Wind and light load multi-stage scene tree generation method based on source load uncertainty feature extraction

Abstract

The invention discloses a wind-light load multi-stage scene tree generation method based on source load uncertainty feature extraction, and the method comprises the steps: extracting wind-light load scene features based on a constructed stacked sparse auto-encoder model, and obtaining a wind-light load scene feature set; clustering the wind-light load scene feature set based on a density peak value improved affinity propagation algorithm to obtain a wind-light load typical day curve as a root node of the scene tree; generating a wind-light load scene tree year by year on the basis of considering different load growth modes according to root nodes of the scene tree; a scene tree is reduced through a scene tree reduction method based on the Sinkhorn distance, so that the scale of the scene tree is reduced. The wind and light load multi-stage scene tree generated by the method can reflect the time sequence information of the new energy output and load, characterize the randomness and volatility of the new energy output and load, can reflect the uncertainty of the long-term increase of the new energy output and load, and is of great significance to the planning and construction of a power grid.

Description

Technical Field

[0001] The present invention belongs to the field of power grid planning and scenario generation, and in particular relates to a method for generating a wind-solar-load multi-stage scenario tree based on source-load uncertainty feature extraction. Background Art

[0002] Renewable energy and loads introduce uncertainty factors into the power grid, increasing the difficulty of system planning and operation decision-making. Currently, the scenario tree generation method is often used to describe the uncertainty of renewable energy output and load on a long time scale. When defining the root node of the scenario tree, the traditional method only considers a single scenario, which often contains only limited data values ​​and cannot represent uncertain information, which is not conducive to the subsequent planning and construction of the power grid.

[0003] Therefore, a new technical solution is needed to solve this problem. Summary of the invention

[0004] Purpose of the invention: In order to overcome the deficiencies in the prior art, a method for generating a wind-solar-load multi-stage scenario tree based on source-load uncertainty feature extraction is provided. The generated wind-solar-load multi-stage scenario tree can reflect the timing information of renewable energy output and load, characterize its randomness and volatility, and reflect the uncertainty of long-term growth of renewable energy output and load, which is of great significance to the planning and construction of power grids.

[0005] Technical solution: To achieve the above purpose, the present invention provides a method for generating a wind-solar-load multi-stage scene tree based on source-load uncertainty feature extraction, comprising the following steps:

[0006] S1: Extract wind-solar-load scene features based on the constructed stacked sparse autoencoder model to obtain a wind-solar-load scene feature set;

[0007] S2: Cluster the wind-solar-load scene feature set based on the improved neighbor propagation algorithm based on the density peak, and obtain the typical daily curve of wind-solar-load as the root node of the scene tree;

[0008] S3: Based on the root node of the scenario tree and considering different load growth patterns, a wind-solar-load scenario tree is generated year by year;

[0009] S4: The scene tree is pruned by the scene tree pruning method based on the Sinkhorn distance to reduce the scale of the scene tree.

[0010] Furthermore, the method for extracting the wind-solar-load scene features in step S1 is specifically as follows:

[0011] A1: Collect historical data of wind, solar and load throughout the year, WT , P PV and P LThey are wind power, photovoltaic and load per unit value data sets, as shown below:

[0012]

[0013]

[0014]

[0015] Where: n um is the total number of days in a year; P WT (i) P PV (i) and P L (i) is a data set consisting of the per-unit values ​​of wind power, photovoltaic power and load at each hour on the i-th day of the whole year; P WT，j , P PV，j and P L,j are the per unit values ​​of wind power, photovoltaic power and load at the kth hour on the i-th day of the whole year respectively;

[0016] s f P WT , P PV and P L The full time sequence scene set composed of the three is as follows:

[0017]

[0018] Where: s f There are n um samples, each sample contains a landscape scene; record s f The i-th sample in m is s f Medium sample dimension;

[0019] A2: Use the built stacked sparse autoencoder SSAE model to s f Feature extraction is performed to obtain the wind, solar and load scene feature extraction results of the scene set.

[0020] The SSAE model in step A2 is composed of M SAEs. SSAE uses the hidden layer of the previous SAE as the input of the next SAE input layer. The training process of SAE includes encoding and decoding. The scene set data features are extracted during the encoding process, as shown in formula (5):

[0021]

[0022] Where: is the output of the hidden layer; W a is the encoding weight matrix; b a is the encoding bias; σ h =1 / (1+e -x) is the activation function;

[0023] During the decoding process, the data is reconstructed and the scene features are extracted, as shown in formula (6):

[0024]

[0025] Where: is the output layer output; W b is the decoding weight matrix; b b is the decoding bias; σ y is the activation function, same as σ h ;

[0026] The decoded output data and the input scene set data constitute an error function, as shown in formula (7):

[0027]

[0028] Where: J SAE is the error function; ρ is the set sparsity coefficient; h j (s f,i ) is the input s f,i The activation value of the hidden layer neuron j when is the average activation of the jth neuron in the hidden layer. f Input into the SSAE model, train the SAEs in the SSAE model one by one according to formula (5-7) until all SAEs have completed training, and obtain s f The feature set y M , the i-th feature sample is

[0029] In the present invention, the stacked sparse autoencoder SSAE can obtain lower-dimensional and deeper features of the wind, solar and load scenes through the stacking of SAE.

