Optimization method considering multi-element cluster energy storage dynamic reconfigurable model

Abstract

The invention belongs to the technical field of power system model aggregation method research, and particularly provides an optimization method considering a multi-element cluster energy storage dynamic reconfigurable model, which comprises the following steps: obtaining a net load fluctuation curve of each node in a system based on an improved conditional generative adversarial network and power flow simulation calculation; variational mode decomposition is carried out on the net load fluctuation curve, and high-frequency and low-frequency components of each load curve are stabilized; minkowski summation is used for obtaining an aggregation model of cluster energy storage, and the minimum stabilizing effect on net load fluctuation of all regions under the minimum economic cost serves as an optimization target. And carrying out output analysis on each energy storage output in the obtained cluster energy storage group, carrying out readjustment and aggregation on a cluster division result, carrying out optimization calculation on the obtained model again, and determining the matching degree of each to-be-utilized energy storage at the provincial domain level and the corresponding region. According to the method, cross-space-time characteristic analysis of energy storage resources and matching evaluation of various types of resources are used as a resource optimization and model basis of a system.

Description

Technical Field

[0001] The present invention belongs to the technical field of research on power system model aggregation methods, and specifically relates to an optimization method considering a dynamic reconfigurable model of multi-cluster energy storage. Background Art

[0002] The scale and diversification of energy storage resources provide more technical means for the flexible operation of the power system and the smoothing of wind and solar fluctuations. The diversified output characteristics of energy storage are complementary at different times. Energy storage resources in different regions have not been fully utilized due to electrical distance limitations. At the same time, due to the integration of energy storage resources by aggregators, the cross-temporal and spatial utilization of energy storage resources in various regions can only be processed globally through provincial-level analysis.

[0003] Therefore, in order to realize the cross-temporal and spatial utilization of energy storage resources at the provincial level and to balance the matching of various types of resources in the process of breaking through the model aggregation barriers, it is urgent to establish a set of model matching and aggregation methods that can effectively target distributed energy storage. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide an optimization method for a dynamic reconfigurable model of multi-cluster energy storage, which provides a resource optimization and model basis for the system by analyzing the cross-temporal and spatial characteristics of energy storage resources and evaluating the matching of various types of resources.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is: an optimization method considering a multi-cluster dynamic reconfigurable energy storage model, comprising the following steps: S1. Based on the historical output data of wind power and photovoltaic power in various regions, the time-series output scenarios of wind power and photovoltaic power in various regions of the province are generated based on the improved conditional generative adversarial network. The typical wind and photovoltaic output scenarios of various regions are obtained by using the synchronous back-substitution reduction method. Considering the controllable resources and load forecast data of various regions in the province, the Newton-Raphson method is used for power flow simulation calculation to comprehensively obtain the net load fluctuation curve of each node in the system. S2. Perform variational modal decomposition on the net load fluctuation curves of each region. According to the fluctuation degree of each frequency band curve, perform time and space matching for the power type and capacity type of energy storage, and consider using power type and capacity type energy storage to smooth the high-frequency and low-frequency components of each load curve respectively; S3. Use Minkowski summation to obtain the aggregation model of cluster energy storage at different time and space, and take the minimum effect of smoothing net load fluctuations in each region under the minimum economic cost as the optimization goal; S4. Analyze the output of each energy storage within the obtained cluster energy storage group, readjust and aggregate the cluster division results, and optimize the obtained model again until the net load smoothing effect of each cluster energy storage on each region is balanced, and finally determine the matching degree between each energy storage to be used and the corresponding region at the provincial level.

[0006] In the preferred solution, in step S1, the wind and solar historical data are trained based on the improved conditional generative adversarial network to complete the generation of wind power and photovoltaic output scenes. The improved conditional generative network is composed of a generator and a discriminator. A gradient descent layer is added after the convolution layer inside the discriminator to encourage the generator to generate more diverse samples. The specific formula is: (1); In the formula, x and y Represent the samples of input discriminator D respectively x and the internal conditions of the discriminator y , represents the neural network output of the discriminator, represents the internal activation function, It is the feature calculated by the Minibatch Discrimination layer, which can process the entire minibatch data to generate a feature that is similar to the current sample. x Additional information about .

