Large-scale power system high-dimensional probability security scene reduction method and system

Abstract

The invention discloses a large-scale power system high-dimensional probability security scene reduction method and system, and the method comprises the steps: carrying out the sorting and combination of component sets of a multi-dimensional random variable X describing an original scene set according to columns, constructing a distribution matrix V, and calculating a corresponding marginal probability distribution matrix P; dividing the distribution matrix V according to columns and calculating a Cartesian product, obtaining a recombination matrix # imgabs0 # and a marginal probability distribution matrix # imgabs1 #, carrying out descending order arrangement and sampling on the recombination matrix # imgabs2 #, obtaining a compact scene set # imgabs3 #, calculating the order of magnitude of the probability corresponding to each scene in the compact scene set # imgabs4 #, carrying out scene classification, and calculating the probability of each scene in the compact scene set # imgabs4 #; and a scene set # imgabs5 # of each order of magnitude is obtained, an optimization model is constructed by taking minimization of a Wasserstein distance as a target, adaptive selection is performed on the scene set # imgabs6 #, and a union set is taken from the selected sets to obtain an optimal reduced scene set. According to the invention, the scene reduction of any dimension scene can be realized, the representativeness of the reduced scene and the complete coverage of the probability space are ensured, and the application value is extremely high.

Description

Technical Field

[0001] The present invention relates to the technical field of power systems, and in particular to a method and system for reducing high-dimensional probabilistic preservation scenarios of large-scale power systems. Background Art

[0002] With the large-scale integration of renewable energy into the power system, the operation and management of the power system are facing unprecedented challenges. Traditional power system planning and analysis mainly rely on deterministic models and methods. However, with the volatility and uncertainty of renewable energy (such as wind power and solar energy), the operation of the power system has become more and more dependent on the accurate understanding and modeling of randomness. Therefore, the importance of stochastic programming and reliability analysis in the field of power systems has become increasingly prominent.

[0003] The selection of representative scenarios is a key step in stochastic programming and reliability analysis. These scenarios can capture the main characteristics of the probability distribution of the original scenarios and provide a basis for subsequent power system operation, optimization, and decision-making. However, for large-scale power systems containing a large number of random variables (such as wind power output, solar irradiance, load demand, etc.), generating representative scenarios becomes a challenge, especially in high-dimensional probability spaces.

[0004] In order to make the distribution of representative scenes close to the distribution of the original random variables, the best choice of representative scenes from the scene set is reduced to an optimization problem. Existing representative scene selection methods usually include optimization-based methods and sampling-based methods. Optimization-based methods mainly focus on simplifying the optimization process to achieve a near-optimal solution; sampling-based methods use discrete samples based on probability sampling to model the distribution of random variables. However, these methods have significant limitations when dealing with high-dimensional distributions. On the one hand, most optimization-based methods show low computational efficiency or even incomputation when dealing with large scene sets in the context of high-dimensional distributions; on the other hand, sampling-based methods are not restricted by dimensionality, but the subsequent results often deviate significantly from the optimal value. Summary of the invention

[0005] In order to solve the above problems, the present invention proposes a method and system for reducing high-dimensional probabilistic scenarios of large-scale power systems, which can realize scenario reduction of arbitrary-dimensional scenarios of power systems and ensure the representativeness of the reduced scenarios and full coverage of the probability space.

[0006] In order to achieve the above object, the present invention is implemented by the following technical solutions:

[0007] The high-dimensional probability preservation scenario reduction method for a large-scale power system of the present invention comprises the following operations: recording D uncertain objects of the power system as D-dimensional random variables X={X1,...X j ,…X D}, the uncertainty value of each variable is discrete into H intervals, denoted as x j ={x 1j ,x 2j ,...,x Hj}, combine the D-dimensional random variable information into a distribution matrix V, where the rows represent the random variables X of each dimension j The uncertainty value of is expressed as a D-dimensional random variable, and the corresponding marginal probability distribution matrix P is calculated;

[0008] Divide the distribution matrix V by columns and calculate the Cartesian product to obtain the reorganized matrix And the corresponding marginal probability distribution matrix Reorganize the matrix Sort in descending order and sample to obtain a compact scene set Computing compact scene sets The order of magnitude of the probability corresponding to each scene in the , the scene classification is performed according to the order of magnitude, and the scene set of each order of magnitude is obtained The optimization model is constructed with the goal of minimizing the Wasserstein distance. Adaptive selection is performed, and the union of the selected sets is taken to obtain the optimal reduced scene set.