[0030] Furthermore, in step S2, f The feature set Y M The clustering index is clustered, and the generated wind-solar load typical time series scene set s c As the root node of the multi-stage scene tree of wind, solar and load, it specifically includes the following steps:

[0031] B1: Similarity calculation: and Y M The i'th and j'th samples in the , calculate the similarity between the samples, and form a similarity matrix

[0032]

[0033] B2: Density peak calculation: Calculate the density peak γ between sample i' and other samples based on local density and relative distance i' ;

[0034] B3: Update the similarity matrix: Update the calculated density peak γ i' is the preference value, which is the diagonal element in the similarity matrix

[0035] s i'i' =-γ i' (9)

[0036] B4: Calculate the attribution matrix and the attraction matrix Elements

[0037]

[0038]

[0039] Where: a i'k' 、s i'k' is the attribution and similarity value of other samples to the i'th sample except the j'th sample; r i”j' is the attractiveness value obtained from other samples;

[0040] B5: Introducing a damping factor to prevent oscillation during iteration

[0041]

[0042]

[0043] Where: d is the number of iterations;

[0044] B6: Determine the cluster center: According to formula (14) Is it a cluster center? If formula (14) is greater than 0, then Corresponding full-time scenario f,i Can be used as cluster center

[0045]

[0046] B7: Determine whether to terminate the iteration: When: 1) the set number of iterations is reached; 2) the cluster center no longer changes, the iteration can be terminated; if not, return to step B5.

[0047] The above method can obtain the cluster centers c1, c2, ..., c K A collection of several landscape scenes K is the number of cluster centers. The cluster center is the typical day scene of wind, solar and load throughout the year, which is the root node of the scene tree. In addition, the number of each scene set obtained by clustering is counted, and the frequency is calculated as the probability of this type of scene set. After clustering is completed, three commonly used clustering indicators, namely silhouette coefficient, CH index and DB index, are selected to analyze the clustering results and verify the effectiveness of clustering.

[0048] Furthermore, the step S3 is specifically as follows:

[0049] C1: Considering the uncertainty of load growth, three different load growth plans are set: high, medium and low;

[0050] C2: Based on the K typical daily scenarios at the root node of the scenario tree, assuming that the load increases year by year and shows high, medium and low growth modes with different probabilities, scenarios are gradually generated to obtain a scenario tree. Each point of the scenario tree represents a typical day 72-dimensional wind-solar load scenario. The probability of each scenario is determined by the probability of each scenario at the root node and the different growth probabilities of the load; the generated wind-solar load scenario tree generated year by year has K possible typical day wind-solar load scenarios in the first year, 3K possible typical day wind-solar load scenarios in the second year, and so on. The scale of the scenario tree grows proportionally with the number of stages, and the scenario tree needs to be cut to reduce its scale.

[0051] Furthermore, in step S4, the scene tree is reduced by using the scene tree reduction method of Sinkhorn distance, which specifically includes the following steps:

[0052] D1: Each node of the original scene tree is recorded as T m,τ , each node represents a 72-dimensional wind, solar and load scene, where τ is the time step of the original scene tree, and m is the number of each node under τ; T m,τ The scene probability corresponding to any node in is recorded as p m,τ , the sum of the scene probabilities in each column is 1; year The planning stage is divided into years, denoted as t, the root node of the scenario tree is denoted as "stage 1", and the following stages are numbered 2, 3, etc. The sum of the scenario probabilities of each stage is also 1; each stage has multiple wind-solar-load scenario sets to be simplified, and an optimal subset is found so that the new probability of each scenario in the subset can be as close as possible to the probability of the original scenario set to reduce the scale of the scenario tree; at any stage t, the number of wind-solar-load scenario sets that need to be simplified is equal to the number of nodes in the previous stage, and the wind-solar-load scenario set to be simplified in each stage is Indicates that t is the number of stages, t=1,2,…,T; w=1,2,…,n t-1 represents the index of the scene set to be reduced corresponding to each node in the previous stage, n t-1is the number of nodes in the previous stage. By default, n1 is the number of typical day wind-solar-load scenarios generated in Section 1. Suppose the number of leaf nodes generated by each node in the previous stage is Then the number of leaf nodes in the next stage is

[0053] D2: When t = 2, the number of nodes in the previous stage is n1, so there are n1 scene sets to be simplified at this time, and each scene set to be simplified The superset is composed of the scenes in the column of the original scene tree belonging to the second stage, and its row number is the number corresponding to each node in the previous stage. The SD (Sinkhorn distance, SD) method is used to convert the superset Reduced to Thus, the scenes in each cluster are obtained and the corresponding probability; when t≥3, each set of scenes to be simplified The scene tree is composed of the scenes in the column of the tth stage in the original scene tree. Its row number is the number corresponding to each point in each cluster in the previous stage. The leaf nodes are output stage by stage. The scene tree and the corresponding probability A multi-stage scene tree is formed.

[0054] Furthermore, the operation of the SD method in step D2 specifically includes the following steps:

[0055] E1: Regularization coefficient γ and entropy of transportation plan H = -∑ i,j π i,j [log(π i,j )-1] introduces the objective function to improve the computational efficiency of finding the optimal wind-solar-load scenario set; the model is shown in formula (15)

[0056]

[0057] Where: π i,j is the element in the optimal transportation plan matrix Π, representing the element (x i ~P(x)) to reduce the scene set (y j ~Q(y)); the transportation matrix Π is an M×N decision matrix, where M and N are x i The scene collection X and y i The size of the scene set Y; c i,j To transfer element x i to y i The cost is usually calculated using the 2-norm; a i and b j are the probabilities of each scene in the scene sets X and Y respectively; the optimal target value is recorded as S D ;

[0058] E2: To solve equation (15), the Sinkhorn-knopp algorithm is used and the Lagrange multiplier is introduced to transform the constrained optimization problem of equation (15) into an unconstrained problem.

