Active power distribution network comprehensive optimization method and system

Abstract

The invention discloses an active power distribution network comprehensive optimization method and system, and the method comprises the steps: carrying out the scene clustering of an obtained power load prediction output curve, a photovoltaic device prediction output curve and a wind power device prediction output curve through employing an improved ISODATA algorithm, and obtaining a scene set; decomposing variables of the active power distribution network into a first type of variables and a second type of variables based on the scene set, constructing a first objective function and a first constraint condition of a first stage model based on the first type of variables, and constructing a second objective function and a second constraint condition of a second stage model based on the second type of variables, and respectively solving the first-stage model and the second-stage model to obtain a first-class variable value and a second-class variable value, constructing a distribution robust optimization model by using the first-class variable value and the second-class variable value, and solving the distribution robust optimization model to obtain an optimization result, thereby effectively improving the reliability of comprehensive optimization of the active power distribution network.

Description

Technical Field

[0001] The present invention relates to the technical field of electrical automation, and particularly to an integrated optimization method and system for an active distribution network. Background Art

[0002] In the field of power systems, clustering analysis, as a key data processing method, is of great significance for analyzing the load characteristics of power users. It can accurately classify power users according to their load characteristics, providing strong support for power system planning, operation, and market strategy formulation.

[0003] The traditional K-means clustering algorithm is widely used in power market research. It aims to divide user groups with different load characteristics by analyzing user load data, providing a basis for power pricing, resource allocation, and power grid planning. However, this algorithm has many defects. When facing massive high-dimensional data, it is difficult to determine the k value, and different k values will produce quite different clustering results; the randomly selected initial clustering centers may lead to slow convergence speed and poor clustering effect; it is sensitive to noise and outliers and is not suitable for tasks with strong discreteness and imbalanced sample categories; the distance measurement method is single. For example, the commonly used Euclidean distance cannot effectively capture high-dimensional features such as the temporal variation of load curves, and cannot distinguish the importance of load data at different times.

[0004] The ISODATA (Iterative Self-Organizing Data Analysis Technique Algorithm) algorithm improves the deficiencies of the K-means algorithm to a certain extent. Its k value can change dynamically during the clustering process, and its adaptability to different data distributions is improved through splitting and merging operations. However, the ISODATA algorithm still has defects. For example, the randomly selected initial clustering centers affect the clustering effect and convergence speed, the default Euclidean distance measurement cannot effectively extract high-dimensional features of load curves, and more parameters need to be determined in advance, increasing the difficulty and complexity of algorithm application.

[0005] With the wide access of distributed generation (DG) to the distribution network, the contradiction between the uncertainty of DG output and the system flexibility poses a huge challenge to system security. In dealing with the uncertainty of DG, there are problems such as statistical optimality and conservatism in stochastic optimization and robust optimization. Although distributionally robust optimization can balance the contradiction between the two, when describing the uncertainty of DG and load by the scenario method, scenario generation is crucial. In existing research, insufficient consideration is given to the correlation of sample data during scenario generation, the determination of the number of clusters lacks rationality, and in the model constraints considering energy storage, micro gas turbines, and network topology, there are problems such as dynamic constraints affecting the solution efficiency and insufficient consideration of temporal coupling constraints, resulting in the optimization result being only a local optimum and affecting the engineering practical value. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to provide an active distribution network comprehensive optimization method and system, which can effectively improve the reliability of the active distribution network comprehensive optimization.

[0007] To solve the above technical problem, a technical solution adopted by the present invention is:

[0008] An active distribution network comprehensive optimization method, comprising the steps of:

[0009] Using an improved iterative self-organizing data analysis algorithm to perform scenario clustering on the obtained power load prediction output curve, photovoltaic device prediction output curve, and wind power device prediction output curve to obtain a scenario set. The improved iterative self-organizing data analysis algorithm is used to optimize the selection strategy of the initial clustering center to reduce the number of algorithm iterations, and combines the kernel method to capture the characteristics of the high-dimensional space;

[0010] Based on the scenario set, decompose the variables of the active distribution network into first-class variables and second-class variables. The first-class variables include continuous variables related to micro gas turbines and energy storage, and the second-class variables are discrete variables related to network topology;

[0011] Based on the first-class variables, construct the first objective function and the first constraint condition of the first-stage model, and based on the second-class variables, construct the second objective function and the second constraint condition of the second-stage model;

[0012] Based on the first objective function and the first constraint condition, solve the first-stage model to obtain the first-class variable values, and based on the first-class variable values, solve the second-stage model according to the second objective function and the second constraint condition to obtain the second-class variable values;

[0013] Use the first-class variable values and the second-class variable values to construct a distributionally robust optimization model, and solve the distributionally robust optimization model to obtain the optimization result.

[0014] To solve the above technical problem, another technical solution adopted by the present invention is:

[0015] An active distribution network comprehensive optimization system, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the following steps are implemented:

[0016] The improved iterative self-organizing data analysis algorithm is used to perform scenario clustering on the obtained power load prediction output curve, photovoltaic device prediction output curve, and wind power device prediction output curve to obtain a scenario set. The improved iterative self-organizing data analysis algorithm is used to optimize the selection strategy of the initial clustering center to reduce the number of algorithm iterations and capture the characteristics of the high-dimensional space in combination with the kernel method;

[0017] Based on the scenario set, the variables of the active distribution network are decomposed into first-class variables and second-class variables. The first-class variables include continuous variables related to micro gas turbines and energy storage, and the second-class variables are discrete variables related to network topology;

[0018] Based on the first-class variables, the first objective function and the first constraint condition of the first-stage model are constructed, and based on the second-class variables, the second objective function and the second constraint condition of the second-stage model are constructed;

[0019] Based on the first objective function and the first constraint condition, the first-stage model is solved to obtain the first-class variable values, and based on the first-class variable values, the second-stage model is solved according to the second objective function and the second constraint condition to obtain the second-class variable values;

[0020] The first-class variable values and the second-class variable values are used to construct a distributionally robust optimization model, and the distributionally robust optimization model is solved to obtain an optimization result.