[0007] In the preferred solution, in step S1, the generated scenery scene is reduced using a synchronous back-substitution reduction method: first, the data of the generated scene is used as the initial scene, and the loop is iterated, and the closest scene is eliminated each time, and the Kantorovich distance is defined as the distance index, and the expression is: (2) In the formula, Represents a scene collection With scene collection The distance between the scenes, the indicator indicates that the sum of the distances between the scenes in the scene set is the smallest; N Indicates the number of scenes in the collection, and Represents a collection of scenes and Specific scenes inside; and Indicates respectively and The probability of the scenario occurring; express and Euclidean distance of the scene.

[0008] In a preferred solution, in step S2, the net load fluctuation of nodes in each region is decomposed into frequency bands using variational mode decomposition, which is expanded by minimizing the following variational problem: (3); In the formula, K Indicates the decomposed mode The number of modes, each with a center frequency , Indicates time t The derivative of is the Dirac function, j represents the imaginary unit, represents the complex exponential function that shifts the mode to baseband, represents the modal components to be decomposed.

[0009] In the preferred solution, in step S2, the multivariate energy storage in each region is divided into power type and capacity type, and formulas (4) and (5) are used as indicators for evaluating the power type and capacity type. The hierarchical analysis method is combined with the entropy weight method to form a subjective and objective hierarchical analysis method to obtain a comprehensive evaluation result. Then, the evaluation result is aggregated using the K-means algorithm to obtain a preliminary cluster division for each energy storage to determine the number of modes to be decomposed. Different frequency bands correspond to corresponding types of energy storage output. Frequency bands with larger net load amplitudes correspond to power-type energy storage, while frequency bands with smaller amplitudes but greater volatility correspond to power-type energy storage. The matching degree between energy storage and net load curve is measured from the aspects of power support and duration. The expression is as follows: (4); (5); In the formula, Indicates the maximum net load of each node; Indicates the maximum value of charge and discharge of energy storage; and They represent the peak and valley values ​​of the net load respectively. represents the maximum available capacity of energy storage, and Indicates the discharge and charging efficiency of the corresponding energy storage, and They respectively represent the power support capacity of each node energy storage for the corresponding net load and the relative sustainable working time under the corresponding charging and discharging states.

[0010] In the preferred solution, in step S3, the optimization goal is to achieve maximum suppression of power fluctuations with minimum economic operation cost of the overall energy storage system, and the objective function is: (5); In the formula, and Respectively represent energy storage S exist t The charging and discharging power at each moment; F (·) represents the optimization objective function; The net load curve of the system and the energy storage output cost are the two influencing factors. The net load curve represents the overall output effect of the energy storage, and the energy storage output cost represents the minimum overall operating cost of the energy storage system. represents the sub-function with the minimum energy storage operation cost, and The energy storage charging and discharging output is the independent variable.

[0011] In the preferred solution, in step S4, through simulation optimization, according to the smoothing effect of each energy storage within the cluster on the local net load fluctuation, the energy storage monomers that need to be re-divided and aggregated are determined, and step S3 is repeated, and iteration is performed based on the Nash equilibrium principle: the energy storage under each node of the system is taken as a participant, the overall minimum net load fluctuation in each region of the system is taken as the Nash equilibrium point, and the division results between different energy storages are taken as the strategy set. By looping the step S3 process, each energy storage cluster division is adjusted until the output status of all energy storages is approximately consistent, and the system Nash equilibrium is achieved.

[0012] The present invention provides an optimization method considering a multi-element cluster energy storage dynamic reconfigurable model, which has the following beneficial effects: 1. In step S1 of the present invention, compared with the traditional application of generating various wind and solar scenes, this method also combines the system's power flow simulation calculation, based on the refined analysis of nodes in each region, and uses the synchronous back-generation reduction method to reduce the output scene of each node, thereby establishing a data analysis foundation for the cross-temporal and spatial complementary analysis of energy storage in the later stage.