[0059]

[0060] Where: α M×1 and β N×1 are the Lagrange multiplier vectors corresponding to the equality constraints; Π is the matrix [π i,j ];

[0061] E3: Derivative (16) to obtain

[0062]

[0063] Solving for π i,j , get the expression of the optimal transportation plan

[0064]

[0065] Where: π i,j is an element in the optimal transport matrix Π; exp(-α i / γ) and exp(-β j / γ) are the diagonal elements of U and V in the diagonal matrix respectively; exp(-c i,j / γ) is an element in the positive definite kernel cost matrix M;

[0066] E4: Then calculate the vectors u and v, which are constructed from the diagonal elements of U and V respectively. The corresponding iterative algorithm is as follows:

[0067]

[0068] Where: g is the number of iterations. When g = 1, v (g) All elements in are 1; the error associated with vectors u and v is given by

[0069]

[0070] E5: The vectors u and v continue to be iteratively calculated until the error term ε a and ε b Below the specified error threshold At this time, the u and v vectors are used as the main diagonal elements to form matrices U and V. The elements in the optimal transportation plan matrix are calculated by formula (18), thus obtaining S D The calculation formula

[0071] S D =∑c i,j π i,j +γ∑πi,j [log(π i,j )-1] (21)

[0072] Furthermore, the optimal wind-solar-load scenario set generation in step S4 includes taking the original wind-solar-load scenario set I as input, simplifying it to generate an optimal subset S containing r' scenarios. * , the specific calculation process is as follows:

[0073] F1: The wind-solar-load scene set to be simplified is I, the size of the scene set after simplification is r', the error threshold β2, the number of initialization iterations, d'=1;

[0074] F2: Define the wind and solar load subset. When d'=1, S (d) is r' scenes randomly selected from superset I. When the number of iterations d ≥ 2, S (d') =S (d'-1) ;

[0075] F3: Calculate I and S using formula (18-20) (d') The Sinkhorn distance S D (d') and the optimal transportation plan matrix Π (d') ;

[0076] F4: By finding the transport matrix Π (d') The column index corresponding to the maximum value of each row can determine the cluster of each scene in the scene set I, thus surrounding S (d') The surrounding clusters C1, C2, ..., C r' ;

[0077] F5: Calculate relative error. When d'=1, continue to step F3. When d'≥2, calculate relative error.

[0078] F6: If the relative error is greater than the error threshold, traverse each wind-solar-load scene in the cluster as the center of the cluster to calculate the transportation distance, and get the minimum intra-cluster transportation distance as the new intra-cluster center, so as to obtain the updated subset S (d') Proceed to the next iteration;

[0079] F7: When the relative error is less than the error threshold, the final reduced subset S is obtained. * :=S (d') , and the corresponding clusters C1,C2,...C r' ;

[0080] F8: Set the probability of occurrence of all scenes in I, and calculate the optimal subset S * Probability:

[0081]

[0082] F9: Output the final optimal output subset S * Contains r' wind, solar and load scenes, each cluster C1, C2, ... C r' and the corresponding probability vector.

[0083] Furthermore, in step S4, the process distance (PD) is introduced to measure the similarity between the wind-solar load scene tree generated after reduction and the original wind-solar load scene tree, so as to evaluate the superiority of the proposed SD-based scene tree reduction method. Based on the introduction of process distance to measure the similarity between the original wind-solar load scene tree and the simplified scene tree, it is defined by formula (22):

[0084]

[0085] Where: N T and N' T are all the node sets of the two scene trees respectively; P1 and P2 are the probability sets of the two scene trees respectively; the symbols f→j and e→i are the predecessors of the intermediate nodes f and e, respectively, of the leaf nodes j and i; p i , p' j , p e and p' f are the probabilities of reaching nodes i, j, e and f respectively; Formula (22) is a linear programming problem, which can be solved by moving from one stage to another in a decomposition manner.

[0086] The present invention proposes a method for generating a wind-solar-load multi-stage scenario tree based on the extraction of source load uncertainty features. The generated wind-solar-load multi-stage scenario tree can reflect the uncertainty of medium- and long-term wind-solar output and load growth. In order to improve the clustering efficiency of wind-solar-load scenarios, a wind-solar-load scenario feature extraction method based on stacked sparse autoencoders and a clustering method based on nearest neighbor propagation improved by density peak are proposed to obtain a typical wind-solar-load curve as the root node of the scenario tree. In order to avoid the generated scenario tree from being too large in scale, a scene tree reduction method based on Sinkhorn distance is proposed to reduce the scale of the scene tree.

[0087] The present invention provides a method for generating a wind-solar-load multi-stage scenario tree based on complex feature extraction and Sinkhorn distance, which can effectively reflect the time series information of new energy output and load, characterize its randomness and volatility, and reflect the uncertainty of new energy output and load on a long time scale. It can assist in making various decisions in system planning and operation, thereby reducing possible safety and economic risks, and further provide support for applications such as system operation control, source-load dynamic planning, improving new energy absorption capacity, and reducing the impact of multi-dimensional uncertainty.

[0088] Beneficial effect: Compared with the prior art, the present invention improves the traditional method of only considering a single scene when defining the scene tree root node, and the scene often only contains limited data values ​​and cannot represent uncertain information. The wind, solar and load multi-stage scene tree generated by the present invention can effectively reflect the timing information of new energy output and load, characterize its randomness and volatility, and reflect the uncertainty of new energy output and load on a long time scale, further providing support for applications such as system operation control, source and load dynamic planning, improving new energy absorption capacity, and reducing the impact of multi-dimensional uncertainty. BRIEF DESCRIPTION OF THE DRAWINGS

[0089] Figure 1 is a flow chart of the method of the present invention;

[0090] Figure 2 Schematic diagram of the structure of the stacked sparse autoencoder model in the present invention;

[0091] Figure 3 A schematic diagram of generating a scene tree of scenery, light and lotus year by year in the present invention;

[0092] Figure 4 A schematic diagram of a simple 3-stage scene tree generated for reduction in the present invention;

[0093] Figure 5 The three-stage wind-solar-load scene tree result diagram generated by the method of the present invention;

[0094] Figure 6 This is a result diagram of the wind-solar-load scenario represented by each point in the three-stage scenario tree generated by the method of the present invention. DETAILED DESCRIPTION

[0095] The present invention is further explained below in conjunction with the accompanying drawings and specific embodiments. It should be understood that these embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention. After reading the present invention, various equivalent forms of modifications to the present invention by those skilled in the art all fall within the scope defined by the claims attached to this application.