[0021] The beneficial effects of the present invention are as follows: The improved ISODATA algorithm is used to perform scenario clustering on the obtained power load prediction output curve, photovoltaic device prediction output curve, and wind power device prediction output curve to obtain a scenario set. Based on the scenario set, the variables of the active distribution network are decomposed into first-class variables and second-class variables. Based on the first-class variables, the first objective function and the first constraint condition of the first-stage model are constructed, and based on the second-class variables, the second objective function and the second constraint condition of the second-stage model are constructed. The first-stage model and the second-stage model are solved respectively to obtain the first-class variable values and the second-class variable values. The first-class variable values and the second-class variable values are used to construct a distributionally robust optimization model and solve it to obtain an optimization result. The improved ISODATA algorithm optimizes the selection strategy of the initial clustering center compared with the existing ISODATA algorithm to reduce the number of algorithm iterations, and captures the characteristics of the high-dimensional space in combination with the kernel method. The clustering error is small and the robustness is strong, greatly improving the accuracy and reliability of scenario construction. At the same time, the first-stage model and the second-stage model are established respectively, enhancing the adaptability of the model to different types of variables. Using the obtained first-class variable values and second-class variable values to construct a distributionally robust optimization model and solve it, thereby effectively improving the reliability of the comprehensive optimization of the active distribution network. Description of the Drawings

[0022] Figure 1 This is the flowchart of the steps of an active distribution network comprehensive optimization method according to an embodiment of the present invention;

[0023] Figure 2 This is the schematic structural diagram of an active distribution network comprehensive optimization system according to an embodiment of the present invention;

[0024] Figure 3 This is the clustering flowchart in the active distribution network comprehensive optimization method according to an embodiment of the present invention;

[0025] Figure 4 This is the schematic diagram of the clustering result in the active distribution network comprehensive optimization method according to an embodiment of the present invention;

[0026] Figure 4 In (a), it is the schematic diagram of the first type of center obtained by clustering in the active distribution network comprehensive optimization method according to an embodiment of the present invention;

[0027] Figure 4 In (b), it is the schematic diagram of the second type of center obtained by clustering in the active distribution network comprehensive optimization method according to an embodiment of the present invention;

[0028] Figure 4 In (c), it is the schematic diagram of the third type of center obtained by clustering in the active distribution network comprehensive optimization method according to an embodiment of the present invention;

[0029] Figure 4 In (d), it is the schematic diagram of the fourth type of center obtained by clustering in the active distribution network comprehensive optimization method according to an embodiment of the present invention. Detailed implementation manner

[0030] To illustrate the technical content, achieved purpose and effects of the present invention in detail, the following is described in conjunction with the embodiments and accompanied by the drawings.

[0031] The above-mentioned active distribution network comprehensive optimization method and system of the present invention can be applied to the active distribution network, which is described as follows through specific embodiments:

[0032] Please refer to Figure 1 、 Figure 3 and Figure 4 , the first embodiment of the present invention is:

[0033] An active distribution network comprehensive optimization method, including the steps:

[0034] S1. Use the improved iterative self-organizing data analysis algorithm to perform scenario clustering on the obtained power load prediction output curve, photovoltaic device prediction output curve, and wind power device prediction output curve to obtain a scenario set. The improved iterative self-organizing data analysis algorithm is used to optimize the selection strategy of the initial clustering center to reduce the number of algorithm iterations, and combines the kernel method to capture the characteristics in the high-dimensional space, such as Figure 3 shown in the following, specifically including S11 - S18:

[0035] The objectives of the above scenario clustering include: (1) Identifying the load and new energy output patterns: Mining the output change laws of the power load, photovoltaic, and wind power devices through cluster analysis, and revealing their internal patterns and characteristics; (2) Optimizing the system operation: Based on the clustering results, optimizing the operation strategy of the power system to improve the operation efficiency and economic benefits of the system; (3) Reasonably allocating resources: According to different load and new energy output patterns, reasonably allocating power resources to ensure the stability and reliability of the power system operation. Perform the subsequent steps based on this objective.

[0036] S11. Use the feature extraction model based on the gradient boosting tree algorithm to extract the key features of the obtained power load prediction output curve, photovoltaic device prediction output curve, and wind power device prediction output curve.

[0037] In an alternative embodiment, before S11, it may further include:

[0038] Establish an initial feature extraction model based on the gradient boosting tree algorithm;

[0039] Train the initial feature extraction model based on the gradient boosting tree algorithm to obtain a feature extraction model based on the gradient boosting tree algorithm.

[0040] Among them, the establishment of the initial feature extraction model based on the gradient boosting tree algorithm includes:

[0041] Define the model structure composed of multiple decision trees based on the boosting idea; in the t-th iteration, the t-th decision tree learns the prediction error of the previous t - 1 trees, that is, the residual. The output of the model is:

[0042]

[0043] In the formula, F(x) represents the predicted value of the model, α t represents the weight coefficient of each tree, h t (x) represents the prediction result of the t-th decision tree, and T represents the total number of decision trees.

[0044] The training of the initial feature extraction model based on the gradient boosting tree algorithm to obtain a feature extraction model based on the gradient boosting tree algorithm includes:

[0045] Determine the training objective, and use the negative gradient of the loss function under the initial feature extraction model based on the gradient boosting tree algorithm as the residual of the boosting tree;

[0046] The loss function L(y, F(x)) represents the error between the actual value y and the predicted value F(x), and the negative gradient as the new residual is:

[0047]

[0048] In the formula, r i (t) represents the residual of the i-th sample at the t-th iteration, y i represents the actual value of the i-th sample, F(x i ) represents the predicted value of the current model for the i-th sample;

[0049] Train a single decision tree, select the splitting feature from the feature set according to the principle of minimizing the error, and divide the data into two child nodes;

[0050] In each iteration, train a new decision tree h t (x), with the goal of minimizing the current residual r i (t) Select the splitting point from the feature set, y i is the actual value, is the predicted value, and the best splitting feature is determined by minimizing the mean squared error (MSE) after splitting. The formula is:

[0051]

[0052] Iteratively update the initial feature extraction model based on the gradient boosting tree algorithm, and successively train multiple decision trees based on the new residuals;

[0053] The residuals learned by each tree are added to the initial feature extraction model based on the gradient boosting tree algorithm to update the model predicted value. Specifically:

[0054] F (t) (x) = F (t-1) (x) + θ t h t (x);

[0055] In the formula, θ t represents the learning rate, which is used to control the contribution of each tree to the final model;

[0056] Evaluate the feature importance, and quantify the contribution of the feature by counting the number of times the feature is selected in the splitting nodes;

[0057] The importance of the feature is quantified by counting the number of times each feature is selected in the decision tree splitting nodes. For feature f j the importance calculation formula is:

[0058]

[0059] In the formula, I(f j ) represents the importance of feature f j is the set of nodes where feature f j is used for splitting in the t-th tree, represents the reduction in loss caused by the split of node s.

[0060] In an optional implementation, before S11, it may further include:

[0061] Collect the power load prediction output curve, the photovoltaic device prediction output curve, and the wind power device prediction output curve; In an optional implementation, it may further include: collecting time series data and related features (such as weather conditions, seasons, etc.).