[0009] A further improvement of the present invention is that: D-dimensional random variable information is combined into a distribution matrix V, wherein the rows represent random variables X of each dimension. j The uncertainty value of is represented as a D-dimensional random variable, and the corresponding marginal probability distribution matrix P is calculated, which specifically includes the following operations:

[0010] Based on x with discrete values j ={x 1j ,x 2j ,...,x Hj} of a multidimensional random variable X={X1,…X j ,…X D}, the distribution matrix V is defined as (x ij ) H×D , where D is the dimension of the random variable, H is the discrete interval of the random variable, and x ij is an element in the distribution matrix V, representing the power value of the uncertainty object j when it takes the i-th interval, where the uncertainty objects include but are not limited to the wind turbine output, photovoltaic output, and load power consumption at a certain moment;

[0011] Calculate the element x in the distribution matrix V ij The marginal distribution probability f j (x ij ), define the marginal probability distribution matrix P as y ij ) H×D , where the element yij =f j (x ij ).

[0012] A further improvement of the present invention is that: the distribution matrix V is divided by columns, and the Cartesian product is calculated to obtain the reorganized matrix And the corresponding marginal probability distribution matrix Reorganize the matrix Sort in descending order to obtain a compact scene set The specific operations include:

[0013] Divide the distribution matrix V into v sub-matrices {V t (t=1,2,...,v)}, the first v-1 sub-matrices each have u columns, if the v-th sub-matrix does not have u columns, the remaining uv-D columns are supplemented with discrete uniform distribution;

[0014] Calculate the submatrix V t The Cartesian product of the elements of u different columns in where k = 1, 2, ... H u , get the recombinant matrix Recombinant matrix according to Splicing into a recombinant matrix The corresponding marginal probability distribution matrix is ​​recorded as Reorganize the matrix The elements in each column are arranged in descending order according to the corresponding probability, and K rows are randomly selected from them to obtain a compact scene set in i1=1,2,...,K, where represents the i1th combined scene, Indicates the value of the element in the vth column in the i1th combination scenario.

[0015] A further improvement of the present invention is to calculate a compact scene set The order of magnitude of the probability corresponding to each scene in the , the scene classification is performed according to the order of magnitude, and the scene set of each order of magnitude is obtained Specifically include:

[0016] For any scene where n1, n2, ..., n D ∈{1,2,...,K}, where n represents the set of scenes The nth combined scene in Indicates the specific combination of the nth combination scene, An element representing the nth combined scene;

[0017] Computing scenario X n The joint probability

[0018] Defines the calculation function of the order of magnitude of the scene in Indicates rounding down to calculate the compact scene set The order of magnitude of each scene Get the maximum magnitude and the span between the maximum and minimum magnitudes Divide scenes of the same magnitude into the same subset i2∈

[0019] {1,2,…,l} represents the index of the sub-scenario set divided according to the order of magnitude.

[0020] A further improvement of the present invention is to construct an optimization model with the goal of minimizing the Wasserstein distance, Adaptive selection is performed, and the union of the selected sets is taken to obtain the optimal reduced scene set, which specifically includes:

[0021] Define the selection step variable set S = {s1, s2, ..., s l}, where s l represents the input variables of the adaptively selected optimization model, that is, from the scene set Select Scene to Scene The step size is used to construct an optimization model for adaptive selection. in

[0022] Represents a collection of scenes Adaptive selection to scene set The scene in Representation scene set The size of Denotes a random variable X j and The Wasserstein distance between Represents a reduced set of scenes The distribution formed is the marginal distribution of the jth uncertain object;

[0023] According to the adaptive selection algorithm, the optimization model of adaptive selection is solved to obtain the approximate optimal step size set According to the corresponding right Make a selection Given hyperparameters α, β∈Z, 1≤α, β≤D, Z is a set of integers, we get the optimal reduced scene set