[0096] The present invention provides a method for generating a wind-solar-load multi-stage scene tree based on the uncertainty feature extraction of source load, such as Figure 1 As shown, it includes the following steps:

[0097] S1: Extract the wind-solar-load scene features based on the constructed stacked sparse autoencoder model to obtain the wind-solar-load scene feature set, specifically:

[0098] A1: Collect historical data of wind, solar and load throughout the year, WT , P PV and P L They are wind power, photovoltaic and load per unit value data sets, as shown below:

[0099]

[0100]

[0101]

[0102] Where: n um is the total number of days in a year; P WT (i) P PV (i) and P L (i) is a data set consisting of the per-unit values ​​of wind power, photovoltaic power and load at each hour on the i-th day of the whole year; P WT,j , P PV,j and P L,j are the per unit values ​​of wind power, photovoltaic power and load at the kth hour on the i-th day of the whole year respectively;

[0103] s f P WT , P PV and P L The full time sequence scene set composed of the three is as follows:

[0104]

[0105] Where: s f There are n um samples, each sample contains a landscape scene; record s f The i-th sample in m is s f Medium sample dimension;

[0106] A2: Use the built stacked sparse autoencoder SSAE model to s f Feature extraction is performed to obtain the wind, solar and load scene feature extraction results of the scene set.

[0107] like Figure 2 As shown in Figure 5, the SSAE model consists of M SAEs. SSAE uses the hidden layer of the previous SAE as the input of the next SAE input layer. The training process of SAE includes encoding and decoding. The scene set data features are extracted during the encoding process, as shown in formula (5):

[0108]

[0109] Where: is the output of the hidden layer; W a is the encoding weight matrix; b a is the encoding bias; σ h =1 / (1+e -x ) is the activation function;

[0110] During the decoding process, the data is reconstructed and the scene features are extracted, as shown in formula (6):

[0111]

[0112] Where: is the output layer output; W b is the decoding weight matrix; b b is the decoding bias; σ y is the activation function, same as σ h ;

[0113] The decoded output data and the input scene set data constitute an error function, as shown in formula (7):

[0114]

[0115] Where: J SAE is the error function; ρ is the set sparsity coefficient; h j (s f,i ) is the input s f,i The activation value of the hidden layer neuron j when is the average activation of the jth neuron in the hidden layer. f Input into the SSAE model, train the SAEs in the SSAE model one by one according to formula (5-7) until all SAEs have completed training, and obtain s f The feature set y M , the i-th feature sample is

[0116] In the present invention, the stacked sparse autoencoder SSAE can obtain lower-dimensional and deeper features of the wind, solar and load scenes through the stacking of SAE.

[0117] S2: Cluster the wind-solar-load scene feature set based on the improved neighbor propagation algorithm based on the density peak, and obtain the typical daily curve of wind-solar-load as the root node of the scene tree;

[0118] In s f The feature set Y M The clustering index is clustered, and the generated wind-solar load typical time series scene set s c As the root node of the multi-stage scene tree of wind, solar and load, it specifically includes the following steps:

[0119] B1: Similarity calculation: and Y M The i'th and j'th samples in the , calculate the similarity between the samples, and form a similarity matrix

[0120]

[0121] B2: Density peak calculation: Calculate the density peak γ between sample i' and other samples based on local density and relative distance i' ;

[0122] B3: Update the similarity matrix: Update the calculated density peak γ i' is the preference value, which is the diagonal element in the similarity matrix

[0123] s i'i' =-γ i' (9)

[0124] B4: Calculate the attribution matrix and the attraction matrix Elements

[0125]

[0126]

[0127] Where: a i'k' 、s i'k' is the attribution and similarity value of other samples to the i'th sample except the j'th sample; r i”j' is the attractiveness value obtained from other samples;

[0128] B5: Introducing a damping factor to prevent oscillation during iteration

[0129]

[0130]

[0131] Where: d is the number of iterations;

[0132] B6: Determine the cluster center: According to formula (14) Is it a cluster center? If formula (14) is greater than 0, then Corresponding full-time scenario f,i Can be used as cluster center

[0133]

[0134] B7: Determine whether to terminate the iteration: When: 1) the set number of iterations is reached; 2) the cluster center no longer changes, the iteration can be terminated; if not, return to step B5.

[0135] The above method can obtain the cluster centers c1, c2, ..., c K A collection of several landscape scenes K is the number of cluster centers. The cluster center is the typical day scene of wind, solar and load throughout the year, which is the root node of the scene tree. In addition, the number of each scene set obtained by clustering is counted, and the frequency is calculated as the probability of this type of scene set. After clustering is completed, three commonly used clustering indicators, namely silhouette coefficient, CH index and DB index, are selected to analyze the clustering results and verify the effectiveness of clustering.