[0062] Unify the data formats and align the timestamps of the power load prediction output curve, the photovoltaic device prediction output curve, and the wind power device prediction output curve to obtain the aligned power load prediction output curve, photovoltaic device prediction output curve, and wind power device prediction output curve;

[0063] Perform integrity and consistency checks on the aligned power load prediction output curve, photovoltaic device prediction output curve, and wind power device prediction output curve to ensure that the data can be used for subsequent analysis;

[0064] Identify the missing values in the aligned power load prediction output curve, photovoltaic device prediction output curve, and wind power device prediction output curve, and perform filling or deletion processing on the missing values to obtain the processed power load prediction output curve, photovoltaic device prediction output curve, and wind power device prediction output curve;

[0065] Perform normalization processing on the processed power load prediction output curve, photovoltaic device prediction output curve, and wind power device prediction output curve to eliminate the dimension difference and make the data on the same scale, obtaining the normalized power load prediction output curve, photovoltaic device prediction output curve, and wind power device prediction output curve. The subsequent steps can be executed using the normalized power load prediction output curve, photovoltaic device prediction output curve, and wind power device prediction output curve.

[0066] Among them, the performing filling or deletion processing on the missing values to obtain the processed power load prediction output curve, photovoltaic device prediction output curve, and wind power device prediction output curve includes:

[0067] If the missing value is a data segment with a relatively high missing ratio or data that has little impact on the analysis, delete the missing value to obtain the processed power load prediction output curve, photovoltaic device prediction output curve, and wind power device prediction output curve;

[0068] If the missing value is data with a relatively low missing ratio, perform filling processing on the missing value to obtain the processed power load prediction output curve, photovoltaic device prediction output curve, and wind power device prediction output curve.

[0069] The filling process includes the following steps (1)-(2):

[0070] (1) For a sample with data to be filled, according to a predetermined distance metric method, find the k samples in the sample space that are closest to it, and record these k samples and their corresponding distances d1, d2, d3,..., d k ;

[0071] In an alternative embodiment, the predetermined distance metric method is the Euclidean distance. The Euclidean distance d(x i =(x i1 , x i2 , x i3 ,..., x in ) and x j =(x j1 , x j2 , x j3 ,..., x jn ) is given by: i , x j ) is:

[0072]

[0073] (2) Perform a weighted average calculation on these k nearest samples according to the distance values, and the resulting value is used as the filled data y. The relationship between the filled data y and these k nearest samples can be expressed as:

[0074]

[0075] where w i represents the weight value of the i-th sample, and y i represents the value of the i-th nearest sample.

[0076] The normalization process is to scale the original data to the range of [0, 1] through a linear transformation, specifically:

[0077]

[0078] Wherein, X represents the original data sequence, X norm represents the normalized data, X min represents the minimum value in the original data sequence, X max represents the maximum value in the original data sequence.

[0079] S12. Select the initial clustering centers, specifically including S121 - S125:

[0080] S121. Randomly select a first clustering center from the dataset composed of the power load prediction output curve, the photovoltaic device prediction output curve, and the wind power device prediction output curve.

[0081] S122. For each sample in the dataset, calculate the distance from the sample to each clustering center, and select the shortest distance among the distances.

[0082] S123. Determine the probability that the sample is selected as the next clustering center according to the shortest distance.

[0083] The specific probability is:

[0084]

[0085] Wherein, d(x i ) represents the shortest distance among the distances from the sample x i to each clustering center, and D represents the sample set.

[0086] The greater the probability of the sample, the higher the probability of its being selected.

[0087] S124. Select the subsequent clustering centers from the dataset according to the probability.

[0088] S125. Obtain the initial clustering centers according to the first clustering center and the subsequent clustering centers.

[0089] When selecting the subsequent clustering centers, assume that n initial clustering centers have been selected (0 < n < k). When selecting the (n + 1)-th clustering center, by increasing the probability of selecting points farther away from the current n clustering centers, it is ensured that the k initial clustering centers are kept as far apart as possible, improving the clustering effect. By reducing the situation where the randomly distributed initial points are too concentrated, it avoids problems such as slow convergence of the clustering algorithm and contingency of results caused by randomly selecting the initial clustering centers, reduces the number of iterations, and improves the convergence speed and clustering quality.

[0090] The value of k for the cluster centers in the present invention is a value that dynamically changes during the program iteration process. During the iteration process of the algorithm, the k value will be dynamically adjusted according to the actual situation, and finally an appropriate k value will be found. This is also the difference from the traditional K-means algorithm. The traditional K-means algorithm requires a person to give the k value, and once given, it will not change.

[0091] S13. Use the Gaussian kernel function based on the initial cluster centers to cluster the key features to obtain a clustering result.

[0092] Specifically, the data in the original input space (such as the power load prediction output curve) is non-linearly mapped to a high-dimensional feature space to improve the linear separability of data points.

[0093] Suppose X is the input space and H is a certain high-dimensional space. There exists a mapping function φ(x) that maps a point x in the input space to a point h in the high-dimensional space H, that is, h = φ(x). If there exists a function Θ(x, y) that satisfies the following conditions for all x, y ∈ X:

[0094] Θ(x, y) = φ(x) · φ(y);

[0095] In the formula, φ(x) · φ(y) represents the inner product of the mapped points φ(x) and φ(y), then Θ(x, y) is called a kernel function. Through the kernel function, the inner product of two points after mapping can be directly calculated in the original space, thus avoiding explicitly calculating the mapping in the high-dimensional space.

[0096] Among them, the kernel function is a Gaussian kernel function, specifically:

[0097] Θ(x, y) = exp(-γ||x - y|| 2 );

[0098] In the formula, γ represents the parameter in the Gaussian kernel function, which controls the influence range of a single training sample. The larger the γ value, the smaller the influence range of the sample, and the faster the kernel function value decays. The smaller the γ value, the larger the influence range of the sample, and the greater the influence of a single sample on the whole.

[0099] Based on the Gaussian kernel function, use a clustering algorithm in the new high-dimensional space to complete clustering by calculating the distance between samples.

[0100] Among them, the distance d(x i , x j ) between two samples in the high-dimensional space is expressed by the kernel function as:

[0101] d(x i , x j ) = ||φ(x i ) - φ(xj )|| 2

[0102] = φ(x i ) 2 - 2φ(x i )φ(x j ) + φ(x j ) 2

[0103] = Θ(x i , x i ) - 2Θ(x i , x j ) + Θ(x j , x j );

[0104] In a high - dimensional space, the distance d(x i to each cluster center μ j ) is expressed as: i , μ j )

[0105] d(x i , μ j ) = ||φ(x i ) - φ(μ j )|| 2

[0106] = φ(x i ) 2 - 2φ(x i )φ(μ j ) + φ(μ j ) 2 ;

[0107] The cluster center φ(μ j ) in the high - dimensional space is expressed as:

[0108]

[0109] In the formula, C j represents a cluster;

[0110] The sample x i and the distances to each cluster center μ j can be further expressed as:

[0111]

[0112] In an alternative embodiment, after S13, as Figure 3 shown, it may further include:

[0113] The obtained evaluation results are acquired by evaluating the clustering results using the Davies-Bouldin index evaluation metric and the Dunn index.