[0013] 2. In step S2 of the present invention, variational mode decomposition is used to decompose the net load curve of each region. Compared with the traditional time-complementary analysis and evaluation of resources, the electrical impact of energy storage output in each time period on the local and adjacent regions, that is, the spatial complementarity capability, is also analyzed, breaking through the limitation that the aggregation model can only integrate the resources in the local area and the internal resources are not fully utilized.

[0014] 3. In step S3 of the present invention, a two-layer optimization model is established that takes into account both the energy storage operation cost and the operation effect. The traditional Minkowski sum model aggregation method is used as the basis for model aggregation. The focus is still on whether the resource regulation and utilization is sufficient. By comparing the two, the adjustment space of energy storage resources in the aggregation process is analyzed.

[0015] 4. In step S4 of the present invention, the division and aggregation process of resources is a process of overall planning. Compared with the previous consideration of the division and utilization of resources only from a certain region, the provincial level has a larger spatial span and a wide variety of resources, so multiple iterative analyses are required to determine the matching effect of resources. Therefore, the Nash equilibrium idea is taken as the main line, and the division results of each cluster in the iterative process are used as the iterative direction until the output effects of each cluster tend to be consistent to achieve balanced use of energy storage resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The present invention will be further described below in conjunction with the accompanying drawings and embodiments: Figure 1 A flow chart of the dynamic division and aggregation of distributed energy storage resources of the present invention; Figure 2 A reference diagram for cross-space adjustment of energy storage resources according to the present invention; Figure 3 A reference diagram for adjusting energy storage resources across time as described in the present invention; Figure 4 This is a typical output scenario diagram of wind power and photovoltaic power; Figure 5 Schematic diagram for comparative analysis of regulation optimization and aggregation models. DETAILED DESCRIPTION

[0017] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0018] Combination Figure 1~Figure 5 The specific embodiments of the present invention are described in further detail.

[0019] In response to the problems mentioned in the above background technology, this embodiment mainly focuses on the cross-temporal and spatial division and model aggregation method of cluster energy storage under the decentralized layout of multi-heterogeneous energy storage at the provincial level, mainly from four steps: generation of typical cross-temporal and spatial scenarios of wind and solar power, analysis of cross-temporal and spatial output conditions of multi-heterogeneous energy storage, model optimization aggregation and cluster Nash equilibrium. Among them, the generation of typical cross-temporal and spatial scenarios of wind and solar power is mainly obtained through conditional generative adversarial networks and synchronous back-substitution reduction methods, and finally the net load fluctuation curves of various regions are obtained through power flow simulation calculations; the cross-temporal and spatial output condition analysis of multi-heterogeneous energy storage mainly uses variational mode decomposition to determine the output period of energy storage in each region, and determines the effective utilization period of each cluster energy storage at the spatial level and the time rolling level. After further comprehensive consideration, the cross-temporal and spatial waiting-for-demand output space of energy storage is obtained; since energy storage is a controllable resource, the aggregation process can consider the influence of the "barrel effect" to aggregate the model, so the Minkowski summation is first used to perform a preliminary aggregation of the energy storage model, and the energy storage waiting-for-output space obtained by calculation is re-aggregated and adjusted internally, and the aggregation goal is to smooth the fluctuation of electricity at the minimum economic cost; at the same time, in order to ensure the consistency of the operating status of each cluster energy storage, considering the situation that the instructions in the aggregation model cannot be fully issued, the energy storage that has not been fully utilized in each cluster energy storage is re-divided and aggregated based on the Nash equilibrium idea until the output of each cluster is optimal.