[0136] S3: Based on the root node of the scenario tree, the wind-solar-load scenario tree is generated year by year based on different load growth patterns, such as Figure 3 As shown, specifically:

[0137] C1: Considering the uncertainty of load growth, three different load growth plans are set: high, medium and low;

[0138] C2: Based on the K typical daily scenarios at the root node of the scenario tree, assuming that the load increases year by year and shows high, medium and low growth modes with different probabilities, scenarios are gradually generated to obtain a scenario tree. Each point of the scenario tree represents a typical day 72-dimensional wind-solar load scenario. The probability of each scenario is determined by the probability of each scenario at the root node and the different growth probabilities of the load; the generated wind-solar load scenario tree generated year by year has K possible typical day wind-solar load scenarios in the first year, 3K possible typical day wind-solar load scenarios in the second year, and so on. The scale of the scenario tree grows proportionally with the number of stages, and the scenario tree needs to be cut to reduce its scale.

[0139] S4: The scene tree is reduced by the scene tree reduction method of Sinkhorn distance to reduce the scale of the scene tree;

[0140] like Figure 4 A simple 3-stage scene tree generation is shown, refer to Figure 4 , the scene tree is reduced by the scene tree reduction method of Sinkhorn distance, which specifically includes the following steps:

[0141] D1: Each node of the original scene tree is recorded as T m,τ , each node represents a 72-dimensional wind, solar and load scene, where τ is the time step of the original scene tree, and m is the number of each node under τ; T m,τ The scene probability corresponding to any node in is recorded as p m,τ , the sum of the scene probabilities in each column is 1; yearThe planning stage is divided into years, denoted as t, the root node of the scenario tree is denoted as "stage 1", and the following stages are numbered 2, 3, etc. The sum of the scenario probabilities of each stage is also 1; each stage has multiple wind-solar-load scenario sets to be simplified, and an optimal subset is found so that the new probability of each scenario in the subset can be as close as possible to the probability of the original scenario set to reduce the scale of the scenario tree; at any stage t, the number of wind-solar-load scenario sets that need to be simplified is equal to the number of nodes in the previous stage, and the wind-solar-load scenario set to be simplified in each stage is Indicates that t is the number of stages, t=1,2,…,T; w=1,2,…,n t-1 represents the index of the scene set to be reduced corresponding to each node in the previous stage, n t-1 is the number of nodes in the previous stage. By default, n1 is the number of typical day wind-solar-load scenarios generated in Section 1. Suppose the number of leaf nodes generated by each node in the previous stage is Then the number of leaf nodes in the next stage is

[0142] D2: When t = 2, the number of nodes in the previous stage is n1, so there are n1 scene sets to be simplified at this time, and each scene set to be simplified The superset is composed of the scenes in the column of the original scene tree belonging to the second stage, and its row number is the number corresponding to each node in the previous stage. The SD (Sinkhorn distance, SD) method is used to convert the superset Reduced to Thus, the scenes in each cluster are obtained and the corresponding probability; when t≥3, each set of scenes to be simplified The scene tree is composed of the scenes in the column of the tth stage in the original scene tree. Its row number is the number corresponding to each point in each cluster in the previous stage. The leaf nodes are output stage by stage. The scene tree and the corresponding probability A multi-stage scene tree is formed.

[0143] The operation of the SD method specifically includes the following steps:

[0144] E1: Regularization coefficient γ and entropy of transportation plan H = -∑ i,j π i,j [log(π i,j )-1] introduces the objective function to improve the computational efficiency of finding the optimal wind-solar-load scenario set; the model is shown in formula (15)

[0145]

[0146] Where: π i,j is the element in the optimal transportation plan matrix Π, representing the element (x i~P(x)) to reduce the scene set (y j ~Q(y)); the transportation matrix Π is an M×N decision matrix, where M and N are x i The scene collection X and y i The size of the scene set Y; c i,j To transfer element x i to y i The cost is usually calculated using the 2-norm; a i and b j are the probabilities of each scene in the scene sets X and Y respectively; the optimal target value is recorded as S D ;

[0147] E2: To solve equation (15), the Sinkhorn-knopp algorithm is used and the Lagrange multiplier is introduced to transform the constrained optimization problem of equation (15) into an unconstrained problem.

[0148]

[0149] Where: α M×1 and β N×1 are the Lagrange multiplier vectors corresponding to the equality constraints; Π is the matrix [π i,j ];

[0150] E3: Derivative (16) to obtain

[0151]

[0152] Solving for π i,j , get the expression of the optimal transportation plan

[0153]

[0154] Where: π i,j is an element in the optimal transport matrix Π; exp(-α i / γ) and exp(-β j / γ) are the diagonal elements of U and V in the diagonal matrix respectively; exp(-c i,j / γ) is an element in the positive definite kernel cost matrix M;

[0155] E4: Then calculate the vectors u and v, which are constructed from the diagonal elements of U and V respectively. The corresponding iterative algorithm is as follows:

[0156]

[0157] Where: g is the number of iterations. When g = 1, v (g) All elements in are 1; the error associated with vectors u and v is given by

[0158]

[0159] E5: The vectors u and v continue to be iteratively calculated until the error term ε a and ε b Below the specified error threshold At this time, the u and v vectors are used as the main diagonal elements to form matrices U and V. The elements in the optimal transportation plan matrix are calculated by formula (18), thus obtaining S D The calculation formula

[0160] S D =∑c i,j π i,j +γ∑π i,j [log(π i,j )-1] (21)

[0161] Based on this optimal wind-solar-load scenario set, the original wind-solar-load scenario set I is taken as input and simplified to generate an optimal subset S containing r' scenarios. * ) is as follows:

[0162] F1: The wind-solar-load scene set to be simplified is I, the size of the scene set after simplification is r', the error threshold β2, the number of initialization iterations, d'=1;

[0163] F2: Define the wind and solar load subset. When d'=1, S (d) is r' scenes randomly selected from superset I. When the number of iterations d ≥ 2, S (d') =S (d'-1) ;

[0164] F3: Calculate I and S using formula (18-20) (d') The Sinkhorn distance S D (d') and the optimal transportation plan matrix Π (d') ;

[0165] F4: By finding the transport matrix Π (d') The column index corresponding to the maximum value of each row can determine the cluster of each scene in the scene set I, thus surrounding S (d') The surrounding clusters C1, C2, ..., C r' ;

[0166] F5: Calculate relative error. When d'=1, continue to step F3. When d'≥2, calculate relative error.