[0024] The high-dimensional probabilistic preservation scenario reduction system for a large-scale power system of the present invention comprises:

[0025] The distribution matrix building module is used to record the D uncertain objects of the power system as D-dimensional random variables X = {X1, ...X j ,…X D}, the uncertainty value of each variable is discrete into H intervals, denoted as x j ={x 1j ,x 2j ,...,x Hj}, combine the D-dimensional random variable information into a distribution matrix V, where the rows represent the random variables X of each dimension j The uncertainty value of is expressed as a D-dimensional random variable, and the corresponding marginal probability distribution matrix P is calculated;

[0026] The compact scene set acquisition module is used to divide the distribution matrix V by column and calculate the Cartesian product to obtain the reorganized matrix And the corresponding marginal probability distribution matrix Reorganize the matrix Sort in descending order to obtain a compact scene set

[0027] Probability magnitude scene set acquisition module, used to calculate compact scene sets The order of magnitude of the probability corresponding to each scene in the , the scene classification is performed according to the order of magnitude, and the scene set of each order of magnitude is obtained Adaptive selection module, used to build an optimization model with the goal of minimizing the Wasserstein distance, for scene sets Adaptive selection is performed, and the union of the selected sets is taken to obtain the optimal reduced scene set.

[0028] A further improvement of the present invention is that the operations performed by the distribution matrix building module include:

[0029] Based on x with discrete values j ={x 1j ,x 2j ,...,x Hj} of a multidimensional random variable X={X1,…X j ,…X D}, the distribution matrix V is defined as (x ij ) H×D , where D is the dimension of the random variable, H is the discrete interval of the random variable, and x ij is an element in the distribution matrix V, representing the power value of the uncertainty object j when it takes the i-th interval, where the uncertainty objects include but are not limited to the wind turbine output, photovoltaic output, and load power consumption at a certain moment;

[0030] Calculate the element x in the distribution matrix V ij The marginal distribution probability f j (x ij ), define the marginal probability distribution matrix P as y ij ) H×D , where the element y ij =f j (x ij ).

[0031] A further improvement of the present invention is that the operations performed by the compact scene set acquisition module include:

[0032] Divide the distribution matrix V into v sub-matrices {V t (t=1,2,...,v)}, the first v-1 sub-matrices each have u columns, if the v-th sub-matrix does not have u columns, the remaining uv-D columns are supplemented with discrete uniform distribution;

[0033] Calculate the submatrix V t The Cartesian product of the elements of u different columns in where k = 1, 2, ... H u , get the recombinant matrix Recombinant matrix according to Splicing into recombinant matrix The corresponding marginal probability distribution matrix is ​​recorded as Reorganize the matrix The elements in each column are arranged in descending order according to the corresponding probability, and K rows are randomly selected from them to obtain a compact scene set in i1=

[0034] 1,2,...,K, where represents the i1th combined scene, Indicates the value of the element in the vth column in the i1th combination scenario.

[0035] A further improvement of the present invention is that the operation performed by the probability magnitude scene set acquisition module includes: where n1, n2, ..., n D ∈{1,2,...,H}, where n represents the set of scenes The nth combined scene in Indicates the specific combination of the nth combination scene, An element representing the nth combined scene;

[0036] Computing scenario X n The joint probability

[0037] Defines the calculation function of the order of magnitude of the scene in Indicates rounding down to calculate the compact scene set The order of magnitude of each scene Get the maximum magnitude and the span between the maximum and minimum magnitudes Divide scenes of the same magnitude into the same subset i2∈

[0038] {1,2,…,l} represents the index of the sub-scenario set divided according to the order of magnitude.