[0020] like Figure 1 As shown, an optimization method considering a multi-element cluster energy storage dynamic reconfigurable model includes the following steps: S1. Based on the improved conditional generative adversarial network (CGAN), the historical data of wind and solar power are trained to generate wind power and photovoltaic output scenarios. The conditional generative adversarial network is mainly composed of a generator and a discriminator. This method mainly adds a gradient descent layer after the convolution layer inside the discriminator to encourage the generator to generate more diverse samples. The specific formula is: (1); In the formula, x and y Represent the samples of input discriminator D respectively x and the internal conditions of the discriminator y , represents the neural network output of the discriminator, represents the internal activation function, It is the feature calculated by the Minibatch Discrimination layer, which can process the entire minibatch data to generate a feature that is similar to the current sample. xThe purpose of this layer is to enable the discriminator to make decisions based on the similarity of samples within a small batch, helping the discriminator to better distinguish between generated samples and real samples, thereby improving the diversity of the generated model. Then, the typical scene of wind and solar power output is obtained through the synchronous back-substitution reduction method, the uncertainty representation of wind and solar power output in various regions is determined, and the net load fluctuation curve of each node in the system is obtained as a whole by using power flow simulation and the maximum output calculation of controllable resources except energy storage.

[0021] Use the synchronous back-generation reduction method to reduce the generated scenery scene: first, use the data of the generated scene as the initial scene, iterate in a loop, remove the closest scene each time, and define the Kantorovich distance as the distance indicator: (2) In the formula, Represents a scene collection With scene collection The distance between N Indicates the number of scenes in the collection, and Represents a collection of scenes and Specific scenes inside; and Indicates respectively and The probability of the scenario occurring; express and The Euclidean distance of the scene. The indicator indicates that the sum of the distances between the scenes in the scene set is the smallest. The scene with the smallest distance is eliminated and the probability of the retained scene is updated to obtain the typical wind and solar output scene, as shown in the following example. Figure 4 Afterwards, the Newton-Raphson method is used to perform power flow analysis on the system.

[0022] S2. Use variational mode decomposition to decompose the net load fluctuation of nodes in each region by frequency band, mainly by minimizing the following variational problem: (3); In the formula, K Indicates the decomposed mode The number of modes, each with a center frequency , Indicates time t The derivative of is the Dirac function, j represents the imaginary unit, represents the complex exponential function that shifts the mode to baseband, represents the modal components to be decomposed.

[0023] The time scale smoothing effect of each energy storage in different scenarios can be referred to Figure 3 As shown. The multi-energy storage in each region is divided into power type and capacity type, and formulas (4) and (5) are used as indicators for evaluating power type and capacity type. The hierarchical analysis method is combined with the entropy weight method to form a subjective and objective hierarchical analysis method to obtain a comprehensive evaluation result. Then, the K-means algorithm is used to aggregate the evaluation results to obtain the preliminary clustering division of each energy storage to determine the number of modes to be decomposed. Different frequency bands correspond to corresponding types of energy storage outputs. Frequency bands with larger net load amplitudes correspond to power-type energy storage, and frequency bands with smaller amplitudes but greater volatility correspond to power-type energy storage. Specifically, the amplitudes (the difference between the peak and the trough) are arranged from small to large, and correspond one-to-one with the divided energy storage. Frequency bands with large amplitudes correspond to cluster energy storage with large power support scores, and frequency bands with smaller amplitudes but greater volatility correspond to energy storage with large support assessment scores. Overall, they correspond one-to-one.

[0024] The matching degree between energy storage and net load curve is measured from the aspects of power support and duration. The expression is as follows: (4); (5); In the formula, Indicates the maximum net load of each node; Indicates the maximum value of charge and discharge of energy storage; and They represent the moment values ​​of net load in peak and valley states respectively. represents the maximum available capacity of energy storage, and Indicates the discharge and charge efficiency of the corresponding energy storage. and They respectively represent the power support capacity of each node energy storage for the corresponding net load and the relative sustainable working time under the corresponding charging and discharging states.

[0025] In different regions, the optimal state of energy storage and the effect of energy storage output on the node under the consideration of electrical distance are mainly considered. Figure 2 As shown. The optimal energy state of all energy storage is defined as 50% of its capacity. The maximum output during the peak and valley periods is selected, and the system flow simulation calculation is performed for each energy storage. It is required that when the energy storage output satisfies the net load fluctuation, this point is directly set to zero. The standard deviation is used to calculate the fluctuation of the power of each node before and after the energy storage is incorporated.