[0167] F6: If the relative error is greater than the error threshold, traverse each wind-solar-load scene in the cluster as the center of the cluster to calculate the transportation distance, and get the minimum intra-cluster transportation distance as the new intra-cluster center, so as to obtain the updated subset S (d') Proceed to the next iteration;

[0168] F7: When the relative error is less than the error threshold, the final reduced subset S is obtained. * :=S (d') , and the corresponding clusters C1,C2,...C r' ;

[0169] F8: Set the probability of occurrence of all scenes in I, and calculate the optimal subset S * Probability:

[0170]

[0171] F9: Output the final optimal output subset S * Contains r' wind, solar and load scenes, each cluster C1, C2, ... C r' and the corresponding probability vector.

[0172] In step S4, the process distance (PD) is introduced to measure the similarity between the wind-solar load scene tree generated after reduction and the original wind-solar load scene tree. Based on the introduction of the process distance to measure the similarity between the original wind-solar load scene tree and the simplified scene tree, it is defined by formula (22):

[0173]

[0174] Where: N T and N' T are all the node sets of the two scene trees respectively; P1 and P2 are the probability sets of the two scene trees respectively; the symbols f→j and e→i are the predecessors of the intermediate nodes f and e, respectively, of the leaf nodes j and i; p i , p' j , p e and p' f are the probabilities of reaching nodes i, j, e and f respectively; Formula (22) is a linear programming problem, which can be solved by moving from one stage to another in a decomposition manner.

[0175] Based on the above-mentioned wind-solar-load multi-stage scenario tree generation method for extracting source-load uncertainty features, a simulation analysis is performed in this embodiment, as follows:

[0176] This embodiment uses 8760h wind power, photovoltaic power and load per unit value in 2023 to form a full-time scenario of wind, photovoltaic and load, and uses the clustering results obtained by the SSAE and improved AP clustering algorithms proposed in this invention. In order to further compare the advantages and disadvantages of clustering effects, three clustering validity test indicators of original AP clustering, improved AP clustering, traditional k-means clustering, PCA+improved AP clustering and SSAE+improved AP clustering are calculated respectively, and the results are shown in Table 1.

[0177] Table 1 Comparison of clustering results

[0178]

[0179] The original AP clustering algorithm obtained 23 clusters, which is a large number. The optimal number of clusters obtained by the improved AP clustering algorithm is 3, which greatly reduces the number of clusters. Among the three clustering indicators, the larger the silhouette coefficient, the larger the CH index, and the smaller the DB index, the tighter the cluster itself, the more dispersed the clusters, and the better the clustering effect. Combined with Table 2, it is found that the clustering effect after feature extraction is better than the result before feature extraction. In addition, the clustering effect after SSAE feature extraction is better than PCA, which shows that compared with PCA, the method proposed in the present invention has the ability to extract scene features more effectively and improves the clustering quality of the improved AP algorithm.

[0180] According to the root node of the scenario tree generated above, assuming that the load increases year by year, with probabilities of 0.22, 0.55 and 0.23, it is expressed as high-speed (4.5%), medium-speed (4%) and low-speed (2%) growth modes, and gradually generates scenarios to obtain a 5-year wind-solar-load scenario tree. Each point of the scenario tree represents a typical day 72-dimensional wind-solar-load scenario. And according to the method proposed by the present invention, the large-scale wind-solar-load scenario tree generated above is reduced to obtain a 3-stage wind-solar-load scenario tree. The scenario sequence number and probability are as follows: Figure 5 As shown in the figure, each node represents the scene as follows Figure 6 As shown. In addition, in order to evaluate the superiority of the Sinkhorn reduction scene tree method proposed in the present invention, a linear programming method based on traditional optimal transmission theory is used as a comparison method to evaluate the two methods in terms of the calculation accuracy and calculation efficiency of the reduction. The process distance and total calculation time calculated by the two methods are shown in Table 2 below. As can be seen from Table 2, the method of the present invention has a certain improvement in calculation accuracy and calculation efficiency.

[0181] Table 2 Process distance and time calculated by 2 methods

[0182]

[0183] The above embodiments are only for illustrating the technical idea of ​​the present invention, and cannot be used to limit the protection scope of the present invention. Any changes made on the basis of the technical solution in accordance with the technical idea proposed by the present invention shall fall within the protection scope of the present invention; any technology not involved in the present invention can be realized by existing technologies.

Claims

1. A method for generating a multi-stage scene tree of wind and solar load based on uncertainty feature extraction of source load, characterized in that: The steps include: S1: Extract wind-solar-load scene features based on the constructed stacked sparse autoencoder model to obtain a wind-solar-load scene feature set; S2: Cluster the wind-solar-load scene feature set based on the improved neighbor propagation algorithm based on the density peak, and obtain the typical daily curve of wind-solar-load as the root node of the scene tree; S3: Based on the root node of the scenario tree and considering different load growth patterns, a wind-solar-load scenario tree is generated year by year; S4: The scene tree is pruned by the scene tree pruning method based on the Sinkhorn distance to reduce the scale of the scene tree.