[0114] Specifically, assuming the clustering result is C = {C1, C2, C3, … C k}, the Davies-Bouldin (DB) index is expressed as:

[0115]

[0116] d cen (C i , C j ) = dist(μ i , μ j );

[0117] In the formula, DBI represents the Davies-Bouldin index, avg(C) represents the average distance between samples in cluster C, including avg(C i ) and avg(C j ), dist(x i , x j ) represents the distance between samples x i and x j , and d cen (C i , C j ) represents the distance between the cluster centers of cluster C i and cluster C j .

[0118] The Dunn (Dunn) index is expressed as:

[0119]

[0120] In the formula, DI represents the Dunn index, d min (C i , C j ) represents the minimum distance between samples in cluster C i and cluster C j , and diam(C) represents the maximum distance between samples in cluster C.

[0121] For DBI, the smaller avg(C) is, the higher the similarity within the cluster, and the larger d cen (C i , C j ) is, the lower the similarity between clusters. Therefore, the smaller the value of DBI, the better, indicating a higher similarity within the cluster and a lower similarity between clusters.

[0122] For DI, where d min (C i , C j)The larger it is, the lower the similarity between clusters. And the smaller diam(C) is, the higher the similarity within the cluster. Therefore, the larger the value of DI, the better. A larger value indicates lower similarity between clusters and higher similarity within the cluster.

[0123] S14. Return to execute S11 - S13 for iteration until the clustering result no longer changes.

[0124] S15. If the current iteration count reaches the preset maximum iteration count, perform a merging operation on the cluster centers in the clustering result, as Figure 3 shown.

[0125] Among them, the steps of the merging operation are as follows (1) - (2):

[0126] (1) Calculate the distances between the cluster centers of all categories pairwise, and arrange these distances in ascending order.

[0127] (2) Check whether the distance between pairwise cluster centers is less than the threshold δ0. If the condition is met, merge the corresponding two categories into a new category.

[0128] The new cluster center μ new is calculated by the weighted average of the cluster centers of the two categories, specifically:

[0129]

[0130] S16. If the current number of clusters in the clustering result is less than or equal to the first preset number, it indicates that the current number of clusters is too small, then perform a splitting operation on the cluster centers in the clustering result.

[0131] In an optional implementation manner, the first preset number is determined according to the number of initial cluster centers and is k / 2.

[0132] Among them, the steps of the splitting operation are as follows (1) - (3):

[0133] (1) Calculate the standard deviation vector of the samples in each category C i ;

[0134] (2) Check whether there exists a certain category C i , in which a certain component in its standard deviation vector is greater than the standard deviation parameter σ0 and satisfies one of the following two conditions:

[0135] (a) and |C i | > 2(σ0 + 1);

[0136] (b)

[0137] If any of the above conditions is satisfied, perform a splitting operation on this category; otherwise, return directly;

[0138] (3) Split the category C that meets the conditions i into two subcategories. Assume the clustering centers of the two new subcategories are μ1 and μ2 respectively, and their calculation methods are as follows:

[0139] (a) Set a splitting parameter that satisfies and can be set to 0.5;

[0140] (b) Define the vector where y is non-zero only in the components of σ0;

[0141] (c) Calculate the two new clustering centers as follows:

[0142] μ1 = μ i + y;

[0143] μ2 = μ i - y.

[0144] S17. If the current iteration number is even, or the current number of clusters in the clustering result is greater than or equal to the second preset number, perform a merging operation on the clustering centers in the clustering result; otherwise, perform a splitting operation on the clustering centers in the clustering result.

[0145] In an alternative embodiment, the second preset number is determined according to the number of initial clustering centers and is 2k.

[0146] S18. Return to S11 - S17 for iteration until the clustering result no longer changes or the current iteration number reaches the preset maximum iteration number, and obtain the scenario set as Figure 3 shown.

[0147] Through the above steps, during scenario clustering, the key features are clustered using the Gaussian kernel function based on the initial clustering centers to obtain the clustering result, which can capture the high-dimensional features of the power load prediction output curve, the photovoltaic device prediction output curve, and the wind power device prediction output curve, thereby improving the accuracy of scenario clustering.

[0148] In addition, the efficient clustering and accurate analysis of the prediction output curves of power loads, photovoltaics, and wind power devices are realized. The method of the present invention ensures the comprehensive mining and reasonable classification of data features through gradient boosting tree feature extraction, improved selection of initial clustering centers, and clustering analysis using the Gaussian kernel function. At the same time, by using splitting and merging operations and multi-round iterative optimization, the clustering effect is continuously improved, and finally the classification and segmentation of the prediction output curves of power loads, photovoltaics, and wind power devices are realized, meeting the performance requirements of the analysis and improving the robustness and reliability of the results.

[0149] S2. Decompose the variables of the active distribution network into the first - type variables and the second - type variables based on the scenario set. The first - type variables include continuous variables related to micro - gas turbines and energy storage, and the second - type variables are discrete variables related to network topology.

[0150] The present invention introduces energy storage and micro - gas turbines to participate in optimization, effectively coping with the uncertainties of wind, light, and load output, reducing the operating cost, and improving the system flexibility and economy.

[0151] S3. Construct the first objective function and the first constraint conditions of the first - stage model based on the first - type variables, and construct the second objective function and the second constraint conditions of the second - stage model based on the second - type variables, specifically including S31 - S36:

[0152] S31. Establish the first objective function of the first - stage model by minimizing the start - up cost of the gas turbine and the constant term of the operating cost, specifically as follows:

[0153] min(C ST +C RUN );

[0154]

[0155] In the formula, C ST represents the start - up cost of the gas turbine, C RUN represents the constant term of the operating cost of the gas turbine, T represents the number of time periods, Ω G represents the set of nodes containing gas turbines, represents the start - up state of the gas turbine at node j in the t - th time period, and its value of 1 indicates that the unit starts up, and 0 indicates that the unit does not start up, represents the start - up cost of the gas turbine at node j in the t - th time period, represents the first constant given for the gas turbine at node j, represents the second constant given for the gas turbine at node j, represents the continuous outage time of the gas turbine at node j in the (t - 1) - th time period, represents the time constant of the gas turbine at node j, represents the operating state of the gas turbine at node j in the t - th time period, and its value of 1 indicates that the unit is operating, and 0 indicates that it is not operating, represents the coefficient of the constant term of the operating cost of the gas turbine at node j.