[0039] A further improvement of the present invention is that the operations performed by the adaptive selection module include:

[0040] Define the selection step variable set S = {s1, s2, ..., s l}, where s l represents the input variables of the adaptively selected optimization model, that is, from the scene set Select Scene to Scene The step size is used to construct an optimization model for adaptive selection. in

[0041] Represents a collection of scenes Adaptive selection to scene set The scene in Representation scene set The size of Denotes a random variable X j and The Wasserstein distance between Represents a reduced set of scenes The distribution formed is the marginal distribution of the jth uncertain object;

[0042] According to the adaptive selection algorithm, the optimization model of adaptive selection is solved to obtain the approximate optimal step size set According to the corresponding right Make a selection Given hyperparameters α, β∈Z, 1≤α, β≤D, Z is a set of integers, we get the optimal reduced scene set

[0043] The beneficial effects of the present invention are as follows: the present invention takes advantage of the sampling-based method and the optimization-based method, proposes a scene reduction framework, and is capable of selecting representative scenes that cannot be obtained by traditional sampling methods in high-dimensional probability space. At the same time, the present invention proposes a distribution matrix representation method, which can implicitly represent the original scene set of the ultra-high-dimensional probability space, and stores the original scene set of any dimension in the form of Cartesian product.

[0044] The present invention obtains a compact scene set that can completely cover the probability space by reorganizing and sampling the distribution matrix. The sampling of this method is not affected by the probability of discrete random variables, and contains potential representative scenes that can be used for subsequent algorithm selection. At the same time, this method overcomes the defect of repeated scenes generated by probability sampling methods, and directly obtains samples through matrix reorganization, thereby improving computational efficiency.

[0045] The present invention is based on an adaptive selection algorithm of distribution matrix reorganization. Through an optimization-based method, the compact scene set obtained after the distribution matrix reorganization is adaptively selected to obtain a representative reduced scene set, which not only ensures the applicability in high-dimensional scenarios, but also ensures the excellent representativeness of the reduced scene set to the original scene set. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 is a schematic diagram of the scene reduction framework of the present invention and its comparison with the traditional method;

[0047] Figure 2 is an example diagram of distribution matrix reorganization in an embodiment of the present invention;

[0048] Figure 3 is an example diagram of adaptive selection in an embodiment of the present invention;

[0049] Figure 4 It is a schematic diagram of the relationship between the number of repeated samples and the dimensions of the random variables in the scene obtained by sampling using the Monte Carlo method in an embodiment of the present invention;

[0050] Figure 5 Schematic diagram of the relationship between the number of repeated samples and the dimensions of the random variables in the scene obtained by sampling using the D-hypercube method in the embodiment of the present invention;

[0051] Figure 6 is a comparative schematic diagram of the joint probability magnitude distribution histograms of sampling scenarios in different sampling methods in different probability spaces in an embodiment of the present invention;

[0052] Figure 7 4 is a schematic diagram of comparison results of different scene reduction methods in an embodiment of the present invention. DETAILED DESCRIPTION

[0053] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0054] like Figure 1 As shown, the high-dimensional probabilistic preservation scenario reduction method for a large-scale power system of the present invention comprises the following steps:

[0055] Step 1: record the D uncertain objects of the power system as D-dimensional random variables X = {X1, ...X j ,…X D}, the uncertainty value of each variable is discrete into H intervals, denoted as x j ={x 1j ,x 2j ,...,x Hj}, combine the D-dimensional random variable information into a distribution matrix V, where the rows represent the random variables X of each dimension j The uncertainty value of is represented as a D-dimensional random variable, and the corresponding marginal probability distribution matrix P is calculated.

[0056] Based on x with discrete values j ={x 1j ,x 2j ,...,x Hj} of a multidimensional random variable X={X1,…X j ,…X D},

[0057] X1 X2...X D

[0058] The distribution matrix V is defined as It can be expressed as: (x ij ) H×D , where D is the dimension of the random variable, i.e., the uncertainty object, H is the discrete interval of the random variable, and x ij is an element in the distribution matrix V, indicating the power value of the uncertainty object j when it takes the i-th interval, where the uncertainty objects include but are not limited to the wind turbine output, photovoltaic output, and load power consumption at a certain moment.

[0059] Calculate the element x in the distribution matrix V ij The marginal distribution probability f j (x ij ), define the marginal probability distribution matrix P as: That is, (y ij ) H×D, where the element y ij =f j (x ij ).

[0060] Step 2: Divide the distribution matrix V by columns and calculate the Cartesian product to obtain the reorganized matrix And the corresponding marginal probability distribution matrix Reorganize the matrix Sort in descending order to obtain a compact scene set The reorganization matrix It has the same properties as the distribution matrix V.

