atypical days.

[0060] Using the extracted ideal output normalized curve of typical day, the curve of atypical day can be calculated by linear interpolation method, as shown in the formula: i represents an atypical day; m and n are the two typical days closest to the i-th day; Z * (t * ) is the ideal output normalized curve on the tth day (considering the annual periodicity of the statistical characteristics of photovoltaic output, the position of the typical day cycles annually); this method can be used to obtain the ideal output normalized curve for each day of the year.

[0061] Next, the horizontal axis range of the curve is restored to the real-time photovoltaic output time range; based on the solar-terrestrial motion model in solar physics, given the longitude and latitude of the photovoltaic power station, the daily sunrise and sunset times and the daytime duration T can be calculated. i,day , based on this, the ideal output normalized curve The horizontal coordinate of the real-time ideal output normalized curve Z is obtained by stretching and inverse transformation. Real,i (t), as shown in the formula: Comparison of measured data and generated real-time ideal output normalized curve Figure 1 shown.

[0062] (2) Amplitude parameter sequence generation method based on vector autoregression (VAR): Before random process modeling, the sequence must first be subjected to variance stabilization transformation. Variance stabilization transformation refers to finding a mapping function f to map the original time series x into a new sequence y = f(x) so that the variance of the new sequence y is stable. The method adopted by the present invention is to map the probability distribution of the amplitude parameter with the probability distribution of the normal distribution one by one, and transform the amplitude parameter sequence in the original domain into the normal domain, so as to achieve the purpose of stabilization. Each sample point x in the original time series has a unique cumulative probability density value distributed on [0,1]. Then, according to the cumulative probability density function curve of the normal distribution, the corresponding normal domain value x is found. * ; Perform this mapping on each point in the world sequence to obtain a new sequence X * That is, the mapping X that completes the variance stabilization transformation * =f(x); the transformed normal domain sequence obeys the normal distribution and achieves variance stabilization.

[0063] After the time series stability test and variance stabilization transformation, the amplitude parameter normal domain sequence can be modeled; from the above analysis, it can be seen that the autocorrelation function of the amplitude parameter sequence is tailing and the partial autocorrelation function is truncated, so it can be determined as an autoregressive AR model, and its expression is as follows: In the formula, Z t represents the value of the amplitude parameter sequence on day t; Z t-i is the value of the previous i days; is the coefficient to be estimated; a t is the white noise component whose variance is to be estimated; μ is the mean of the amplitude parameter; the AR model is applicable to stationary random processes, which requires ACF tailing and PACF truncation, and its truncation order is the order of the AR model; according to the above analysis, the ACF and PACF of the amplitude parameter sequence meet the modeling requirements of the AR model and can be defined as a first-order autoregressive model.

[0064] If the mutual correlation between multiple time series is considered, that is, the spatial correlation of photovoltaic output, Z in formula (4) t Expanded to a vector, the AR model will evolve into a vector autoregression model VAR; as shown below: vector The different components of represent the amplitude parameters of different photovoltaic power stations, and the correlation between them is reflected through the connection of the matrix Φ.

[0065] For the first-order VAR model considering two photovoltaic power plants, the above formula can be written as: The parameters to be estimated in this model are the coefficient matrix Φ and the white noise component The covariance matrix Σ of ; after derivation, its expression can be obtained as follows: Among them, Γ(k) is the k-order lag covariance matrix, which is expressed as: Where γ ij (k) is: ij (k) = E(Z i,t -μ i )(Z j,t+k -μ j )(9), based on the vector autoregression model, the amplitude parameter sequences of multiple photovoltaic power stations are modeled and generated, all of which have third-order tailing characteristics; the generation model based on VAR can effectively retain the time correlation and spatial correlation of the original sequence.

[0066] (3) Random component generation: For any random variable x, its cumulative distribution function is F. Since F is a monotonically non-decreasing function with a range of [0,1], the inverse function of the cumulative distribution function is F -1 is a function defined on [0,1]; it can be proved that the cumulative distribution function F(x) corresponding to any distribution x obeys uniform distribution. Therefore, by generating uniformly distributed random numbers on [0,1] and then performing inverse function transformation, we can obtain random numbers that obey the given cumulative distribution function F(x). Therefore, the random components that obey TLS distribution can be sampled and generated through the following process:

[0067] Step 1: randomly generate a variable u uniformly distributed on [0, 1]. In the present invention, a function for generating random numbers provided by MATLAB is used;

[0068] Step 2: Based on the probability density function, integrate to get the corresponding cumulative distribution function, and find the corresponding sample point x with the cumulative distribution function value u, that is, x = F -1 (u);

[0069] Step 3: By repeatedly sampling using the above method, we can obtain a random number sequence X that obeys the cumulative distribution function F.

[0070] The present invention adopts the above-mentioned random sampling generation method to generate the random component part. According to the random component separated from the measured data, its probability distribution TLS fitting parameters, the inverse Gaussian fitting parameters of the duration of the flat part and the inverse Gaussian fitting parameters of the duration of the fluctuating part are respectively counted, and a random component sequence of a specified length is sampled and generated according to the following process.

[0071] According to the above process, a random component sequence of any length can be generated. On this basis, the complete random component sequence can be obtained by replacing the time before sunrise and after sunset with zero. The generated random component sequence is compared with the random component sequence of the original data. Figure 3 shown.

[0072] ③ Load scenario generation method, specifically:

[0073] (1) Joint distribution model of daily load characteristic indicators: The daily load characteristic indicators are not completely independent, but have a certain correlation. By calculating the Pearson coefficients between the daily load characteristic indicators, it can be found that the daily peak-to-valley difference rate has a strong correlation with the daily load rate. The Copula function expresses the marginal distribution and correlation of random variables separately, which is easy to describe the complex correlation between random variables.