[0156] S32. Establish the upper and lower bound constraints of the unit operation range, the ramp-up and ramp-down constraints within the unit operation range, and the minimum continuous operation time constraint of the unit, and generate the unit operation constraint conditions according to the upper and lower bound constraints of the unit operation range, the ramp-up and ramp-down constraints within the unit operation range, and the minimum continuous operation time constraint of the unit.

[0157] Among them, the upper and lower bound constraints of the unit operation range are specifically:

[0158]

[0159] In the formula, represents the minimum technical output of the gas turbine at node j in period t, represents the lower limit of the output of the gas turbine at node j in period t, represents the upper limit of the output of the gas turbine at node j in period t, represents the maximum technical output of the gas turbine at node j in period t.

[0160] The minimum continuous operation time constraint of the unit is specifically:

[0161]

[0162] In the formula, represents the operating state of the gas turbine at node j in period t - 1, represents the operating state of the gas turbine at node j in period γ, represents the continuous operation time of the gas turbine at node j, represents the continuous outage time of the gas turbine at node j.

[0163] The ramp-up and ramp-down constraints within the unit operation range are specifically:

[0164]

[0165] In the formula, represents the lower limit of the output of the gas turbine at node j in period t - 1, and Δt represents the time interval, represents the maximum ramp-up power of the gas turbine at node j in period t, represents the upper limit of the output of the gas turbine at node j in period t - 1, represents the maximum ramp-down power of the gas turbine at node j in period t.

[0166] S33. Establish the charge and discharge state constraints of the energy storage, the upper and lower bound constraints of the energy storage operation range, and the energy storage capacity constraint, and generate the energy storage operation constraint conditions according to the charge and discharge state constraints of the energy storage, the upper and lower bound constraints of the energy storage operation range, and the energy storage capacity constraint.

[0167] Among them, the charge and discharge state constraints of the energy storage are specifically:

[0168]

[0169] In the formula, represents the charging state of the energy storage at node j during period t, represents the discharging state of the energy storage at node j during period t, Ω E represents the set of nodes containing the energy storage.

[0170] The upper and lower bound constraints of the energy storage operation domain are specifically as follows:

[0171]

[0172] In the formula, represents the lower limit of the charging power of the energy storage at node j during period t, represents the upper limit of the charging power of the energy storage at node j during period t, represents the maximum charging power of the energy storage at node j, represents the lower limit of the discharging power of the energy storage at node j during period t, represents the upper limit of the discharging power of the energy storage at node j during period t, represents the maximum discharging power of the energy storage at node j.

[0173] The energy storage capacity constraint is specifically as follows:

[0174]

[0175] In the formula, represents the self-discharge efficiency of the energy storage at node j, represents the charging efficiency of the energy storage at node j, represents the lower limit of the charging power of the energy storage at node j during period σ, represents the discharging efficiency of the energy storage at node j, represents the upper limit of the discharging power of the energy storage at node j during period σ, represents the initial energy of the energy storage at node j, represents the upper limit of the charging power of the energy storage at node j during period σ, represents the lower limit of the discharging power of the energy storage at node j during period σ, represents the maximum stored energy of the energy storage at node j.

[0176] S34. Obtain the first constraint condition of the first-stage model according to the unit operation constraint condition and the energy storage operation constraint condition.

[0177] In this way, the first objective function of the first-stage model is established to minimize the constant terms of the gas turbine startup cost and the operation cost. Constraints such as the upper and lower bounds of the unit operation domain, the ramp-up constraint within the unit operation domain, the minimum continuous operation time constraint of the unit, the charge and discharge state constraint of the energy storage, the upper and lower bounds of the energy storage operation domain, and the energy storage capacity constraint are established, which can effectively improve the system economy and operation efficiency.

[0178] S35. The second objective function of the second-stage model is established by minimizing the sum of the gas turbine unit operation cost, the energy storage aging cost, the network loss cost, the wind and light abandonment cost, and the main network power purchase cost, specifically as follows:

[0179]

[0180]

[0181] In the formula, C fuel represents the gas turbine operation cost, C dge represents the energy storage aging cost, C Ploss represents the network loss cost, C Dloss represents the wind and light abandonment cost, C grid represents the main network power purchase cost, represents the linear term coefficient of the gas turbine operation cost, represents the output of the gas turbine at node j at time t, represents the quadratic term coefficient of the gas turbine operation cost, represents the aging cost per unit charge and discharge energy of the energy storage at node j, represents the charging power of the energy storage at node j at time t, represents the discharging power of the energy storage at node j at time t, represents the replacement cost of the energy storage at node j, represents the logarithmic function of the number of energy storage cycles at node j with respect to the depth of discharge, represents the depth of discharge of the energy storage at node j, represents the maximum stored energy of the energy storage at node j, C L represents the unit network loss cost, Ω L represents the set of all branches of the system, r ij represents the resistance value of branch (i, j), represents the upper limit of the current flowing through branch (i, j) at time t, I ij,t represents the current value flowing through branch (i, j) at time t. The positive direction of the power flow of branch (i, j) is from node i to node j, C D represents the unit wind and light abandonment penalty cost, Ω S represents the set of nodes of the photovoltaic power generation units, represents the predicted output of the photovoltaic power generation unit at node j at time t, Denotes the actual output of the photovoltaic power generation unit at node j during period t, Ω W Denotes the set of nodes of the wind turbine generator Denotes the predicted output of the wind turbine generator at node j during period t Denotes the actual output of the wind turbine generator at node j during period t Denotes the purchase cost of the main grid per unit of electricity during period t, Ω sub Denotes the set of substation nodes Denotes the power input from the substation at node j to the power grid during period t

[0182] S36. Establish power flow constraints, security constraints, distributed power source operation constraints, gas turbine operation constraints, energy storage operation constraints, and reactive power compensation device constraints, and generate the second constraint condition of the second-stage model according to the power flow constraints, the security constraints, the distributed power source operation constraints, the gas turbine operation constraints, the energy storage operation constraints, and the reactive power compensation device constraints

[0183] Among them, the power flow constraint is specifically:

[0184]

[0185] In the formula, P j,t Denotes the active power injected into node j during period t, φ(j) denotes the set of all branch head nodes with j as the end node, P jr,t Denotes the active power passing through branch (j, r) during period t, P ij,t Denotes the active power passing through branch (i, j) during period t, Ω N Denotes the set of all nodes, Q j,t Denotes the reactive power injected into node j during period t, Q jr,t Denotes the reactive power passing through branch (j, r) during period t, Q ij,t Denotes the reactive power passing through branch (i, j) during period t, x ij Denotes the resistance value of branch (i, j) Denotes the upper limit of the voltage value at node j during period t Denotes the upper limit of the voltage value at node i during period t Denotes the active load value at node j during period t Denotes the reactive power injection amount of wind power generation at node j during period t Denotes the reactive power injection amount of the gas turbine at node j during period t Denotes the reactive power injection amount of the reactive power compensation device at node j during period t Denotes the reactive power injection power of the substation at node j during period t Denotes the reactive load value at node j during period t

[0186] The specific safety constraints are as follows:

[0187]

[0188] In the formula, V j,t represents the lower limit of the voltage value at node j in period t, I ij,t represents the lower limit of the current flowing through branch (i, j) in period t, represents the upper limit of the active power output of the substation at node j in period t, represents the upper limit of the reactive power output of the substation at node j in period t.