[0061] This embodiment uses a distribution matrix reorganization algorithm to reorganize and sample the distribution matrix V in blocks to achieve complete coverage of the original distribution probability space. Figure 2 As shown, the distribution matrix reorganization algorithm performs the distribution matrix V

[0062] Divide the columns into blocks, reorganize the elements of each block according to the Cartesian product, and splice each block together to form a reorganized matrix Reorganize the matrix Sampling to get a compact scene set Specifically include:

[0063] Step 2.1: Divide the distribution matrix V into v sub-matrices {V t (t=1,2,...,v)}, the first v-1 sub-matrices each have u columns, if the v-th sub-matrix does not have u columns, the remaining uv-D columns are supplemented with discrete uniform distribution;

[0064] Step 2.2, calculate the submatrix V t The Cartesian product of the elements of u different columns in where k = 1, 2, ... H u , get the recombinant matrix Recombinant matrix according to Splicing into a recombinant matrix The corresponding marginal probability distribution matrix is ​​recorded as

[0065] Step 2.3, restructure the matrix The elements in each column are arranged in descending order according to the corresponding probability, and K rows are randomly selected from them to obtain a compact scene set in

[0066] i1=1,2,...,K, where represents the i1th combined scene, Indicates the value of the element in the vth column in the i1th combination scenario.

[0067] Step 3: Calculate the compact scene set The order of magnitude of probability corresponding to each scene in the , the scene classification is carried out according to the order of magnitude, and the scene set of each probability order of magnitude is obtained

[0068] For compact scene sets The scene set can be obtained by the distribution matrix reorganization algorithm in step 2, or it can be a scene set obtained by Monte Carlo sampling, Markov chain Monte Carlo sampling or Latin hypercube sampling, depending on the sampling method selected in the adaptive selection framework. Specifically, it includes:

[0069] Step 3.1, for any scene where n1, n2, ..., n D ∈{1,2,...,K}, where n represents the set of scenes The nth combined scene in Indicates the specific combination of the nth combination scene, Represents the element of the nth combined scene. Calculate scene X n The joint probability

[0070] Step 3.2, define the calculation function of the probability magnitude of the scenario in Indicates rounding down to calculate the compact scene set The order of magnitude of each scene Get the maximum magnitude and the span between the maximum and minimum magnitudes

[0071] Step 3.3: Divide the scenarios with the same probability magnitude into the same subset 1≤i2≤l, i2∈Z, represents the index of the sub-scenario set divided according to the order of magnitude, and Z is an integer set.

[0072] Step 4: Build an optimization model with the goal of minimizing the Wasserstein distance for the scene set Perform adaptive selection and take the union of the selected sets to obtain the final reduced scene set

[0073] Combination Figure 3 As shown, this embodiment selects the step vector S as the decision variable of the optimization model, solves the optimization model through an adaptive selection algorithm, and obtains an approximate optimal solution. The specific steps include:

[0074] Step 4.1: Define the selection step variable set Among them, s l represents the input variables of the adaptively selected optimization model, that is, from the scene set Select Scene to Scene The step size is used to construct an optimization model for adaptive selection. in Represents a collection of scenes Adaptive selection to scene set The scene in Representation scene set The size of Denotes a random variable X j and The Wasserstein distance between Represents a reduced set of scenes The distribution formed is the marginal distribution of the jth uncertain object;

[0075] Step 4.2: According to the adaptive selection algorithm, solve the optimization model of adaptive selection to obtain the approximate optimal step size set According to the corresponding right Make a selection Step 4.3, given hyperparameters α, β∈Z, 1≤α, β≤D, Z is an integer set, get the reduced scene set

[0076] The above method is implemented by a large-scale power system high-dimensional probabilistic preservation scenario reduction system of this embodiment, and the system includes:

[0077] The distribution matrix building module is used to record the D uncertain objects of the power system as D-dimensional random variables X = {X1, ...X j ,…X D}, the uncertainty value of each variable is discrete into H intervals, denoted as x j ={x 1j ,x 2j ,...,x Hj}, combine the D-dimensional random variable information into a distribution matrix V, where the rows represent the random variables X of each dimension j The uncertainty value of is expressed as a D-dimensional random variable, and the corresponding marginal probability distribution matrix P is calculated;