[0074] The Copula function connects the multivariate joint probability distribution function with the marginal distribution functions of multiple variables according to Sklar's theorem, and takes into account the correlation between multiple variables; taking the ternary joint probability distribution function as an example, according to Sklar's theorem, if the marginal distribution functions F(x1), F(x2), and F(x3) of three variables x1, x2, and x3 are known, there must be a Copula function C such that the joint probability distribution function G: G(x1, x2, x3) = C(F(x1), F(x2), F(x3)); the present invention uses the library function GaussianMultivariate in the Python copulas library to input different types of daily load characteristic matrices respectively. GaussianMultivariate first fits the three characteristic indicators respectively and determines the best probability density fitting function, integrates the probability density fitting function, obtains their marginal distribution functions respectively, and then uses the Copula function with the best fitting accuracy to model the joint probability distribution, and randomly samples through the triple standard deviation principle to obtain new daily load characteristic indicators.

[0075] (2) Daily load scenario generation method: In order to ensure that the change trend of the generated sequence is consistent with that of the historical sequence, the daily load sequence that is closest to the new daily load characteristic index is extracted from the historical sequence as the benchmark sequence, and a daily load sequence optimization model is established; the steps are as follows:

[0076] 1) Randomly generate a typical daily scenario for the first day and randomly extract new daily load characteristic indicators from the joint probability distribution model of the corresponding class;

[0077] 2) Search the historical sequence for the sequence that is closest to the daily load characteristic index as the benchmark sequence. The standard is the historical sequence with the smallest Euclidean distance to the newly generated daily load characteristic index;

[0078] 3) Establish a daily load optimization model. The objective function is to minimize the sum of square errors between the generated sequence and the benchmark sequence at each moment, that is: In the formula, x(t) is the load value of the generated sequence at each moment; l(t) is the load value of the benchmark sequence at each moment; the daily load rate constraint is used to ensure that the daily load rate of the generated sequence is equal to the daily load rate of the random sample: The daily peak-to-valley difference constraint is used to ensure that the time when the minimum load occurs is the same as the historical sequence, and the daily minimum load is consistent with the randomly selected index: x(t min )=(1-β)γ(12), x(t)≥x(t min )(13); where t min is the time corresponding to the minimum load value of the benchmark sequence; α is the daily load rate proportional factor, which is used to adjust the overall load level of the generated sequence to meet the specific daily load rate requirements. It may be used to determine the proportional relationship between the average load of the generated sequence and the benchmark load; β is the daily peak-to-valley difference proportional factor, which is used to adjust the minimum load value of the generated sequence. By controlling β / tanβ, it can be ensured that the generated minimum load value has a certain proportional relationship with the minimum load of the benchmark load sequence; γ is the benchmark load scalar, which is used to calibrate the load value of the generated sequence. It usually represents the maximum load value determined in historical data or sampled data, so as to calibrate the load level of the generated sequence;

[0079] The daily maximum load constraint is used to ensure that the time of the maximum load is consistent with the historical sequence, and the daily maximum load is consistent with the randomly selected index: x(t max )=γ(14), x(t)≤x(t max )(15), where: t max is the time corresponding to the maximum load value of the benchmark sequence; the fluctuation constraint is to ensure that the absolute value of the fluctuation of the generated sequence is greater than the minimum absolute value of the fluctuation of the historical sequence to prevent the occurrence of flat peaks: Δx(t)|≥|Δl(t)| min (16), where: Δx(t) represents the load increment or change of the generated sequence at time t, which is usually used to describe the load change of the generated load sequence between adjacent moments, that is, the change of x(t) relative to the previous moment; Δl(t) represents the load increment or change of the benchmark sequence at time t, and similarly, it is used to describe the load change of the benchmark load sequence between adjacent moments.

[0080] 4) Call the Python library function minimize to solve the daily load optimization model. The initial value is set to the load value of the benchmark curve, and the load value at each moment is obtained.

[0081] Step 2: Construction method of typical scenario sets and extreme scenario sets for supply and consumption;

[0082] In the present invention, specifically: ① Typical scenario construction method: scenario reduction is to replace the original scenario set with a small number of scenario sets according to the maximum probability. According to the historical data of wind power, photovoltaic and load, the intra-cycle clustering reduction method is used for wind, photovoltaic and load with intra-day periodic variation characteristics, daytime similarity characteristics and seasonal regularity characteristics. K clustering centers are obtained as wind power and load scenario sets by kmeans clustering technology, and then the scenarios are fused to form a typical scenario set of the system; synchronous back generation elimination method is one of the commonly used scenario reduction technologies, and its steps are as follows:

[0083] 1) For each scene wi, calculate the scene wj with the shortest distance to it as follows: In the formula, pro j is the occurrence probability of scenario wj; d(w i ,w j ) is the Euclidean distance between scenes wi and wj; S is the number of scenes.

[0084] 2) Determine the scene wi to be deleted according to the following formula:

[0085] 3) Modify the number of remaining scenes to S=S-1. To ensure that the sum of the probabilities of all scenes is 1, the probability of the deleted scene is added to the scene closest to it.

[0086] 4) Repeat the above steps until the number of remaining scenes reaches the desired set value NS.