[0026] S3, use Minkowski summation to aggregate the operation models of energy storage in the same cluster, establish an optimization model, and the optimization goal is to achieve the maximum smoothing of power fluctuations with the minimum economic operation cost of the overall energy storage system. The objective function is: (5); In the formula, and Respectively represent energy storage S The charge and discharge power at time t, F (·) represents the optimization objective function; The net load curve of the system and the energy storage output cost are the two influencing factors. The net load curve represents the overall output effect of the energy storage, and the energy storage output cost represents the minimum overall operating cost of the energy storage system. represents the sub-function with the minimum energy storage operation cost, and The energy storage charging and discharging output is the independent variable.

[0027] The operation of other controllable units in the system is not considered, and the operation constraints of energy storage are the same as traditional operation constraints. Only the effect of smoothing net load fluctuations is considered from the perspective of energy storage. By calculating the aggregated energy storage model and comparing it with the original aggregate model, the difference between the output of each cluster and the boundary of the original model is compared to determine the energy storage to be adjusted within the cluster.

[0028] S4. Through simulation optimization, according to the smoothing effect of each energy storage within the cluster on the local net load fluctuation, determine the energy storage units that need to be re-divided and aggregated. The energy storage output effect in one iteration is as follows: Figure 5 As shown, the above step S3 is performed again.

[0029] Iterate based on the Nash equilibrium principle: the energy storage at each node of the system is taken as a participant, the overall minimum net load fluctuation in each region of the system is taken as the Nash equilibrium point, and the division results between different energy storages are taken as the strategy set. Through the loop step S3 process, each energy storage cluster division is adjusted until all energy storage output states are approximately consistent, thus reaching the system Nash equilibrium.

[0030] like Figure 5 As shown in the figure, the boundaries of the aggregated energy storage model are characterized on the right to determine the boundaries of energy storage parameters at the power and energy levels. The boundaries of the aggregated model are involved in the scheduling process to obtain the day-ahead time-series output curve of the energy storage. It can be seen that the upper and lower boundaries of the power and capacity of the model effectively limit the energy storage output, indicating that the division and aggregation of clustered energy storage can effectively participate in system scheduling.

[0031] It will be easily understood by those skilled in the art that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. An optimization method considering a multi-element cluster energy storage dynamic reconfigurable model, characterized in that: The following steps are involved: S1. Based on the historical output data of wind power and photovoltaic power in various regions, the time-series output scenarios of wind power and photovoltaic power in various regions of the province are generated based on the improved conditional generative adversarial network. The typical wind and photovoltaic output scenarios of various regions are obtained by using the synchronous back-substitution reduction method. Considering the controllable resources and load forecast data of various regions in the province, the Newton-Raphson method is used for power flow simulation calculation to comprehensively obtain the net load fluctuation curve of each node in the system. S2. Perform variational modal decomposition on the net load fluctuation curves of each region. According to the fluctuation degree of each frequency band curve, perform time and space matching for the power type and capacity type of energy storage, and consider using power type and capacity type energy storage to smooth the high-frequency and low-frequency components of each load curve respectively; S3. Use Minkowski summation to obtain the aggregation model of cluster energy storage at different time and space, and take the minimum effect of smoothing net load fluctuations in each region under the minimum economic cost as the optimization goal; S4. Analyze the output of each energy storage within the obtained cluster energy storage group, readjust and aggregate the cluster division results, and optimize the obtained model again until the net load smoothing effect of each cluster energy storage on each region is balanced, and finally determine the matching degree between each energy storage to be used and the corresponding region at the provincial level.

2. The optimization method of the multi-element cluster energy storage dynamic reconfigurable model according to claim 1 is characterized in that: In step S1, the wind and solar historical data are trained based on the improved conditional generative adversarial network to complete the generation of wind power and photovoltaic output scenes. The conditional generative adversarial network consists of a generator and a discriminator. A gradient descent layer is added after the convolution layer inside the discriminator to encourage the generator to generate more diverse samples. The specific formula is: (1)； In the formula, x and y Represent the samples of input discriminator D respectively x and the internal conditions of the discriminator y , represents the neural network output of the discriminator, represents the internal activation function, It is the feature calculated by the Minibatch Discrimination layer, which can process the entire minibatch data to generate a feature that is similar to the current sample. x Additional information about .