[0078] The compact scene set acquisition module is used to divide the distribution matrix V by column and calculate the Cartesian product to obtain the reorganized matrix And the corresponding marginal probability distribution matrix Reorganize the matrix Sort in descending order to obtain a compact scene set

[0079] Probability magnitude scene set acquisition module, used to calculate compact scene sets The order of magnitude of the probability corresponding to each scene in the , the scene classification is performed according to the order of magnitude, and the scene set of each order of magnitude is obtained Adaptive selection module, used to build an optimization model with the goal of minimizing the Wasserstein distance, for scene sets Perform adaptive selection and take the union of the selected sets to obtain the final reduced scene set

[0080] In order to verify the present invention, this embodiment uses a test case to obtain the practical results of scenario reduction through the wind power generation process and Python in different dimensions and multiple random variables. It also shows the coverage of the scenarios obtained by the present invention in the entire probability space. In addition, the practical results of other scenario reduction methods are simulated, including traditional Monte Carlo sampling, Markov chain Monte Carlo sampling and Latin hypercube sampling, and they are combined with the adaptive selection framework to obtain Monte Carlo optimization and Latin hypercube optimization methods. Figure 4 and Figure 5 It can be seen that when the dimension D of the random variable and the number of discrete values ​​H of the random variable increase, the number of repeated values ​​decreases. When 10≤D≤16, even in the case of a larger probability space such as 8^12 (H=8, D=12), the number of repetitions of the Monte Carlo and Latin hypercube methods is also large. The present invention uses the Cartesian product of random variables, and the distribution matrix reorganization effectively reduces the occurrence of potential repeated values, making it more conducive to the subsequent optimization method without excluding repeated values.

[0081] like Figure 6 As shown in the figure, the distribution matrix reorganization parameters set in the figure include H = 10 and u = 6, and the number of samples is 10 6 , the number of samples is 10 6 ,The Monte Carlo, Latin Hypercube and Markov Chain Monte Carlo methods shown in the figure are less effective in covering the entire probability space, and are mainly concentrated on high probability samples.,In contrast, the distribution matrix reorganization of the present invention always,continuously covers the entire probability space, even in high-dimensional scenarios such as,D,=48,.

[0082] like Figure 7As shown, the scenarios obtained by using the Monte Carlo method, the Latin Hypercube method and the Markov Monte Carlo method have a high nesting distance, but they are all improved after optimization by the adaptive selection framework. The recombination-selection method proposed in the present invention always maintains an extremely low nesting distance even when the number of random variables increases to 100, and shows an almost linear growth. The present invention comprehensively covers the entire probability space through distribution matrix recombination, which improves the effectiveness of the adaptive selection process, in contrast to the Monte Carlo optimization and Latin Hypercube optimization methods that only cover a limited probability space, especially in ultra-high-dimensional scenarios.

[0083] It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as those generally understood by those skilled in the art in the art to which the present invention belongs. It should also be understood that terms such as those defined in common dictionaries should be understood to have the same meaning as in the context of the prior art, and will not be interpreted with an idealized or overly formal meaning unless defined similarly here.

[0084] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A high-dimensional probabilistic preservation scenario reduction method for large-scale power systems, characterized by: The following operations are included: record the D uncertain objects of the power system as D-dimensional random variables X = {X1, ...X j , ...X D }, the uncertainty value of each variable is discrete into H intervals, denoted as x j ={x 1j , x 2j , ..., x Hj }, combine the D-dimensional random variable information into a distribution matrix V, where the rows represent the random variables X of each dimension j The uncertainty value of is expressed as a D-dimensional random variable, and the corresponding marginal probability distribution matrix P is calculated; Divide the distribution matrix V by columns and calculate the Cartesian product to obtain the reorganized matrix And the corresponding marginal probability distribution matrix Reorganize the matrix Sort in descending order and sample to obtain a compact scene set Computing compact scene sets The order of magnitude of the probability corresponding to each scene in the , the scene classification is performed according to the order of magnitude, and the scene set of each order of magnitude is obtained The optimization model is constructed with the goal of minimizing the Wasserstein distance. Adaptive selection is performed, and the union of the selected sets is taken to obtain the optimal reduced scene set.