[0087] ② Extreme scenario construction method: In order to construct an extreme scenario set based on historical data of wind, solar and load, it is necessary to model the set of uncertain variables (wind power, photovoltaic and load curve). Among the existing uncertain set modeling, the most common one is the Cubic Set. Ellipsoid set Its expression is: In the formula, ω is the n-dimensional vector represented by the uncertain parameters, that is, there are n random variables; R n Indicates that the vector is a real number; Formula (19) is the general expression of the box set, where is the lower bound / upper bound of the n-dimensional random variable, and is a known parameter; Formula (20) is the general expression of the ellipsoidal uncertainty set, and the ellipsoid is a generalized n-dimensional ellipsoid; where the matrix Q is a positive definite matrix, representing the angle of the high-dimensional ellipsoid deviating from the positive direction of the coordinate axis and the length of each symmetry axis; the vector c represents the coordinates of the center point of the high-dimensional ellipsoid, and Q and c are both known quantities; the following will analyze the two separately: It can be seen from (19) that the box set expression is concise, but it does not consider the correlation between random variables, such as Figure 4As shown in the figure, it can be seen that when the box set is used to describe uncertain parameters, the box set will additionally consider some non-existent scenarios (such as Figure 4 ), which leads to unnecessary conservatism; in the context of multiple wind farms, the time and spatial correlation of their active power generation cannot be ignored. If the box set is still used to describe the uncertain set, the conservatism of the decision will be improved. For the ellipsoid in the form of the formula, the formula takes the correlation into consideration. From the perspective of spatial correlation, taking the output of two wind farms in a single period as an example (two-dimensional random variables), the present invention collects the output data of two wind farms in Henan Province, and plots the historical values ​​of the two random variables in rectangular coordinates, such as Figure 5 As shown, since the mathematical expression of the ellipsoid set is in quadratic form, the subsequent optimization may be difficult to solve.

[0088] Based on the above analysis, the present invention proposes a data-driven robust uncertain set modeling method, which can fully consider the correlation between random variables and adopts the representation form of generalized convex hull. The mathematical expression is: in, is a representative scenario extracted from the historical data scenario, which is defined as an extreme scenario in the present invention; i represents the weight coefficient of the representative scene; The modeling process is as follows:

[0089] 1) Collection and collation of historical output data of new energy stations: The collected historical data are organized into column vectors, and a group of historical data is recorded as a historical scene. The collected historical scenes are recorded as Where N h is the number of historical scenes collected;

[0090] 2) Construction of a high-dimensional ellipsoid set based on historical data: Assume the following prerequisites: when the amount of historical data collected is sufficient, the historical output data is representative of the output data of the period of interest, that is, if there is a closed set that can completely cover the historical output scene, then the output data of the period of interest is also in the closed set; the present invention uses a high-dimensional closed ellipsoid algorithm, first solves a high-dimensional ellipsoid to surround all historical scenes, and the form of the obtained ellipsoid is shown in formula (20). Solving the high-dimensional ellipsoid parameters Q and c is equivalent to solving the following optimization: where ρ n is a constant representing the volume of the n-dimensional unit sphere; ω h,1 ,ω h,2...is the historical data collected in step 1). As for the form of optimization formula (22), the optimization is a convex optimization form, which can be solved quickly in polynomial time;

[0091] 3) Construction of uncertain set driven by initial data: Since robust equivalent transformation is required in robust optimization, the quadratic form in formula (20) will cause the mathematical properties of the original model to change after equivalent transformation. If a polyhedron set can be constructed on the basis of formula (20) to describe the uncertain set, the equivalent transformation process will not change the mathematical properties of the original model. However, this polyhedron needs to meet two conditions: ① It can be simply described by mathematical expressions; ② Considering the computing time requirements, the number of vertices should not be too large. Therefore, in the present invention, the convex set surrounded by the 2n vertices of the high-dimensional ellipsoid described by formula (20) is selected as the initial uncertain set. The method for determining the vertex coordinates is as follows:

[0092] First, the positive definite matrix Q is orthogonalized and decomposed: Q = P T DP=P -1 DP, where D is a diagonal matrix, and all the values ​​on the diagonal are positive numbers, P is the transformation matrix. To obtain the vertices corresponding to the high-dimensional ellipsoid, the ellipsoid is rotated and translated so that its axis of symmetry coincides with the coordinate axis. The translation and rotation transformation equation is: ω′=P×(ω-c)(23), where ω′ is the coordinate value of the random variable ω in the rotated coordinate system. The mathematical expression of the high-dimensional ellipsoid E′ obtained after rotation is: E′(D)={ω′∈R n |ω′ T Dω′≤1}(24), since the matrix D is a diagonal matrix, the vertex coordinates of the high-dimensional ellipsoid E′ are: Among them, ω v ' ,i is the coordinate value of the i-th vertex of the high-dimensional ellipsoid E′; N e is the number of vertices, that is, N e =2n; using the inverse transformation of formula (23) ω = c + P -1 ω′(26) can get the vertex equations of the original high-dimensional ellipsoid, and finally the polyhedron equation enclosed by the vertices E of the high-dimensional ellipsoid is:

[0093] 4) Correction of the initial convex hull and extraction of extreme scenes: Since the polyhedron enclosed by the vertices of the selected high-dimensional ellipsoid is not sufficient to cover all historical scenes, it is necessary to introduce a magnification factor to scale the convex hull of the polyhedron, such as Figure 6 As shown; the corrected axial ellipsoid vertex coordinates can be expressed as: Assume that the final uncertain set is: where ωe,i is the limit scene obtained after scaling the convex hull; combining equations (26) and (28), we can get its vertex ω′ with the high-dimensional ellipsoid E′ v,i Existence relationship: ω e,i =c+k * P -1 ω′ v,i (30), where k * This is the magnification factor in the convex hull scaling, which is determined as follows:

[0094] First, the historical scene h,1 ,ω h,2 ...The new scene is obtained by rotation and translation through equation (23), denoted as ω′ h,1 ,ω′ h,2 ...; From the convex optimization theory, we know that: if the historical scenario ω′ h,1 It can be covered by the convex hull formed by the vertices of the high-dimensional ellipsoid, then the following formula holds: where α j,i is a positive coefficient. If a scene is outside the convex hull, the convex hull needs to be enlarged, that is, the following formula needs to be satisfied: where k i is the magnification factor, and rearranging formula (32) yields: where β j,i is a positive coefficient; it can be seen that as the distance between the scene and the convex hull increases, the coefficient k i will also increase accordingly; therefore, an optimization can be established to determine the positional relationship between the historical scene and the convex hull: For N h There are optimizations of the above form in all historical scenarios, and there is no correlation between the optimizations. Therefore, we can consider N h The optimizations are combined into: For N h Historical scenes, we can get N by formula (35) h The maximum value in the sequence is the k in formula (29). * ; The final data-driven robust uncertainty set is represented by the polyhedron described in formula (29); in addition, since the scaling factor k in formula (29) is determined by itself, and according to the relevant knowledge of robust optimization, the larger the area contained in the uncertain set, the more conservative the decision, so the scaling factor can be modified according to current needs.

[0095] ③Flexibility power aggregation method of power grid: Using the calculation method based on electrical distance, the aggregated nodes at different voltage levels are divided into different target aggregation node areas according to the size of the equivalent transmission impedance; then the overall approximate solution technology based on the virtual power plant is used to aggregate the previously divided different target aggregation nodes through the virtual battery and virtual generator models to adjust the power domain, and the operating characteristics, costs and network information of the flexibility resources are comprehensively considered to achieve the optimal utilization of distributed energy and improve the reliability and economy of the power system; the details are as follows:

[0096] (1) Aggregation area division principle: According to different node connection methods, three types of connection methods are specified as follows: Method 1: A node at the voltage level that needs to be aggregated is directly connected to only one target aggregation node; Method 2: A node at the voltage level that needs to be aggregated is directly connected to two or more target aggregation nodes; Method 3: A node at the voltage level that needs to be aggregated is indirectly connected to one or more target aggregation nodes through other nodes.

[0097] It is stipulated here that for the aggregated nodes of connection mode 1, they are directly aggregated to the target aggregation node connected to them; for the aggregated nodes of connection modes 2 and 3, it is necessary to determine the electrical distances between them and different target aggregation nodes, and aggregate them to the target aggregation node with a short electrical distance.

[0098] Since the electrical distance between nodes is usually measured by transmission impedance, all operating parameters are allocated according to the transmission impedance when the power supply and network topology are fixed. The size of the transmission impedance is related to the specific connection method between the impedances. Taking a single-power supply and single-load network as an example, the equivalent transmission impedance satisfies the series-parallel relationship, and the transmission impedance between branches ik can be obtained by the series relationship: ik =z ij +z jk , the transmission impedance between branches ik can be obtained by the parallel relationship: Through the above method, all nodes that need to be aggregated can be divided into aggregation areas and divided into different target aggregation nodes. Then, based on the overall approximate solution method of the virtual power plant, multiple nodes on each target aggregation node can be aggregated into adjustable power domains.

[0099] (2) Flexibility resource aggregation method: The overall approximate solution based on the virtual power plant is obtained by mapping the adjustable power domain of the virtual power plant onto a high-dimensional convex polyhedron, and then selecting a new special high-dimensional polyhedron to approximate it from the outside or inside to maximize the accuracy of the model; the mathematical models used to approximate the high-dimensional convex polyhedron can be summarized into two categories: one is based on the virtual battery model, and the other is based on the virtual generator model; the mathematical model based on the virtual battery is obtained according to the following formula: in, An adjustable power domain for aggregated nodes described by a virtual battery model; is a column vector, and the virtual power plant adjusts the power at each time in the dispatch period T constitute; is the lower power limit of the virtual battery model; is the power upper limit of the virtual battery model; is the lower limit of the electric energy of the virtual battery model; It is the upper limit of the electric energy of the virtual battery model.

[0100] A high-dimensional isosceles right pyramid is selected to perform internal approximation on the adjustable power domain of the virtual power plant. The inscribed isosceles right pyramid with the longest side length in the high-dimensional convex polyhedron is obtained according to the following formula: Among them, P max Indicates the maximum adjustable power; p 0 (t) represents the power in each time period corresponding to the vertex of the inscribed isosceles right-angled pyramid; Ω N Represents N-dimensional constraint space; according to the formula p 0 (t)+P max =p t (t) Obtain the power boundary in the virtual battery model; According to the formula Obtain the energy boundary in the virtual battery model.

[0101] The mathematical model based on the virtual generator is obtained according to the following formula: in, is the adjustable power domain of the virtual power plant described by the virtual generator model; is the lower power limit of the virtual generator model; is the power upper limit of the virtual generator model; is the lower limit of the climbing of the virtual generator model; is the climbing upper limit of the virtual generator model.

[0102] A high-dimensional cube is selected to perform internal approximation on the adjustable power domain of the virtual power plant, and the inscribed cube with the longest side length in the high-dimensional convex polyhedron is obtained according to the following formula: Among them, P maxIndicates the maximum adjustable power; p 1 (t) represents the power of each time period corresponding to the “lower left” vertex of the multidimensional cube; p 2 (t) represents the power of each time period corresponding to the "upper right" vertex of the multidimensional cube; Ω N Represents an N-dimensional constraint space.

[0103] According to the formula The power boundary in the virtual generator model is obtained; the comprehensive characterization model is obtained according to the following formula: in, Adjustable power domain for virtual power plants; is a column vector element, and the virtual power plant adjusts the power p at each time in the dispatch period T (agg) (t) constitute; is the upper power limit; is the lower power limit; is the upper limit of energy storage capacity; is the lower limit of energy storage capacity; is the upper limit of climbing; The lower limit of climbing.