3. The optimization method of the multi-element cluster energy storage dynamic reconfigurable model according to claim 1 is characterized in that: In step S1, the generated scenery scene is reduced by using a synchronous back-generation reduction method: first, the data of the generated scene is used as the initial scene, and the loop is iterated, and the closest scene is eliminated each time, and the Kantorovich distance is defined as the distance index, and the expression is: (2) In the formula, Represents a scene collection With scene collection The distance between the scenes, the indicator indicates that the sum of the distances between the scenes in the scene set is the smallest; N Indicates the number of scenes in the collection, and Represents a collection of scenes and Specific scenes inside; and Indicates respectively and The probability of the scenario occurring; express and Euclidean distance of the scene.

4. The optimization method of the multi-element cluster energy storage dynamic reconfigurable model according to claim 1 is characterized in that: In step S2, the net load fluctuation of each regional node is decomposed into frequency bands using variational mode decomposition, which is expanded by minimizing the following variational problem: (3)； In the formula, K Indicates the decomposed mode The number of modes, each with a center frequency , Indicates time t The derivative of is the Dirac function, j represents the imaginary unit, represents the complex exponential function that shifts the mode to baseband, represents the modal components to be decomposed.

5. The optimization method of the multi-element cluster energy storage dynamic reconfigurable model according to claim 1 is characterized in that: In step S2, the multivariate energy storage in each region is divided into power type and capacity type, and formulas (4) and (5) are used as indicators for evaluating power type and capacity type. The hierarchical analysis method is combined with the entropy weight method to form a subjective and objective hierarchical analysis method to obtain a comprehensive evaluation result. Then, the evaluation results are aggregated using the K-means algorithm to obtain a preliminary cluster division for each energy storage to determine the number of modes to be decomposed. Different frequency bands correspond to corresponding types of energy storage output. Frequency bands with larger net load amplitudes correspond to power-type energy storage, while frequency bands with smaller amplitudes but greater volatility correspond to power-type energy storage. The matching degree between energy storage and net load curve is measured from the aspects of power support and duration. The expression is as follows: (4)； (5)； In the formula, Indicates the maximum net load of each node; Indicates the maximum value of charge and discharge of energy storage; and They represent the peak and valley values ​​of the net load respectively. represents the maximum available capacity of energy storage, and Indicates the discharge and charging efficiency of the corresponding energy storage, and They respectively represent the power support capacity of each node energy storage for the corresponding net load and the relative sustainable working time under the corresponding charging and discharging states.

6. The optimization method of the multi-element cluster energy storage dynamic reconfigurable model according to claim 1, characterized in that: In step S3, the optimization goal is to achieve maximum suppression of power fluctuations while minimizing the economic operation cost of the overall energy storage system. The objective function is: (5)； In the formula, and Respectively represent energy storage S exist t The charging and discharging power at each moment; F (·) represents the optimization objective function; The net load curve of the system and the energy storage output cost are the two influencing factors. The net load curve represents the overall output effect of the energy storage, and the energy storage output cost represents the minimum overall operating cost of the energy storage system. represents the sub-function with the minimum energy storage operation cost, and The energy storage charging and discharging output is the independent variable.

7. The optimization method of the multi-element cluster energy storage dynamic reconfigurable model according to claim 1, characterized in that: In step S4, through simulation optimization, according to the smoothing effect of each energy storage within the cluster on the local net load fluctuation, the energy storage monomers that need to be re-divided and aggregated are determined, and step S3 is repeatedly executed, and iteration is performed based on the Nash equilibrium principle: the energy storage under each node of the system is taken as a participant, the overall minimum net load fluctuation in each region of the system is taken as the Nash equilibrium point, and the division results between different energy storages are taken as the strategy set. By looping the step S3 process, each energy storage cluster division is adjusted until all energy storage output states are approximately consistent, and the system Nash equilibrium is achieved.

Images