2. The high-dimensional probabilistic preservation scenario reduction method for a large-scale power system according to claim 1 is characterized by: Combine the D-dimensional random variable information into a distribution matrix V, where the rows represent the random variables X in each dimension j The uncertainty value of is represented as a D-dimensional random variable, and the corresponding marginal probability distribution matrix P is calculated, which specifically includes the following operations: Based on x with discrete values j ={x 1j , x 2j , ..., x Hj } of a multidimensional random variable X={X1,...X j , ...X D }, the distribution matrix V is defined as (x ij ) H×D , where D is the dimension of the random variable, H is the discrete interval of the random variable, and x ij is an element in the distribution matrix V, representing the power value of the uncertainty object j when it takes the i-th interval, where the uncertainty objects include but are not limited to the wind turbine output, photovoltaic output, and load power consumption at a certain moment; Calculate the element x in the distribution matrix V ij The marginal distribution probability f j (x ij ), define the marginal probability distribution matrix P as (y ij ) H×D , where the element y ij =f j (x ij ).

3. The method for reducing high-dimensional probabilistic scenarios of large-scale power systems according to claim 1 is characterized in that: The distribution matrix V is divided by columns, and the Cartesian product is calculated to obtain the reorganized matrix And the corresponding marginal probability distribution matrix Reorganize the matrix Sort in descending order and sample to obtain a compact scene set The specific operations include: The distribution matrix V is divided into v sub-matrices {V t (t=1,2,...,v)}, the first v-1 sub-matrices each have u columns, if the v-th sub-matrix does not have u columns, the remaining uv-D columns are supplemented with discrete uniform distribution; Calculate the submatrix V t The Cartesian product of the elements of u different columns in where k = 1, 2, ... H u , get the recombinant matrix Recombination matrix according to Splicing into recombinant matrix The corresponding marginal probability distribution matrix is ​​recorded as Reorganize the matrix The elements in each column are arranged in descending order according to the corresponding probability, and K rows are randomly selected from them to obtain a compact scene set in in represents the i1th combined scene, Indicates the value of the element in the vth column in the i1th combination scenario.

4. The high-dimensional probabilistic preservation scenario reduction method for a large-scale power system according to claim 3 is characterized by: The computationally compact scene set The order of magnitude of the probability corresponding to each scene in the , the scene classification is performed according to the order of magnitude, and the scene set of each order of magnitude is obtained Specifically include: For any scene where n1, n2, ..., n D ∈{1, 2, .., K}, where n represents the set of scenes The nth combined scene in Indicates the specific combination of the nth combination scene, An element representing the nth combined scene; Computing scenario X n The joint probability Defines the calculation function of the order of magnitude of the scene in Indicates rounding down to calculate the compact scene set The order of magnitude of each scene Get the maximum magnitude and the span between the maximum and minimum magnitudes Divide scenes of the same magnitude into the same subset An indicator representing a set of sub-scenario sets divided according to magnitude.

5. The method for reducing high-dimensional probabilistic scenarios of large-scale power systems according to claim 4 is characterized in that: The optimization model is constructed with the goal of minimizing the Wasserstein distance. Perform adaptive selection, take the union of the selected sets to obtain the optimal reduced scene set, specifically Including: defining a selection step variable set S = {s1, s2, ..., s l }, where s l represents the input variables of the adaptively selected optimization model, that is, from the scene set Select Scene to Scene The step size is used to construct an optimization model for adaptive selection. in Represents a collection of scenes Adaptive selection to scene set The scene in Representation scene set The size of Denotes a random variable X j and The Wasserstein distance between Represents a reduced set of scenes The distribution formed is the marginal distribution of the jth uncertain object; According to the adaptive selection algorithm, the optimization model of adaptive selection is solved to obtain the approximate optimal step size set S*= According to the corresponding right Make a selection Given hyperparameters α, β∈Z, 1≤α, β≤D, Z is a set of integers, we get the optimal reduced scene set 6. A high-dimensional probabilistic preservation scenario reduction system for large-scale power systems, characterized by: The system comprises: a distribution matrix building module for recording D uncertain objects of the power system as D-dimensional random variables X={X1,...X j , ...X D }, the uncertainty value of each variable is discrete into H intervals, denoted as x j ={x 1j , x 2j , ..., x Hj }, combine the D-dimensional random variable information into a distribution matrix V, where the rows represent the random variables X of each dimension j The uncertainty value of is expressed as a D-dimensional random variable, and the corresponding marginal probability distribution matrix P is calculated; The compact scene set acquisition module is used to divide the distribution matrix V by column and calculate the Cartesian product to obtain the reorganized matrix And the corresponding marginal probability distribution matrix Reorganize the matrix Sort in descending order to obtain a compact scene set Probability magnitude scene set acquisition module, used to calculate compact scene sets The order of magnitude of the probability corresponding to each scene in the , the scene classification is performed according to the order of magnitude, and the scene set of each order of magnitude is obtained Adaptive selection module, used to build an optimization model with the goal of minimizing the Wasserstein distance, for scene sets Adaptive selection is performed, and the union of the selected sets is taken to obtain the optimal reduced scene set.