[0104] The upper and lower limits of power, upper and lower limits of ramping, and upper and lower limits of energy storage in the comprehensive characterization model are obtained according to the following solution model: in, The weight coefficients corresponding to the upper and lower limits of power respectively; The weight coefficients corresponding to the upper and lower limits of energy storage capacity respectively; They correspond to the weight coefficients of the upper and lower limits of the climbing slope respectively.

[0105] The constraints for solving the model include power system power balance constraints, power constraints of various flexibility resources, and network information constraints; the specific constraints are as follows:

[0106] 1) Parameter constraints: Where: T is the set of economic dispatch time periods;

[0107] 2) Unit combination state constraints: Where: The binary decision variable corresponding to the unit combination state of equipment i at time t; is the startup variable of device i at time t; S D Assemble for the crew;

[0108] 3) Equipment boundary constraints and climbing constraints; Where: P i,min , P i,max are the minimum and maximum output values ​​of the ith device respectively; R i,u , R i,dare the up-climbing and down-climbing powers of the i-th device respectively;

[0109] 4) Constraints on new energy units: Where: P pre is the output prediction value of the new energy unit; S R Assemble for new energy units;

[0110] 5) Energy storage system constraints: Where: are the charging and discharging power of energy storage device i respectively; E i,min 、E i,max are the minimum and maximum storage energies of energy storage device i, respectively; represents the stored energy of energy storage; η c,i , η dc,i are the charging and discharging efficiencies of energy storage device i; s i is the loss rate of energy storage device i; S ES is a collection of energy storage systems;

[0111] 6) Energy balance constraints: Among them, S G , S L They are generator nodes and load nodes respectively; this constraint realizes the power balance of electricity supply and demand.

[0112] In summary, the overall approximate solution technology based on virtual power plants not only takes into account the power constraints of flexible resources, but also the influence of network information, breaking through the limitations of approximate linear summation technology based on flexible resources. Therefore, this type of method is suitable for scenarios in virtual power plants where the distribution of flexible resources is divergent, the capacity is large, and the types are diverse.

[0113] Step 3: Typical case analysis: including analysis of wind and solar load characteristics, construction of typical scenario sets, construction of extreme scenario sets, and scenario analysis for supply and consumption.

[0114] In the present invention, ① analysis of wind-solar-load characteristics: the wind power generation curve shows a certain periodic trend within 92 days in summer. Specifically, from about 20:00 in the evening to about 5:00 in the morning of the next day, wind power generation presents a higher output level, generally reaching a peak at 22:00; from about 5:00 the next day, wind power output begins to decrease, generally reaching the lowest value of the day at about 8:00, and then is at a lower level within the day from 8:00 to 20:00.

[0115] Wind power generation reached a minimum value of about 2.3 million kilowatts at around 8:00 every day, and a peak value of about 11 million kilowatts at around 22:00 every day; at the same time, in June, the peak of wind power generation was between 5 and 11 million kilowatts on most days (27 days) out of a total of 30 days, and the minimum value was between 400,000 and 4 million kilowatts; in July, there were 31 days in total, and the daily peak of wind power generation decreased, with a peak value of less than 9 million kilowatts on 30 days, and even from July 26 to July 31, the peak value dropped to between 1.9 and 3.8 million kilowatts, and the minimum wind power generation was similar to that in June, between 300,000 and 4 million kilowatts; the overall trend in August was similar to that in June.

[0116] The regularity of the photovoltaic power generation output curve is more obvious, and it changes in a consistent manner with the light intensity. The typical day of June 21 (summer solstice) was selected for analysis. The output of photovoltaic power generation began to increase from 0 at around 5:30, when it was sunrise. The photovoltaic output increased with the increase of light intensity until it reached a peak of 8.9 million kilowatts at around 12:00. Then, the photovoltaic output began to decrease slowly in the afternoon until it reached 0 at sunset at around 19:30. Looking at the entire summer, the photovoltaic output began to increase from 0 at around 6:00, reached a peak at around 12:00, and then began to decrease until it decreased to 0 at around 19:00. The peak value of photovoltaic output throughout the summer basically fluctuated between 5 million and 10.328 million kilowatts, with the maximum peak value of 10.328 million kilowatts reached on July 21. The peak values ​​of a few days were below 4 million kilowatts, which may be caused by weak light intensity on cloudy days.

[0117] The provincial grid load level in summer fluctuates between 25.8 and 65 million kilowatts, of which the average load between June 1 and June 15 was around 40 million kilowatts, and the average load between June 16 and June 30 increased to around 50 million kilowatts; the average load level was around 40 million kilowatts throughout the middle and early July; from the end of July, the average load level increased to around 50 million kilowatts, and then fell back to around 35 million kilowatts in late August; a typical day, June 19, was selected for analysis, and the load of the entire network began to climb around 5:00 in the morning, reaching the noon peak around 12:00 and maintaining a stable trend until 18:00; at 18:00, the load began to decrease until 19:00 (after get off work), and then quickly climbed again until it reached a peak at 21:30 (mainly the peak of the day), which was concentrated between 21:00 and 23:00 and lasted for about two hours.