[0189] The specific operating constraints of the distributed power source are as follows:

[0190]

[0191] In the formula, represents the power factor angle of the wind turbine.

[0192] The specific operating constraints of the gas turbine are as follows:

[0193]

[0194] In the formula, represents the lower limit of the reactive power output of the gas turbine at node j in period t, represents the upper limit of the reactive power output of the gas turbine at node j in period t.

[0195] The specific operating constraints of the energy storage are as follows:

[0196]

[0197] The constraints of the reactive power compensation device are as follows:

[0198]

[0199] In the formula, represents the lower limit of the output power of the reactive power compensation device at node j in period t, represents the upper limit of the output power of the reactive power compensation device at node j in period t, Ω C represents the set of nodes containing the reactive power compensation device.

[0200] In this way, the second objective function of the second-stage model is established to minimize the sum of the operating cost of the gas turbine unit, the aging cost of the energy storage, the network loss cost, the cost of curtailed wind and solar power, and the main network power purchase cost. The power flow constraints, security constraints, distributed power generation operation constraints, gas turbine operation constraints, energy storage operation constraints, and reactive power compensation device constraints are established as the second constraint conditions of the second-stage model. By introducing energy storage and gas turbines into the active distribution network, not only the system flexibility is improved, effectively coping with the uncertainties of wind, solar, and load outputs, but also the operating cost is reduced, comprehensively enhancing the system performance.

[0201] S4. Solve the first-stage model based on the first objective function and the first constraint conditions to obtain the first type of variable values, and solve the second-stage model based on the first type of variable values according to the second objective function and the second constraint conditions to obtain the second type of variable values.

[0202] S5. Use the first type of variable values and the second type of variable values to construct a distributionally robust optimization model, and solve the distributionally robust optimization model to obtain the optimization results, specifically including S51 - S53:

[0203] S51. Use the first type of variable values and the second type of variable values to construct the third objective function and the third constraint conditions of the distributionally robust optimization model.

[0204] The third objective function is specifically:

[0205]

[0206] In the formula, y represents the first type of variable values, z s represents the second type of variable values under scenario s, Z s represents the set of the second type of variable values under scenario s, a represents the cost coefficient corresponding to the first type of variable values, N s represents the number of scenario clusters, p s represents the occurrence probability of scenario s, Λ represents the quadratic term cost coefficient in the objective function, and b represents the linear term cost coefficient in the objective function.

[0207] The third constraint conditions are specifically:

[0208] Cy ≤ f;

[0209] Xy + Hz s = q;

[0210] ‖Qy + o||2 ≤ c T y + d;

[0211] Dz s ≤ x;

[0212] In the formula, C, X, H, f, q, Q, o, c, d, D, and x respectively represent matrices or vectors corresponding to the values of the corresponding variables.

[0213] S52. Split the distributionally robust optimization model into a master problem and a sub-problem. The master problem is to solve the optimal solution that satisfies the system economy on the premise of a known scenario probability distribution, and the sub-problem is a two-layer structure of max-min.

[0214] Among them, the master problem is specifically:

[0215]

[0216]

[0217] In the formula, L represents the master problem, λ represents a preset threshold, W represents the total number of model iterations, p s (w)* represents the optimal solution of the occurrence probability of scenario s in the w-th iteration, z s (w)* represents the optimal solution of the values of the second type of variables under scenario s in the w-th iteration, z s (w) represents the values of the second type of variables under scenario s in the w-th iteration.

[0218] The sub-problem is specifically:

[0219]

[0220] In this way, the complex distributionally robust optimization model is split into a master problem and a sub-problem for solution, reducing the complexity of the problem, making each part easier to solve, and effectively improving the solution efficiency.

[0221] S53. Use an iterative method to alternately solve the master problem and the sub-problem to obtain the optimization result.

[0222] Among them, by solving the master problem, the variable y * and the lower bound L M of the model are obtained. When the variable z s can be flexibly adjusted with the scenario and the solution of the master problem is known as y * , find the worst-case scenario probability distribution in the confidence interval. Since the constraint ranges of the outer and inner problems are not related, the sub-problem can be solved in two steps: first solve the inner optimization problem, and then find the maximum value of the outer layer, which can be expressed as:

[0223]

[0224] In the formula, h s represents the inner optimization problem, and U represents the outer optimization problem.

[0225] h s Obtained from the solution result of the main problem, the constraints after equivalent transformation are as follows:

[0226]

[0227] In the formula, p s + represents the probability distribution p of scenario s s relative to p s0 positive offset, p s - represents the probability distribution p of scenario s s relative to p s0 negative offset, θ1 represents the allowable deviation value based on the 1-norm probability distribution, θ ∞ represents the allowable deviation value based on the ∞-norm probability distribution, represents the 0-1 flag variable that generates a positive offset for p s represents the 0-1 flag variable that generates a negative offset for p s

[0228] After these steps, the model is transformed into a mixed linear programming problem, and the optimal solution p of the occurrence probability of scenario s can be quickly solved by a solver s * , which is substituted into the main problem for the next iteration optimization to obtain the upper limit value U of the model M .

[0229] In an alternative implementation, the optimization result may include the network topology structure and the output of each DG. Specifically, it includes the switch status of the unit, the output of the unit, the output of new energy, the amount of wind and light abandonment, and the change of energy storage. It is presented in the form of economic cost, that is, it includes network loss cost, power purchase cost, energy storage cost, gas turbine cost and total cost, as shown in Table 1

[0230] In this way, the third objective function and the third constraint condition of the distributionally robust optimization model are constructed using the first type of variable value and the second type of variable value. On the basis of the stochastic optimization model, the confidence interval of the probability distribution is innovatively considered, and the system economy and conservatism are skillfully balanced. Compared with the GA (genetic algorithm) and the classical distributionally robust optimization model, while the optimization result is more ideal, the solution efficiency is significantly improved, and it can provide an optimization strategy for the operation of the active distribution network faster and more accurately, greatly improving the overall efficiency of the system

[0231] 160 groups of curves were selected from the predicted output curves of power load, photovoltaic and wind power devices for clustering. The improved ISODATA algorithm divides the scenarios of load and wind-solar output into 4 categories, and the scenario curves of the clustering center of each category are as Figure 4 ​​As shown in (a), (b), (c), and (d) therein, where the abscissa is the load sampling point and the ordinate is the scenario curve after normalization of the linear function. The DBI and DI are used to evaluate the clustering effect, which is used to measure the quality of the clustering result. The DBI value of this clustering result is 0.52487, and the DI value is 0.67893. It can be seen that the clustering effect is quite good.