2. According to claim 1, a method for generating a wind-solar-load multi-stage scene tree based on source-load uncertainty feature extraction is characterized in that: The method for extracting the wind-solar-load scene features in step S1 is specifically as follows: A1: Collect historical data of wind, solar and load throughout the year, WT , P PV and P L They are wind power, photovoltaic and load per unit value data sets, as shown below: Where: n um is the total number of days in a year; P WT (i) P PV (i) and P L (i) is a data set consisting of the per-unit values ​​of wind power, photovoltaic power and load at each hour on the i-th day of the whole year; P WT,j , P PV,j and P L,j are the per unit values ​​of wind power, photovoltaic power and load at the kth hour on the i-th day of the whole year respectively; s f P WT , P PV and P L The full time sequence scene set composed of the three is as follows: Where: s f There are n um samples, each sample contains a landscape scene; record s f The i-th sample in m is s f Medium sample dimension; A2: Use the built stacked sparse autoencoder SSAE model to s f Feature extraction is performed to obtain the wind, solar and load scene feature extraction results of the scene set.

3. The method for generating a wind-solar-load multi-stage scene tree based on source-load uncertainty feature extraction according to claim 2 is characterized in that: The SSAE model in step A2 is composed of M SAEs. The SSAE uses the hidden layer of the previous SAE as the input of the next SAE input layer. The training process of SAE includes encoding and decoding. The scene set data features are extracted during the encoding process, as shown in formula (5): Where: is the output of the hidden layer; W a is the encoding weight matrix; b a is the encoding bias; σ h =1 / (1+e -x ) is the activation function; During the decoding process, the data is reconstructed and the scene features are extracted, as shown in formula (6): Where: is the output layer output; W b is the decoding weight matrix; b b is the decoding bias; σ y is the activation function, same as σ h ; The decoded output data and the input scene set data constitute an error function, as shown in formula (7): Where: J SAE is the error function; ρ is the set sparsity coefficient; h j (s f,i ) is the input s f,i The activation value of the hidden layer neuron j when is the average activation of the jth neuron in the hidden layer. f Input into the SSAE model, train the SAEs in the SSAE model one by one according to formula (5-7) until all SAEs have completed training, and obtain s f The feature set y M , the i-th feature sample is 4. The method for generating a wind-solar-load multi-stage scene tree based on source-load uncertainty feature extraction according to claim 3 is characterized in that: In step S2, s f The feature set Y M The clustering index is clustered, and the generated wind-solar load typical time series scene set s c As the root node of the multi-stage scene tree of wind, solar and load, it specifically includes the following steps: B1: Similarity calculation: and Y M The i'th and j'th samples in the , calculate the similarity between the samples, and form a similarity matrix B2: Density peak calculation: Calculate the density peak γ between sample i' and other samples based on local density and relative distance i' ; B3: Update the similarity matrix: Update the calculated density peak γ i' is the preference value, which is the diagonal element in the similarity matrix s i'i' =-γ i' (9) B4: Calculate the attribution matrix and the attraction matrix Elements Where: a i'k' 、s i'k' is the attribution and similarity value of other samples to the i'th sample except the j'th sample; r i”j' is the attractiveness value obtained from other samples; B5: Introducing a damping factor to prevent oscillation during iteration Where: d is the number of iterations; B6: Determine the cluster center: According to formula (14) Is it a cluster center? If formula (14) is greater than 0, then Corresponding full-time scenario f,i Can be used as cluster center B7: Determine whether to terminate the iteration: When: 1) the set number of iterations is reached; 2) The cluster center no longer changes, and the iteration can be terminated; if not, return to step B5.

5. The method for generating a wind-solar-load multi-stage scene tree based on source-load uncertainty feature extraction according to claim 4 is characterized in that: The step S3 is specifically as follows: C1: Considering the uncertainty of load growth, three different load growth plans are set: high, medium and low; C2: Based on the K typical daily scenarios at the root node of the scenario tree, assuming that the load increases year by year and shows high, medium and low growth modes with different probabilities, scenarios are gradually generated to obtain a scenario tree. Each point of the scenario tree represents a typical day 72-dimensional wind-solar-load scenario. The probability of each scenario is determined by the probability of each scenario at the root node and the different growth probabilities of the load; the generated year-by-year wind-solar-load scenario tree has K possible typical day wind-solar-load scenarios in the first year, 3K possible typical day wind-solar-load scenarios in the second year, and so on.

6. The method for generating a wind-solar-load multi-stage scene tree based on source-load uncertainty feature extraction according to claim 5 is characterized in that: The scene tree is reduced by the scene tree reduction method of Sinkhorn distance in step S4, which specifically includes the following steps: D1: Each node of the original scene tree is recorded as T m,τ , each node represents a 72-dimensional wind, solar and load scene, where τ is the time step of the original scene tree, and m is the number of each node under τ; T m,τ The scene probability corresponding to any node in is recorded as p m,τ , the sum of the scene probabilities in each column is 1; year The planning stage is divided into years, denoted as t, and the root node of the scenario tree is denoted as "stage 1". The sum of the scenario probabilities of each stage is also 1; in each stage, there are multiple wind-solar-load scenario sets to be simplified, and an optimal subset is found so that the new probability of each scenario in the subset can be as close as possible to the probability of the original scenario set to reduce the scale of the scenario tree; at any stage t, the number of wind-solar-load scenario sets that need to be simplified is equal to the number of nodes in the previous stage, and the wind-solar-load scenario set to be simplified in each stage is Indicates that t is the number of stages, t=1,2,…,T; w=1,2,…,n t-1 represents the index of the scene set to be reduced corresponding to each node in the previous stage, n t-1 is the number of nodes in the previous stage. By default, n1 is the number of typical day wind-solar-load scenarios generated in Section 1. Suppose the number of leaf nodes generated by each node in the previous stage is Then the number of leaf nodes in the next stage is D2: When t = 2, the number of nodes in the previous stage is n1, so there are n1 scene sets to be simplified at this time, and each scene set to be simplified The superset is composed of the scenes in the column belonging to the second stage in the original scene tree, and its row number is the number corresponding to each node in the previous stage. The SD-based method is used to convert the superset Reduced to Thus, the scenes in each cluster are obtained and the corresponding probability; when t≥3, each set of scenes to be simplified The scene tree is composed of the scenes in the column of the tth stage in the original scene tree. Its row number is the number corresponding to each point in each cluster in the previous stage. The leaf nodes are output stage by stage. The scene tree and the corresponding probability A multi-stage scene tree is formed.