[0118] ② Construction of typical scenario sets: (1) Wind power scenario: The values ​​of the power system time series scenario generated by the four seasons of wind power in spring, summer, autumn and winter are not independent of each other, but there are time series correlations on different time scales. The time series correlation degree of random variables in the power system time series scenario can be verified using the correlation coefficient. Common correlation coefficients include Pearson coefficient, Spearman coefficient, Kendall coefficient, etc.; the autocorrelation coefficient is often selected in the Pearson coefficient to describe the linear correlation of the time series. The calculation of the autocorrelation coefficient ACF describes the autocorrelation relationship between two output sequences within different time delays. The maximum value of ACF is 1. The larger the value, the stronger the correlation. The formula for calculating the autocorrelation coefficient is as follows: In the formula, S t is the output value of the sample at time t; μ is the mean of the entire sequence; the ACF of wind power and photovoltaic generation and historical data are calculated respectively when the time delay is 7. From the ACF of the four seasons, it can be concluded that the generated sequences in the four seasons are well fitted with the ACF of the historical sequences, especially in autumn and winter; the correlation coefficients of the historical scenes can be well included in the generated scenes, and the time correlation between the two is consistent at the same time delay, that is, the correlation is higher in a shorter time interval, and the longer the time interval, the lower the correlation, which is in line with the longitudinal similarity of new energy output.

[0119] K-means clustering is used to generate the day-ahead wind power output scenario for each season. The data used for clustering can select seasonal generated scenarios or seasonal historical scenarios. In order to ensure the authenticity of typical day-ahead scenarios, the present invention uses wind power seasonal historical scenarios for clustering, and clustering is performed based on features such as the mean wind power daily output, the extreme difference of wind power daily output, and the influencing factors of wind power daily output shape.

[0120] Table 1 Clustering silhouette coefficient effect of wind power in four seasons typical days

[0121]

[0122] The silhouette coefficient method is used to determine the optimal number of clusters. The silhouette coefficient is defined as follows: the data to be classified is divided into k clusters by the K-means algorithm. Each cluster contains N sample points. The silhouette coefficients of each sample point in the cluster are calculated separately. For one sample point x i For example: define a(i) as the vector from i to all other points x in the cluster to which it belongs n The mean of the Euclidean distance is: Where N is the total number of samples in each cluster.

[0123] Define b(i) as the vector from i to all sample points y in other clusters n ,...,z n The minimum average distance of: Then the vector silhouette coefficient is: It can be seen that the value of the silhouette coefficient is between [-1,1]. The closer it is to 1, the better the cohesion and separation are. The sample center points of different clusters in each season are taken as the typical day scene set, and the proportion of the cluster in which each typical day is located is regarded as the probability of occurrence of this type of typical day scene.

[0124] (2) Photovoltaic scenario: The method for verifying the time correlation of wind power series is the same as that for calculating the quarterly curve ACF and fitting. The ACF fitting of the photovoltaic spring, summer, autumn and winter seasons shows that the time correlation of the generated scenario is consistent with the historical scenario. The K-means clustering method is used to aggregate the photovoltaic day-ahead output scenario, which is classified according to the daily output amplitude, daily output mean and daytime output change mean.

[0125] Table 2 Clustering silhouette coefficient effects of typical days in four seasons for photovoltaic

[0126]

[0127] From the clustering results, it can be seen that the silhouette coefficients of the four seasons clusters are all high (>0.6), indicating that the clustering effect is better. In spring and summer, the overall sunshine time is longer, the daily average output is higher, and the power fluctuation is relatively stable; in autumn and winter, the daytime output of some scenes increases, and the fluctuation is relatively violent; the sample center points of different clusters in each season are taken as the typical day scene set, and the proportion of the cluster in which each typical day is located is regarded as the probability that the scene before the day will become the typical day.

[0128] (3) Load scenario: The cumulative distribution function (CDF) can describe the probability distribution of random variables. The cumulative distribution functions of the generated data and historical data are fitted respectively. The CDF fitting of the four seasons shows that the generated sequence can accurately fit the probability distribution characteristics of the historical sequence. The statistical characteristics of the generated sequence are consistent with those of the historical sequence. The method of verifying the time correlation of the new energy sequence is the same. The quarterly curve ACF is calculated and fitted. Since the load curve is relatively stable, the proposed scenario reduction method is used to identify the typical load scenario in the fourth quarter.

[0129] ② Construction of extreme scenario sets: First, the proposed scenario generation method is used to analyze the historical wind-solar-load statistical characteristics of 18 cities in a province, capture the probability distribution characteristics of wind-solar-load scenarios in each city, and generate 500 sets of historical scenarios for each city; then, based on the generated historical scenario set, the high-dimensional diamond convex hull method is used to construct extreme scenarios for "double protection", and then the proposed generation and reduction method is used to aggregate the extreme scenarios and reduce them to 10 extreme scenarios.

[0130] ③ Scenario analysis for supply guarantee and consumption: The total installed capacity of thermal power in a certain province is 75,217 MW, the total installed capacity of wind farms is 15,965.1 MW, and the total installed capacity of photovoltaic power stations is 6,258.02 MW.

[0131] Table 3 Installed capacity

[0132]

[0133]

[0134] (1) “Supply guarantee” scenario analysis: The above data is based on the analysis that the maximum power generation of thermal power units is equal to their designed installed capacity. Considering the retirement of thermal power units in the future, which will lead to a significant reduction in installed capacity, there will inevitably be supply guarantee problems. Therefore, the maximum output of thermal power units is adjusted to 70% of the installed capacity to simulate the partial retirement of thermal power units and observe the load shedding situation. Overall, the summer load shedding situation, that is, the supply guarantee problem, only occurs in mid-to-late June, mid-July and mid-to-early August when the average load level of the entire network increases. At the same time, the peak value of load reduction will vary with the peak value of the load level, and most of it is around 5-10 million kilowatts. In other periods, due to the low average load level, the output of thermal power units and wind and solar power can meet the load demand, and there will be no load shedding.