[0232] Taking the IEEE 33-node as an example to verify the effectiveness of the above method of the present invention. The GA algorithm, the classical distributionally robust optimization model and the above method of the present invention are respectively used for the optimization result analysis, and the results are shown in Table 1.

[0233] Table 1 Optimization results of different methods

[0234]

[0235]

[0236] It can be seen from Table 1 that the optimization result obtained by the GA algorithm is not the best, while the results obtained by the classical distributionally robust optimization model and the method of the present invention are relatively ideal. However, since the method of the present invention eliminates the dynamic constraints in the sub-problems, reduces the complexity of the model, and avoids the problem of too long calculation time of the commercial solver when dealing with mixed variables, the calculation time of the method of the present invention is significantly less than that of the classical distributionally robust optimization model, thus greatly improving the calculation efficiency of the model, and further improving the efficiency of the integrated optimization of the active distribution network.

[0237] Please refer to Figure 2 , the second embodiment of the present invention is:

[0238] An active distribution network integrated optimization system, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, each step in the active distribution network integrated optimization method in the first embodiment is implemented.

[0239] In summary, the present invention provides a comprehensive optimization method and system for an active distribution network. The improved ISODATA algorithm is used to perform scenario clustering on the obtained power load prediction output curve, photovoltaic device prediction output curve, and wind power device prediction output curve to obtain a scenario set. Based on the scenario set, the variables of the active distribution network are decomposed into first-class variables and second-class variables. The first objective function and first constraint conditions of the first-stage model are constructed based on the first-class variables, and the second objective function and second constraint conditions of the second-stage model are constructed based on the second-class variables. The first-stage model and the second-stage model are solved respectively to obtain the first-class variable values and the second-class variable values. The first-class variable values and the second-class variable values are used to construct a distributionally robust optimization model, and it is solved to obtain the optimization result. Compared with the existing ISODATA algorithm, the improved ISODATA algorithm optimizes the selection strategy of the initial clustering center to reduce the number of algorithm iterations, combines the kernel method to capture the characteristics of the high-dimensional space, has small clustering error and strong robustness, greatly improves the accuracy and reliability of scenario construction. At the same time, the first-stage model and the second-stage model are established respectively, enhancing the adaptability of the model to different types of variables. The first-class variable values and the second-class variable values obtained by solving are used to construct a distributionally robust optimization model and solve it, thus effectively improving the reliability of the comprehensive optimization of the active distribution network. In addition, the third objective function and third constraint conditions of the distributionally robust optimization model are constructed using the first-class variable values and the second-class variable values. On the basis of the stochastic optimization model, the confidence interval of the probability distribution is innovatively considered, skillfully balancing the system economy and conservatism. Compared with the GA and the classical distributionally robust optimization model, while the optimization result is more ideal, the solution efficiency is significantly improved, and it can provide an optimization strategy for the operation of the active distribution network faster and more accurately, greatly enhancing the overall system efficiency.

[0240] The above are only the embodiments of the present invention, and do not limit the patent scope of the present invention accordingly. Any equivalent transformation made by using the content of the specification and drawings of the present invention, or directly or indirectly applied in the relevant technical field, shall be equally included in the patent protection scope of the present invention.

Claims

1. An integrated optimization method for an active distribution network, characterized in that, Including the steps: Using an improved iterative self-organizing data analysis algorithm to perform scenario clustering on the obtained power load prediction output curve, photovoltaic device prediction output curve, and wind power device prediction output curve, to obtain a scenario set. The improved iterative self-organizing data analysis algorithm is used to optimize the selection strategy of the initial clustering center to reduce the number of algorithm iterations, and combines the kernel method to capture the characteristics in the high-dimensional space; Based on the scenario set, decomposing the variables of the active distribution network into first-class variables and second-class variables. The first-class variables include continuous variables related to micro gas turbines and energy storage, and the second-class variables are discrete variables related to the network topology; Based on the first-class variables, constructing the first objective function and the first constraint condition of the first-stage model, and based on the second-class variables, constructing the second objective function and the second constraint condition of the second-stage model; Based on the first objective function and the first constraint condition, solving the first-stage model to obtain the first-class variable values, and based on the first-class variable values, solving the second-stage model according to the second objective function and the second constraint condition to obtain the second-class variable values; Using the first-class variable values and the second-class variable values to construct a distributionally robust optimization model, and solving the distributionally robust optimization model to obtain the optimization result.

2. The integrated optimization method for an active distribution network according to claim 1, wherein The using an improved iterative self-organizing data analysis algorithm to perform scenario clustering on the obtained power load prediction output curve, photovoltaic device prediction output curve, and wind power device prediction output curve, to obtain a scenario set includes: Using a feature extraction model based on the gradient boosting tree algorithm to extract the key features of the obtained power load prediction output curve, photovoltaic device prediction output curve, and wind power device prediction output curve; Selecting the initial clustering center; Based on the initial clustering center, using the Gaussian kernel function to cluster the key features to obtain a clustering result; Returning to execute the step of using a feature extraction model based on the gradient boosting tree algorithm to extract the key features of the obtained power load prediction output curve, photovoltaic device prediction output curve, and wind power device prediction output curve for iteration until the clustering result no longer changes; If the current iteration number reaches the preset maximum iteration number, performing a merging operation on the clustering centers in the clustering result; If the current number of clusters in the clustering result is less than or equal to the first preset number, performing a splitting operation on the clustering centers in the clustering result; If the current iteration number is even, or the current number of clusters in the clustering result is greater than or equal to the second preset number, performing a merging operation on the clustering centers in the clustering result, otherwise, performing a splitting operation on the clustering centers in the clustering result; Returning to execute the step of using a feature extraction model based on the gradient boosting tree algorithm to extract the key features of the obtained power load prediction output curve, photovoltaic device prediction output curve, and wind power device prediction output curve for iteration until the clustering result no longer changes or the current iteration number reaches the preset maximum iteration number, to obtain a scenario set.

3. The integrated optimization method for an active distribution network according to claim 2, wherein The selecting the initial clustering center includes: Randomly select a first clustering center from the dataset composed of the power load prediction output curve, the photovoltaic device prediction output curve, and the wind power device prediction output curve; For each sample in the dataset, calculate the distance from the sample to each clustering center, and select the shortest distance among the distances; Determine the probability that the sample is selected as the next clustering center according to the shortest distance; Select subsequent clustering centers from the dataset according to the probability; Obtain the initial clustering center according to the first clustering center and the subsequent clustering centers; The probability is specifically: where d(x i ) represents the shortest distance among the distances from the sample x i to each cluster center, and D represents the sample set.