7. The method for generating a wind-solar-load multi-stage scene tree based on source-load uncertainty feature extraction according to claim 6 is characterized in that: The operation of the SD method in step D2 specifically includes the following steps: E1: Regularization coefficient γ and entropy of transportation plan H = -∑ i,j π i,j [log(π i,j )-1] introduces the objective function to improve the computational efficiency of finding the optimal wind-solar-load scenario set; the model is shown in formula (15) Where: π i,j is the element in the optimal transportation plan matrix Π, representing the element (x i ~P(x)) to reduce the scene set (y j ~Q(y)); the transportation matrix Π is an M×N decision matrix, where M and N are x i The scene collection X and y i The size of the scene set Y; c i,j To transfer element x i to y i The cost of i and b j are the probabilities of each scene in the scene sets X and Y respectively; the optimal target value is recorded as S D ; E2: To solve equation (15), the Sinkhorn-knopp algorithm is used and the Lagrange multiplier is introduced to transform the constrained optimization problem of equation (15) into an unconstrained problem. Where: α M×1 and β N×1 are the Lagrange multiplier vectors corresponding to the equality constraints; Π is the matrix [π i,j ]; E3: Derivative (16) to obtain Solving for π i,j , get the expression of the optimal transportation plan Where: π i,j is an element in the optimal transport matrix Π; exp(-α i / γ) and exp(-β j / γ) are the diagonal elements of U and V in the diagonal matrix respectively; exp(-c i,j / γ) is an element in the positive definite kernel cost matrix M; E4: Then calculate the vectors u and v, which are constructed from the diagonal elements of U and V respectively. The corresponding iterative algorithm is as follows: Where: g is the number of iterations. When g = 1, v (g) All elements in are 1; the error associated with vectors u and v is given by E5: The vectors u and v continue to be iteratively calculated until the error term ε a and ε b Below the specified error threshold At this time, the u and v vectors are used as the main diagonal elements to form matrices U and V. The elements in the optimal transportation plan matrix are calculated by formula (18), thus obtaining S D The calculation formula S D =∑c i,j p i,j +γ∑π i,j [log(π i,j )-1] (21).

8. The method for generating a wind-solar-load multi-stage scene tree based on source-load uncertainty feature extraction according to claim 7 is characterized in that: The generation of the optimal wind-solar-load scenario set in step S4 includes taking the original wind-solar-load scenario set I as input, simplifying it to generate an optimal subset S containing r' scenarios * , the specific calculation process is as follows: F1: The wind-solar-load scene set to be simplified is I, the size of the scene set after simplification is r', the error threshold β2, the number of initialization iterations, d'=1; F2: Define the wind and solar load subset. When d'=1, S (d) is r' scenes randomly selected from superset I. When the number of iterations d ≥ 2, S (d') =S (d'-1) ; F3: Calculate I and S using formula (18-20) (d’) The Sinkhorn distance S D (d’) and the optimal transportation plan matrix Π (d’) ; F4: By finding the transport matrix Π (d’) The column index corresponding to the maximum value of each row can determine the cluster of each scene in the scene set I, thus surrounding S (d’) The surrounding clusters C1, C2, ..., C r' ; F5: Calculate relative error. When d'=1, continue to step F3. When d'≥2, calculate relative error. F6: If the relative error is greater than the error threshold, traverse each wind-solar-load scene in the cluster as the center of the cluster to calculate the transportation distance, and get the minimum intra-cluster transportation distance as the new intra-cluster center, so as to obtain the updated subset S (d’) Proceed to the next iteration; F7: When the relative error is less than the error threshold, the final reduced subset S is obtained. * :=S (d’) , and the corresponding clusters C1,C2,...C r' ; F8: Set the probability of occurrence of all scenes in I, and calculate the optimal subset S * Probability: F9: Output the final optimal output subset S * Contains r' wind, solar and load scenes, each cluster C1, C2, ...C r' and the corresponding probability vector.

9. The method for generating a wind-solar-load multi-stage scene tree based on source-load uncertainty feature extraction according to claim 1 is characterized in that: In step S4, the process distance is introduced to measure the similarity between the wind-solar-load scene tree generated after the reduction and the original wind-solar-load scene tree.

10. The method for generating a wind-solar-load multi-stage scene tree based on source-load uncertainty feature extraction according to claim 9 is characterized in that: In step S4, the similarity between the original scene tree and the simplified scene tree is measured based on the introduced process distance, which is defined by formula (22): Where: N T and N' T are all the node sets of the two scene trees respectively; P1 and P2 are the probability sets of the two scene trees respectively; the symbols f→j and e→i are the predecessors of the intermediate nodes f and e, respectively, of the leaf nodes j and i; p i , p' j , p e and p' f are the probabilities of reaching nodes i, j, e and f respectively; Formula (22) is a linear programming problem, which can be solved by moving from one stage to another in a decomposition manner.

Images