[0135] A typical day, June 22, was selected to analyze the load shedding situation of the entire network on that day. The trend of load reduction was similar to the trend of some curves of load level. The load reduction began to increase from 16:00 and reached the first peak of about 4.2 million kilowatts around 18:00; then it reached the second peak of about 8 million kilowatts around 21:00, which was a period of high electricity consumption, and maintained for 2 hours; then as the load level decreased, the load reduction also gradually decreased; further, the maximum output of the thermal power units was adjusted to 60% of the installed capacity, and the load shedding situation was observed. It can be seen that: at this time, there are several changes in the load reduction of the entire network: first, the number of days with load shedding has increased significantly; second, the peak value of load reduction has increased, and some peak values ​​have increased from the previous 5-10 million kilowatts to about 1500 kilowatts; for periods with high load levels, the load shedding time may become longer, and there may even be a situation where the load is shelved most of the day.

[0136] (2) “Guaranteed consumption” scenario analysis: Considering that the installed capacity of wind power and photovoltaic power will increase in the future, the power system may face the problem of “guaranteed consumption”. According to the main achievements of a province’s planned energy development, the average annual growth rate of wind power and photovoltaic installed capacity is 66% and 88%. Therefore, it is predicted that in five years, the installed capacity of wind power will increase by about three times, and the capacity of photovoltaic power will increase by about four times. Simulating the increase in new energy capacity, the time series production is used to simulate the wind and solar power abandonment in a province’s power system throughout the summer. After increasing the installed capacity of wind power and photovoltaic power, the whole province will experience wind and solar power abandonment in the summer. The typical day of June 9 was selected for analysis. Most of the wind and solar power abandonment occurred in the afternoon. Since photovoltaic power generation was large at this time and could not be absorbed, the amount of solar power abandoned reached its peak at 13:00 and 15:00, and the amount of wind abandoned reached its peak at 14:00.

[0137] The present invention is a method for identifying and constructing typical scenarios of source-load timing matching. In use, the present invention proposes a multi-time-scale wind-light-load scenario generation method, which can better reflect the timing characteristics of wind power, photovoltaics and load demand and their dynamic changes at different time scales. Compared with the single-scale analysis in the existing methods, the present invention captures the fluctuation characteristics of wind power, photovoltaics and load at the minute, hour and day levels through multi-time-scale modeling, and realizes a refined description of source-load matching. This can significantly improve the system's scheduling accuracy when a high proportion of renewable energy is connected, ensure that the source-load matching solution is more in line with actual operation needs, and thus effectively improve the stability and economy of the power grid; the present invention adopts a dual-scale approach oriented towards supply guarantee and consumption. The method for constructing a scenario set based on demand can effectively balance the system's supply guarantee demand under extreme load and the consumption demand during the peak period of new energy; through the construction of typical scenario sets and extreme scenario sets, the present invention, on the basis of systematically analyzing historical data, adds predictions of extreme situations that may occur in the future, thereby generating a more representative and forward-looking scenario set. This method overcomes the limitation of over-reliance on historical data in the prior art, so that the generated scenario set can cover a variety of states from peak load to extreme situations of new energy output, ensuring the safe operation and flexibility of the power grid in a complex environment; the present invention has the advantages of multi-time scale modeling, better reflection of dynamic changes, effective balance of supply guarantee and consumption demand, and improved stability and economy.

Claims

1. A method for identifying and constructing a typical scenario of source-load timing matching, characterized in that: The method comprises the following steps: Step 1: Multi-time scale wind-solar-load scenario generation method with source-load matching temporal characteristics; Step 2: Construction method of typical scenario sets and extreme scenario sets for supply and consumption; Step 3: Typical case analysis: including analysis of wind and solar load characteristics, construction of typical scenario sets, construction of extreme scenario sets, and scenario analysis for supply and consumption.

2. A method for identifying and constructing a typical scenario of source-load timing matching according to claim 1, characterized in that: The step 1 specifically comprises the following steps: Step 1.1: Wind power scenario generation method; Step 1.2: Photovoltaic scene generation method; Step 1.3: Load scenario generation method.

3. A method for identifying and constructing a typical scenario of source-load timing matching according to claim 2, characterized in that: The wind power scenario generation method in step 1.1 specifically comprises the following steps: Step 1.11: Generate the state change matrix: Set all diagonal elements of the state transfer matrix to zero, and then calculate the ratio of each remaining element to the sum of all elements in the row as the new probability value; Step 1.12: Generate wind power state sequence based on MCMC method: Based on the state change matrix, the cumulative state transition probability matrix P cum The calculation method is: If P r The dimension of is N×N, then P cum The dimension is N×(N+1); the first column of the cumulative state transition probability matrix is ​​all zero, and starting from the second column, each element p cum,mn The value of (m rows and n columns) is the matrix P r The sum of the elements in the corresponding mth row and column number less than n; Step 1.13: Determine the duration of each state using the duration distribution characteristics: First, generate a random variable set of the duration of each state according to the fitted inverse Gaussian distribution; traverse the generated wind power state sequence, if the current state is m, randomly select an element x from the duration variable set of state m and set it to be the length of m; if the length of the generated time series Σx i If it is greater than or equal to the required wind power time series length l, the simulation ends; Step 1.14: Superimpose the fluctuation component: The wind power fluctuation value satisfies the tLocation-Scale distribution. Based on this feature, random variables that satisfy the tLocation-Scale distribution with specific parameters can be superimposed on the basic discrete sequence; therefore, it is necessary to first generate random variables that satisfy the tLocation-Scale distribution with specific parameters, and the fluctuation component in the wind power time series to be generated is generated from this part of the random variables.