4. The integrated optimization method for an active distribution network according to claim 1, wherein The first objective function and the first constraint conditions for constructing the first-stage model based on the first type of variables include: Establish the first objective function of the first-stage model by minimizing the gas turbine start-up cost and the constant term of the operating cost; Establish the upper and lower bounds of the unit operating range, the in-range ramp constraint of the unit, and the minimum continuous operation time constraint of the unit, and generate the unit operation constraint conditions according to the upper and lower bounds of the unit operating range, the in-range ramp constraint of the unit, and the minimum continuous operation time constraint of the unit; Establish the energy storage charge and discharge state constraint, the upper and lower bounds of the energy storage operating range, and the energy storage capacity constraint, and generate the energy storage operation constraint conditions according to the energy storage charge and discharge state constraint, the upper and lower bounds of the energy storage operating range, and the energy storage capacity constraint; Obtain the first constraint condition of the first-stage model according to the unit operation constraint conditions and the energy storage operation constraint conditions.

5. The integrated optimization method for an active distribution network according to claim 4, characterized in that The first objective function for establishing the first-stage model by minimizing the gas turbine start-up cost and the constant term of the operating cost includes: min(C ST +C RUN ); Where, C ST represents the start-up cost of the gas turbine, C RUN represents the constant term of the operating cost of the gas turbine, T represents the number of time periods, Ω G represents the set of nodes containing the gas turbine, represents the start-up status of the gas turbine at node j in time period t, represents the start-up cost of the gas turbine at node j in time period t, represents the first constant given for the gas turbine at node j, represents the second constant given for the gas turbine at node j, represents the continuous outage time of the gas turbine at node j in time period t - 1, represents the time constant of the gas turbine at node j, represents the operating status of the gas turbine at node j in time period t, represents the coefficient of the constant term of the operating cost of the gas turbine at node j.

6. The active distribution network comprehensive optimization method according to claim 1, wherein The second objective function and the second constraint conditions for constructing the second-stage model based on the second type of variables include: Establish the second objective function of the second-stage model by minimizing the sum of the gas turbine unit operating cost, the energy storage aging cost, the network loss cost, the wind and light abandonment cost, and the main grid power purchase cost; Establish the power flow constraint, the security constraint, the distributed power source operation constraint, the gas turbine operation constraint, the energy storage operation constraint, and the reactive power compensation device constraint, and generate the second constraint condition of the second-stage model according to the power flow constraint, the security constraint, the distributed power source operation constraint, the gas turbine operation constraint, the energy storage operation constraint, and the reactive power compensation device constraint.

7. An integrated optimization method for an active distribution network according to claim 6, characterized in that, The second objective function for establishing the second-stage model by minimizing the sum of the gas turbine unit operating cost, the energy storage aging cost, the network loss cost, the wind and light abandonment cost, and the main grid power purchase cost includes: min(C fuel +C dge +C Ploss +C Dloss +C grid ); where, C fuel represents the operating cost of the gas turbine, C dge represents the aging cost of the energy storage, C Ploss represents the network loss cost, C Dloss represents the cost of curtailed wind and solar power, C grid represents the cost of purchasing electricity from the main grid, T represents the number of time periods, Ω G represents the set of nodes containing gas turbines, Ω E represents the set of nodes containing energy storage, represents the linear term coefficient of the gas turbine operating cost, represents the output of the gas turbine at node j in period t, represents the quadratic term coefficient of the gas turbine operating cost, represents the aging cost per unit charge and discharge energy of the energy storage at node j, represents the charging power of the energy storage at node j in period t, represents the discharging power of the energy storage at node j in period t, represents the replacement cost of the energy storage at node j, represents the logarithmic function of the number of cycles of the energy storage at node j with respect to the depth of discharge, represents the depth of discharge of the energy storage at node j, represents the maximum stored energy of the energy storage at node j, C L represents the unit network loss cost, Ω L represents the set of all branches in the system, r ij represents the resistance value of branch (i, j), represents the upper limit of the current flowing through branch (i, j) in period t, I ik,t represents the current value flowing through branch (i, j) in period t, with the positive direction of the power flow from node i to node j being the positive direction of branch (i, j), C D represents the unit penalty cost for curtailed wind and solar power, Ω S represents the set of nodes of the photovoltaic power generation units, represents the predicted output of the photovoltaic power generation unit at node j in period t, represents the actual output of the photovoltaic power generation unit at node j in period t, Ω W represents the set of nodes of the wind power generation units, represents the predicted output of the wind power generation unit at node j in period t, represents the actual output of the wind power generation unit at node j in period t, represents the cost of purchasing electricity from the main grid per unit of electricity in period t, Ω sub represents the set of substation nodes, represents the power input from the substation at node j to the power grid in period t.

8. The integrated optimization method for an active distribution network according to claim 1, characterized in that The construction of the distributionally robust optimization model using the values of the first type of variables and the values of the second type of variables includes: Construct the third objective function and the third constraint conditions of the distributionally robust optimization model using the values of the first type of variables and the values of the second type of variables; The third objective function is specifically: In the formula, y represents the value of the first type of variable, and z s represents the value of the second type of variable under scenario s, and Z s represents the set of values of the second type of variable under scenario s, a represents the cost coefficient corresponding to the value of the first type of variable, and N s represents the number of scenario clusters, and p s represents the occurrence probability of scenario s, Λ represents the quadratic term cost coefficient in the objective function, and b represents the linear term cost coefficient in the objective function; The third constraint condition is specifically: Cy ≤ f; Xy + Hz s = q; ||Qy + o||² ≤ c T y + d; Dz s ≤ x; In the formula, C, X, H, f, q, Q, o, c, d, D, and x respectively represent matrices or vectors corresponding to the corresponding variable values.

9. An integrated optimization method for an active distribution network according to claim 8, characterized in that The solution of the distributionally robust optimization model to obtain the optimization result includes: Split the distributionally robust optimization model into a master problem and a subproblem; The master problem and the sub-problem are alternately solved iteratively to obtain an optimization result; The master problem specifically is: Wherein, L represents the main problem, λ represents a preset threshold, W represents the total number of model iterations, and p s (w)* represents the optimal solution of the occurrence probability of the scenario s in the w-th iteration, and z s (w)* represents the optimal solution of the value of the second type of variable under the scenario s in the w-th iteration, and z s (w) represents the value of the second type of variable under the scenario s in the w-th iteration; The sub-problem specifically is:

10. An active distribution network comprehensive optimization system, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements each step in an active distribution network comprehensive optimization method described in any one of claims 1 to 9.

Images