7. The high-dimensional probabilistic protection scenario reduction system for large-scale power systems according to claim 6, characterized in that: The operations performed by the distribution matrix building module include: Based on x with discrete values j ={x 1j , x 2j , ..., x Hj } of a multidimensional random variable X={X1,...X j , ...X D }, the distribution matrix V is defined as (x ij ) H×D , where D is the dimension of the random variable, H is the discrete interval of the random variable, and x ij is an element in the distribution matrix V, representing the power value of the uncertainty object j when it takes the i-th interval, where the uncertainty objects include but are not limited to the wind turbine output, photovoltaic output, and load power consumption at a certain moment; Calculate the element x in the distribution matrix V ij The marginal distribution probability f j (x ij ), define the marginal probability distribution matrix P as (y ij ) H×D , where the element y ij =f j (x ij ).

8. The high-dimensional probabilistic protection scenario reduction system for large-scale power systems according to claim 6, characterized in that: The operations performed by the compact scene set acquisition module include: The distribution matrix V is divided into v sub-matrices {V t (t=1,2,...,v)}, the first v-1 sub-matrices each have u columns, if the v-th sub-matrix does not have u columns, the remaining uv-D columns are supplemented with discrete uniform distribution; Calculate the submatrix V t The Cartesian product of the elements of u different columns in where k = 1, 2, ... H u , get the recombinant matrix Recombination matrix according to Splicing into recombinant matrix The corresponding marginal probability distribution matrix is ​​recorded as Reorganize the matrix The elements in each column are arranged in descending order according to the corresponding probability, and K rows are randomly selected from them to obtain a compact scene set in in represents the i1th combined scene, Indicates the value of the element in the vth column in the i1th combination scenario.

9. The high-dimensional probabilistic protection scenario reduction system for large-scale power systems according to claim 8, characterized in that: The operations performed by the probability magnitude scene set acquisition module include: For any scene where n1, n2, ..., n D ∈{1, 2, .., K}, where n represents the set of scenes The nth combined scene in Indicates the specific combination of the nth combination scene, An element representing the nth combined scene; Computing scenario X n The joint probability Defines the calculation function of the order of magnitude of the scene in Indicates rounding down to calculate the compact scene set The order of magnitude of each scene Get the maximum magnitude and the span between the maximum and minimum magnitudes Divide scenes of the same magnitude into the same subset An indicator representing a set of sub-scenario sets divided according to magnitude.

10. The high-dimensional probabilistic protection scenario reduction system for large-scale power systems according to claim 6, characterized in that: The operations performed by the adaptive selection module include: Define the selection step variable set S = {s1, s2, ..., s l }, where s l represents the input variables of the adaptively selected optimization model, that is, from the scene set Select Scene to Scene The step size is used to construct an optimization model for adaptive selection. in Represents a collection of scenes Adaptive selection to scene set The scene in Representation scene set The size of Denotes a random variable X j and The Wasserstein distance between Represents a reduced set of scenes The distribution formed is the marginal distribution of the jth uncertain object; According to the adaptive selection algorithm, the optimization model of adaptive selection is solved to obtain the approximate optimal step size set According to the corresponding right Make a selection Given hyperparameters α, β∈Z, 1≤α, β≤D, Z is a set of integers, we get the optimal reduced scene set

Images