4. A method for identifying and constructing a typical scenario of source-load timing matching according to claim 2, characterized in that: The photovoltaic scene generation method in step 1.2 specifically includes the following steps: Step 1.21: Ideal output curve generation method: Through the extraction method based on measured data, the ideal output normalized curve of all typical days can be obtained. Using the extracted ideal output normalized curve of typical days, the curve of atypical days can be calculated by linear interpolation method: i represents an atypical day; m and n are the two typical days closest to the i-th day; Z * (t * ) is the ideal output normalized curve of day t; this method can be used to obtain the ideal output normalized curve of every day throughout the year; Step 1.22: Amplitude parameter sequence generation method based on vector autoregression: Map the probability distribution of the amplitude parameter to the probability distribution of the normal distribution one by one, and transform the amplitude parameter sequence of the original domain into the normal domain; each sample point x in the original time series has a unique cumulative probability density value distributed on [0,1], and then find the corresponding normal domain value x according to the cumulative probability density function curve of the normal distribution * ; Perform this mapping on each point in the world sequence to obtain a new sequence X * That is, the mapping X that completes the variance stabilization transformation * =f(x); Step 1.23: Random component generation: For any random variable x, its cumulative distribution function is F. Since F is a monotone non-decreasing function with a range of [0,1], the inverse function of the cumulative distribution function is F -1 is a function defined on [0,1]. It can be proved that the cumulative distribution function F(x) corresponding to any distribution x obeys the uniform distribution. Therefore, by generating uniformly distributed random numbers on [0,1] and then performing the inverse function transformation, we can obtain random numbers that obey the given cumulative distribution function F(x).

5. A method for identifying and constructing a typical scenario of source-load timing matching according to claim 2, characterized in that: The load scenario generation method in step 1.3 specifically includes the following steps: Step 1.31: Joint distribution model of daily load characteristic index: Taking the ternary joint probability distribution function as an example, according to Sklar theorem, if the marginal distribution functions F(x1), F(x2), F(x3) of three variables x1, x2, x3 are known, then there must be a Copula function C such that the joint probability distribution function G: G(x1, x2, x3) = C(F(x1), F(x2), F(x3)); Step 1.32: Daily load scenario generation method: Extract the daily load sequence that is closest to the new daily load characteristic index from the historical sequence as the benchmark sequence, and establish a daily load sequence optimization model.

6. A method for identifying and constructing a typical scenario of source-load timing matching according to claim 1, characterized in that: The step 2 specifically includes the following steps: Step 2.1: Typical scene construction method; Step 2.2: Extreme scenario construction method; Step 2.3: Grid flexibility power aggregation methods.

7. A method for identifying and constructing a typical scenario of source-load timing matching according to claim 6, characterized in that: The typical scenario construction method in step 2.1 adopts synchronous back-generation elimination method, which specifically includes the following steps: Step 2.11: For each scene wi, calculate the scene wj with the shortest distance to it as follows: In the formula, pro j is the occurrence probability of scenario wj; d(w i ,w j ) is the Euclidean distance between scenes wi and wj; S is the number of scenes; Step 2.12: Determine the scene wi to be deleted according to the following formula: Step 2.13: Modify the number of remaining scenes to S = S-1. To ensure that the sum of the probabilities of all scenes is 1, add the probability of the deleted scene to the scene closest to it. Step 2.14: Repeat the above steps until the number of remaining scenes reaches the desired set value NS.

8. A method for identifying and constructing a typical scenario of source-load timing matching according to claim 6, characterized in that: The extreme scenario construction method in step 2.2 is specifically as follows: In order to construct an extreme scenario set based on historical data of wind and solar load, it is necessary to model the uncertainty variable set, including the box set Ellipsoid uncertainty set Its expression is: Where ω is the n-dimensional vector represented by the uncertain parameters; R n The vector is a real number; the matrix Q is a positive definite matrix; the vector c represents the coordinates of the center point of the high-dimensional ellipsoid; the data-driven robust uncertain set modeling method is adopted, which can fully consider the correlation between random variables and adopt the representation form of generalized convex hull. The mathematical expression is: in, is a representative scenario extracted from historical data scenarios, which is defined as an extreme scenario; i Represents the weight coefficient of the representative scene.

9. A method for identifying and constructing a typical scenario of source-load timing matching according to claim 8, characterized in that: The data-driven uncertainty set in step 2.2 The modeling process is as follows: Step 2.21: Collection and organization of historical output data of new energy stations: The collected historical data are organized into column vectors, and a group of historical data is recorded as a historical scene. The collected historical scenes are recorded as Where N h is the number of historical scenes collected; Step 2.22: Construction of high-dimensional ellipsoid set based on historical data; Step 2.23: Initial data-driven uncertainty set construction; Step 2.24: Correction of the initial convex hull and extraction of the extreme scene.

10. A method for identifying and constructing a typical scenario of source-load timing matching according to claim 6, characterized in that: The method for aggregating the flexible power of the power grid in step 2.3 specifically comprises the following steps: Step 2.31: Aggregation area division principle: Method 1: A node at the voltage level that needs to be aggregated is directly connected to only one target aggregation node; Method 2: A node at the voltage level that needs to be aggregated is directly connected to two or more target aggregation nodes; Method 3: A node at the voltage level that needs to be aggregated is indirectly connected to one or more target aggregation nodes through other nodes; Step 2.32: Flexibility Resource Aggregation Method: Comprehensive Representation Model: in, Adjustable power domain for virtual power plants; is a column vector element, and the virtual power plant adjusts the power p at each time in the dispatch period T (agg) (t) constitute; is the upper power limit; is the lower power limit; The upper limit of energy storage capacity; is the lower limit of energy storage capacity; is the upper limit of climbing; The lower limit of climbing.

Images